Datasets:

Hendrik-a

/

arxiv

like 0

we study a construction of quantum ldpc codes proposed by mackay , mitchison and shokrollahi . it is based on the cayley graph of together with a set of generators regarded as the columns of the parity check matrix of a classical code . we give a general lower bound on the minimum distance of the quantum code in where is the minimum distance of the classical code . when the classical code is the $ ] repetition code , we are able to compute the exact parameters of the associated quantum code which are $ ] . * msc : * 94c15 , 05c99 , 94b99 * key words : * quantum codes , ldpc codes , cayley graphs , graph covers . * notes . * the material in this paper was presented in part at isit 2011 . this article is published in _ ieee transactions on information theory _ . we point out that the second step of the proof of proposition vi.2 in the published version ( proposition [ prop : wrong_proof ] in the present version and proposition 18 in the isit extended abstract ) is not strictly correct . this issue is addressed in the present version .

1206.2656

classical ldpc codes , it hardly needs to be recalled , come together with very efficient and fast decoding algorithms and overall display extremely good performance for a variety of channels . quantum error - correcting codes on the other hand , under the guise of the css @xcite scheme , are in some ways strikingly similar to classical codes , and in particular can be decoded with purely classical means . it is therefore natural to try to import the classical ldpc know - how to the quantum setting . there is however a structural obstacle . a quantum css code is defined by two binary parity - check matrices whose row - spaces must be orthogonal to each other . to have a quantum ldpc code decodable by message - passing these two matrices should be sparse , as in the classical case . therefore , randomly choosing these matrices , the generic method which works very well in the classical case , is simply not an option in the quantum case , because the probability of finding two sparse row - orthogonal matrices is extremely small . a number of constructions have been suggested by classical coding theorists nevertheless @xcite but they do not produce families of quantum ldpc codes with a minimum distance growing with the blocklength . while this may be tolerable for practical constructions of fixed size , this is clearly an undesirable feature of any asymptotic construction and it raises the intriguing theoretical question of how large can the minimum distance of sparse ( or ldpc ) css codes be . families of sparse css codes with a growing minimum distance do exist , the most well - known of these being kitaev s toric code @xcite , which has been generalised to codes based on tesselations of surfaces ( see _ e.g. _ @xcite ) and higher - dimensional objects . these constructions exhibit minimum distances that scale at most as a square root of the blocklength @xmath5 ( to be precise , @xmath6 is achieved in @xcite ) though this often comes at the cost of a very low dimension ( recall that the dimension of the toric code is @xmath7 ) . it is an open question as to whether families of sparse css codes exist with a minimum distance that grows at least as @xmath8 for @xmath9 , even for quantum codes with dimension @xmath10 . the recent construction @xcite manages to reconcile a minimum distance of the order of @xmath11 with a dimension linear in the blocklength . all these constructions borrow ideas from topology and can be seen as some generalisation of kitaev s toric code . in a follow - up to the paper @xcite mackay , mitchison and shokrollahi @xcite proposed a construction that seemingly owes very little to the topological approach . they noticed that the adjacency matrix of any cayley graph over @xmath12 with an even set of generators is self - dual and can therefore be used to define a sparse css code . experiments with some cayley graphs were encouraging . in the present work we take up the theoretical study of the parameters of these css codes which was left open by mackay _ et al . _ the quantum code in the construction is defined by a classical @xmath13 $ ] linear binary code where @xmath14 must be even . its length is @xmath15 , and the row - weight of the parity - check matrix is @xmath14 . the dimension and the minimum distance of the quantum code does not depend solely on the classical code s parameters , but depend more subtly on its structure . we solve the problem in the first non - trivial case , which was an explicit question of mackay _ et al . _ , namely the case when the classical code is the @xmath3 $ ] repetition code . computing the parameters of the associated quantum code turns out to be not easy , even in this apparently simple case . our main result , theorem [ thm : main ] , gives the exact parameters for this quantum code , namely : @xmath16.\ ] ] the construction therefore hits the @xmath11 barrier for the minimum distance , but it is quite noteworthy that it does so using a construction that breaks significantly with the topological connection . for quantum codes based on more complicated classical @xmath13 $ ] structures , similarly precise results seem quite difficult to obtain , but we managed to prove a lower bound on the quantum minimum distance of the form @xmath17 for some constant @xmath18 ( theorem [ thm : min_dist ] ) . notice that the constructed quantum ldpc codes have not a constant row - weight . indeed , this weight is logarithmic in the blocklength . this has its drawbacks since decoding will be slightly more complex , we remark however that the best families of classical ldpc codes ( i.e. capacity - achieving ldpc codes ) all have row weights that grow logarithmically in the block length . we note also that it was recently proved in @xcite that quantum ldpc stabilizer codes can not achieve the capacity of the quantum erasure channel if their stabilizer matrices have constant row weight . some prerequisites on quantum and quantum css codes together with some basic notions on cayley graphs are recalled in section [ sec : prelim ] . in sections [ sec : basic ] , we describe some basic properties of cayley graphs associated to the group @xmath0 . in section [ sec : hamm ] , we focus on the properties of the _ hamming hypercube _ , that is the cayley graph @xmath19 , where @xmath20 denotes the canonical basis . in particular , we observe some nice property : for almost all families @xmath21 of generators of @xmath22 , the cayley graph @xmath23 looks locally like the hamming hypercube of dimension @xmath24 . in section [ sec : dist ] , we study the minimum distance of a quantum code associated to a cayley graph of @xmath0 and show that this distance is at least quadratic in @xmath14 . finally in section [ sec : repet ] , we focus on the example studied by mitchison _ _ in @xcite and give the exact parameters of this family of quantum codes . in this article all codes , classical and quantum , are binary . a classical code @xmath25 is said to be _ self - orthogonal _ if @xmath26 . it is said to be self - dual if @xmath27 . for convenience s sake , we also say that a binary @xmath28 matrix @xmath29 is _ self - orthogonal _ ( resp . _ self - dual _ ) if @xmath30 ( resp . @xmath31 and @xmath32 ) . classical self - orthogonal codes provide a way of constructing quantum codes through a particular case of the css construction @xcite . let us just recall that if @xmath33 is self - orthogonal with classical parameters @xmath13 $ ] , then it yields a quantum code with parameters @xmath34 $ ] , where @xmath35 , @xmath36 and where @xmath37 is the minimum weight of a codeword in @xmath38 . notice that this last characterization of @xmath37 implies that @xmath39 where @xmath40 denotes the dual distance of @xmath33 . one way of obtaining quantum codes with good parameters is therefore simply to use classical self - orthogonal codes with a large dual distance : this approach has been used repeatedly to obtain record parameters . however , our purpose is to construct css codes with a low - density stabilizer ( parity - check ) matrix , meaning that we need a _ sparse _ self - orthogonal matrix @xmath29 . since we have @xmath41 for the self - orthogonal code @xmath33 generated by the rows of @xmath29 , the bound @xmath39 is of little use because it can not bound @xmath37 from below by anything more than the ( low ) weight of the rows of @xmath29 . obtaining a better lower bound on the quantum code s minimum distance @xmath37 can be quite challenging . in the present work we shall develop a method to obtain improved lower bounds on @xmath37 for some quantum codes based on sparse self - orthogonal matrices . we focus on mackay et al.s construction based on the adjacency matrices of some cayley graphs . let us first recall some basic notions on cayley graphs . let @xmath42 be a group and @xmath21 be a subset of @xmath42 . the cayley graph @xmath43 or @xmath44 , when there is no possible confusion , is the graph whose vertex - set equals @xmath42 and such that two vertices @xmath45 are connected by an edge if there exists @xmath46 such that @xmath47 . [ rem : oriented_symmetric ] the graph @xmath48 is oriented unless @xmath49 . in addition , if @xmath50 , then , the adjacency matrix of the graph is symmetric . the graph @xmath48 is connected if and only of @xmath21 generates @xmath42 . our point is to get pairs @xmath51 such that the adjacency matrix @xmath29 of @xmath48 is self orthogonal , i.e. such that @xmath30 . notice that @xmath30 happens if and only if both conditions are satisfied . 1 . [ item : row_self ] each row of @xmath29 is self - orthogonal , i.e. has even weight ; 2 . [ item : row_orth ] any pair of distinct rows of @xmath29 are orthogonal , i.e. any two distinct rows of @xmath29 have an even number of @xmath10 s in common . the following proposition translates the above conditions in terms of the pair @xmath51 . [ prop : how_to ] let @xmath42 be a finite group and @xmath21 be a system of generators of @xmath42 . assume that a. [ item : even ] @xmath24 is even ; b. [ item : expressions ] for all @xmath52 , there is an even number of distinct expressions of @xmath53 of the form @xmath54 , with @xmath55 . then , the adjacency matrix of the cayley graph @xmath48 is self - orthogonal . condition ( [ item : even ] ) entails obviously ( [ item : row_self ] ) . now , let @xmath56 be two distinct elements of @xmath42 and @xmath57 the corresponding rows of the adjacency matrix of @xmath48 . the rows have a @xmath10 in common if and only if @xmath58 for some pair @xmath59 . this equality is equivalent with @xmath60 . thus , ( [ item : expressions ] ) naturally entails ( [ item : row_orth ] ) . if @xmath61 , then the graph in undirected , its adjacency matrix is symmetric and ( [ item : expressions ] ) can be replaced by 1 . for all @xmath52 , there is an even number of distinct expressions of @xmath53 of the form @xmath62 , with @xmath55 . it is worth noting that if @xmath63 and @xmath64 commute , then @xmath62 and @xmath65 correspond to distinct expressions . we still consider a pair @xmath51 , where @xmath42 is a group and @xmath21 is a generating set of @xmath42 . recall that the _ group algebra of @xmath42 over @xmath66 _ denoted by @xmath67 $ ] is the @xmath66vector space with a basis @xmath68 in one to one correspondence with elements of @xmath42 together with a multiplication law induced by the group law , i.e. @xmath69 . [ nota : pi_s ] given a pair @xmath70 , where @xmath42 is a group and @xmath21 a generating set . we denote respectively by @xmath71 and @xmath72 the elements of @xmath73 $ ] , @xmath74 clearly , the two elements are equal when @xmath75 . [ lem : mult_pi ] the adjacency matrix @xmath29 of @xmath48 represents the right multiplication by @xmath71 i.e. the application @xmath76 & \rightarrow & { \mathbb{f}}_2 [ g ] \\ f & \mapsto & f \pi_s \end{array } \right . .\ ] ] in addition , the matrix @xmath77 represents the right multiplication by @xmath72 . * caution . * in lemma [ lem : mult_pi ] above , we suppose that matrices act on row - vectors , i.e. an @xmath78 binary matrix @xmath79 corresponds to an endomorphism of @xmath0 by @xmath80 , where @xmath81 is represented by a row - vector . [ lem : get_nilp ] the adjacency matrix @xmath29 of @xmath43 is self orthogonal if and only if @xmath82 . in particular , if @xmath75 , then @xmath29 is self - orthogonal if and only if @xmath83 . in particular , the problem of finding sparse self - orthogonal matrices is equivalent with that of finding a @xmath7-nilpotent element of @xmath67 $ ] having a low weight compared to @xmath84 . let @xmath42 be the group @xmath85 and @xmath21 be the set @xmath86 . then , the adjacency matrix of @xmath87 is self - orthogonal . the corresponding group algebra is isomorphic to @xmath88/(x^{2n}-1 , y^{2n}-1)$ ] and the element @xmath71 equals @xmath89 . motivated by mackay et al.s draft @xcite , the group we will focus on in the rest of the paper is @xmath90 . since we are dealing with an abelian group we denote group operations additively rather than multiplicatively . [ ex : ex_cube ] @xmath90 and @xmath21 is any system of generators with an even number of elements . the corresponding group algebra is isomorphic to @xmath91/(x_1 ^ 2 - 1 , \ldots , x_n^2 - 1)$ ] , in which one sees easily that any element of even weight satisfies @xmath92 . as we have just seen in the last example , any even number of generators of @xmath0 defines a cayley graph whose adjacency matrix is a self - orthogonal matrix , from which we have a quantum code . the row weight of the matrix is equal to the cardinality of the set of generators : when this cardinality is chosen proportional to @xmath14 , we have a row weight that is logarithmic in the row length , hence the ldpc character of the quantum code . as put forward in @xcite , note also that the matrix is @xmath93 , i.e. has a highly redundant number of rows , which is beneficial for decoding . it also makes the computation of its rank , and hence the dimension of the quantum code , non - trivial . the present paper strives to compute or estimate parameters , dimension and minimum distance , of the resulting quantum ldpc code . one of the main difficulties of the following work is that we juggle with different kinds of classical codes . roughly speaking , we deal with _ small _ codes of length @xmath14 and _ big _ codes of length @xmath94 . this is the reason why we first need to describe carefully the landscape and the notation we choose . for a positive integer @xmath14 , the canonical basis of @xmath0 is denoted by @xmath95 . in what follows , words of @xmath0 are denoted by letters in lower case such as @xmath96 or @xmath97 . such words are referred as _ small words _ and subspaces of @xmath0 are referred as _ small codes_. given a set @xmath21 of generators of @xmath0 we denote by @xmath98 , or @xmath99 when no confusion is possible , an adjacency matrix of the cayley graph @xmath100 . from proposition [ prop : how_to ] , if @xmath24 is even , then @xmath99 is self orthogonal . we denote by @xmath101 or @xmath102 the code with generator matrix @xmath99 . words of this code or more generally of its ambient space , namely @xmath103 will be denoted by letters in gothic font such as @xmath104 or @xmath105 . in what follows and to help the reader , we frequently refer to _ big words _ and _ big codes _ when dealing with such words or codes . gothic fonts are dedicated to _ big _ objects , such as the matrices @xmath99 , the cayley graphs @xmath44 , the corresponding big codes @xmath102 and so on ... in a graph @xmath106 , we say that two connected vertices have distance @xmath107 if the the shortest path between them consists of @xmath107 edges . this defines a natural metric on @xmath106 . for this distance , a ball centred at a vertex @xmath97 of radius @xmath108 is denoted by @xmath109 , it is the set of vertices at distance @xmath110 of @xmath97 . a sphere of centre @xmath97 and radius @xmath108 is denoted by @xmath111 . we will say that a graph @xmath112 is a _ cover _ or a _ lift _ of @xmath106 if it comes together with a surjective map @xmath113 called a _ covering map _ such that for any vertex @xmath114 of @xmath112 , the map @xmath115 , restricted to the set of neighbours of @xmath114 , is a one - to - one mapping onto the set of neighbours of @xmath116 . the covering map @xmath115 is a local isomorphism . it can be shown that when @xmath106 is connected , the cardinality of the preimage of any vertex is constant : we will refer to this number as the _ degree _ of the cover . consider the particular case when @xmath117 for @xmath118 some set of generators of @xmath22 . a natural covering map of @xmath106 is @xmath119 which can be thought of as removing linear dependencies between elements of @xmath118 ( see [ section : hyper_cover ] ) . any cayley graph associated to @xmath22 is therefore locally isomorphic to some hypercube @xmath120 . this covering construction was used by tillich and friedman in @xcite . starting with a code @xmath33 of generating matrix @xmath79 , they used the set @xmath118 of columns of @xmath79 to define a graph @xmath117 : relating the eigenvalues of @xmath106 to those of its cover they derived upper bounds on the minimum distance of @xmath33 . here we shall rather view the set of generators @xmath118 as the set of columns of a code @xmath33 s parity - check matrix ( rather than a generating matrix ) . the minimum distance @xmath2 of @xmath33 is therefore the minimum weight of a linear relation between generators of @xmath118 , and for @xmath121 the balls @xmath109 in @xmath112 and @xmath122 in @xmath106 are isomorphic . we keep the notation of [ subsubsection : small_and_big ] . it is worth noting that elements of the ambient space of @xmath102 are in one to one correspondence with subsets of the vertex set of @xmath44 . in what follows , we frequently allow ourselves to regard big words as sets of vertices , while vertices are nothing but elements of @xmath0 . in particular we allow ourselves notation such as `` @xmath123 '' , where @xmath124 and @xmath125 . from this point of view , we frequently use the elementary lemma below . recall that , given two subsets @xmath126 of a set @xmath127 , the symmetric difference of @xmath128 and @xmath129 is defined by @xmath130 . this operation is associative . [ lem : dict ] regarding elements of the ambient space of @xmath131 as subsets of the vertex set of @xmath132 , 1 . a row of @xmath98 is nothing but a sphere @xmath133 of centre @xmath124 and radius @xmath10 , where @xmath97 is the index of the row ; 2 . a word of @xmath131 is a symmetric difference of spheres of radius @xmath10 , or equivalently an @xmath66formal sum of such spheres ; 3 . [ item : geo_dual ] a word @xmath134 is a set of vertices such that for every sphere @xmath133 of radius @xmath10 , the intersection @xmath135 has even cardinality . given a positive integer @xmath14 , recall that the hamming isometries @xmath136 are of the form @xmath137 , where @xmath138 is a permutation of the coordinates and @xmath139 is the affine translation @xmath140 for some fixed @xmath141 . let @xmath21 be a family of generators of @xmath0 and @xmath142 be a hamming isometry of @xmath0 , then @xmath142 induces a permutation @xmath143 of @xmath103 which is an automorphism of @xmath44 and an element of the permutation group of @xmath102 ( and hence in that of @xmath144 ) . for all small word @xmath124 , the sphere @xmath133 is the big word whose nonzero entries are the small words @xmath145 with @xmath46 . the code @xmath102 is generated by the @xmath133 s for @xmath146 and one sees easily that @xmath147 . [ cor : vertex_trans ] let @xmath148 , if there exists a nonzero big word @xmath104 in @xmath102 ( resp . @xmath144 ) , then , there exists a big word @xmath149 ( resp . @xmath144 ) with the same weight and which contains the small word @xmath150 . in this section , @xmath14 denotes an even integer and we study the properties the cayley graph @xmath19 . recall that @xmath20 denotes the canonical basis of @xmath0 . first , we show that @xmath151 , which means that the corresponding big code is self dual and hence that the corresponding css code is trivial . however , the properties of @xmath19 are of interest because of its role in the covering construction . [ prop : rk_an ] let @xmath14 be an even integer . the adjacency matrix @xmath152 of @xmath19 , satisfies @xmath153 therefore , @xmath154 , or equivalently @xmath155 , is self - dual . the group algebra of @xmath0 is @xmath156\simeq { \mathbb{f}}_2 [ x_1 , \ldots , x_n]/(x_1 ^ 2 -1 , \ldots , x_n^2 -1)$ ] . using _ notation _ [ nota : pi_s ] , the element @xmath157 is @xmath158 . thus , the cokernel of the endomorphism @xmath159 is @xmath160/(x_1 ^ 2 -1 , \ldots , x_n^2 -1 , x_1+\cdots + x_n).\ ] ] this last algebra is isomorphic to @xmath161/(x_1 ^ 2 -1 , \ldots , x_{n-1}^2 -1 , ( x_1+\cdots + x_{n-1})^2 - 1)$ ] and one sees easily that if @xmath14 is even , then @xmath162 . thus , this cokernel is isomorphic to @xmath163/(x_1 ^ 2 -1 , \ldots , x_{n-1}^2 -1 ) \cong { \mathbb{f}}_2[{\mathbb{f}}_2^{n-1}]$ ] whose @xmath66dimension is exactly the half of that of @xmath156 $ ] . another very useful and nice property of this family of graphs is given by the following statement . [ prop : bipartite ] consider the partition of @xmath0 by @xmath164 of small words of having respectively even and odd hamming weight . then , @xmath19 is bipartite , i.e. any edge links an element of @xmath165 with one of @xmath166 . for all @xmath124 and all @xmath167 , the small words @xmath97 and @xmath168 have weights of distinct parities . in matrix terms , this means , that , for a suitable ordering of the elements of @xmath0 , there exists a @xmath169 binary matrix @xmath170 such that @xmath171 in addition , one shows easily by induction on @xmath14 that @xmath172 . the former result has interesting consequences on the code @xmath155 for @xmath14 even . let @xmath14 be an even integer . the code @xmath173 splits in a direct sum of two isomorphic subcodes with disjoint supports @xmath174 corresponding to big words whose supports are the small words of even and odd weight respectively . both subcodes are self dual . [ cor : split_code ] the two codes come respectively from the upper and lower halves of the row - set of @xmath154 in ( [ eq : mat ] ) . they are obviously isomorphic since they have the same generator matrix @xmath170 . the self orthogonality is clear since @xmath175 entails @xmath176 . in addition , it is clear that @xmath177 , which yields self duality . using the notation of proposition [ prop : bipartite ] and corollary [ cor : split_code ] , a big word @xmath178 whose support is contained in @xmath165 ( resp . @xmath166 ) is in @xmath179 ( resp . @xmath180 ) if and only if it is orthogonal to any sphere @xmath181 where @xmath97 is a small word of odd ( resp . even ) weight . since @xmath155 is self - dual , a big word is in @xmath155 if and only if it is orthogonal to any sphere of radius @xmath10 . if @xmath97 is a small word of even weight , then the elements of @xmath133 have odd weight and hence is obviously orthogonal to any big word supported in @xmath165 . thus , a big word with support in @xmath165 ( resp . in @xmath166 ) is in @xmath155 if and only if it is orthogonal to any sphere of radius @xmath10 centred at a small word of even ( resp . odd ) weight . consequently , the graph @xmath182 can be regarded as a tanner graph for @xmath179 where @xmath165 is the set of bit nodes and @xmath166 the set of check nodes . it can conversely be regarded as a tanner graph for @xmath180 by switching bit and check nodes . [ rem : bipart ] actually , this property of being bipartite is satisfied by any cayley graph @xmath132 as soon as for all @xmath124 and all @xmath46 , the weights of the small words @xmath97 and @xmath145 have distinct parities . it holds for instance for @xmath183 , where @xmath150 is odd . the following statement is crucial in the study of the minimum distance of quantum codes from graphs covered by @xmath182 . [ prop : local_sum ] let @xmath104 be a codeword in the row - space of @xmath154 . regarding @xmath104 as a subset of the vertex set of @xmath182 , assume that @xmath104 is contained in the ball @xmath184 for some vertex @xmath124 and some integer @xmath185 . then @xmath104 is a sum of rows of @xmath154 with support contained in @xmath184 . equivalently , @xmath104 is the @xmath66formal sum of spheres of radius @xmath10 contained in @xmath184 . from corollary [ cor : vertex_trans ] , one can assume that @xmath186 and hence @xmath187 . let us prove the result by induction on @xmath107 . we will consider the extremal points of @xmath104 , that is the vertices of @xmath104 whose distance @xmath107 to 0 is maximal . for every extremal vertex @xmath188 of @xmath104 , we will add a sphere included in the ball @xmath189 to @xmath104 to obtain a new codeword @xmath190 which does not contain the vertex @xmath188 . this procedure will lead to a decomposition of @xmath104 as a sum of spheres included in the ball @xmath189 . if @xmath191 , then @xmath104 is either the zero codeword or the unique big word with support equal to the vertex @xmath192 . but the big word of @xmath103 with support equal to the vertex @xmath193 can not be in @xmath155 . indeed , this big word has weight @xmath10 and since @xmath154 is self dual , if it had such a big word in its row space , the word would lie in its kernel . thus , @xmath154 would have a zero column which is impossible . thus , @xmath104 is the big word zero which is the empty formal sum of spheres of radius @xmath10 . let @xmath194 and assume that the result holds for all radius @xmath195 . _ claim . _ let @xmath196 be the least integer such that @xmath197 . if @xmath198 , then , clearly , @xmath199 is nonempty . then , for all @xmath200 , there exists @xmath201 whose @xmath202-th entry is nonzero . _ proof of the claim . _ assume the claim is false . without loss of generality , one can assume that the @xmath14th entry of any element of @xmath199 is zero . thus , the elements of @xmath199 are of the form @xmath203 , where the @xmath204 s @xmath205 and the `` @xmath206 '' denotes the concatenation . from , proposition [ prop : rk_an ] , we have @xmath207 . thus , regarding @xmath104 as an element of @xmath208 and using lemma [ lem : dict]([item : geo_dual ] ) , we see that the intersection of @xmath104 with any sphere of radius @xmath10 has an even cardinality . however , the spheres @xmath209 contain one and only one element of @xmath104 , namely @xmath210 . this yields the contradiction . thanks to the claim , we know that there exists at least one element of @xmath199 with a nonzero @xmath14th entry . let @xmath211 be these elements . clearly , the small words @xmath212 have weight @xmath213 and hence the spheres @xmath214 are contained in @xmath215 . for all @xmath202 , the only element of @xmath216 whose @xmath14th entry is nonzero is @xmath217 . thus , the big word @xmath218 is contained in @xmath215 and the elements @xmath219 have all a zero @xmath14th entry . indeed , the @xmath217 s have been cancelled and no other element of the form @xmath220 have been added while adding the spheres of radius @xmath10 . the claim entails that @xmath221 . by the induction hypothesis , @xmath105 is a sum of spheres of radius @xmath10 contained in @xmath222 . since the spheres @xmath214 are also contained in @xmath215 , equation ( [ eq : d_c ] ) yields the result . in what follows , @xmath150 denotes an integer . recall that @xmath223 denotes the canonical basis of @xmath22 . let @xmath224 be a family of distinct nonzero elements of @xmath225 with cardinality @xmath226 and assume that @xmath227 is even . from proposition [ prop : how_to ] , the code @xmath228 is self - orthogonal and hence provides a quantum css code with parameters @xmath229 $ ] , where @xmath37 is the minimum weight of a codeword of @xmath230 . regarding the elements of @xmath231 as column vectors , we introduce the binary @xmath232 matrix @xmath233 whose columns correspond to the elements of @xmath234 , that is @xmath235{ccc } & & \\ & i_m & \\ & & \end{array } \right| \ p(w)\ \right),\ ] ] where @xmath236 denotes the @xmath237 identity matrix and @xmath238 is the matrix whose columns are the elements of @xmath224 . [ thm : d_ball ] let @xmath239 be the code with parity check matrix @xmath233 . there is a natural graph cover @xmath240 the degree of @xmath241 is @xmath242 . in addition , denoting by @xmath2 the minimum distance of @xmath239 , the restriction of @xmath243 to any ball of radius @xmath244 is an isomorphism of graphs . recall that we denote the elements of the canonical basis @xmath223 by @xmath245 . denote by @xmath246 the elements of @xmath224 . consider the linear map @xmath247 that sends @xmath248 . the covering map @xmath241 is naturally constructed from the above map . one sees easily that the fibre ( preimage ) of a vertex @xmath249 of @xmath250 is nothing but the coset @xmath251 and hence has cardinality @xmath242 . to conclude , consider a ball of @xmath252 of radius @xmath244 . notice that two vertices @xmath253 of @xmath252 have the same image by @xmath241 if and only if @xmath254 , i.e. if and only if @xmath255 with @xmath256 . in particular , two such vertices have the same image only if their distance is @xmath257 . since the distance between any two vertices in a ball of radius @xmath244 is @xmath258 then , they have distinct images by @xmath241 . thus , the restriction of @xmath241 to the ball is an isomorphism . we keep the notation of [ section : hyper_cover ] . given a set of generators @xmath234 of @xmath22 as before , our point is to bound below the minimum distance of the corresponding css quantum code , that is the minimum weight of the set @xmath259 . [ prop : unbound ] we keep the notation of theorem [ thm : d_ball ] . a codeword in @xmath260 is not contained in a ball of radius @xmath261 . first , let us quickly sketch this proof . set @xmath262 . assume that @xmath263 is contained in a ball of radius @xmath264 . then , using theorem [ thm : d_ball ] , we construct a _ lift _ @xmath265 of @xmath104 satisfying 1 . [ item : orth_c ] @xmath266 ; 2 . [ item : bound_c ] @xmath265 is contained in a ball of radius @xmath264 ; 3 . [ item : gamma_c ] @xmath267 , where @xmath241 is the graph covering map introduced in theorem [ thm : d_ball ] . from proposition [ prop : rk_an ] , the code @xmath268 is self orthogonal and hence @xmath269 . from proposition [ prop : local_sum ] , @xmath265 is a sum of spheres contained in the ball of radius @xmath264 . from theorem [ thm : d_ball ] , the covering map @xmath241 restricted to a ball of radius @xmath270 is an isomorphism . thus , @xmath104 is a sum of spheres and hence is a codeword of @xmath271 which leads to a contradiction . the non - obvious part of the proof is the construction of the lift @xmath265 . it is worth noting that , despite @xmath241 inducing an isomorphism between balls of radius @xmath64 , it is however not possible to lift all such big words in a ball of radius @xmath272 . a counter - example is given in example [ ex : non_lift ] . let us prove the existence of such a lift . without loss of generality , one can assume that @xmath104 is contained in the ball @xmath273 . clearly @xmath241 induces an isomorphism between this ball and the ball centred at zero of radius @xmath264 of @xmath274 . let @xmath265 be the inverse image of @xmath104 by this isomorphism . the above conditions ( [ item : bound_c ] ) and ( [ item : gamma_c ] ) are obviously satisfied . there remains to prove that @xmath265 has an even number of common elements with any sphere of radius @xmath10 . since @xmath275 , any sphere which is not contained in @xmath276 has an empty intersection with @xmath265 . on the other hand , any sphere of radius @xmath10 contained in @xmath276 corresponds thanks to @xmath241 and theorem [ thm : d_ball ] to a unique sphere of radius @xmath10 contained in the ball of radius @xmath64 centred at @xmath193 of @xmath274 . thanks to this ball - isomorphism and by definition of @xmath104 , it is clear that @xmath265 has an even number of common elements with such a sphere . this yields ( [ item : orth_c ] ) . [ ex : non_lift ] suppose that @xmath277 and @xmath278 . theorem [ thm : d_ball ] asserts the existence of a graph covering map @xmath279 . the classical code @xmath239 defined in theorem [ thm : d_ball ] is nothing but the pure repetition code of length @xmath280 which has minimum distance @xmath280 . therefore , @xmath115 induces isomorphisms between balls of radius @xmath7 . let us show that some big words @xmath281 contained in @xmath282 in @xmath283 can not lift as in the previous proof as a word @xmath284 . let @xmath285 let us show that @xmath281 . let @xmath286 , we have to prove that @xmath287 is orthogonal to @xmath104 , that is has an even number of common elements with @xmath104 . notice that for the graph @xmath288 , we have @xmath289 it is clear that if @xmath290 or @xmath291 , then @xmath287 contains no element of weight @xmath7 and hence is obviously orthogonal to @xmath104 . if @xmath292 , then @xmath293 for some @xmath202 and @xmath294 contains four elements of weight @xmath7 , namely all the @xmath295 with @xmath296 . if @xmath297 , that it @xmath298 for @xmath299 distinct to each other , then @xmath287 contains also four elements of weight @xmath7 , namely @xmath300 and @xmath301 , where @xmath302 . thus , @xmath281 . from theorem [ thm : d_ball ] , the map @xmath115 induces an isomorphism from the ball @xmath303 of @xmath304 and the ball @xmath282 of @xmath288 . let us consider the lift of @xmath104 by this isomorphism @xmath305 this big word is not an element of @xmath306 ( which equals @xmath307 ) . indeed , let @xmath308 be a small word of weight @xmath309 . then , @xmath287 has exactly three common elements with @xmath265 , namely @xmath310 and @xmath311 . thus , @xmath265 is not orthogonal to this sphere of radius @xmath10 . [ thm : min_dist ] let @xmath224 be a family of @xmath312 vectors of @xmath22 with @xmath313 . let @xmath233 be as in ( [ eq : ms ] ) and @xmath239 be the small code of length @xmath314 and parity check matrix @xmath233 . recall that @xmath223 denotes the canonical basis of @xmath22 . let @xmath315 be the minimum distance of @xmath239 . then , the minimum distance @xmath37 of @xmath316 and hence of the corresponding quantum code satisfies @xmath317 this theorem is proved further thanks to the following technical lemma . [ lem : card_ball ] let @xmath150 and @xmath224 be as in theorem [ thm : min_dist ] . assume moreover that the minimum distance of the small code @xmath239 is at least @xmath318 . let @xmath104 be a big word of minimum weight in @xmath230 and let @xmath123 then @xmath319 from corollary [ cor : vertex_trans ] , one can assume that the small word @xmath97 of the statement is @xmath193 . from theorem [ thm : d_ball ] , the ball @xmath320 of @xmath321 is isomorphic to that of @xmath322 . therefore , as soon as we stay inside @xmath323 , we can reason as if we were inside that of @xmath322 . thus , set @xmath314 and let us reason in @xmath182 . _ step 1 . _ first , it is important to notice that @xmath104 is supposed to have a minimum weight as a word of @xmath260 , therefore , @xmath324 _ step 2 . _ since by assumption @xmath325 and @xmath326 , this word must be orthogonal to any sphere of radius @xmath10 . denote by @xmath327 the elements of the canonical basis @xmath20 of @xmath0 . then @xmath104 must be orthogonal to the spheres @xmath328 . @xmath329 thus , @xmath104 contains at least @xmath330 small words in @xmath331 . now , consider the maximal subset of elements of @xmath332 with disjoint supports . after reordering the indexes , one can assume that these elements are @xmath333 , for some @xmath334 . we will get the result by considering separately the situations `` @xmath335 is large '' and `` @xmath335 is small '' . _ _ if @xmath336 , then for all odd @xmath337 and @xmath338 , consider the sphere @xmath339 . since @xmath104 is orthogonal to any sphere of radius @xmath10 and contains @xmath340 , it should contain at least one other element of @xmath339 . this other element is either in @xmath341 or in @xmath342 . * if this other element of @xmath343 is in @xmath344 , then it is either @xmath345 or @xmath346 . this additional element is in at most one other sphere of the form @xmath347 with @xmath348 odd , @xmath349 and @xmath350 . * if this other element is in @xmath342 , then it is of the form @xmath351 for some @xmath352 . for obvious degree reasons this additional element is in at most @xmath353 spheres of degree @xmath10 centred at a small word of weight @xmath309 . finally , since there are @xmath354 spheres of the form @xmath347 with @xmath348 odd , @xmath349 and @xmath350 , there are at least @xmath355 additional elements in @xmath104 lying in @xmath356 . therefore , considering also @xmath193 and the elements @xmath357 we get @xmath358 since , by assumption @xmath336 we conclude that @xmath359 _ step 4 . _ now , assume that @xmath360 . from ( [ eq : exists ] ) and by maximality of the set @xmath361 , for all @xmath362 , there exists at least an integer @xmath349 such that @xmath363 . for all @xmath364 let us choose such an integer @xmath365 such that @xmath366 and @xmath367 . _ claim . _ for all @xmath368 , there exist at least @xmath369 integers @xmath370 such that @xmath371 . indeed , if there were @xmath372 integers @xmath373 such that @xmath374 , then @xmath375 which contradicts ( [ eq : weight_c ] ) . let us consider the spheres @xmath376 for @xmath368 , @xmath377 and @xmath378 . thanks to the previous claim , we know that there exists at least @xmath379 such spheres . by definition @xmath104 is orthogonal to any sphere of radius @xmath10 . in particular @xmath104 is orthogonal to @xmath376 which contains @xmath380 . consequently , @xmath104 contains at least another vertex of @xmath376 . since , by assumption @xmath371 , the additional vertex is of the form a. [ item : ls ] either @xmath381 ; b. [ item : ilst ] or @xmath382 for some integer @xmath383 . case ( [ item : ls ] ) can not happen since it would contradict the maximality of the set @xmath384 . since , for obvious degree reasons , a vertex of type ( [ item : ilst ] ) is contained in at most @xmath353 spheres of radius @xmath10 centred at a small word of weight @xmath309 , then the spheres of the form @xmath376 ( whose number is at least @xmath379 ) provide at least @xmath385 additional vertices . considering the vertex @xmath193 and the @xmath386 vertices @xmath357 together with the above - described set of additional vertices , we get @xmath387 _ final step . _ compare ( [ eq : klarge ] ) and ( [ eq : smallk ] ) . for all @xmath14 , we have @xmath388 and hence @xmath389 in @xcite , the statement ( * ? ? ? * lemma 8) corresponding to lemma [ lem : card_ball ] of the present article is false since it refers to big words of @xmath173 and not @xmath390 . but the result does not hold for @xmath173 . indeed , spheres of radius @xmath10 are elements of @xmath173 which can not have a weight quadratic in @xmath14 . it is then necessary to state the result for elements of @xmath271 even if in the proof we reason locally and can do as if we worked in @xmath19 . let @xmath104 be a minimum weight codeword in @xmath260 . from corollary [ cor : vertex_trans ] , one can assume that @xmath325 . set @xmath391 . from proposition [ prop : unbound ] , this word is not contained in @xmath273 . thus , in the worst case @xmath392 . _ claim . _ there are no two consecutive integers @xmath393 such that @xmath394 . indeed , if both sets were empty , then @xmath104 would split into two disjoint sets @xmath395 where @xmath396 and @xmath397 . since the distance between @xmath398 and @xmath399 is at least @xmath7 , any sphere of radius @xmath10 intersects at most one of the words @xmath400 . hence , since @xmath104 is orthogonal to any sphere of radius @xmath10 , so are @xmath398 and @xmath399 . thus , @xmath401 and , by definition of @xmath104 , at least one of them is not in @xmath402 . this contradicts the assumption `` @xmath104 has minimum weight in @xmath260 '' . thanks to the claim and proposition [ prop : unbound ] , one shows that there exist at least @xmath403 disjoint balls of radius @xmath353 centred at an element of @xmath104 . indeed , the worst case is sketched as follows : @xmath320 covers every element of @xmath104 of weight @xmath404 , then , from the claim , there exists at least one element @xmath405 in @xmath406 . the worst case is when @xmath407 has weight @xmath408 . then consider the ball @xmath409 , which is clearly disjoint from @xmath320 . by the same manner , one uses the claim to assert the existence of an element @xmath410 of @xmath104 of weight @xmath411 or @xmath412 and consider @xmath413 and so on ... one shows easily that @xmath414 . then , using lemma [ lem : card_ball ] , we get the result . in this section we answer a question raised by mitchison _ _ in @xcite . namely , we give the exact parameters of the quantum code @xmath415 associated to the classical pure repetition code . that is the quantum code associated to the cayley graph @xmath416 , where @xmath14 denotes an odd integer and @xmath20 denotes the canonical basis of @xmath0 . in what follows , @xmath417 is an odd integer , @xmath418 denotes the set of generators @xmath419 and @xmath420 is the @xmath421 parity check matrix of the @xmath3 $ ] repetition code whose columns consist of the elements of @xmath418 . @xmath422 our goal is to prove the following theorem . [ thm : main ] the quantum code associated to the repetition code , i.e. to @xmath416 has parameters : @xmath16.\ ] ] the matrix @xmath416 has row weight @xmath423 that is logarithmic in the length of the quantum code . this proves the ldpc character of the quantum code . actually , the parameters of this family of quantum codes can be slightly improved since , from remark [ rem : bipart ] , the cayley graph @xmath424 is bipartite . considering only the vertices corresponding to small words of even weight , one obtains another quantum code whose length is divided by @xmath7 and which has the same rate and the same minimum distance . that is , we get a quantum code with parameters @xmath425.\ ] ] in the previous sections , most of the proofs were combinatorial and involved set of vertices of the cayley graph . for such a task the terminology of big codes as sets of sets of vertices of the cayley graph was adapted . in what follows , we will reason on the matrices @xmath99 . for this reason we will use preferentially the terminology of kernels and images of the matrix instead those of the big codes @xmath102 . notice that , if @xmath24 is even , then @xmath99 is self orthogonal . it is also always symmetric and hence @xmath426 sorting the vectors of @xmath428 in the lexicographic order , one obtains @xmath429 [ rem : rankms3 ] this matrix has rank 2 . therefore , the associated classical code has dimension 6 and the quantum code encodes 4 qubits . for every positive integer @xmath63 , let @xmath430 be the @xmath431 matrix defined by @xmath432 to avoid heavy notation and for convenience sake , we frequently remove the index , which can be easily guessed thanks to the context , and just write @xmath433 . for the same reason , we frequently write @xmath434 for the identity matrix , without any index . the matrices @xmath435 can be built recursively in the following manner . [ lem : recursion_matrix ] let @xmath417 be an integer , sorting the elements of @xmath0 and @xmath436 in the lexicographic order , we get @xmath437 from _ remark _ [ rem : oriented_symmetric ] , the matrix is symmetric and hence it is sufficient to prove the result for the upper half of the rows , that is the first @xmath94 rows . these rows correspond to vectors of @xmath436 whose @xmath438th coordinate is zero . using this lexicographic order , elements of @xmath436 are in one to one correspondence with the integers @xmath439 , each integer corresponding to the word yielding its dyadic expansion . _ before reasoning recursively , notice that for all @xmath14 , all the anti - diagonal entries of @xmath435 are equal to @xmath10 . indeed , @xmath418 contains the word @xmath440 and for all @xmath441 , the word @xmath442 is the word obtained from @xmath97 by swapping @xmath193 s and @xmath10 s . in term of the above - described correspondence , if @xmath443 is the dyadic expansion of @xmath444 , then @xmath442 corresponds to @xmath445 . this yields the terms @xmath433 in the top right - hand and bottom left - hand blocks of ( [ eq : recursion ] ) . _ for all @xmath441 whose @xmath438th entry is @xmath193 . the corresponding integer @xmath446 is in @xmath447 and @xmath448 corresponds to @xmath449 . this yields the term @xmath434 in the top right - hand block . _ the top left - hand block of @xmath450 is similar to the matrix @xmath435 with only one difference , the contribution of @xmath451 in @xmath418 should be removed from the block of @xmath450 . this explains the term @xmath433 in the top left - hand corner . * caution . * the matrix @xmath435 is self orthogonal only for @xmath452 even , that is for @xmath14 odd . however , because of this recursive approach , it is necessary to consider also the matrices @xmath435 with even @xmath14 which do not provide a quantum code . this is the reason why lemma [ lem : recursion_matrix ] is stated for any integer @xmath417 and not only for odd such integers . [ lem : jsquare ] for all integer @xmath63 , we have @xmath453 . using the symmetries of the matrix @xmath435 we obtain the following lemmas . [ lem : conjugation ] for all odd integer @xmath417 , we have : a. [ item : conj ] @xmath454 ; b. [ item : ker ] @xmath455 ; c. [ item : im ] @xmath456 ; from lemma [ lem : jsquare ] , the left - hand term of ( [ item : conj ] ) is the conjugation of @xmath435 by @xmath433 . conjugating a matrix by @xmath433 is nothing but changing the basis by reversing the sorting of its elements , i.e. reversing the sorting of the rows and the columns . in terms of small words it corresponds to apply the permutation of @xmath0 given by the affine automorphism @xmath457 , which is a hamming isometry . from corollary [ cor : vertex_trans ] , the permutation @xmath142 is a graph automorphism of @xmath458 thus considering the elements @xmath459 of @xmath0 sorted by the lexicographic order or sorted as @xmath460 provides the same adjacency matrix . another way to prove the assertion is to look at ( [ eq : ms3 ] ) and observe that it is true for @xmath461 . then to prove the result by induction on @xmath14 using lemma [ lem : recursion_matrix ] . assertions ( [ item : ker ] ) and ( [ item : im ] ) are straightforward consequences of ( [ item : conj ] ) . [ lem : caracterisation ] let @xmath462 where @xmath463 are vectors of @xmath464 . then , we have @xmath465 if and only if : @xmath466 by the recursion formula of lemma [ lem : recursion_matrix ] , we have : @xmath467 this gives a characterisation of the vectors of the kernel @xmath468 in function of @xmath469 . we have @xmath470 if and only if @xmath471 for @xmath14 odd , we have : @xmath472 we prove the result by induction on @xmath417 odd . from remark [ rem : rankms3 ] , the case of @xmath473 is done . assume now that the result holds for some @xmath417 . if @xmath465 , then the characterisation of lemma [ lem : caracterisation ] provides @xmath474 such that @xmath475 and @xmath476 are in the image of the matrix @xmath435 . we will show that given such a pair @xmath477 together with a couple of elements of @xmath478 one can construct any element of @xmath479 . first , to study these couples @xmath480 , let us introduce the map @xmath481 from lemma [ lem : conjugation ] , @xmath475 and @xmath476 are both in @xmath482 if and only if @xmath480 is in the kernel of @xmath483 . given such a couple @xmath480 , we can construct a codeword in @xmath484 by choosing arbitrary pre - images of @xmath475 and @xmath476 for @xmath398 and @xmath399 . from this , one can construct @xmath484 from @xmath485 and @xmath478 . let us choose an arbitrary linear section @xmath486 of the map @xmath487 defined by the matrix @xmath435 . that is , @xmath486 is a linear map @xmath488 such that @xmath489 for all @xmath490 . let us introduce the map @xmath491 this map is injective since @xmath492 implies @xmath493 and @xmath494 , which entails @xmath495 and @xmath496 . then @xmath497 yields @xmath498 now , let us show that @xmath499 is surjective . let @xmath500 . from the characterisation ( [ eq : characterisation ] ) , one gets @xmath501 and @xmath502 and @xmath503 means that @xmath398 is congruent to @xmath504 modulo @xmath478 , which yields @xmath505 . one gets @xmath506 by the very same manner . we proved that @xmath499 is an isomorphism . thus , @xmath507 by the rank - nullity theorem , we get @xmath508 , since @xmath483 is surjective . finally we have : @xmath509 we know that that the number of encoded qubits is @xmath510 . from the above proposition , we deduce the dimension of the quantum code . to compute the minimum distance of the quantum code , we examine the weight of the vectors of @xmath511 . that is the set @xmath512 . [ lem : representative ] every word @xmath104 of @xmath513 satisfies one of the following assertions . a. @xmath104 is of the form @xmath514 with @xmath515 ; b. @xmath104 has a representative modulo @xmath482 of the form @xmath516 with @xmath517 . let @xmath104 be a vector of @xmath484 . from lemma [ lem : caracterisation ] , @xmath104 is of the form @xmath514 where @xmath474 . thus , there only remains to prove the statement when either @xmath501 or @xmath502 are in @xmath482 . actually , if one of them is an element of @xmath482 , then so is the other one . indeed , assume that @xmath518 , then using @xmath519 and lemma [ lem : conjugation]([item : im ] ) , we see that @xmath520 . thus , assume that @xmath521 and let @xmath522 be respective premimages of them . thanks to the recursive description ( [ eq:2recurs ] ) , we know that the vectors of @xmath523 are of the form : @xmath524 where @xmath525 . therefore , set @xmath526 , @xmath527 and @xmath528 . the vector @xmath529 . thus , replacing @xmath104 by @xmath530 , which does not changes its class in @xmath531 , one can assume that @xmath532 . thus , from now on , @xmath104 is for the form @xmath533 and @xmath534 . set @xmath535 and @xmath536 , the vector @xmath537 is in @xmath523 . thus , replacing @xmath104 by @xmath538 we get a representative of the form @xmath539 with @xmath517 . [ prop : wrong_proof ] the minimum distance of the quantum code @xmath415 is : @xmath540 for @xmath541 , using ( [ eq : ms3 ] ) , we can see that the distance of the quantum code is @xmath7 . indeed every non zero codeword has weight at least 2 and for instance the word @xmath542 is in the kernel of @xmath473 although it is not a sum of rows . we show the result by induction on @xmath14 for @xmath417 odd . we proceed as follows , first , we show that the distance of @xmath543 is bounded below by @xmath544 . for that , we consider separately the two situations described by lemma [ lem : representative ] . then , we show that our lower bound for the minimum distance is reached . _ let @xmath465 . assume that we are in the first case of lemma [ lem : representative ] . that is @xmath104 is of the form @xmath514 , where @xmath501 and @xmath502 are in @xmath545 . by induction and by definition of the quantum distance , we have @xmath546 . using the triangle inequality for the hamming distance , we get : @xmath547 applying the same reasoning to @xmath502 , we show that the weight of @xmath104 is at least @xmath544 . _ now , assume that we are in the second case of lemma [ lem : representative ] , that is @xmath104 has a representative of the form @xmath539 with @xmath517 . we first show that the result holds for the representative and then show the general case . assume that @xmath548 . let @xmath549 be a pre - image of @xmath398 . set @xmath550 and @xmath551 , using ( [ eq : description_im ] ) , we see that @xmath104 is in the image of @xmath468 and hence its weight is not involved in the computation of the minimum distance . otherwise @xmath552 and , by definition of the distance , the weight of @xmath104 is at least twice the minimum distance of @xmath553 , that is , by induction hypothesis , @xmath544 . now , let us show that the bound holds while adding an element of @xmath523 to @xmath104 . let @xmath554 be an element of @xmath523 . from ( [ eq : description_im ] ) , it is of the form @xmath555 for @xmath556 . we look at the weight of the two first components of @xmath557 : @xmath558 the two other components can be treated by the very same manner . notice that @xmath433 is a permutation matrix and hence a hamming automorphism of @xmath103 , then , using lemmas [ lem : jsquare ] and [ lem : conjugation]([item : conj ] ) @xmath559 combining ( [ eq : truc ] ) and ( [ eq : bidule ] ) and using the triangle inequality , we get @xmath560 since @xmath561 , then , by induction hypothesis , we have @xmath562 . thus , @xmath563 . _ final step . _ we now have a lower bound for the minimum distance . actually , the distance of @xmath564 is exactly @xmath544 . indeed , let @xmath477 be two minimum weight words in @xmath553 , by induction hypothesis , their weights are @xmath565 . the vector @xmath566 is in @xmath484 and its weight is exactly @xmath544 . this vector is not in @xmath482 otherwise , using ( [ eq : description_im ] ) we would get @xmath567 , which yields a contradiction . * we have proved a lower bound on the minimum distance of the quantum code associated with a classical code with mackay et al.s construction . this bound is in @xmath568 where @xmath14 is the length and @xmath2 is the minimum distance of the classical code . this result is based on the enumeration of the minimum number of vertices of a big codeword of @xmath230 restricted to a ball of radius 4 . we have found it difficult to extend this enumeration process to larger balls . we conjecture however that the minimum distance of the quantum code is in fact exponential in @xmath2 . * this family of quantum codes shares some characteristics with topological codes @xcite . the minimum distance of a stabilizer code defined on a square lattice in two dimensions is subjected to the upper bound @xmath569 . mackay et al s construction can be seen as a topological code defined on a lattice with growing dimension . such a stabilizer code is not _ a priori _ limited by the bravyi and terhal bound @xcite . * some quantum ldpc codes lend themselves to fault - tolerant quantum computing @xcite and this provides extra motivation for their study . it would be worthwhile to investigate the potential of the ldpc family investigated in this paper for such a purpose , and it would therefore be desirable to understand what kind of logical operations can be implemented on the encoded data without decoding . a starting point for this research could be based on lemma [ lem : representative ] that leads to a representation of the encoded qubits for the quantum code associated with the repetition code . this work was supported by the french anr defis program under contract anr-08-emer-003 ( cocq project ) . we acknowledge support from the dlgation gnrale pour larmement ( dga ) and from the centre national de la recherche scientifique ( cnrs ) . t. camara , h. ollivier , and j .- tillich . a class of quantum ldpc codes : construction and performances under iterative decoding . in _ proc . of ieee international symposium on information theory , isit 2007 _ , pages 811 815 , june 2007 . tillich and g. zmor , quantum ldpc codes with positive rate and minimum distance proportional to @xmath571 . in _ proc . of ieee international symposium on information theory , isit 2009 _ , page 799804 , july 2009 . g. zmor . on cayley graphs , surface codes , and the limits of homological coding for quantum error correction . in _ proc . of the 2nd international workshop on coding and cryptology , iwcc 09 _ , pages 259273 , berlin , heidelberg , 2009 . springer - verlag .

the 22 ghz ho maser in orion kl has shown extraordinary burst events in 1979 - 1985 and 1998 - 1999 , sometimes called supermaser . we have conducted monitoring observations of the supermaser in orion kl using vera , vlbi exploration of radio astrometry , in the current third burst since 2011 march . three flux maxima are detected in 2011 and 2012 with rising and falling timescales of 2 - 7 months . time variations of the supermaser seem symmetric for all of the active phases . the maximum total flux density of 135000 jy is observed in 2012 june while it is still one order of magnitude lower than those in previous bursts . the supermaser consists of two spatially different components at different velocities . they are elongated along a northwest - southeast direction perpendicular to the low - velocity outflow driven by source i. proper motions of the supermaser features with respect to source i are measured toward west and southwest directions , almost parallel to the low - velocity outflow . the flux density and linewidth show an anti - correlation as expected for an unsaturated maser emission . the supermaser is located close to the methylformate ( hcooch ) line and continuum emission peaks in the orion compact ridge detected by alma , atacama large millimeter / submillimeter array . the broader velocity range of the weak hcooch emission at the supermaser position would be an evidence of a shock front . on the other hand , the 321 ghz ho line is not detected at the position of the supermaser . it can be explained qualitatively by one of the theoretical ho excitation models without extraordinary conditions . our results support a scenario that the supermaser is excited in the dense gas interacting with the low - velocity outflow in the compact ridge . the extremely high flux density and its symmetric time variation for rising and falling phases could be explained by a beaming effect during the amplification process rather than changes in physical conditions .

1407.2757

orion kl ( kleinmann - low object ) is one of the nearest massive star - forming regions at a distance of 420 pc @xcite and hence , has been the most extensively studied source in the context of star - formation study @xcite . there are powerful outflows and jets possibly associated with young stellar objects ( ysos ) in this region at a wide range of spatial scales from @xmath2100 au to @xmath310000 au . they are traced by various molecular lines such as vibrationally excited sio masers at 43 ghz @xcite , 22 ghz h@xmath0o maser @xcite , millimeter / submillimeter thermal sio @xcite and co lines @xcite , and near - infrared h@xmath0 line @xcite . these observational studies have revealed physical properties of the outflows , jets , and surrounding shocked materials interacting with ambient clouds . among a number of tracers , the 22 ghz h@xmath0o maser is known to be a unique tracer to investigate a three - dimensional velocity structure . thanks to its extremely high brightness , it can be detected by high - resolution interferometers and very long baseline interferometers ( vlbi ) providing a capability of proper motion measurements . as a result , the 22 ghz h@xmath0o maser in orion kl is found to be tracing a low - velocity outflow along a northeast - southwest axis @xcite centered at a strong radio continuum source labeled i @xcite . hereafter we call this source i. one of these h@xmath0o maser features in orion kl has exhibited an enormous outburst , which is sometimes called supermaser . the h@xmath0o maser burst in orion kl was reported in 1979 - 1985 for the first time @xcite followed by 1998 - 1999 @xcite . based on vlbi observations , the supermaser was revealed to have east - west or northwest - southeast elongations in 1979 - 1985 and 1998 - 1999 , respectively . the line - of - sight velocity with respect to the local standard of rest ( lsr ) was stable at @xmath4 8 km s@xmath5 , suggesting a common origin . although the maser burst could be related to star - formation activities in orion kl , either a circumstellar disk @xcite or a jet / outflow driven by a yso @xcite , its origin and mechanism are still under debate . the third burst of the h@xmath0o maser in orion kl has started since february 2011 @xcite . the burst appears to be possibly periodic with an interval of about 13 years ( 1985 , 1998 , and 2011 ) . it could support the hypothesis of a common origin @xcite . we have carried out astrometric monitoring observations with vlbi exploration of radio astrometry ( vera ) since 2011 march @xcite and studied basic properties of the supermaser . main conclusions by @xcite are ( 1 ) the burst occurs at two spatially nearby but distinct components ( separation of @xmath410 mas ) at the lsr velocities of 6.95 and 7.58 km s@xmath5 ( hereafter we call them `` 6.9 km s@xmath5 '' and `` 7.5 km s@xmath5 '' features , respectively , using representative velocities ) as reported in previous bursts @xcite , ( 2 ) these features are elongated along a northwest - southeast direction perpendicular to the low - velocity outflow and their proper motions are parallel to the outflow axis , and ( 3 ) the absolute position of the supermaser is coincident with a shocked molecular gas called the orion compact ridge . all of these characteristics suggest that the supermaser could be excited in an interacting region between the low - velocity outflow and the ambient gas in the compact ridge @xcite . nevertheless , the maximum flux density for the current burst phase is reported to be 44000 jy on may 2011 @xcite . it is much lower than those in previous bursts in which the flux densities of the supermaser reached 10@xmath6 jy @xcite . although the maser bursts could have possible periodicity of 13 years , 1985 , 1998 , and 2011 , it is still unclear whether all of three events are caused by a common origin . in order to understand their origin and to reveal their relationship between star - formation activities in orion kl in relation to outflows , jets , and interaction with ambient clouds , it is important to continue further monitoring observations of the supermaser at a current valuable opportunity . in this paper , we report results of our continued monitoring observations of the supermaser with vera to investigate time variations of their properties . in addition , we present follow - up observations with atacama large millimeter / submillimeter array ( alma ) to identify possible powering source(s ) , as well as to study pumping mechanisms based on multi - transition h@xmath0o lines . although we find that the current burst activity is lower by a factor of 10 than those in previous bursts in terms of the maser flux density as discussed later , we call the current bursting maser feature `` supermaser '' throughout the present paper . observations of the @xmath7-@xmath8 transition of h@xmath0o at 22.235080 ghz @xcite in orion kl were carried out with vera . details of the observations have been described in separate papers @xcite , and we briefly summarize key parameters in this section . vlbi monitoring observations have been conducted regularly since 2011 march with a typical interval of 1 - 4 weeks . the maximum baseline length of vera is 2270 km and the resultant fringe - spacing is 1.2 mas . we employed the dual beam observation mode , in which orion kl and an icrf source j054138.0 - 054149 @xcite were observed simultaneously , to measure the absolute position of the target h@xmath0o maser in orion kl with respect to the positional reference source . the astrometric accuracy is estimated to be 0.4 mas according to post - fit residuals of proper motion measurements . these two sources were observed for 8 hours being tracked from horizon to horizon at the lowest elevation angle of @xmath415 degrees . a typical on - source time was 6 hours for each epoch . in this paper , we report the astrometric and imaging results of the first 18 epochs of observations from 2011/068 to 2012/269 ( hereafter we indicate an epoch of observation as year / day of the year ) . a spectral resolution was set to be 15.625 khz , corresponding to a velocity resolution of 0.21 km s@xmath5 . calibration and synthesis imaging were performed using the nrao astronomical image processing system ( aips ) software package in the same manner employed previously @xcite . we also used other data of vera observations at 22 ghz including orion kl as a fringe - finder to investigate a time variation in spectral profiles . a total number of spectral line monitoring data is 53 epochs from 2008/329 to 2014/025 . on - source integration times of each spectral line monitoring data for orion kl was 5 - 10 minutes . in addition , we used alma data taken in one of the early science projects in cycle 0 ( ads / jao.alma#2011.0.00199.s ) to compare images of submillimeter continuum emissions and molecular lines . observations were carried out at the alma band 7 ( in the range of 320 - 340 ghz ) on 2012 july 16 , august 25 , and october 21 . they were almost coincident with the most active burst phase as discussed in the next section . in this paper , we mainly focus on the @xmath9-@xmath10 h@xmath0o line at 321.225656 ghz @xcite and methyl formate ( hcooch@xmath1 ) lines close to the h@xmath0o line . the maximum baseline length in our alma observations was 400 m , corresponding to a spatial resolution of @xmath40.5 , depending on the uv weighting in the imaging . synthesis imaging and self - calibration were done with the common astronomy software applications ( casa ) package . further details of observations are reported elsewhere @xcite . figure [ fig - sp ] shows spectra of the supermaser observed in 18 epochs of the regular astrometry observations from 2011 to 2012 , and figure [ fig - time ] shows time variations of the flux density , velocity width , and peak velocity for all the observed epochs including both 18 epochs of the regular astrometry and 53 epochs of the spectral line monitoring observations . first , these parameters are determined by fitting a single gaussian to the observed profile . however , there are two distinct velocity components at @xmath46.9 km s@xmath5 and @xmath47.5 km s@xmath5 as reported in @xcite . these components can be clearly seen in spectral profiles such as the double - peaked structure on 2011/121 and 2011/137 and red - shifted shoulder on 2011/208 ( figure [ fig - sp ] ) . this results in a significant deviation from the observed line profile in the fitting . thus , we next fit two gaussians to the observed profiles for these epochs . although these line parameters show slightly larger error bars as shown in figure [ fig - time ] , we can see time variations of flux density , velocity width , and peak velocity for the @xmath47.5 km s@xmath5 component continuously changing even during this phase . in contrast , the velocity component that appeared later at @xmath46.9 km s@xmath5 shows a different time variation . according to characteristics of the time variation in figure [ fig - time ] , we can divide the monitoring period of the current h@xmath0o maser burst into four phases ; a moderately active phase in 2009 followed by a relatively quiescent phase in 2010 , a start of the burst with rapid variation in 2011 , the maximum phase in 2012 , and a quiescent / possible initial phase of the next burst in 2013 . in 2009 , there is a moderately active phase slowly increasing and decreasing in the flux density both in total power and cross power spectra . timescales of both rising and falling phases , which are defined as periods from minimum to maximum and maximum to minimum flux densities , respectively , are almost equal , 10 months and 11 months , respectively ( figure [ fig - flux](a ) ) . during this phase , the peak velocity is also slightly changing from 7.4 to 7.3 km s@xmath5 while the velocity width , @xmath40.8 km s@xmath5 , is almost stable with time . after this moderately active phase in 2009 , the maser activity becomes a quiescent phase , whereas the total and correlated flux densities are as high as 2400 jy and 500 jy , respectively , at the minimum phase on 2010/227 . toward this minimum phase , the velocity width of the total - power spectrum gradually increases although that of the cross - power spectrum shows a decreasing trend . at the minimum phase on 2010/227 , the velocity width and peak velocity change drastically . furthermore , these parameters show significant differences between total - power and cross - power spectra but with larger uncertainties than other epochs . this could be caused by other velocity components as their contributions become more and more significant relative to that of the 7.5 km s@xmath5 component . before and after this minimum phase , there were breaks of monitoring observations each for 100 days due to the maintenance of the vera antennas , and hence , we could not trace time variations in this quiescent phase . the h@xmath0o maser burst has started in the end of 2010 or the beginning of 2011 @xcite . the flux density of the total - power spectrum at @xmath47.5 km s@xmath5 has started to increase since 2010/350 while that of the cross power spectrum is temporarily halted on 2010/360 followed by the brightening from 2011/023 . although there is a lack of maser images , this might reflect a variation in the spatial structure of the emitting region . at the beginning of the current burst phase in 2011 , there are two maximum phases on 2011/60 - 2011/121 and 2011/296 - 2011/311 with the total flux density of about 50000 jy ( figure [ fig - time ] ) . the timescales of rising and falling phases are 7 months and 3 months , respectively , for the maximum phase of 2011/60 - 2011/121 ( figure [ fig - flux](b ) ) , although the flux density in this phase shows a variability rather than a monotonic increase or decrease . the timescales for the subsequent maximum phase of 2011/296 - 2011/311 ( figure [ fig - flux](c ) ) are 2 months for both rising and falling phases . apart from the different timescales , the behaviors in rising and falling phases seem to be symmetric as seen in the active phase in 2009 ( figure [ fig - flux ] ) . as discussed above , the burst appears to occur in two different velocity components at 6.9 km s@xmath5 and 7.5 km s@xmath5 @xcite as seen in figures [ fig - time](b ) and ( c ) . the burst activity is temporarily stopped at the end of 2011 , but more active burst has started in early 2012 . the maximum total flux density of 135000 jy is recorded on 2012/154 at the peak velocity of 7.4 km s@xmath5 . in this most active phase , the flux density monotonically increases with the timescale of 5 months . the velocity width is about 0.60 - 0.75 km s@xmath5 , which is the narrowest value observed in our monitoring as shown in figure [ fig - time](b ) . on 2012/123 , the derived velocity widths show the minimum values of 0.50@xmath110.02 km s@xmath5 and 0.46@xmath110.02 km s@xmath5 for the total - power and cross - power spectra , respectively . only in this epoch , the velocity widths are significantly narrower than those observed just before and after this epoch ( 2012/107 and 2012/145 ) . due to the lack of maser images on 2012/123 , we do not discuss what is the true reason for these narrower velocity widths only at this epoch . the flux density of the maser decreases with the timescale of 4 months after this maximum phase . it should be noted that the time variations in rising and falling phases again seem to be symmetric as seen in figure [ fig - flux](d ) . at the end of the burst phase in 2012/265 and 2012/269 , the peak velocities of cross - power spectra show significant differences from those of the total - power spectra . they are thought to be the results of significant contributions from other velocity components and/or effects of structure changes in maser features . this effect is seen particularly in less active phases such as in 2010 . finally , the burst activity seems to end around 2012/269 while it again appears to be brightening from 2013/138 . it is most likely an ignition of a newly occurred burst event . we will continue monitoring observations with vera of the time variation of the supermaser in orion kl . still , the activity of the 22 ghz h@xmath0o maser burst in 2011 - 2012 is relatively lower by a factor of 10 in comparison with the previous bursts in which the peak flux densities reached 10@xmath6 jy . we note that the fluxes reported in figures [ fig - time ] and [ fig - flux ] are those integrated over the whole region including both the supermaser and other features . we investigate possible contribution from other maser features at almost the same velocity with the supermaser feature in the 18 epochs of our vlbi imaging . although it is difficult to make full channel maps with a wide field of view of an order of 10 , we search for positions of the maser features detected by @xcite . as a result , we detect one maser feature around the position offset of ( @xmath43.175 , @xmath4 - 1.850 ) in right ascension and declination , respectively , with respect to the supermaser feature . @xcite detected two maser features near this position at the velocities of 6.3 km s@xmath5 and 7.6 km s@xmath5 with the flux densities of 21.0 jy and 897.7 jy , respectively . figure [ fig - se ] shows the flux density of this maser feature ( we call this feature `` se feature '' ) and the flux density ratio relative to the supermaser feature . the flux density of the se feature increases with time and the maximum phase is observed on 2012/244 . it is about 2 months after the maximum phase of the supermaser . even around the maximum phase of the se feature , a possible contribution to the total flux density is as large as 8% as shown in figure [ fig - se ] ( b ) . therefore , we conclude that the contamination from other maser features including the se features are not significant in figures [ fig - time ] and [ fig - flux ] . the time variation of the linewidth of the supermaser implies an anti - correlation with the flux density as shown in figure [ fig - dv ] . according to model calculations , the line profile of the maser first tends to be narrower as its intensity increases as long as the maser is unsaturated @xcite . then the maser line starts to rebroaden when it becomes saturated @xcite . under the condition of an unsaturated maser , the logarithm of the flux density of the maser line , @xmath12 , is predicted to be proportional to the inverse square of the linewidth , @xmath13 . in figure [ fig - dv ] , we also show best - fit results ; @xmath14 our results may suggest that the supermaser in the current burst phase is still unsaturated . it is consistent with the observed narrow linewidths , 0.40 - 0.48 km s@xmath5 , of the supermaser in 1998 - 1999 possibly attributed to the line narrowing for an unsaturated maser @xcite . in contrast , no clear relationship between the flux density and linewidth is found in the burst in 1979 - 1985 @xcite , suggesting a saturated maser emission . the difference in the degree of saturation could be attributed to the lower flux density , @xmath410@xmath15 jy in the current burst phase than those in the previous bursts , @xmath310@xmath6 jy @xcite . we note that the line narrowing is predicted to occur even for a saturated maser if a stimulated emission rate is not large compared with the cross - relaxation rate @xcite . this means that the above relation holds when the maser flux density is not enough large regardless of saturated or unsaturated cases . thus , we can not firmly estimate the degree of saturation of the supermaser by judging from the linewidth - flux density relation only . figure [ fig - map ] represents channel maps of the phase - referenced images of the supermaser in 2011/137 . three spatially distinct components are found in the initial phase of the burst from 2011/068 to 2011/137 as reported previously @xcite . it is well known that the 22 ghz h@xmath0o line ( 6@xmath16 - 5@xmath17 ) has a hyperfine structure with 3 main components and that the @xmath18=7 - 6 and 5 - 4 components appear with the separation of + 0.45 km s@xmath5 and -0.58 km s@xmath5 with respect to the central component , @xmath18=6 - 5 @xcite . the velocity separation of -0.58 km s@xmath5 is consistent with that observed for two bursting features . however , our results suggest that the two velocity components observed at 6.9 km s@xmath5 and 7.5 km s@xmath5 in the supermaser are not emitted from the same masing gas clump . thus , these two velocity components are unlikely due to the hyperfine structure but they corresponds to distinct masing gas clumps with different positions and velocities . we conduct two - dimensional gaussian fittings on the phase - referenced images to measure the absolute position of each maser spot . hereafter , we define a `` spot '' as a maser emission identified in a single velocity channel and a `` feature '' as a group of spots that are spatially coincident in two or more consecutive velocity channels . by using these results , we derive the absolute proper motions of the detected spots as summarized in table [ tab - proper ] and figure [ fig - position ] . we find that the northern feature , feature 1 ( or spots 1 , 3 , and 5 ) , is detectable only until 2011/296 ( table [ tab - proper ] ) . the lifetime of the northern feature 1 is about 8 months . given the typical monitoring interval and detection criterion , in which a spot is regarded as real when it is detected for more than three contiguous epochs , we might miss a short - term variation of a maser spot on phase - referenced images within a lifetime of less than @xmath42 months . according to the absolute proper motion measurements , the supermaser features are apparently moving toward south and southwest directions as shown in the gray arrows in figure [ fig - map ] . the derived proper motions are consistent but are more reliable than those previously obtained @xcite because we have continued the proper motion measurements in a longer observing period . when we subtract a modulation of an annual parallax of 2.39 mas @xcite and the absolute proper motion of a possible powering source of the low - velocity outflow , source i ( @xmath19 , @xmath20)=(6.3 mas yr@xmath5 , -4.2 mas yr@xmath5 ) , measured by astrometric observations with the very large array ( vla ) @xcite , the derived proper motion vectors with respect to source i point toward west and southwest directions as drawn in the black arrows in figure [ fig - map ] . the magnitude of the proper motions range from 2.7 to 11.4 mas yr@xmath5 or from 5 to 23 km s@xmath5 ( table [ tab - proper ] ) . similar to the previous reports @xcite the supermaser features are elongated along the northwest - southeast direction perpendicular to the low - velocity outflow @xcite . observed proper motions with respect to source i are almost perpendicular to the major axis of each feature . the size of the maser spot at the peak channel on the maximum epoch , 2012/164 , is 2.265@xmath110.014 mas@xmath210.950@xmath110.006 mas ( 0.95 au@xmath210.40 au ) obtained from a gaussian fitting on the self - calibrated image . the peak flux density is ( 1.029@xmath110.010)@xmath22 jy , corresponding to the brightness temperature of 1.2@xmath23 k. the isotropic luminosity is estimated to be 2.4@xmath2110@xmath24 by assuming a linewidth of 0.6 km s@xmath5 . to study physical and chemical properties of the supermaser , we utilize alma cycle 0 data to make images of submillimeter continuum emissions and molecular lines at band 7 . first , we have searched for any 321.225656 ghz h@xmath0o line signal toward the compact ridge where the supermaser resides for the 321.20 - 321.25 ghz range , as shown in figure [ fig - alma ] . several spectral lines are seen , but they are identified as the hcooch@xmath1 lines . if the 321 ghz h@xmath0o line is excited by the maser action like that at 22 ghz , it should be much more compact than those of other molecular lines . with this in mind , we then investigate spectra of the h@xmath0o lines by using the visibility data with the baseline length longer than 100 k@xmath25 , corresponding to the angular size of the emission region of @xmath22 . as a result , we can resolve out all of the lines in the compact ridge while a significant fraction of the 321 ghz h@xmath0o line . in contrast , the broad 321 ghz h@xmath0o line is clearly seen in source i with a double peaked profile @xcite . the 321 ghz h@xmath0o line remains in source i even with the baselines longer than 200 k@xmath25 suggesting a compact structure @xcite . thus , we conclude that no clear evidence of the 321 ghz h@xmath0o maser is detected at the position of the supermaser in spite of the extraordinary strong emission of the 22 ghz h@xmath0o maser . we note that in the observed spectral range a vibrationally excited ( @xmath26=1 , 5@xmath27 - 6@xmath28 ) h@xmath0o line at 336.227931 ghz is not detected in the compact ridge . non - detection of the 321 ghz and 336 ghz h@xmath0o lines implies that the physical conditions around the compact ridge are certainly different from those in source i @xcite . we next compare the distribution of the molecular gas in the compact ridge which hosts the supermaser . for this purpose , we employ the hcooch@xmath1 line at 321.22338 ghz ( @xmath29=0 , 28@xmath30 - 27@xmath31 a ) , which is close to the frequency of the 321 ghz h@xmath0o line ( figure [ fig - alma ] ) , as a tracer of the dense molecular gas in the compact ridge @xcite . the 321.223 ghz hcooch@xmath1 line is not blended with other hcooch@xmath1 transitions unlike those at 321.229 ghz ( @xmath29=1 ) lines , and hence , it is more suitable to analyze the velocity structure of the molecular gas . figure [ fig - moment](a ) shows the moment 0 ( integrated intensity ) map of the 321.223 ghz hcooch@xmath1 line superposed on the alma band 7 continuum emission . the supermaser is located at 8 southwest of source i and is coincident with the compact ridge . the position of the supermaser is close to the hcooch@xmath1 peak but is rather located at the edge of the hcooch@xmath1 condensation . as can be seen in the moment 1 ( peak velocity ) map in figure [ fig - moment ] ( b ) , the velocity of the supermaser @xmath46.9 - 7.5 km s@xmath5 is slightly blue - shifted compared with that of the hcooch@xmath1 line . furthermore , the moment 2 ( velocity width ) map in figure [ fig - moment ] ( c ) implies that the linewidth of the hcooch@xmath1 line tends to be broader around the supermaser . we stress that the velocity structure seems to change abruptly around the supermaser . these apparent velocity structures are caused by two different velocity components of the molecular gas traced by the hcooch@xmath1 line . in figure [ fig - pv ] , we can see detailed velocity structures in the compact ridge by the position - velocity diagram along the x and y axes as indicated in figure [ fig - moment ] ( a ) . it is clear that there exist two different velocity components at 7.5 km s@xmath5 and 9.0 km s@xmath5 . these two components were identified in the previous observations @xcite , but we can resolve both the velocity and spatial structures by higher spectral and spatial resolutions with alma . the supermaser is thought to be associated with one of them at 7.5 km s@xmath5 . interestingly , the velocity of the weak emission from the hcooch@xmath1 line in figure [ fig - pv ] ranges from @xmath46 to 12 km s@xmath5 , although the linewidths of the hcooch@xmath1 lines are only about 1 km s@xmath5 as shown in figure [ fig - alma ] . the velocity range of the weak hcooch@xmath1 emission is significantly broader at x=0 and y=0 , corresponding to the position of the supermaser ( see figure [ fig - pv ] ) . this is probably an evidence of a shock front at the position close to the supermaser . indeed , possible effects of shocks in the compact ridge have been suggested by the enhancement of organic molecules such as hcooh @xcite , hcooch@xmath1 @xcite , and ch@xmath1och@xmath1 @xcite . figure [ fig - cont ] shows a close - up view of distributions of the alma band 7 continuum emission superposed on the integrated intensity map of the 321.223 ghz hcooch@xmath1 line ( figure [ fig - moment](a ) ) . positions of the 22 ghz h@xmath0o masers @xcite including that of the supermaser are also indicated . we detect the compact continuum emission both at alma bands 6 and 7 @xcite . the accurate position of the supermaser is significantly shifted from the continuum peak by 0.3 southeast . we also indicate positions of other possible counterparts observed previously . their relationship will be discussed in the next section . in this paper , we report results of continued monitoring observations of the supermaser with vera for more than 5 years . we also acquire follow - up observational data with alma of submillimeter continuum emissions and h@xmath0o lines at 321 ghz to reveal characteristics of the supermaser and their environments . in this section , we will address following issues related to the supermaser : what is the powering source of the supermaser and what makes the 6.9 and 7.5 km s@xmath5 features in the compact ridge only such a special burst event ? why there is no bursting h@xmath0o maser feature at 321 ghz despite the extraordinary high flux density in the 22 ghz h@xmath0o maser ? how the drastic burst of the h@xmath0o maser can be explained ? is there a periodicity of the burst event ? according to the absolute position of the supermaser derived from our vera observations , we propose that the supermaser would be excited in the shocked molecular gas interacting between the low - velocity outflow driven by source i and ambient dense gas in the compact ridge @xcite . proper motions and elongations of the supermaser features ( figure [ fig - map ] ) and the broader hcooch@xmath1 line at the position of the supermaser ( figure [ fig - pv ] ) support this scenario . it has also been suggested by the fact that the h@xmath0o maser tends to be distributed in the interface of the shocked molecular gas in the compact ridge traced by hcooh @xcite and hcooch@xmath1 @xcite lines . furthermore , @xcite point out a possible relation between the compact ridge and an explosive outflow traced by the co lines @xcite . indeed , the supermaser is found to be located at the edge of the low - velocity bipolar molecular outflow traced by the thermal sio lines @xcite as well as the explosive outflow @xcite . the latter one is thought to be the result of a recent close encounter of source i and bn object @xcite , although a relationship between these two outflows is still unclear . note that an infrared source irc 5 is located at 1 southeast of the supermaser . the position of irc 5 corresponds to a nearby 3 mm continuum peak , c29 @xcite , and an hcooch@xmath1 peak , mf14 @xcite and hence , it is more closely related to these sources . since irc 5 is thought to be a shocked molecular gas heated externally @xcite , it would not be directly related to the excitation of the supermaser . alternatively , both of them are thought to be heated by the common outflow . as discussed previously @xcite , however , it has been still unknown why only the 6.9 km s@xmath5 and 7.5 km s@xmath5 features show such an extraordinary burst phenomenon . one of the possible explanations is a special physical or geometrical condition of the maser excitation in this region . an anomalous amplification of the h@xmath0o maser burst might need a long path length . such a condition could be achieved only in the systemic velocity component . because it can supply plenty of ambient molecular gas within a narrow velocity range , maser emission would be efficiently amplified only close to the systemic velocity @xcite . millimeter and submillimeter continuum emissions are detected by @xcite , smm1 in @xcite and cb1 in @xcite with spatial resolution of @xmath40.8 - 3 . the mass of the submillimeter core in the compact ridge is estimated to be 4@xmath32 @xcite , and hence , they could provide the dense gas required to form new stars . a similar conclusion is obtained from the alma science verification continuum data that have been analyzed by @xcite . to account for a necessary condition for the supermaser , it is also suggested that the maser burst could occur as a result of an interaction with outflows and a pre - existing yso in the compact ridge @xcite . as discussed in detail by @xcite , there are possible candidate sources physically associated with the supermaser ; a radio continuum source labeled r detected with vla @xcite , 3 mm continuum emission sources detected with the combined array for research in millimeter - wave astronomy ( carma ) @xcite , molecular peaks traced by hcooch@xmath1 lines with the iram plateau de bure interferometer ( pdbi ) @xcite , and optical source parenago 1822 , which is also identified from millimeter , near infrared , optical and x - ray ( see simbad ) , in the compact ridge ( figure [ fig - cont ] ) . @xcite detect a compact 1.3 mm continuum source named hc 438 in the compact ridge with a flux density of 67.8@xmath1114.2 mjy by using carma . we also detect a compact continuum source at alma band 7 ( see contours in the middle of figure [ fig - cont ] ) which is also detected at band 6 @xcite . these observations achieve much higher spatial resolution of @xmath40.5 than those reported above @xcite . the flux density at alma bands 6 and 7 are 57@xmath115 mjy and 129@xmath1114 mjy , respectively @xcite . the alma band 6 result is consistent with that of the carma 230 ghz observation @xcite . the supermaser is located within the submillimeter core detected by alma but is shifted by 0.3 from the peak position . taking into account positional uncertainties , they are identical to hc 438 @xcite , the 3 mm continuum source c32 @xcite and hcooch@xmath1 peak mf1 @xcite . the circumstellar mass of hc 438 is derived to be 0.085 @xmath32 with the assumed dust temperature of 20 k @xcite . if we assume a higher temperature of 100 k derived from the hcooch@xmath1 data @xcite , the smaller mass of 0.01@xmath32 is obtained . these mass estimates would be a lower limit of the dust continuum source possibly associated with the supermaser because most of the flux is filtered out by alma and carma with extended configurations . we speculate that an embedded yso associated with jet / outflow / disk @xcite would play an important role to realize special conditions required for the supermaser . we notice that the h@xmath0o maser features near the supermaser detected with vla @xcite are distributed around the submillimeter continuum peak as seen in figure [ fig - cont ] . it may imply an expanding shell ejected by an embedded yso , which are sometimes seen in massive ysos @xcite . as discussed in section [ sec - vera ] , we tried to search for other maser features in orion kl region previously detected with vla @xcite . we made channel maps of the h@xmath0o maser features with vera within an 1@xmath211 field of view around the supermaser in the compact ridge . however , we found no maser emission because of the limited dynamic range seriously affected by the extremely strong supermaser . further observations of continuum emission and h@xmath0o masers with higher dynamic range will unveil the exact location of the embedded yso in the compact ridge . it is proposed that the pumping mechanism of the h@xmath0o maser is dominated by collisional excitation @xcite . the burst of the h@xmath0o maser in orion kl is proposed as the result of collisional pumping in the magnetohydrodynamic shock wave caused by the outflow @xcite . on the other hand , alternative mechanisms are also considered to account for the extraordinary h@xmath0o maser burst , such as overlapping of two maser gas clumps @xcite and accidental beaming of radiation toward the earth rather than a change in excitation @xcite . to constrain the physical properties of the supermaser , we have carried out observations of the submillimeter h@xmath0o lines which could be masing @xcite by using alma . among a number of h@xmath0o lines , we employ the 10@xmath33 - 9@xmath34 transition at 321 ghz which is proposed to be a good indicator of some key parameters including temperature , density , and h@xmath0o abundance in the masing gas clump @xcite . however , there is no significant emission of the 321 ghz h@xmath0o line at the position around the supermaser . the upper limit of the flux density of the 321 ghz h@xmath0o line is @xmath40.1 jy . if we assume physical properties of the masing gas are common for both the 22 ghz and 321 ghz lines , then a photon luminosity ratio , @xmath35(22 ghz)/@xmath35(321 ghz ) , is identical to the observed flux density ratio , @xmath18(22 ghz)/@xmath18(321 ghz ) . then , an upper limit of @xmath35(22 ghz)/@xmath35(321 ghz)@xmath310@xmath6 is obtained from the flux density of the 22 ghz maser of @xmath310@xmath15 jy . when the beaming angle of the 22 ghz maser is also the same as that of 321 ghz , an emissivity ratio @xmath36(22 ghz)/@xmath37(321 ghz ) @xcite is also equal to the photon luminosity ratio and flux density ratio , @xmath38 . as demonstrated in the model calculations @xcite , the lower limit of @xmath38 can be achieved when the temperature is @xmath4200 - 400 k or the parameter @xmath39 is @xmath40.4 - 1.8 depending on the temperature within the calculated range below 2000 k ( see their figure 3 ) . the parameter @xmath40 is defined as @xmath41(h@xmath0o ) @xmath42/@xmath43 , in which @xmath44(h@xmath0o ) , @xmath45 , and @xmath43 are a fractional abundance of h@xmath0o relative to hydrogen nuclei in units of 10@xmath46 , a number density of hydrogen nuclei in units of 10@xmath47 @xmath48 , and a velocity gradient in units of 10@xmath49 cm s@xmath5 cm@xmath5 @xcite . we note that the model employed here is calculated under the assumption of a saturated maser @xcite and hence , the estimated parameters may be different from an unsaturated case as adopted for the supermaser . furthermore , the model calculation only considers collisional pumping of h@xmath0o molecules ignoring infrared pumping . in addition , the collision rates used in these works are now superseded by newer calculations @xcite . despite these uncertainties , we use the model of @xcite to roughly interpret the supermaser burst at 22 ghz and the non - detection of the 321 ghz line . as a result , we find the supermaser burst at 22 ghz and the non - detection at 321 ghz could still be explained by the theoretical model @xcite without extraordinary physical conditions . thus , under the condition of the unsaturated 22 ghz h@xmath0o maser , the lower temperature and larger value of @xmath40 would also be expected according to a qualitative consideration . to explain the non - detection of the 321 ghz maser line ( high value of @xmath51 ) , either high h@xmath0o abundance , high h@xmath0 density , small velocity gradient , or low temperature of 200 - 400 k would be required . the h@xmath0o abundance in the compact ridge is estimated from the herschel data to be 2.6@xmath2110@xmath52 with the source size of 6 @xcite . if a local abundance enhancement occurs around the supermaser , the higher h@xmath0o abundance up to 6.5@xmath2110@xmath46 as observed in the hot core region @xcite may be possible . the average h@xmath0 number density of the dust continuum source in the compact ridge is estimated to be ( 0.5 - 5)@xmath2110@xmath47 @xmath48 assuming a diameter ( fwhm ) of 0.43 or 90 au @xcite and a total mass of 0.01 - 0.1 @xmath32 @xcite . if we assume a velocity width of 0.6 km s@xmath5 and a path length of 1 au as obtained from a gaussian fitting for a bursting maser spot , the velocity gradient @xmath43 is calculated to be 0.4 . although the above results contain large uncertainties by a factor of 10 or more , these parameters seem reasonable to explain estimated range of @xmath39 of 0.4 - 1.8 unless the h@xmath0o abundance is the lowest estimate of @xmath44(h@xmath0o)=0.026 @xcite . it is also conceivable that the supermaser is the result of strong beaming effects @xcite . if only the 22 ghz maser emission has a smaller beaming angle compared with that of the 321 ghz line , the emissivity ratio @xmath53 is proportional to the flux density ratio , @xmath18(22 ghz)/@xmath18(321 ghz ) , and the ratio of beaming angles , @xmath54(22 ghz)/@xmath54(321 ghz ) , which reduces the upper limit of @xmath53 estimated from the observed flux density ratio . the smaller beaming angle of the 22 ghz maser line than that of 321 ghz by a factor of @xmath4100 would enable a large @xmath54(22 ghz)/@xmath54(321 ghz ) value of @xmath410000 . this may result in the lower @xmath53 value by a factor of 10000 which can be achieved under wider range of physical conditions as seen in figure 3 of @xcite . in spite of the presence or absence of a beaming effect , the condition for the h@xmath0o maser burst only at the 22 ghz transition , @xmath38 , does not require extreme physical properties but rather those typical in hot molecular cores @xcite . we suggest that the symmetric time variation in the flux density ( figure [ fig - flux ] ) could be explained by a change in the beaming angle of the maser rather than changes in the physical properties such as h@xmath0o abundance , h@xmath0 density , and/or temperature . on the other hand , a change in a velocity gradient caused by a length of the maser cloud and/or velocity structure might happen such as those in the overlapping maser model @xcite . nevertheless , it is still uncertain how such an accidental overlapping and/or a beaming effect could occur in two spatially distinct clouds for several times during our monitoring period of @xmath45 years . comparing the basic properties such as the peak lsr velocity , the spatial structure of each feature , and the location coincident with the compact ridge , all of the burst events in 1979 - 1985 , 1998 - 1999 , and 2011 - 2012 could have a common origin . it is also suggested that the h@xmath0o maser burst may have periodicity of 13 years @xcite . however , the current burst shows less activities in terms of the maximum flux density . furthermore , no such burst event is reported in 1973 @xcite , in which the burst event is expected if 13 year - periodicity is valid ( i.e. 1972 - 1973 , 1979 - 1985 , 1998 - 1999 , and 2011 - 2012 ) . in 1973 july , the observed flux density of the h@xmath0o maser at 7.4 km s@xmath5 was only about 100 jy and was weaker than the 11.0 km s@xmath5 component @xcite . this component was probably identical to the second brightest component in the current burst phase ( figure [ fig - sp ] ) . the position of the 7.4 km s@xmath5 component was measured to be r.a.=05h35m14.22s and decl.=-05@xmath552237.0 ( j2000 ) and was consistent with that of the current bursting feature located in the compact ridge . in 1975 - 1976 , the h@xmath0o maser feature at 7.6 km s@xmath5 was detected at the position of r.a.=05h35m14.07s and decl.=-05@xmath55d2238.0 ( j2000 ) . the position of this feature was again in good agreement with that of the current h@xmath0o maser burst , whereas it was rather weak ( much less than 1000 jy ) compared with another strong feature at 10.8 km s@xmath5 . although the h@xmath0o maser feature at @xmath47.5 km s@xmath5 could be excited by a common origin in the compact ridge , it is not very likely that the maser burst occurs regularly at a period of 13 years . more archive data of past observations will be helpful to investigate a long - term variability of the h@xmath0o maser activities since early 1970s . the flux increase started in 2013 is now on - going and hence , further monitoring will also provide some hints to understand a possible relation with previous bursts . along with the 22 ghz h@xmath0o maser lines , it could be useful to continue monitoring observations of the 321 ghz h@xmath0o lines in orion kl with alma in order to see whether the compact ridge and source i , which is a powering source of the outflow interacting with the compact ridge , behave in the same way . figures [ fig - sp]-[fig - flux ] imply that the current burst in 2011 and 2012 sometimes shows variations in the flux density and velocity structure with the timescales of 2 - 7 months . it is comparable to those of typical h@xmath0o masers in orion kl @xcite . although the spectral component of the supermaser seems to survive for more than 5 - 10 years @xcite , a spatial structure of each bursting feature is found to be changing within a timescale of a few months to @xmath22 years at the longest lifetime ( table [ tab - proper ] ) . thus , it is unlikely that the bulk of the bursting maser gas is identical for longer than @xmath42 years . we have carried out astrometric monitoring observations of the bursting 22 ghz h@xmath0o maser feature , called supermaser , in orion kl using vera since 2011 march . combined with spectral line data obtained as a fringe - finder , we reveal a time variation of the supermaser for more than 5 years . in addition , we utilize alma cycle 0 data at band 7 ( 320 - 330 ghz ) for both spectral lines and continuum emissions to investigate physical and chemical properties of the supermaser . the main findings are summarized as follows : 1 . within the monitoring period in 2011 and 2012 , we detect three active phases with rising and falling timescales ranging from 2 to 7 months . we also find a smaller burst in 2009 with a slightly longer rising and falling timescales of @xmath410 months . the flux variations of the supermaser seem to be symmetric in time for all of the active phases . 2 . the maximum phase of the current burst is observed on 2012/154 with the total flux density of 135000 jy , which is one order of magnitude lower than those in previous bursts . although the burst events may occur with a possible periodicity of 13 years , ( i.e. 1979 - 1985 , 1998 - 1999 , and 2011 - 2012 ) , a lower flux density by a factor of 10 in the current burst , as well as no significant burst event in 1973 @xcite would make the claimed periodicity of 13 years still unconvincing . the outburst of the maser emission can be seen in two velocity components at @xmath46.9 km s@xmath5 and @xmath47.5 km s@xmath5 as reported in @xcite and @xcite . the 6.9 km s@xmath5 feature is alive only in 2011 with the lifetime of @xmath48 months . note that we detect two spatially distinct features at 6.9 km s@xmath5 and 7.5 km s@xmath5 , and hence , these two velocity components are unlikely due to the hyperfine structure of the h@xmath0o molecule . 4 . the flux density and linewidth show an anti - correlation as expected for an unsaturated maser emission @xcite , in which the logarithm of the flux density of the maser line , @xmath12 , is proportional to the inverse square of the linewidth , @xmath13 . however , we can not rule out a possibility of a saturated maser case because such a line narrowing is also predicted to occur even for a saturated maser if a stimulated emission rate is not large compared with that of the cross - relaxation @xcite . as reported in previous papers @xcite the bursting maser features are elongated along the northwest - southeast direction perpendicular to the low - velocity outflow driven by source i. absolute proper motions of the bursting maser features are measured based on a longer monitoring period than that in the previous paper @xcite . their proper motions with respect to a powering source of the low - velocity outflow , source i , are almost perpendicular to the elongation of the maser features , suggesting that they are related to the low - velocity outflow . the size of the maser spot at the peak channel on the maximum epoch , 2012/164 , is 2.265 mas@xmath210.950 mas , and the total flux density is 1.029@xmath2110@xmath15 jy . these values correspond to the brightness temperature and the isotropic luminosity of 1.2@xmath23 k and 2.4@xmath2110@xmath24 , respectively . the supermaser is located at the compact ridge region in orion kl . continuum peaks detected by alma bands 6 and 7 are located 0.3 northwest of the supermaser . as suggested by @xcite , there are a radio continuum source labeled r @xcite , 3 mm continuum source c32 @xcite , hcooch@xmath1 core mf1 @xcite , and optical source parenago 1822 , which are probably identical or closely related to the compact continuum sources detected by alma and the supermaser . the position of the supermaser is coincident with the dense gas traced by the hcooch@xmath1 line . the velocity of the supermaser agrees well with one of the two velocity components of the hcooch@xmath1 line detected by alma and pdbi @xcite . a slightly broader velocity range of the weak hcooch@xmath1 emission would imply a shock front at the position of the supermaser . the 321 ghz h@xmath0o line is not detected around the 22 ghz supermaser features , although the strong maser emission is found to be associated with source i @xcite . given the high emissivity ratio , the maser burst at 22 ghz could occur under conditions of the temperature of @xmath2200 - 400 k and/or the parameter @xmath56 , suggesting a high abundance of h@xmath0o , a high number density of h@xmath0 , and/or a small velocity gradient . meanwhile , these physical properties are not extraordinary values but are very similar to typical hot molecular cloud cores . thus , the supermaser burst at 22 ghz with non - detection of the 321 ghz maser emission could be explained by the theoretical model qualitatively @xcite . further detailed calculations would be required to interpret the observed flux ratios of the masers @xcite . alternatively , the supermaser may be caused by an accidental beaming only for the 22 ghz h@xmath0o maser @xcite . the symmetric time variation in the flux density is consistent with a result of the beaming rather than changes in physical properties of the masing gas . however , it still remains a matter of debate how accidental overlapping and/or beaming occur repetitively for more than @xmath430 years since the discovery of the supermaser . we are grateful to alain baudry for useful comments , which is essential to improve our scientific discussion . we thank the staff of the vera stations and alma observatory to support our project . this paper makes use of the following alma data : ads / jao.alma#2011.0.00199.s . alma is a partnership of eso ( representing its member states ) , nsf ( usa ) and nins ( japan ) , together with nrc ( canada ) and nsc and asiaa ( taiwan ) , in cooperation with the republic of chile . the joint alma observatory is operated by eso , aui / nrao and naoj . this research has made use of the simbad database , operated at cds , strasbourg , france . t.h . is supported by the mext / jsps kakenhi grant numbers 21224002 , 24684011 , and 25108005 , and the alma japan research grant of naoj chile observatory , naoj - alma-0006 . is supported by the mext / jsps kakenhi grant numbers 24540242 and 25120007 . allen , d. a. & burton , m. g. 1993 , nature , 363 , 54 bally , j. 2008 , in handbook of star forming regions , vol . 1 , ed . b. reipurth ( asp monograph publications , vol . 4 ; san francisco , ca : asp ) , 459 bally , j. , cunningham , n. j. , moeckel , n. , burton , m. g. , smith , n. , frank , a. , & nordlund , a. 2011 , apj , 727 , 113 baudry , a. , forster , j. r. , & welch , w. j. 1974 , a&a , 36 , 217 beuther , h. , & nissen , h. d. 2008 , apjl , 679 , l121 brouillet , n. , despois , d. , baudry , a. , et al . 2013 , a&a , 550 , 46 brouillet , n. et al . 2014 , in preparation chen , p. , pearson , j. c. , pickett , h. m. , matsuura , s. , & blake , g. a. 2000 , apjs , 128 , 371 churchwell , e. , felli , m. , wood , d. o. s. , & massi , m. 1987 , apj , 321 , 516 daniel , f. & cernicharo , j. 2013 , a&a , 553 , a70 deguchi , s. & watson , w. d. 1989 , apjl , 340 , l17 eisner , j. a. , plambeck , r. l. , carpenter , j. m. , corder , s. a. , qi , c. , & wilner , d. 2008 , apj , 683 , 304 elitzur , m. , hollenbach , d. j. , mckee , c. f. 1989 , apj , 346 , 983 favre , c. , despois , d. , brouillet , n .. , baudry , a. , combes , f. , gulin , m. , wootten , a. , & wlodarczak , g. 2011 , a&a , 532 , a32 felli , m. , churchwell , e. , wilson , t. l. , taylor , g. b. 1993 , a&as , 98 , 137 forster , j. r. , welch , w. j. , wright , m. c. h. , & baudry , a. , 1978 , apj , 221 , 137 friedel , d. n. & widicus weaver , s. l. 2011 , apj , 742 , 64 garay , g. , moran , j. m. , & haschick , a. d. 1989 , 338 , 244 gaume , r. a. , wilson , t. l. , vrba , f. j. , johnston , k. j. , schmid - burgk , j. 1998 , apj , 493 , 940 genzel , r. , reid , m. j. , moran , j. m. , & downes , d. 1981 , apj , 244 , 884 genzel , r. , & stutzki , j. 1989 , araa , 27 , 41 goddi , c. , humphreys , e. m. l. , greenhill , l. j. , chandler , c. j. , & matthews , l. d. 2011 , apj , 728 , 15 goldreich , p. & kwan , j. 1974 , apj , 190 , 27 greenhill , l. j. , goddi , c. , chandler , c. j. , matthews , l. d. , humphreys , e. m. l. 2013 , apj , 770 , 32 hirota , t. , tsuboi , m. , fujisawa , k. et al . 2011 , apj , 739 , l59 hirota , t. , bushimata , t. , choi , y. k. et al . 2007 , pajs , 59 , 897 hirota , t. , kim , m. k. , kurono , y. , & honma , m. 2014a , apjl , 782 , l28 hirota , t. , kim , m. k. , kurono , y. , & honma , m. 2014b , in preparation kaifu , n. , usuda , t. , hayashi , s. et al . 2000 , pasj , 52 , 1 kim , m. k. , hirota , t. , honma , m. , et al . 2008 , pasj , 60 , 991 kim , m. k. , hirota , t. , honma , m. , & kobayashi , h. 2014 , in preparation kukolich , s.g . 1969 , j. chem . phys . , 50 , 3751 liu , s. -y . , girart , j. m. , remijan , a. , & snyder , l. e. 2002 , apj , 576 , 255 ma , c. , arias , e. f. , eubanks , t. m. , et al . 1998 , aj , 116 , 516 matthews , l. d. , greenhill , l. j. , goddi , c. , chandler , c. j. , humphreys , e. m. l. , & kunz , m. w. 2010 , apj , 708 , 80 matveenko , l. i. , graham , d. a. , & diamond , p. j. 1988 , sov . , 14 , 468 matveyenko , l. i. , zhakharin , k. m. , diamond , p. j. , & graham , d. a. 2004 , astron . lett . , 30 , 100 matveyenko , l. i. , demichev , v. a. , ipatov , a. v. , melnikov , a. e. , & surkis , i. f. 2012a , astl , 38 , 575 matveyenko , l. i. , demichev , v. a. , ipatov , a. v. , melnikov , a. e. , & surkis , i. f. 2012b , astl , 38 , 764 menten , k. m. , & reid , m. j. 1995 , apj , 445 , l157 menten , k.m . , reid , m. j. , forbrich , j. , & brunthaler , a. 2007 , a&a , 474 , 515 neill , j. , l. , wang , s. , bergin , e. a. , crockett , n. r. , favre , c. , plume , r. , melnick , g. j. 2013 , apj , 770 , 142 neufeld , d. a. & melnick , g. j. 1990 , apjl , 352 , l9 neufeld , d. a. & melnick , g. j. 1991 , apj , 368 , 215 niederhofer , f. , humphreys , e. m. l. , & goddi , c. 2012 , a&a , 548 , a69 okumura , s. , yamashita , t. , sako , s. , miyata , t. , honda , m. , kataza , h. , okamoto , y. k. 2011 , pasj , 63 , 823 omodaka , t. , maeda , t. , miyoshi , m. , et al . 1999 , pasj , 51 , 333 otto , s. & gaylard , m. j. 2012 , in iau symp . , 287 , cosmic masers from oh to h@xmath57 , ed . r. s. booth , e. m. l. humphreys , & w. h. t. vlemmings ( cambridge : cambridge univ . press ) , 110 plambeck , r. l. , wright , m. c. h. , friedel , d. n. , et al . 2009 , apj , 704 , l25 shimoikura , t. , kobayashi , h. , omodaka , t. , diamond , p. j. , matveyenko , l. i. , & fujisawa , k. 2005 , apj , 634 , 459 tang , y. w. , ho , p. t. p. , koch , p. m. , & rao , r. 2010 , apj , 717 , 1262 torrelles , j. m. et al . 2003 , apjl , 598 , l115 tolmachev , a. 2011 , atel , 3177 zapata , l. a. , rodrguez , l. f. , schmid - burgk , j. , loinard , l. , menten , k. m. , & curiel , s. 2012 , apj , 754 , l17 zapata , l. a.,schmid - burgk , j. , ho , p. t. p. , rodrguez , l. f. , menten , k. m. 2009 , apj , 704 , l45 zapata , l. a. , schmid - burgk , j. , & menten , k. m. 2011 , a&a , 529 , a24 cccccc feature & spot & @xmath58 & @xmath19 & @xmath20 & + id@xmath59 & id@xmath60 & [ km s@xmath5 ] & [ mas yr@xmath5 ] & [ mas yr@xmath5 ] & epoch + 1 & 1 & 6.74 & -2.0(0.4 ) & -1.8(0.4 ) & 2011/068 - 2011/296 + 1 & 3 & 6.95 & -2.2(0.5 ) & -1.7(0.6 ) & 2011/068 - 2011/296 + 1 & 5 & 7.16 & -2.7(1.2 ) & -1.0(1.4 ) & 2011/068 - 2011/250 + 2 & 2 & 6.74 & 5.2(0.9 ) & -10.2(0.8 ) & 2011/296 - 2012/164 + 2 & 4 & 6.95 & 4.8(0.6 ) & -9.7(0.5 ) & 2011/250 - 2012/207 + 2 & 6 & 7.16 & 4.0(0.3 ) & -9.0(0.2 ) & 2011/068 - 2012/207 + 2 & 7 & 7.37 & 4.0(0.2 ) & -9.1(0.2 ) & 2011/068 - 2012/207 + 2 & 8 & 7.58 & 4.0(0.2 ) & -9.1(0.2 ) & 2011/068 - 2012/269 + 2 & 10 & 7.79 & 4.0(0.2 ) & -9.0(0.2 ) & 2011/068 - 2012/269 + 3 & 9 & 7.58 & -2.6(9.2 ) & -4.1(8.3 ) & 2011/121 - 2011/208 + 3 & 11 & 7.79 & 0.6(3.7 ) & -5.5(3.7 ) & 2011/121 - 2011/208 + + +

we present an x - ray study of our nearest starburst galaxy ic 10 , based on _ xmm - newton _ and observations . a list of 73 _ xmm - newton _ and 28 detections of point - like x - ray sources in the field is provided ; a substantial fraction of them are likely stellar objects in the milky way due to the low galactic latitude location of ic 10 . the brightest source in the ic 10 field , x-1 , has a mean 0.3 - 8.0 kev luminosity of and shows a large variation by a factor of up to on time scales of s during the _ xmm - newton _ observation . the x - ray spectra of the source indicate the presence of a multi - color blackbody accretion disk with an inner disk temperature kev . these results are consistent with the interpretation of the source as a stellar mass black hole , probably accreting from a wolf - rayet star companion . we infer the mass of this black hole to be m if it is not spinning , or a factor of up to higher if there is significant spinning . we also detect an apparent diffuse soft x - ray emission component of ic 10 . an effective method is devised to remove the x - ray ccd - readout streaks of x-1 that strongly affect the study of the diffuse component in the _ xmm - newton _ and observations . we find that the diffuse x - ray morphology is oriented along the optical body of the galaxy and is chiefly associated with starburst regions . the diffuse component can be characterized by an optically thin thermal plasma with a mean temperature of k and a 0.5 - 2 kev luminosity of , representing only a small fraction of the expected mechanical energy inputs from massive stars in the galaxy . there is evidence that the hot gas is driving outflows from the starburst regions ; therefore , the bulk of the energy inputs may be released in a galactic wind .

astro-ph0507041

x - ray observations are the most sensitive tool available for the study of massive star end products , both stellar and interstellar . ic 10 is our nearest laboratory for such a study in a dwarf starburst galaxy . this galaxy is particularly known for its large number of wolf - rayet ( w - r ) stars ; the global surface density , @xmath8 w - r kpc@xmath9 over an area of @xmath10 , is about 20 times higher than in the large magellanic cloud ( richer et al . 2001 ; massey & holmes 2002 ) . this large population of w - r stars is very unusual for the low metallicity of ( @xmath11 solar ; e.g. , lequeux et al . 1979 ; crowther et al . the galaxy also contains a number of energetic interstellar shells ; their origins are , however , unclear ( e.g. , wilcots & miller 1998 ; yang & skillman 1993 ) . the presence of an enigmatic radio emission shell of diameter @xmath12 pc around the brightest x - ray source x-1 is particularly interesting @xcite . the radio emission is predominantly nonthermal , as indicated by its steep spectrum . the required presence of an extremely powerful source of relativistic particles is _ not _ expected for a typical shell - like supernova remnant on such a large scale or for a young superbubble produced by an ob association , which would generate overwhelming amounts of ionizing radiation . indeed , the shell shows only weak h@xmath13 ( or thermal ) emission @xcite . it has been suggested that the nonthermal radio shell may be powered by x-1 , similar to the link between the galactic x - ray binary ss 433 and its surrounding nonthermal structure w50 ( diameter @xmath14 pc ) , albeit on greater physical and energetic scales @xcite . x-1 is the focus of the previous x - ray studies by @xcite and @xcite . they find that the source is spatially coincident with the w - r star [ mac92 ] 17a in ( massey , armanroff , & conti 1992 ) and speculate that the accreting compact object is a black hole ( bh ) based on its high x - ray luminosity , strong variability , and the high mass of this likely companion star . it is of great interest to confirm the identity of this compact object , which is comparable in many ways to cyg x-3 a candidate for a rare class of compact star / w - r binaries , which may shed light onto the differing pre - supernova evolutionary scenarios for very massive stars ( clark & crowther 2004 ) . here we present our _ xmm - newton _ observations of , together with a re - analysis of the _ chandra _ data studied by @xcite . these two complementary data sets together give us a comprehensive x - ray view of the galaxy . the data provide us with arcsecond spatial resolution , important for isolating point - like sources , especially in regions close to x-1 . but the data , taken with the detector acis - s in a ( 500 pixel ) sub - array mode , covers only a portion of the galaxy . in addition , the significant pile - up of x-1 adds a degree of spectral complexity@xcite . furthermore , the ccd - readout streak of the source significantly contaminates the `` diffuse '' x - ray component observed . the _ xmm - newton _ observations provide both a higher effective photon - collecting area and a larger field coverage than the observation , albeit with a substantially poorer resolution ( fwhm @xmath15@xmath16 ) . while the pile - up of x-1 is not an issue for the _ xmm - newton _ observations , the ccd - readout streak contamination is serious . therefore , we need to remove these streaks in order to correctly map out the diffuse x - ray component of . in this work , we adopt the distance of to be @xmath17 mpc ( hence @xmath18 = 204 pc ; demers , battinelli , & letarte 2004 and references therein ) . while data points in plots are shown all with @xmath19 error bars , our quoted parameter uncertainty ranges are at the 90% confidence level unless otherwise noted . the _ xmm - newton _ observations were taken on july 3 , 2003 with a total exposure time of @xmath20 45 ks for the epic - mos cameras and @xmath2042 ks for the epic - pn camera . we use only the data from the pn camera ( with a thin optical blocking filter ) because of its high sensitivity to soft x - rays in the study of diffuse emission and include the data from the mos cameras in the analysis of x-1 . the data from the rgs and om provide no useful information here and are thus not presented here . we calibrate the data with the sas software ( version 6.1.0 ) , together with the latest calibration files . for the mos data , we include only events with patterns 0 through 12 , and the standard mos flags ( xmmea_em ) . the pn events are filtered to include only events with patterns 0 through 4 , and the pn flags ( xmmea_ep , xmmea_2 , and xmmea_3 ) . we exclude time intervals with strong flares to reduce the non - cosmic x - ray background , which is due to cosmic - ray induced events , instrumental fluorescence , and soft protons that are funneled towards the detectors by the x - ray mirrors . we find that the flaring is consistent in the 0.3 - 2 kev and 10 - 15 kev bands , in which the light curves of the observations are examined . we thus remove time intervals with significant count rate ( cr ) deviations from the mean in the 0.3 - 2 kev band . the chosen threshold is 0.8 counts s@xmath21 for the mos and 2.0 counts s@xmath21 for the pn data . this filtering leaves a total accepted net exposure time of @xmath20 35.5 ks for each of the two mos cameras and @xmath20 32 ks for the pn camera . for the pn data , we construct maps in the 0.5 - 1 , 1 - 2 , 2 - 4.5 , and 4.5 - 7.5 kev bands . we have also examined the data in the 0.2 - 0.5 kev band , but find that it contains too many artifacts and little x - ray emission from . we generate corresponding effective exposure images , using the sas program @xmath22 . these exposure maps correct for various instrumental artifacts , including bad pixels and columns , the detector quantum efficiency non - uniformity , filter transmission , and telescope vignetting . such a correction is made on a count image after the removal of a background , which is partly due to non - cosmic x - rays not vignetted by the telescope . we generate the background data , using the _ skycast _ program and the `` blank sky '' database source - removed high galactic latitude observations with a total exposure of 352 ks @xcite . the background data are processed with the same filters as used for the observations . the most outstanding artifact in the _ xmm - newton _ data is the southeast - northwest streak ( fig . [ fig : im_pn_raw1 - 4 ] ) , which was caused by the continuous exposure of the pn ccd ( # 4 ) to x-1 during its readout . this streak spreads the counts ( statistically ) evenly into columns along the read - out direction of the ccd . such a streak is typically removed by replacing the affected columns with events randomly generated from an estimated background spectrum . this simple cosmetic repair , however , would also remove any sources and/or extended structures in the columns and would create new artifacts , especially in regions near x-1 . we have developed a more effective streak removal method . we calculate the average intensity distribution as a function of the column number across the ccd chip ( fig . [ fig : pn_st ] ) , after excluding regions that are significantly affected by diffuse or scattered x-1 emission as well as individual detected sources ( to be discussed later ) . the calculated streak contribution is then cast into the sky coordinates . the resultant streak image can then be subtracted from the observed x - ray intensity image . [ fig : sou_pn ] shows such a `` streak - free '' , background - removed , and flat - fielded intensity image in the 0.5 - 4.5 kev band . the streak produces no apparent effect in lower or higher energy bands . a streak image in a sub - energy band can easily be produced with a normalization of the relative intensity of the streak . the observation , taken on march 12 , 2003 with an exposure of 29 ks , has already been described by @xcite . we reprocess the data , using the latest ciao software ( version 3.2.1 ) and calibration database ( version 3.0.0 ) , following the same procedure as detailed in wang et al . we create acis - s images and corresponding exposure ( flat - fielding ) images in the 0.3 - 0.7 , 0.7 - 1.5 , 1.5 - 3 , and 3 - 7 kev bands . a ccd - readout streak similar to that seen in the pn data is also found in the acis - s image in the 0.7 - 3 kev range ( fig . [ fig : acis_st]a ) . we remove this ccd - readout streak in the same way as used for the pn data . the total number of counts contributed by the streak is only about 92 , a @xmath23 excess above the local background . the streak - removed image is shown in fig . [ fig : acis_st]b . the observation has a superb on - axis resolution of 90% eer@xmath24@xmath16within @xmath25 . however , the psf degrades steeply at larger off - axis angles . the _ xmm - newton _ psf is relatively uniform across much of the pn field of view . therefore , our use of the data is limited within the central region as shown in fig . [ fig : sou_acis ] . we use an idl - based program to detect sources in both the acis - s and the _ xmm - newton _ pn data . as in the previous applications ( e.g. , wang et al . 2003 ; wang 2004a ) , this program , optimized to detect point - like sources , uses a combination of detection algorithms : wavelet , sliding - box , and maximum likelihood centroid fitting . first , we construct the ` mexican cap ' wavelet images on scales of 1 , 2 , 4 and 8 image bins ( bin size @xmath26@xmath16for the pn and 0.5@xmath16for the acis - s ) . on each scale we search for source candidates corresponding to local maxima with signal - to - noise ratios @xmath27 . next , we apply a map detection ( ` sliding box ' method ) with a background map produced by excising the source candidates and adaptively smoothing the map to achieve a local count - to - noise ratio greater than 10 . finally , the sources detected with the map method are analyzed by a maximum likelihood algorithm , using both the background map and an approximate gaussian psf . this analysis gives optimal source positions and their @xmath19 errors . the source detection in both the map detection and the maximum likelihood analysis is based on data within the 50% psf energy - encircled radius ( eer ) for the pn data and the 90% eer for the acis - s data . the final accepted sources all have the individual false detection probability @xmath28 ( acis - s ) and @xmath29 ( pn ) . the processed pn data still contain various residual artifacts ( more significant than in the acis - s data ) ; the chosen stringent detection threshold minimizes their effect in the source detection . the source detections are carried out in the broad ( b ) , soft ( s ) , and hard ( h ) bands , defined differently for the acis - s and pn data sets in the notes to tables 1 and 2 . for each data set , the detected sources in the three bands are merged together . multiple detections with overlapping 2@xmath30 centroid error circles are considered to be the same source , and the centroid with the smallest error is adopted . tables 1 and 2 summarize the results from our source detection in the two data sets . for ease of reference , we will refer to x - ray sources detected in the _ xmm - newton _ pn and acis data sets with the prefixes xp and xa , respectively ( e.g. , xp-13 ) . the source locations are marked in the x - ray intensity images ( figs . [ fig : sou_pn ] and [ fig : sou_acis ] ) . the detection reveals many more sources in the data than in the _ xmm - newton _ data in the overlapping region around x-1 . part of this difference is due to the limited spatial resolution of the _ xmm - newton _ instrument ; the source detection sensitivity is very much affected by the psf of x-1 and the presence of the diffuse emission , as well as source confusion . nevertheless , most bright x - ray sources in the field are also detected with the _ xmm - newton _ data . several relatively bright sources , such as xa-13 , xa-17 , and xp-49 ( the pn source next to xa-35 in fig . [ fig : sou_acis ] ) , should have been detected in both data sets . the fact that they appear only in one data set indicates their strong variability from one observation to another . we may further use the _ xmm - newton _ and data sets to provide limited x - ray spectral information on the detected sources . most of the _ xmm - newton _ detected sources are apparently outside the optical / near - ir main body of and are probably unrelated to the galaxy ( fig . [ fig : sou_pn ] ) . the detected sources , on the other hand , are more likely to be associated with the galaxy . we thus extract an accumulated acis - s spectrum of the sources located in the s3 chip , excluding sources that are located at the edges ( i.e. , xa-12 , 16 , and 23 ; fig . [ fig : sou_acis ] ) . this spectrum can be characterized ( @xmath31 ) by a power law with a photon index of 1.2(0.89 - 1.6 ) and an absorption column density @xmath32 ( fig . [ fig : spec_sou_acis ] ) . although the constraints on the spectral parameters are generally weak , the index is consistent with the value ( @xmath33 ) expected for typical high - mass x - ray binaries . the @xmath34 is considerably lower than that expected total absorption toward sources inside ; the galactic absorption alone is @xmath35 ( stark et al . the absorption - corrected luminosity of the sources is @xmath36 in the 0.3 - 8 kev band . therefore , the spectrum , especially in the low energy part , may still significantly be contaminated by galactic objects . for relatively bright sources , we give hardness ratios to constrain individual spectral properties ( tables 1 and 2 ) . most of the calculated ratios are consistent with a power law that has an index in the range of 1 - 3 , together with a foreground absorption @xmath37 . such sources are probably x - ray binaries in or background agns . soft x - ray sources , especially those with hr1 values @xmath38 ( table 1 ) , should be in our galaxy ( 4.1 ) , because the inferred @xmath34 is less than the accepted galactic absorption . most of the other sources are too faint to be constrained individually . a useful , though crude , conversion factor from a count rate ( in either table 1 or 2 ) to an absorption - corrected energy flux in the 0.5 - 8 kev band is @xmath39 @xmath40 for a source with a power law of a photon index @xmath41 and a total x - ray - absorbing gas column density of @xmath42 , reasonable approximations for sources in ( see 4.1 ) . the _ xmm - newton _ observations show a large intensity variation of x-1 , a factor of up to @xmath1 and 4 in the pn 0.5 - 2.0 kev and 2.0 - 7.5 kev bands and on a time scale of @xmath43 s ( fig . [ fig : var ] ) . this amplitude is a factor of @xmath41 greater than that during the observation @xcite . similar variability is also seen in the mos data . the large luminosity dip , as seen in the light curves , could be due to an eclipsing of the accreting compact source by its companion . we detect no apparent periodic signal ; both the fft power spectrum and the period folding analysis show too much confusion from the harmonics and their aliases of the ccd readout period ( 73.4 ms ) . [ fig : spec_x1]a shows that the _ xmm - newton _ spectra of x-1 are nearly featureless ( except for various instrumental features ) . we fit the spectra with various models that are typically used for x - ray binaries , together with a foreground photoelectric absorption . the overall metal abundance of the absorbing gas is allowed to vary , relative to hydrogen , although the relative metal abundances are fixed to the pattern of balucinska - church & mccammon ( 1992 ) . we find that a simple power law model with a best - fit @xmath44 can be ruled out with a null hypothesis probability of @xmath45 ; the spectra appear to have an intrinsically blackbody - like convex shape . but the spectra are reasonably well characterized ( @xmath46 ) by the multi - color disk model ( mcd ; i.e. , _ diskbb _ in the _ xspec _ spectral analysis software package ; arnaud 1996 ; makishima et al . the fitted inner disk temperature and radius are @xmath47 kev and @xmath48 km , where @xmath49 is the inclination angle of the disk , while the absorption column density is @xmath50 with a metal abundance @xmath51 solar ; a fit with the fixed solar metal abundances is not acceptable ( @xmath52 ) . this unusually low metal abundance is probably not physical and may be due to uncertainties in the spectral calibration at low energies . the absorption - corrected luminosity of the source is @xmath53 in the 0.3 - 8.0 kev band . both the spectral shape and the luminosity of x-1 are very similar to those of the well - studied persistent x - ray binaries with stellar mass bhs ( i.e. , lmc x-1 , lmc x-3 , and cyg x-1 ) . to further characterize the putative bh as well as the accretion disk of x-1 , we fit the _ xmm - newton _ spectra with a self - consistent comptonized mcd ( cmcd ) model ( wang et al . 2004 ; yao et al . this model is implemented in xspec as a table of spectra generated from monte - carlo simulations of the comptonization , which assumes a spherically symmetric , thermal corona around an accretion disk . the fitted parameters of the cmcd model are the corona optical depth @xmath54 , effective radius @xmath55 km , electron temperature @xmath56 kev , and disk inclination angle @xmath57 , as well as @xmath58 kev and @xmath59 km . the obtained @xmath34 is almost the same as that obtained for the mcd model . the fit is satisfactory ( @xmath60 ) . following wang et al . ( 2004 ) , we further infer the mass of the bh by assuming that @xmath61 represents the radius of the last stable circular orbit , after correcting for various general relativity spectral hardening and doppler shift effects ( a factor of @xmath41 total correction ) . for a non - spinning bh , our inferred mass is 4.1(3.1 , 4.9 ) m@xmath5 , which depends weakly on @xmath49 ( within its 90% uncertainty range ) . but for an extreme spinning bh , the mass could be a factor of up to 6 greater , sensitively depending on @xmath49 ( see fig . 2 in wang et al . 2004 ) . there is a marginal presence of an emission line at @xmath62 kev , presumably representing the k@xmath13 transition of neutral or weakly ionized iron . the inclusion of a narrow gaussian line with the centroid fixed at this energy leads to a small improvement of the fit ( @xmath63 ) and does not significantly change the parameters of the other spectral components . the luminosity of the line is @xmath64 , corresponding to an equivalent width of @xmath65 ev . to test the spectral variability , we further extract two separate sets of spectra in the time intervals less and greater than 8000 s in fig . [ fig : var ] . however , the total of pn+mos counts in the lower flux interval is only about 470 . the two spectra show no statistically significant difference . for comparison with the _ xmm - newton _ spectra , we further extract a acis - s spectrum of x-1 from a circle of 8@xmath16radius and a local background spectrum from a concentric annulus of radius 32@xmath16 - 67@xmath16 . we use the newly improved ciao routine _ mkacisrmf _ to generate the spectral response matrices and include davis s model for the pile - up correction ( davis 2001 ) . our obtained pile - up fraction is similar to that found in bauer & brandt ( 2004 , who use a different method entirely ) . the same cmcd model without any parameter adjustment gives an acceptable fit to this acis - s spectrum of x-1 ( @xmath66 ) ; there is only a slight excess of the observed soft x - ray flux above the model . simply letting one parameter free [ e.g. , @xmath67 improves the fit [ @xmath68 ; f - test probability for no improvement is @xmath69 ; fig . [ fig : spec_x1]b ] . the inferred 0.3 - 8 kev luminosity is almost exactly the same as the mean value during the _ xmm - newton _ observation . therefore , there is no substantial difference in both the spectral shape and luminosity between the _ xmm - newton _ and spectra of x-1 . [ fig : spec_x1]b shows little evidence for residual , emission line - like features that are apparently present in the x-1 spectrum extracted by @xcite , at somewhat unexpected energies ( e.g. , @xmath70 and 3.65 kev ; their fig . 4 and discussion ) . therefore , these features of the spectra are likely data processing artifacts . there are clear indications for the presence of diffuse soft x - ray emission from . fig . [ fig : xc_c ] shows the large - scale soft x - ray intensity distribution of the field . the distribution is calculated after a subtraction of the blank - sky background , which is normalized to the local source - removed intensity estimated in an annulus of off - axis radius @xmath71 in the pn observation . in this step , we can approximately account for the background variation from one observation to another , particularly important for because of its low galactic latitude position . the soft x - ray structure near x-1 can only be examined in the high resolution observation ( fig . [ fig : acis_xc_c ] ; see also @xcite ) . to quantify the diffuse soft x - ray component , one needs to excise the detected sources and to estimate their residual contamination as well as sources below our detection limit . with the limited spatial resolution of the _ xmm - newton _ data , a clean separation of the sources from the diffuse component is difficult . one might use a large eer to exclude a large fraction of source counts . however , the 90% eer at 1.5 kev , for example , is @xmath72@xmath16on - axis and increases to @xmath73@xmath16at an off - axis angle of 12@xmath74 . the adoption of such an eer for all the sources would leave regions too sparse for a diffuse emission study . a smaller eer is adequate for the bulk of faint sources ( cr @xmath75 ) , which would leave acceptable amounts of contamination . for such a source , our adopted removal radius is twice the source detection radius for both the acis - s and pn data ( or @xmath76 psf eer for the pn data ) . for brighter sources , we scale this radius by a factor of @xmath77 . the adopted source removal regions are outlined in figs . [ fig : sou_pn ] and [ fig : sou_acis ] . of course , the remaining `` diffuse '' enhancement is still contaminated by the psf wing of x-1 and other discrete sources . the on - axis pn psf can be characterized by a king function , @xmath78^{-\alpha},\ ] ] where @xmath79 is the off - source angular distance and @xmath80 is a normalization factor , while the core radius @xmath81 or 52 and index @xmath82 or 1.5 for the the 0.5 - 2 kev or 2 - 4.5 kev bands , respectively . we compare the psf with the radial profiles of the source - removed pn intensity around x-1 by fitting the normalization and local uniform background ( fig . [ fig : x1_rbp ] ) . the psf gives a reasonably good fit to the profile in the 2 - 4.5 kev band ( @xmath83 ) . however , the fit in the 0.5 - 2 kev band is not acceptable ( @xmath84 ) ; an excess of the observed intensity above the psf is apparent at the annulus of the off - source distance between @xmath85 . the total net cr in this annulus is @xmath86 ( 0.005 ) @xmath87 in the 0.5 - 2 kev ( 2 - 4.5 kev ) band . we estimate that @xmath88 in the 0.5 - 2 kev band is due to the residual point - like contribution , assuming that it dominates the 2 - 4.5 kev band cr and has a mean spectrum similar to the best - fit absorbed power law of the detected sources ( 3.1 ) . the remaining @xmath89 in the 0.5 - 2 kev band may be a truly diffuse component . the luminosity of contribution is about the same as that of the detected sources within , excluding x-1 ( 3.1 ) . the faint point source contribution is probably a bit overestimated here , if it is less contaminated by the galactic stellar population ( which tends to contribute more soft x - rays ) than the detected source spectrum . we may also decompose the point - like and diffuse components in the pn data spectrally . we extract a spectrum of the diffuse component in the @xmath90 annulus . from the same region , we also obtain a background spectrum from the re - projected blank - sky data . the use of this background spectrum minimizes the effect of instrumental background variation across the detector . however , the sky background typically varies from one field to another . to account for this variation , we adjust the background spectrum by adding the spectral difference between the and blank - sky data in an annulus between @xmath91 off x-1 . in all of these spectral extractions , the source regions are removed from both the and the blank - sky data . [ fig : spec_d ] shows the background - subtracted net spectrum of the diffuse x - ray component of . the apparent emission line features at @xmath92 and @xmath93 , corresponding to the ne x and mg xi k@xmath13 transitions , indicates that part of the component has a thermal origin . therefore , we fit this spectrum with the combination of an optically thin thermal plasma ( xspec _ mekal _ model ) and a power law to characterize the residual point - like source contribution . the fit is satisfactory ( @xmath94 ; fig . [ fig : spec_d ] ) and gives reasonable model parameters : @xmath95 , metal abundance @xmath96 , power law photon index @xmath97 , and @xmath98 ( assuming the solar metal abundance for the absorbing gas ) . the index is marginally greater than that inferred from the fit to the accumulated source spectrum ( 3.1 ) , which is probably due , in part , to our demand for the same @xmath34 for both the power law and the plasma components . indeed , these two power laws ( one from the source spectrum and the other from the spectral decomposition here ) give comparable predictions of the relative contribution of point - like sources in the 0.5 - 2 kev and 2 - 4.5 kev bands . thus we conclude that the residual point - like source contribution , corrected for the absorption , is @xmath99 in the 0.5 - 2 kev band and the corresponding hot gas luminosity is a factor of @xmath4 higher . if an extrapolation of the best - fit thermal plasma model to energies outside the band is a good approximation , the total cooling rate of the plasma should then be about @xmath100 . following wang ( 2004a ) , we estimate the contribution of background agns in the sources detected from the pn data . this estimate uses a x - ray source log @xmath101 - log @xmath102 relation in the 0.5 - 2 kev band from the deep survey . we assume that these agns have a power law with a typical photon index of @xmath103 and are subject to an average foreground absorption of the galactic column density @xmath104 . accounting for both the sensitivity incompleteness and the eddington bias of our source detection in the 0.5 - 7.5 kev band , we estimate the agn contribution is @xmath105 . if the absorption is higher ( e.g. , including the column contribution from ) , the contribution would be smaller . although this estimate is quite uncertain , both statistically and systematically ( wang 2004 ) , it is clear that the bulk of our detected pn sources are either galactic or associated with . the identification of individual x - ray sources in the field is generally not easy , because of the galaxy s low galactic latitude position ( @xmath106 ) . in addition to the galactic column density , there is evidence for substantial contributions from molecular gas in the galaxy and/or gas inside . the comparison between the radio and h@xmath13 fluxes of regions , for example , indicates @xmath107 ( yang & skillman 1993 ) . from the galactic relation @xmath108 ( bohlin , savage , & drake 1978 ) , we infer @xmath109 , which is quite similar to the x - ray - absorbing gas column density that we have estimated from the above spectral analysis of x-1 and the sources in the main body of . considering the general high extinction expected for the objects in the field , we first use the two micron all sky survey ( 2mass ) all - sky catalog of point sources @xcite to search for potential counterparts . we cross - correlate the spatial positions of the objects in the catalog with those from tables [ pn_source_list ] and [ acis_source_list ] , using a matching radius of 4@xmath16 for pn sources and 2@xmath16 for acis sources . the radius is chosen to be greater than the @xmath19 statistical position uncertainty of almost all the sources . table [ tab_sou_id ] presents the matching results , including the position offset of each match ; there is no match with multiple 2mass objects . the table also includes the j , h , and k@xmath110 magnitudes of the matched 2mass objects ; the @xmath111 limiting sensitivities of the catalog are 17.1 , 16.4 and 15.3 mag in the three bands . the x - ray sources with the 2mass matches tend to have soft x - ray spectral characteristics , having negative hardness ratios and/or being detected preferentially in the s band ( tables [ pn_source_list ] and [ acis_source_list ] ) . these 2mass objects , many of which also show up in optical images , should mostly be galactic stars . the 2mass matches with those few sources with relatively hard x - ray spectral characteristics ( namely xp-16 , xp-34 , and xa-11 ) are all very faint ( e.g. k@xmath112 ) ; they may be stars in distant parts of the galaxy . some of the matches in table [ tab_sou_id ] should represent chance projections of 2mass objects within the matching regions . we estimate the expected number of such chance projections to be 0.25 for the acis sources and 2.6 for the pn sources , based on the surface number density of 2mass objects within annuli of radius 4@xmath16 - 15@xmath16around the x - ray sources . we next use the nasa / ipac extragalactic database and the simbad astronomical database to search for identified objects within 2@xmath16of each x - ray source position . we find that xp-29/xa-9 matches with an unresolved nonthermal radio source ( yang & skillman 1993 ; chyzy et al . 2003 ) , xp-9 with hd 1486 an eclipsing binary of algol type , and xp-25 with hl90 an hii region in . increasing the search radius to 4@xmath16only leads to one additional correspondence , i.e. , xp-57 with the nrao vla sky survey radio source nvss j002108 + 591132 . we have shown that x-1 has a blackbody - like x - ray spectrum . the inferred inner disk temperature ( @xmath113 kev ) is consistent with those found in lmc x-1 , lmc x-3 , and cyg x-1 stellar mass bh x - ray binaries . in contrast , x - ray binaries with intermediate mass bh candidates have substantially lower @xmath114 values ( @xmath115 kev ; e.g. , wang et al . the spectrum of an accreting neutron star in an x - ray binary system is typically dominated by a power law component and/or an optically thin x - ray photo - ionized plasma ( e.g. , paerels et al . 2000 ) . therefore , x-1 is most likely a stellar mass bh associated with the w - r star [ mac92 ] 17a . the only other known candidate for such an x - ray luminous bh / w - r binary system is cyg x-3 at a distance of @xmath116 kpc in our galaxy ( @xmath1177 ) , although the true nature of this source remains uncertain ( e.g. , mitra 1998 ) . the x - ray spectrum of cyg x-3 typically shows strong lines , which are interpreted as the evidence for a tenuous x - ray photo - ionized plasma , presumably in the stellar wind from the w - r companion star ( liedahl & paerels 1996 ) . but occasionally , the line emission weakens substantially , probably during a very high luminosity state ( smale et al . the spectrum in this state is very similar to that of x-1 in terms of both @xmath114 and the x - ray luminosity . longer observations of x-1 are needed to further the comparative study ( e.g. , to capture the low luminosity state of the source ) . despite its projected location near the galactic plane , x-1 is still in a line of sight that is considerabally less confusing and obscured than that of cyg x-3 . therefore , the study of x-1 may provide new insights into the end products of most massive stars ( e.g. , the progenitor mass required to form black holes in close binaries ; e.g. , wellstein & langer 1999 ; clark & crowther 2004 ) . evidence for the diffuse soft x - ray emission in the immediate vicinity of x-1 has also been presented by @xcite . from a comparison of the radial intensity profile around the source with the psf of the , they find that a diffuse emission enhancement with a total acis - s cr of @xmath118 extends @xmath119@xmath16 , a scale comparable to the radius of the nonthermal radio shell discovered by yang & skillman ( 1993 ) . this result is not significantly affected by our removal of the ccd - readout streak , which contributes @xmath120 of the enhancement . but its morphology is changed from the slightly northeast - southwest extension before the removal ( see fig . 1 in @xcite ) to the northwest orientation after the removal ( fig . [ fig : acis_st ] ) . we have further shown that the lower surface brightness diffuse soft x - ray emission extends at least @xmath121 ( fig . [ fig : xc_c ] ) , and possibly as far as @xmath25 ( fig . [ fig : acis_xc_c ] ) , northwest from x-1 . the morphology of the emission seems to represent an extension of recent intense star - forming regions where the bulk of known w - r stars are located . the overall extent of the emission appears to outline the boundaries of prominent h@xmath13 streamers / features ; as illustrated by @xcite with a high - quality h@xmath13 image , these features correspond to the shells 6 ( located slightly southwest to x-1 ) , 7 ( northwest to x-1 ) , 8 ( around x-1 ) , and possibly 4/5 further to the northwest . this morphological similarity suggests an association of the diffuse soft x - ray emission with the h@xmath13 streamers , i.e. , a scaled - up version of the 30 doradus nebula in the large magellanic cloud @xcite . the soft spectral characteristics of the diffuse x - ray emission indicates that it represents diffuse hot gas in . our spectral analysis also shows marginal evidence for enhanced metal abundances . but with a temperature of a few @xmath122 k , as we have inferred , this hot gas can not be confined by the gravity of . without sufficient cooling , the gas is likely to escape . indeed , as in many other active star - forming galaxies ( wang et al . 2001 ) , there is an apparent `` missing '' energy problem ( wang 2004b ) . with a typical mass - loss rate of @xmath123 and a wind speed of @xmath124 , a w - r star is expected to have a mechanical energy luminosity of @xmath125 . the w - r star population ( 1 ) alone could provide sufficient energy to balance the cooling of the diffuse soft x - ray - emitting gas ( @xmath100 ; 3.3 ) . of course , stellar winds from other massive stars as well as supernova blastwaves together should supply substantially more energy . but this energy is not observed in x - ray radiation . most likely , the energy is used to drive the expansion of the various observed energetic structures around recent massive star forming regions . this process is probably happening in the galaxy s southeast portion , where the hot gas is apparently still confined , naturally explaining the overall morphological similarity between the diffuse soft x - ray emission and the h@xmath13 streamer / features . once the hot gas is out of the confinement of the ism , the x - ray radiative cooling becomes even less efficient . this transition has probably occurred around previous star formation regions in the western portion of the galaxy . these regions , traced by large ( probably quiescent ) faint h@xmath13 shells @xcite , show little diffuse x - ray enhancement . we have conducted a careful analysis of the _ xmm - newton _ and x - ray ccd observations of the nearby starburst galaxy . in particular , we have devised an effective method to remove the ccd - readout streaks of the bright source x-1 . the main results of our analysis are as follows : * a list of 73 _ xmm - newton _ and 28 detections of point - like sources in the field is presented , including preliminary multi - wavelength identifications based on existing databases . while a large portion of these sources are likely to be foreground and background objects , those associated with tend to show hard x - ray spectral characteristics consistent with high - mass x - ray binaries . * we confirm that x-1 likely represents a bh / w - r binary system . the mean x-1 luminosity of @xmath0 in the 0.3 - 8.0 kev band is about the same during the _ xmm - newton _ and observations . but the source shows large luminosity variation by a factor of up to @xmath1 on time scales of @xmath2 s. the spectrum of x-1 is well characterized by a comptonized multi - color blackbody accretion disk with an inner disk temperature @xmath3 kev , typical of those persistent x - ray binaries with stellar mass bhs ( i.e. , lmc x-1 , lmc x-3 , and cyg x-1 ) . for x-1 , we infer the mass of the putative bh as @xmath4 m@xmath5 if it is not spinning , or a factor of up to @xmath1 higher if the spinning is important . * the presence of the diffuse soft x - ray emission in is revealed . the emission is morphologically oriented along the optical main body and is greatly enhanced around most active star - forming regions of the galaxy . the extent of the emission resembles that of luminous h@xmath13 streamers , emanating from starburst regions . the diffuse soft x - ray emission most likely arises from chemically - enriched hot gas with a characteristic temperature of @xmath126 kev and a total 0.5 - 2 kev luminosity of @xmath7 . although its total cooling rate ( @xmath127 ) is still quite uncertain , this hot gas could be easily maintained by the mechanical energy input from the observed large w - r star population in the galaxy . but the x - ray radiation is unlikely to be the main cooling mechanism of the hot gas , which apparently powers various energetic ism features observed around the starburst regions and will eventually escape from the galaxy . we thank the referee franz bauer for valuable comments on the work , which is supported by nasa through the grant nag5 - 13582 and the massachusetts space grant . this publication makes use of data products from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology , funded by the national aeronautics and space administration and the national science foundation . arnaud , k. a. 1996 , in astronomical data analysis software and systems v(asp conf . series volume 101 ) , eds . jacoby g. and barnes j. , p17 balucinska - church , m. , & mccammon , d. 1992 , apj , 400 , 699 bauer , f. , & brandt , w. n. 2004 , , 601 , 67 bohlin , r. c. , savage , b. d. , & drake , j. f. 1978 , , 224 , 132 brandt , w. n. , ward , m. j. , fabian , a.c . , & hodge , p. w. 1997 , mnras , 291 , 709 chyzy , k. t. , et al . 2003 , a&a , 405 , 513 clark , j. s. , & crowther , p. a. 2004 , a&a , 414 , 45 cutri , r. m. , et al . 2003 , vizier on - line data catalog crowther , p. a. , drissen , l. , abbott , j. b. , royer , p. , & smartt , s. j. 2003 , a&a , 404 , 483 davis , j. e. 2001 , apj , 562 , 575 demers , s. , battinelli , p. , & letarte , b. 2004 , a&a , 424 , 125 gil de paz , a. , madore , b. f. , & pevunova , o. 2003 , apjs , 147 , 29 jarrett , t. h. , chester , t. , cutri , r. , schneider , s. e. , & huchra , j. p. 2003 , aj , 125 , 525 lequeux , j. , et al . 1979 , a&a , 80 , 155 liedahl , d. a. , & paerels , f. 1996 , apjl , 468 , 33 makishima , k. , et al . 1986 , , 308 , 635 massey , p. , armanroff , t. e. , & conti , p. s. 1992 , aj , 103 , 1159 massey , p. , & holmes , s. 2002 , apj , 580 , 35 mitra , a. 1998 , apj , 499 , 385 paerels , f. , et al . 2000 , apjl , 533 , 135 read , a. m. , & ponman , t. j. 2003 , a&a , 409 , 395 richer , m. g. , et al . 2001 , a&a , 370 , 34 smale , a. p. , et al . 1993 , apj , 418 , 894 stark , a. a. , et al . 1992 , apjs , 79 , 77 thurow , j. c. , & wilcots , e. m. 2005 , aj , 129 , 745 wang , q. d. 1999 , apjl , 510 , 139 wang , q. d. , et al . 2001 , apjl , 555 , 99 wang , q. d. , chaves , t. , & irwin , j. 2003 , apj , 598 , 969 wang , q. d. 2004a , apj , 612 , 156 wang , q. d. 2004b , in extraplanar gas , ed . r. braun ( astro - ph/0410035 ) wang , q. d. , et al . 2004 , apj , 609 , 113 wellstein , s. , & langer , n. 1999 , a&a , 350 , 148 wilcots , e. m. , & miller , b. w. 1998 , aj , 116 , 2363 yang , h. , & skillman , e. d. 1993 , aj , 106 , 1448 yao , y. , & wang , q. d. 2005 , mnras , submitted lrrrrrrrr 1 & j001834.79 + 591946.8 & 1.8 & @xmath128 & & @xmath129 & b , s + 2 & j001845.81 + 592331.9 & 3.1 & @xmath130 & & @xmath131 & s , b + 3 & j001855.81 + 591240.1 & 1.5 & @xmath132 & & @xmath133 & s , b + 4 & j001857.44 + 590458.5 & 3.4 & @xmath134 & & & b + 5 & j001902.38 + 591623.3 & 2.2 & @xmath135 & & @xmath136 & b , s + 6 & j001909.72 + 591916.0 & 2.7 & @xmath137 & & & b , s + 7 & j001910.26 + 592455.9 & 3.7 & @xmath138 & & & b , s + 8 & j001917.94 + 591231.9 & 2.2 & @xmath139 & & @xmath140 & b , s + 9 & j001918.64 + 590820.7 & 0.3 & @xmath141 & @xmath142 & @xmath143 & b , s , h + 10 & j001922.60 + 591541.7 & 3.3 & @xmath144 & & & b + 11 & j001925.54 + 591417.5 & 2.1 & @xmath145 & & @xmath146 & s , b + 12 & j001926.58 + 591103.5 & 3.2 & @xmath147 & & & s , b + 13 & j001927.16 + 592423.6 & 3.1 & @xmath148 & & & b , h + 14 & j001929.33 + 590420.9 & 3.0 & @xmath149 & & @xmath150 & b , s + 15 & j001938.11 + 591652.6 & 2.9 & @xmath151 & & & b , s + 16 & j001944.06 + 591815.9 & 2.5 & @xmath152 & @xmath153 & & h , b + 17 & j001947.84 + 591711.5 & 2.0 & @xmath154 & @xmath155 & & b , h + 18 & j001949.48 + 590620.6 & 3.0 & @xmath156 & & & b , s + 19 & j001949.69 + 591334.3 & 2.5 & @xmath157 & & & b + 20 & j001950.60 + 590440.4 & 3.5 & @xmath158 & & & b , s + 21 & j001951.82 + 591328.5 & 2.2 & @xmath159 & & @xmath160 & s + 22 & j001952.36 + 590704.5 & 3.3 & @xmath161 & & & b , s + 23 & j001952.73 + 590854.1 & 2.7 & @xmath162 & @xmath163 & & b , h + 24 & j002008.34 + 592144.5 & 2.0 & @xmath164 & @xmath165 & & b , h , s + 25 & j002008.36 + 591953.2 & 2.5 & @xmath166 & & & b + 26 & j002008.84 + 591338.5 & 2.1 & @xmath167 & & @xmath168 & b , s + 27 & j002010.78 + 592547.2 & 2.0 & @xmath169 & @xmath170 & & b , h , s + 28 & j002012.48 + 591500.4 & 1.1 & @xmath171 & @xmath172 & @xmath173 & b , s , h + 29 & j002015.10 + 591854.0 & 1.4 & @xmath174 & & @xmath175 & b , s + 30 & j002015.84 + 592637.2 & 2.8 & @xmath176 & & & b , h + 31 & j002018.03 + 590732.0 & 1.7 & @xmath177 & @xmath178 & @xmath179 & b , h , s + 32 & j002018.31 + 591834.5 & 2.8 & @xmath180 & & & h + 33 & j002021.57 + 591900.1 & 2.0 & @xmath181 & & & b , h , s + 34 & j002023.16 + 590812.8 & 2.0 & @xmath182 & & @xmath183 & b , s , h + 35 & j002023.46 + 591445.5 & 1.3 & @xmath184 & & @xmath185 & s , b + 36 & j002024.47 + 591525.7 & 0.5 & @xmath186 & @xmath187 & @xmath188 & b , s , h + 37 & j002025.84 + 591844.6 & 1.6 & @xmath189 & & @xmath190 & b , s , h + 38 & j002027.03 + 590613.8 & 3.0 & @xmath191 & & & b , s + 39 & j002029.10 + 591651.1 & 0.1 & @xmath192 & @xmath193 & @xmath194 & b , s , h + 40 & j002033.97 + 591112.0 & 3.2 & @xmath195 & & & s + 41 & j002034.07 + 591031.9 & 2.0 & @xmath196 & @xmath197 & & b , h + 42 & j002034.12 + 591555.6 & 1.2 & @xmath198 & & @xmath199 & b , s + 43 & j002036.20 + 592402.6 & 2.9 & @xmath200 & & & b , s + 44 & j002042.27 + 591709.6 & 2.6 & @xmath201 & & & b , h + 45 & j002042.63 + 591850.3 & 2.9 & @xmath202 & & & b + 46 & j002047.38 + 591935.4 & 2.1 & @xmath203 & @xmath165 & & b , h , s + 47 & j002050.24 + 591528.3 & 2.9 & @xmath204 & & & s , b + 48 & j002050.26 + 591505.1 & 2.7 & @xmath205 & & & b + 49 & j002052.96 + 591647.4 & 2.4 & @xmath206 & & & b , h + 50 & j002053.78 + 592106.4 & 3.3 & @xmath207 & & & s + 51 & j002058.62 + 590833.8 & 3.0 & @xmath208 & & & b + 52 & j002100.75 + 591104.9 & 3.0 & @xmath209 & & & s , b + 53 & j002101.69 + 591519.5 & 2.1 & @xmath210 & & @xmath211 & s , b + 54 & j002102.62 + 592158.0 & 1.8 & @xmath212 & @xmath213 & @xmath214 & b , s , h + 55 & j002103.82 + 590957.4 & 2.2 & @xmath215 & @xmath216 & & h , b + 56 & j002104.70 + 591540.2 & 1.8 & @xmath217 & & @xmath218 & s , b + 57 & j002108.72 + 591134.4 & 2.6 & @xmath219 & & & b , s , h + 58 & j002114.93 + 590844.5 & 1.7 & @xmath220 & & @xmath221 & b , s + 59 & j002117.67 + 591839.9 & 3.4 & @xmath222 & & & b + 60 & j002121.44 + 590905.7 & 0.9 & @xmath223 & & @xmath224 & b , s + 61 & j002125.43 + 591904.2 & 1.8 & @xmath225 & & @xmath226 & b , s + 62 & j002131.27 + 591428.0 & 2.6 & @xmath227 & & & b , s + 63 & j002132.53 + 591313.6 & 3.1 & @xmath228 & & & b + 64 & j002134.56 + 591443.6 & 2.8 & @xmath229 & & & b , s + 65 & j002139.04 + 592243.4 & 1.9 & @xmath230 & & @xmath231 & s , b + 66 & j002142.96 + 591207.3 & 3.4 & @xmath232 & & & b + 67 & j002144.80 + 591730.8 & 3.4 & @xmath233 & & & b + 68 & j002144.98 + 590736.2 & 3.3 & @xmath234 & & & b + 69 & j002151.67 + 590820.1 & 3.5 & @xmath235 & & & b , h + 70 & j002155.38 + 591008.5 & 2.6 & @xmath236 & & & b , s + 71 & j002158.87 + 591326.7 & 1.3 & @xmath237 & & @xmath238 & b , s + 72 & j002203.62 + 591515.4 & 3.9 & @xmath239 & & & b + 73 & j002209.37 + 591357.6 & 1.7 & @xmath240 & & @xmath241 & b , s + lrrrrrrrr 1 & j001954.74 + 591721.5 & 1.7 & @xmath242 & & & s , b + 2 & j002004.91 + 591507.4 & 0.7 & @xmath243 & & & b , h + 3 & j002008.41 + 591540.0 & 0.4 & @xmath244 & & & b , h + 4 & j002009.23 + 591818.3 & 0.9 & @xmath245 & & & b , h + 5 & j002011.49 + 591627.0 & 0.6 & @xmath246 & & & b , h + 6 & j002012.54 + 591757.8 & 0.3 & @xmath247 & & & h , b + 7 & j002012.75 + 591500.8 & 0.1 & @xmath248 & @xmath249 & @xmath250 & b , s , h + 8 & j002013.75 + 591626.6 & 0.2 & @xmath251 & & @xmath252 & b , s , h + 9 & j002015.04 + 591853.5 & 0.2 & @xmath253 & & @xmath254 & b , s + 10 & j002016.75 + 591556.2 & 0.8 & @xmath255 & & & h + 11 & j002020.99 + 591758.6 & 0.1 & @xmath256 & @xmath257 & @xmath258 & h , b , s + 12 & j002021.59 + 591901.5 & 0.5 & @xmath259 & & & b , h + 13 & j002022.92 + 591538.4 & 0.7 & @xmath260 & & & b + 14 & j002023.59 + 591444.0 & 0.1 & @xmath261 & & @xmath262 & b , s + 15 & j002024.54 + 591524.9 & 0.1 & @xmath263 & @xmath264 & @xmath265 & b , h , s + 16 & j002026.11 + 591844.7 & 0.5 & @xmath266 & & & s , b + 17 & j002026.68 + 591454.7 & 0.3 & @xmath267 & & & s , b + 18 & j002029.17 + 591651.2 & 0.0 & @xmath268 & @xmath269 & @xmath270 & b , h , s + 19 & j002030.74 + 591743.8 & 0.7 & @xmath271 & & & b , h + 20 & j002034.67 + 591556.8 & 0.2 & @xmath272 & & @xmath273 & b , s , h + 21 & j002038.76 + 591740.7 & 0.3 & @xmath274 & & & b , h + 22 & j002040.14 + 591435.3 & 0.8 & @xmath275 & & & s + 23 & j002041.41 + 591712.4 & 0.9 & @xmath276 & & & h + 24 & j002046.45 + 591731.4 & 0.4 & @xmath277 & & & s , b + 25 & j002047.30 + 591320.2 & 0.7 & @xmath278 & & & b , s + 26 & j002050.06 + 591504.9 & 2.3 & @xmath279 & & & b + 27 & j002050.18 + 591408.7 & 0.8 & @xmath280 & & & b , h + 28 & j002052.81 + 591648.5 & 0.4 & @xmath281 & & & b , h , s + lccrrrrr xp-1 & j001834.91 + 591947.6 & 1.2 & 12.4 11.7 11.5 + xp-3 & j001855.89 + 591240.6 & 0.8 & 12.8 12.1 11.9 + xp-9 & j001918.74 + 590820.3 & 0.9 & 6.9 6.8 6.8 + xp-11 & j001925.62 + 591419.1 & 1.7 & 14.0 13.3 13.1 + xp-12 & j001926.70 + 591105.4 & 2.1 & 14.2 13.7 13.6 + xp-16 & j001944.20 + 591812.5 & 3.6 & 15.3 14.9 14.7 + xp-19 & j001949.37 + 591333.7 & 2.6 & 15.5 14.7 14.6 + xp-21 & j001951.75 + 591327.1 & 1.5 & 11.6 11.2 11.1 + xp-34 & j002023.22 + 590814.0 & 1.2 & 16.2 15.7 14.5 + xp-35/xa-14 & j002023.51 + 591444.7 & 1.0 & 8.8 8.5 8.5 + xp-37/xa-16 & j002025.98 + 591845.3 & 1.3 & 15.4 14.4 14.7 + xp- 38 & j002026.57 + 590615.6 & 4.0 & 15.1 14.6 14.3 + xp-40 & j002034.39 + 591114.1 & 3.8 & 15.1 14.6 14.3 + xp-41 & j002033.80 + 591033.5 & 2.6 & 13.0 12.2 12.0 + xp-42/xa-20 & j002034.58 + 591557.3 & 3.9 & 13.9 13.3 13.1 + xp-43 & j002036.24 + 592405.6 & 3.0 & 16.4 15.8 15.4 + xp-45 & j002042.67 + 591851.4 & 1.1 & 15.1 14.5 14.4 + xp-47 & j002050.13 + 591525.9 & 2.6 & 15.3 14.6 14.3 + xp-50 & j002053.89 + 592108.2 & 2.0 & 12.2 11.6 11.4 + xp-52 & j002101.03 + 591104.5 & 2.2 & 14.9 14.1 14.0 + xp-53 & j002101.55 + 591518.1 & 1.8 & 10.6 9.9 9.7 + xp-56 & j002104.59 + 591539.9 & 0.9 & 11.2 10.9 10.7 + xp-58 & j002114.77 + 590845.5 & 1.6 & 10.9 10.6 10.5 + xp-60 & j002121.38 + 590905.3 & 0.6 & 11.0 10.6 10.5 + xp-61 & j002125.69 + 591904.9 & 2.1 & 14.0 13.5 13.2 + xp-65 & j002139.00 + 592243.3 & 0.3 & 14.4 13.8 13.5 + xp-71 & j002158.82 + 591327.2 & 0.7 & 11.8 11.4 11.2 + xp-73 & j002209.58 + 591401.1 & 3.8 & 12.4 12.0 11.9 + xa-11 & j002020.90 + 591759.0 & 0.8 & 16.4 15.4 15.5 + xa-13 & j002022.80 + 591539.0 & 1.1 & 13.5 13.2 13.1 + xa-17 & j002026.61 + 591455.9 & 1.3 & 13.7 13.0 12.7 + xa-22 & j002040.10 + 591435.5 & 0.3 & 10.8 10.4 10.2 + xa-24 & j002046.44 + 591731.8 & 0.4 & 15.4 14.8 14.4 + xa-25 & j002047.26 + 591321.2 & 1.1 & 14.8 14.1 13.9 +

we consider convex sets and functions over idempotent semifields , like the max - plus semifield . we show that if is a conditionally complete idempotent semifield , with completion , a convex function which is lower semi - continuous in the order topology is the upper hull of supporting functions defined as residuated differences of affine functions . this result is proved using a separation theorem for closed convex subsets of , which extends earlier results of zimmermann , samborski , and shpiz .

math0308166

in this paper , we consider convex subsets of semimodules over semirings with an idempotent addition , like the max - plus semifield @xmath4 , which is the set @xmath5 , with @xmath6 as addition , and @xmath7 as multiplication . convex subsets @xmath8 , or , satisfy @xmath9 where the operation `` @xmath10''should be understood componentwise , and where @xmath11 for @xmath12 . we say that a function @xmath13 is if its epigraph is max - plus convex . an example of max - plus convex function is depicted in figure [ fig : fconvex ] ( further explanations will be given in [ sec - fconvex ] ) . and its supporting half spaces ] motivations to study semimodules and convex sets over idempotent semirings arise from several fields . first , semimodules over idempotent semirings , which include as special cases sup - semilattices with a bottom element ( which are semimodules over the boolean semiring ) , are natural objects in lattice theory . a second motivation arises from dynamic programming and discrete optimization . early results in this direction are due to cuninghame - green ( see @xcite ) , vorobyev @xcite , romanovski @xcite , k. zimmermann @xcite . the role of max - plus algebra in hamilton - jacobi equations and quasi - classical asymptotics , discovered by maslov ( * ? ? ? vii ) led to the development of an `` idempotent analysis '' , by kolokoltsov , litvinov , maslov , samborski , shpiz , and others ( see @xcite and the references therein ) . a third motivation arises from the algebraic approach of discrete event systems @xcite : control problems for discrete event systems are naturally expressed in terms of invariant spaces @xcite . another motivation , directly related to the present work , comes from abstract convex analysis @xcite : a basic result of convex analysis states that convex lower semi - continuous functions are upper hulls of affine maps , which means precisely that the set of convex functions is the max - plus ( complete ) semimodule generated by linear maps . in the theory of generalized conjugacies , linear maps are replaced by a general family of maps , and the set of convex functions is replaced by a general semimodule . more precisely , given an abstract class of convex sets and functions , a basic issue is to find a class of elementary functions with which convex sets and functions can be represented . this can be formalized in terms of @xcite . if @xmath14 is a set and @xmath15 , a function @xmath16 is called @xmath17-_convex _ if there exists a subset @xmath18 of @xmath17 such that @xmath19 a subset @xmath20 of @xmath14 is said to be @xmath17-_convex _ @xcite if for each @xmath21 we can find a map @xmath22 such that @xmath23 in this paper , we address the problem of finding the set @xmath17 adapted to max - plus convex sets and functions . the analogy with classical algebra suggests to introduce functions : @xmath24 with @xmath25 , and functions , which are of the form @xmath26 where @xmath27 . in the max - plus case , we can not take for @xmath17 the set of affine or linear functions , because any sup of max - plus affine ( resp . linear ) functions remains max - plus affine ( resp . this is illustrated in the last ( bottom right ) picture in figure [ fig : affine ] , which shows the graph of a generic affine function in dimension @xmath28 ( the graph is the black broken line , see table [ tab:1 ] in [ sec - fconvex ] for details ) . it is geometrically obvious that we can not obtain the convex function of figure [ fig : fconvex ] as the sup of affine functions . ( linear functions , however , lead to an interesting theory if we consider max - plus functions instead of max - plus convex functions , see rubinov and singer @xcite , and downward sets instead of max - plus convex sets , see martnez - legaz , rubinov , and singer @xcite . ) we show here that for max - plus convex functions , and more generally for convex functions over conditionally complete idempotent semifields , the appropriate @xmath17 consists of of affine functions , which are of the form @xmath29 , where @xmath30 are affine functions , and @xmath31 denotes the residuated law of the semiring addition , defined in below . theorem [ tconverse ] shows that lower semi - continuous convex functions are precisely upper hulls of residuated differences of affine functions , and corollary [ cor - uconvex ] shows that the corresponding @xmath17-convex sets are precisely the closed convex sets . as an illustration , in the case of the max - plus semiring , in dimension @xmath28 , there are 4 kinds of residuated differences of affine functions , as shown in figure [ fig : affine ] , and table [ tab:1 ] : one of these types consists of affine functions ( bottom right , already discussed ) , another of these types consists only of the identically @xmath32 function ( top left , not visible ) , whereas the top right and bottom left plots yield new shapes , which yield `` supporting half - spaces '' for the convex function of figure [ fig : fconvex ] . the main device in the proof of these results is a separation theorem for closed convex sets ( theorem [ th-1 ] ) . we consider a convex subset @xmath20 of @xmath3 , where @xmath0 is an idempotent semifield that is conditionally complete for its natural order . then , we show that if @xmath20 is stable under taking sups of ( bounded ) directed subsets and infs of ( bounded ) filtered subsets , or equivalently , if @xmath20 is closed in birkhoff s order topology , and if @xmath33 , there exists an affine hyperplane @xmath34 with @xmath30 as in , containing @xmath20 and not @xmath35 . when @xmath36 , birkhoff s order topology coincides with the usual one , and we get a separation theorem for convex subsets of @xmath37 which are closed in the usual sense . a key discrepancy , by comparison with usual convex sets , is that a two sided equation @xmath38 is needed . theorem [ th-1 ] extends or refines earlier results by zimmermann @xcite , samborski and shpiz @xcite , and by the three first authors @xcite . some metric assumptions on the semifield , which were used in @xcite , are eliminated , and the proof of theorem [ th-1 ] is in our view simpler ( with a direct geometric interpretation in terms of projections ) . the method of @xcite only applies to the case where the vector @xmath35 does not have entries equal to the bottom element . this restriction is removed in theorem [ th-1 ] ( see example [ ex - ncex ] below for details ) . by comparison with @xcite , the difference is that we work here in conditionally complete semifields ( without a top element ) , whereas the result of @xcite applies to the case of complete semirings ( which necessarily have a top element ) . when the top element is a coefficient of an affine equation defining an hyperplane , the hyperplane need not be closed in the order topology , and a key part of the proof of theorem [ th-1 ] is precisely to eliminate the top element from the equation defining the hyperplanes . in many applications , birkhoff s order topology is the natural one , so that the present theorem [ th-1 ] gives a useful refinement of the universal separation result of @xcite . we finally point out additional references in which semimodules over idempotent semirings or related structures appear : @xcite . in this section , we recall some basic notions about partially ordered sets , residuation , idempotent semirings and semimodules . see @xcite for more details . by , we will mean throughout the paper a set equipped with a order . we say that an ordered set @xmath39 is if any subset @xmath40 has a least upper bound ( denoted by @xmath41 ) . in particular , @xmath42 has both a minimal ( bottom ) element @xmath43 , and a maximal ( top ) element @xmath44 . since the greatest lower bound of a subset @xmath40 can be defined by @xmath45 , @xmath42 is a complete lattice . we shall also consider the case where @xmath42 is only , which means that any subset of @xmath42 bounded from above has a least upper bound and that any subset of @xmath42 bounded from below has a greatest lower bound . if @xmath39 and @xmath46 are ordered sets , we say that a map @xmath47 is if there exists a map @xmath48 such that @xmath49 which means that for all @xmath50 , the set @xmath51 has a maximal element , @xmath52 . if @xmath53 is an ordered set , we denote by @xmath54 the ordered set , for which @xmath55 . due to the symmetry of the defining property , it is clear that if @xmath56 is residuated , then @xmath57 is also residuated . when @xmath58 are complete ordered sets , there is a simple characterization of residuated maps . we say that a map @xmath56 if for all @xmath59 , @xmath60 , where @xmath61 . in particular , when @xmath62 , we get @xmath63 . one easily checks that if @xmath39 and @xmath46 are complete ordered sets , then , a map @xmath56 is residuated if , and only if , it preserves arbitrary sups ( see ( * ? ? ? * th . 5.2 ) , or ( * ? ? 4.50 ) ) . in particular , a residuated map @xmath64 is , @xmath65 , which , together with , yields @xmath66 and @xmath67 , where @xmath68 ( resp . @xmath69 ) denotes the identity map of @xmath70 ( resp . @xmath42 ) . this also implies that : @xmath71 we now apply these notions to idempotent semirings and semimodules . recall that a is a set @xmath72 equipped with an addition @xmath73 and a multiplication @xmath74 , such that @xmath72 is a commutative monoid for addition , @xmath72 is a monoid for multiplication , multiplication left and right distributes over addition , and the zero element of addition , @xmath75 , is absorbing for multiplication . we denote by @xmath76 the neutral element of multiplication ( unit ) . we say that @xmath72 is when @xmath77 . all the semirings considered in the sequel will be idempotent . we shall adopt the usual conventions , and write for instance @xmath78 instead of @xmath79 . an idempotent monoid @xmath80 can be equipped with the order relation , @xmath81 , for which @xmath82 , and @xmath83 . we say that the semiring @xmath72 is ( resp . ) if it is complete ( resp . conditionally complete ) as a naturally ordered set , and if for all @xmath84 , the left and right multiplications operators , @xmath85 , @xmath86 , and @xmath87 , respectively , preserve arbitrary sups ( resp . preserves sups of bounded from above sets ) . an is an idempotent semiring whose nonzero elements are invertible . an idempotent semifield @xmath72 can not be complete , unless @xmath72 is the two - element , @xmath88 . however , a conditionally complete semifield @xmath72 can be embedded in a complete semiring @xmath89 , which is obtained by adjoining to @xmath72 a top element , @xmath90 , and setting @xmath91 , @xmath92 , and @xmath93 for @xmath94 . then , we say that @xmath89 is the of @xmath72 ( @xmath89 was called the of @xmath72 in @xcite , and the of @xmath72 in @xcite ) . for instance , the @xmath4 , defined in the introduction , can be embedded in the @xmath95 , whose set of elements is @xmath96 . a ( right ) @xmath72- @xmath14 is a commutative monoid @xmath97 , equipped with a map @xmath98 , @xmath99 ( right action ) , that satisfies @xmath100 , @xmath101 , @xmath102 , @xmath103 , and @xmath104 , for all @xmath105 , @xmath106 , see @xcite for more details . since @xmath107 is idempotent , @xmath108 is idempotent , so that @xmath73 coincides with the @xmath109 law for the natural order of @xmath14 . all the semimodules that we shall consider will be right semimodules over idempotent semirings . if @xmath72 is a complete semiring , we shall say that a @xmath72-semimodule @xmath14 is if it is complete as a naturally ordered set , and if , for all @xmath110 and @xmath111 , the left and right multiplications , @xmath112 , @xmath113 , and @xmath114 , @xmath115 , respectively , preserve arbitrary sups . we shall say that @xmath116 is a of @xmath14 if @xmath117 is a subsemimodule of @xmath14 stable under arbitrary sups . a basic example of semimodule over an idempotent semiring @xmath72 is the @xmath118 , or more generally the semimodule @xmath119 of functions from an arbitrary set @xmath120 to @xmath72 , which is complete when @xmath72 is complete . for @xmath121 and @xmath122 , we denote , as usual , by @xmath123 the @xmath124-th entry of @xmath125 . in a complete semimodule @xmath14 , we define , for all @xmath105 , @xmath126 where we write @xmath127 for the least upper bound to emphasize the fact that the set has a top element . in other words , @xmath128 is the residuated map of @xmath129 . specializing , we get @xmath130 for instance , when @xmath131 , @xmath132 , and @xmath133 if @xmath134 takes other values ( @xmath72 being thought of as a semimodule over itself ) . more generally , if @xmath72 is any complete semiring , the law `` @xmath135 '' of the semimodule @xmath118 can be computed from the law `` @xmath135 '' of @xmath72 by @xmath136 here , @xmath135 has a higher priority than @xmath137 , so that the right hand side of reads @xmath138 . if the addition of @xmath72 distributes over arbitrary infs ( this is the case in particular if @xmath72 is a semifield , or a completed semifield , see ( * ? ? ? 12 , th . 25 ) ) , for all @xmath111 , the translation by @xmath139 , @xmath140 , defines a residuated map @xmath141 , and we set : @xmath142 where we write @xmath143 for the greatest lower bound to emphasize the fact that the set has a bottom element . when @xmath131 , we have ( see e.g. @xcite ) : @xmath144 dualizing the definition of residuated maps , we get : @xmath145 we next recall the general separation theorem of @xcite . by , we mean throughout the section a complete semimodule over a complete idempotent semiring @xmath72 . let @xmath117 denote a complete subsemimodule of a complete semimodule @xmath14 . we call onto @xmath117 the map @xmath146 ( the least upper bound of @xmath147 belongs to the set because @xmath117 is complete ) . thus , @xmath148 is the residuated map of the canonical injection @xmath149 , @xmath148 is surjective , and @xmath150 . if @xmath151 is an arbitrary family , we set @xmath152 we say that @xmath153 is a of a complete subsemimodule @xmath117 if any element @xmath154 can be written as @xmath155 , for some @xmath156 . if @xmath117 is a complete subsemimodule of @xmath14 with generating family @xmath153 , then @xmath157 see ( * ? ? ? * th . 5 ) . [ th - separ ] let @xmath116 denote a complete subsemimodule , and let @xmath158 . then , the set @xmath159 contains @xmath117 and not @xmath35 . seeing @xmath160 as a `` scalar product '' , @xmath161 can be seen as the `` hyperplane '' of vectors @xmath125 `` orthogonal '' to @xmath162 . as shown in @xcite , the `` hyperplane '' @xmath161 is a complete subsemimodule of @xmath14 , even if it is defined by a nonlinear equation . in order to give a linear defining equation for this hyperplane , we have to make additional assumptions on the semiring @xmath72 . in this paper , we shall assume that @xmath163 is the completed semiring of a conditionally complete idempotent semifield @xmath0 . consider the semimodule of functions @xmath164 . when @xmath165 , we define @xmath166 and @xmath167 that is , @xmath168 ( for instance , when @xmath36 , @xmath169 . ) we have @xmath170 , and @xmath171 is bijective @xmath172 , which allows us to write @xmath161 linearly : @xmath173 ( see @xcite for generalizations to more general semirings , called semirings ) . theorem [ th - separ ] yields a separation result for convex sets as a corollary . we recall that a subset @xmath20 of a complete semimodule @xmath14 over a complete semiring @xmath72 is @xcite ( resp . @xcite ) if for all finite ( resp . arbitrary ) families @xmath174 and @xmath175 , such that @xmath176 , we have that @xmath177 . for example , every subsemimodule of @xmath14 is convex , and every complete subsemimodule of @xmath14 is complete convex . [ cor - convex ] if @xmath20 is a complete convex subset of a complete semimodule @xmath14 , and if @xmath21 , then the set @xmath178 with @xmath179 contains @xmath20 and not @xmath35 . recall our convention explained after equation [ e - def - lres ] , that @xmath135 has a higher priority than @xmath137 , so that for instance @xmath180 . when @xmath163 is a completed idempotent semifield , and @xmath164 , @xmath161 can be rewritten linearly : @xmath181 [ rk - ineq ] since @xmath182 , and @xmath183 , we have @xmath184 and @xmath185 , and hence , by definition of the natural order @xmath186 , we can write equivalently @xmath161 as @xmath187 the same remark applies , mutatis mutandis , to , , and . we complement the results of @xcite by giving a geometric interpretation to the vector @xmath188 and scalar @xmath189 which define the separating hyperplane @xmath161 . if @xmath20 is any subset of @xmath14 , we call , denoted by @xmath190 , the set of linear combinations @xmath191 we also denote by @xmath192 the generated by @xmath20 . the term `` shadow '' can be interpreted geometrically : when for instance @xmath193 , @xmath190 is the shadow of @xmath20 if the sun light comes from the top - right corner of the plane , see figure [ fig : shadowup ] and example [ ex - sh ] below . [ theo - qc ] if @xmath20 is a complete convex subset of a complete semimodule @xmath14 , then , for all @xmath194 , @xmath195 if @xmath196 , @xmath197 if @xmath189 is invertible , @xmath198 belongs to @xmath20 . thus , theorem [ theo - qc ] shows that @xmath199 is a projector which sends @xmath14 to @xmath190 , and @xmath200 to @xmath20 . moreover , when @xmath189 is invertible , @xmath198 can be considered as the projection of @xmath35 onto @xmath20 . since @xmath201 , @xmath202 if @xmath196 , we have @xmath203 for some @xmath204 , hence , @xmath205 ( by ) , which implies that @xmath206 . since @xmath207 holds trivially , we have proved that @xmath208 , so that @xmath209 consider now any element @xmath210 , @xmath211 , with @xmath212 , @xmath213 , @xmath214 , and assume that @xmath215 . then , @xmath216 , so that @xmath217 ( by ) , and since @xmath214 , @xmath218 . summing over all @xmath219 , we get @xmath220 . together with , this shows . since we also proved , this shows a fortiori . finally , if @xmath189 is invertible , we see from that @xmath198 is of the form @xmath221 with @xmath222 , hence @xmath198 belongs to @xmath20 . [ ex - sh ] in figure [ fig : shadowup ] , the convex @xmath20 generated by three points @xmath223 in @xmath224 is displayed , together with its shadow and upper set . the cases of @xmath35 belonging to @xmath200 and of @xmath225 are illustrated . when @xmath163 is a completed idempotent semifield , and @xmath20 is complete and convex , then @xmath226 indeed , let @xmath227 denote the set in the right hand side of . the inclusion @xmath228 is trivial . to show the other inclusion , take any @xmath229 , which can be written as a linear combination @xmath230 , for some @xmath231 , @xmath213 , @xmath214 , with @xmath232 a possibly infinite set . when @xmath233 , @xmath234 trivially . when @xmath235 , @xmath236 for some @xmath237 , so that @xmath238 , and since @xmath239 and @xmath72 is a completed idempotent semifield , @xmath240 is invertible . writing @xmath241 , and observing that @xmath242 belongs to @xmath20 because @xmath20 is complete and convex , we see that @xmath234 . [ cex ] to illustrate the previous results , consider the convex set @xmath193 generated by the two points @xmath243 and @xmath244 . thus , @xmath20 is the set of points of the form @xmath245 , with @xmath246 . since @xmath20 is generated by a finite number of points of @xmath247 , @xmath20 is complete convex . the set @xmath20 is the broken dark segment between the points @xmath243 and @xmath244 , in figure [ fig - todo ] . in order to represent points with @xmath32 coordinates , we use exponential coordinates in figure [ fig - todo ] , that is , the point @xmath248 is represented by the point of the positive quadrant of coordinates @xmath249 . consider now @xmath250 , for any @xmath251 , and let us separate @xmath35 from @xmath20 using corollary [ cor - convex ] . since @xmath252 , @xmath196 , and we get from that @xmath253 . one also easily checks that @xmath254 . when @xmath255 , the separating hyperplane @xmath161 of becomes : @xmath256 the point @xmath257 , together with @xmath188 and the separating hyperplane @xmath161 ( light grey zone ) are depicted at the left of figure [ fig - todo ] . when @xmath258 , is it easily checked that the separating hyperplane is the union of the half space @xmath259 , and of the interval @xmath260\times \{-\infty\}$ ] . unlike in the case of a finite @xmath261 , @xmath161 is not closed for the usual topology , which implies that the max - plus linear forms which define @xmath161 are _ not _ continuous for the usual topology . we next recall some basic facts about birkhoff s order topology ( * ? ? ? 10 , 9 ) , and establish some properties of closed convex sets . see @xcite for more background on topologies on lattices and lattices ordered groups . recall that a nonempty ordered set @xmath262 is if any finite subset of @xmath262 has an upper bound in @xmath262 , and that a nonempty ordered set @xmath263 is if any finite subset of @xmath263 has a lower bound in @xmath263 . we say that a subset @xmath14 of a conditionally complete ordered set @xmath42 is ( resp . ) if for all directed ( resp filtered ) subsets @xmath264 ( resp . @xmath265 ) bounded from above ( resp . below ) , @xmath266 ( resp . @xmath267 ) . when @xmath268 , where @xmath269 is a conditionally complete idempotent semifield , the condition that @xmath263 is bounded from below can be dispensed with , since any @xmath263 is bounded from below by @xmath75 . recall that a with values in a conditionally complete ordered set @xmath42 is a family @xmath270 indexed by elements of a directed set @xmath271 . we say that a net @xmath272 bounded from above and from below to @xmath273 , if @xmath274 , where @xmath275 , and @xmath276 . we say that @xmath40 is if for all nets @xmath277 order converging to some @xmath273 , @xmath110 . the set @xmath278 of order - closed subsets of @xmath42 defines the birkhoff s . in particular , if @xmath262 ( resp . @xmath263 ) is a directed ( resp . filtered ) subset of @xmath14 , @xmath279 ( resp . @xmath280 ) is a net which order converges to @xmath281 ( resp . @xmath282 ) , so that any order closed set is stable under directed sups and filtered infs . we warn the reader that a net which is order convergent is convergent for the order topology , but that the converse need not hold , see ( * ? ? ? 10 , 9 ) . however , both notions coincide when @xmath268 if @xmath0 is a conditionally complete semifield which is a continuous lattice @xcite . when @xmath283 , the order topology is the usual topology on @xmath284 . the following result applies in particular to convex subsets of semimodules . a subset @xmath285 stable under finite sups is closed for the order topology if and only if it is stable under directed sups and filtered infs . assume that @xmath20 is stable under directed sups and filtered infs , and let @xmath174 denote a net order converging to @xmath273 . we have @xmath286 , where @xmath287 . let @xmath288 denote the set of finite subsets of @xmath289 , and for all @xmath290 , define @xmath291 . since @xmath20 is stable under finite sups , @xmath292 . since @xmath293 is directed and bounded from above , and since @xmath20 is stable under directed sups , @xmath294 . since @xmath295 is filtered and bounded from below , and since @xmath20 is stable under filtered infs , @xmath296 , which shows that @xmath20 is closed for the order topology . this shows the `` if '' part of the result . conversely , if @xmath262 ( resp . @xmath263 ) is a directed ( resp . filtered ) subset of @xmath14 , then @xmath279 ( resp . @xmath297 ) is a net which order converges to @xmath281 ( resp . @xmath282 ) , so that any order closed set is stable under directed sups and filtered infs . we shall use repeatedly the following lemma in the sequel . [ lem - birkhoff ] if @xmath0 is a conditionally complete semifield , if @xmath298 order converges to @xmath299 , and @xmath300 order converges to @xmath301 , then @xmath302 order converges to @xmath303 , @xmath304 order converges to @xmath305 , and @xmath306 order converges to @xmath307 . in fact , the result of @xcite is stated only for elements of @xmath308 , but the extension to @xmath0 is plain , since @xmath309 and @xmath310 for all @xmath311 . however , lemma [ lem - birkhoff ] does _ not _ extend to @xmath1 : for instance , in @xmath95 , @xmath312 order converges to @xmath32 , but @xmath313 , which is the constant sequence with value @xmath314 , does not order converge to @xmath315 . this is precisely why the separating hyperplane provided by the universal separation theorem need not be closed , see example [ cex ] above . [ cor - order ] if @xmath316 , @xmath317 , if @xmath318 order converges to @xmath319 , and if @xmath320 order converges to @xmath321 , then , @xmath322 order converges to @xmath323 , @xmath324 order converges to @xmath325 , and @xmath326 order converges to @xmath327 . by , @xmath328 , and by , @xmath329 , where @xmath330 , so the corollary follows from lemma [ lem - birkhoff ] . we shall need the following basic property : [ l-1 ] if @xmath20 is a convex subset ( resp . a subsemimodule ) of @xmath3 , then , its closure for the order topology is a convex subset ( resp . a subsemimodule ) of @xmath3 . we derive this from lemma [ lem - birkhoff ] ( the only unusual point is that the order convergence need not coincide with the convergence for the order topology ) . assume that @xmath20 is convex ( the case when @xmath20 is a semimodule is similar ) . recall that if @xmath64 is a continuous self - map of a topological space @xmath14 , then @xmath331 holds for all @xmath332 , where @xmath333 denotes the closure of a subset of @xmath14 . fix @xmath334 such that @xmath335 , and consider @xmath336 , @xmath337 . we claim that for all @xmath338 , the map @xmath339 is continuous in the order topology . indeed , let @xmath340 denote a subset of @xmath3 that is closed in the order topology , and let us show that the pre - image by @xmath339 of @xmath340 , @xmath341 , is also closed in the order topology . if @xmath342 is any net in @xmath343 converging to some @xmath344 , we have @xmath345 , for all @xmath346 , and it follows from lemma [ lem - birkhoff ] that @xmath347 order converges to @xmath348 . since @xmath340 is closed in the order topology , @xmath349 , so @xmath350 , which shows that @xmath343 is closed in the order topology . thus , @xmath339 is continuous , and so @xmath351 . since @xmath20 is stable under convex combinations , @xmath352 , hence , @xmath353 . pick now any @xmath354 . since @xmath355 , for all @xmath356 , and since @xmath357 is continuous , @xmath358 . since this holds for all @xmath354 , we have shown that @xmath359 , i.e. , @xmath360 is stable under convex combinations . we conclude this section with properties which hold more generally in semimodules of functions . for all @xmath361 , we denote by @xmath362 the set of arbitrary convex combinations of elements of @xmath20 : @xmath363 ( @xmath232 denotes an arbitrary - possibly infinite - index set ) . [ prop - cbar1 ] if @xmath20 is a convex subset of @xmath364 which is closed in the order topology , then @xmath365 consider an element @xmath366 , with @xmath367 . assume , without loss of generality , that @xmath236 , for all @xmath346 . let @xmath262 denote the set of finite subsets of @xmath232 , and for all @xmath368 , let @xmath369 , and @xmath370 . by construction , the net @xmath371 order converges to @xmath372 , and the net @xmath373 order converges to @xmath76 . hence , by lemma [ lem - birkhoff ] , @xmath374 order converges to @xmath372 . but @xmath375 , and since @xmath20 is closed for the order topology , @xmath204 , which shows . when @xmath20 is a semimodule , the condition that @xmath20 is stable under filtered infs , which is implied by the condition that @xmath20 is closed in the order topology , can be dispensed with . [ prop - cbar2 ] if @xmath20 is a subsemimodule of @xmath364 which is stable under directed sups , holds . any element @xmath376 can be written as @xmath377 , for some @xmath378 . setting @xmath379 , we get @xmath380 , and we only need to know that @xmath20 is stable under directed sups to conclude that @xmath204 . the following example shows that we can not derive proposition [ prop - cbar2 ] from proposition [ prop - cbar1 ] . the set @xmath381 is a subsemimodule of @xmath382 , which is stable under directed sups , but not stable under filtered infs ( for instance @xmath383 ) , and hence not closed in the order topology . we saw in example [ cex ] that , when @xmath36 , the separating set given by the universal separation theorem need not be closed for the usual topology . in this section , we refine the universal separation theorem in order to separate a point from a _ closed _ convex set by a _ closed _ hyperplane . from now on , we assume that @xmath0 is a conditionally complete idempotent semifield , whose completed semiring is denoted by @xmath1 . as a preparation for the main result of [ newsepar ] ( theorem [ th-1 ] below ) , we derive from corollary [ cor - convex ] a separation result for order closed convex sets @xmath20 and elements @xmath21 of the semimodule of functions @xmath385 , satisfying an archimedean condition . this archimedean condition will be suppressed in theorem [ th-1 ] , assuming that @xmath120 is finite . [ dhyperpl ] we call of @xmath364 a subset of @xmath364 of the form @xmath386 with @xmath387 , and @xmath388 . we shall say that @xmath161 is a if @xmath389 . ( when @xmath120 is infinite , @xmath390 and @xmath391 may be equal to @xmath392 . ) we have already encountered `` hyperplanes '' of @xmath393 of the above form . indeed , @xmath394 , with @xmath161 of , is of the form , with @xmath395 the main point in definition [ dhyperpl ] is the requirements that @xmath396 , and @xmath397 which need not be satisfied in ( [ two26 ] ) ; indeed , for @xmath398 having a coordinate @xmath399 , by , we have @xmath400 , so that @xmath401 ( see e.g. example [ cex ] ) . given @xmath402 , the question is whether we can find an affine hyperplane of @xmath364 containing @xmath20 and not @xmath35 . we shall need the following on @xmath20 and @xmath35 : @xmath403 for all @xmath404 and @xmath405 , define @xmath406 one readily checks that @xmath407 for some @xmath408 implies that @xmath409 , and that when @xmath120 is finite , the converse implication holds ( indeed , if @xmath410 , take any @xmath321 smaller than @xmath411 , a quantity which is in @xmath412 when @xmath120 is finite ) . thus , assumption ( a ) implies that @xmath413 and it is equivalent to when @xmath120 is finite . [ th - n2 ] let @xmath20 be a convex subset of @xmath364 , and @xmath402 . assume that @xmath20 is closed for the order topology of @xmath364 , and that assumption ( a ) is satisfied . then , there is an affine hyperplane of @xmath364 which contains @xmath20 and not @xmath35 . [ rk - wpgeqws ] the separating hyperplanes constructed in the proof of proposition [ th - n2 ] can be written as , with @xmath414 and @xmath415 , so that , by the same argument as in remark [ rk - ineq ] , @xmath161 in may be rewritten as @xmath416 . first , we can assume that @xmath417 , which , by , means that @xmath418 otherwise , there is an index @xmath122 such that @xmath419 and @xmath420 , for all @xmath204 , so that the hyperplane of equation @xmath420 contains @xmath20 and not @xmath35 . we can also assume that @xmath421 indeed , if @xmath422 , we set @xmath423 , and consider the restriction map @xmath424 , which sends a vector @xmath425 to @xmath426 . we have @xmath427 . assuming that the theorem is proved when holds , we get vectors @xmath428 and scalars @xmath397 such that the affine hyperplane @xmath429 contains @xmath430 and not @xmath431 . let @xmath432 and @xmath433 denote the vectors obtained by completing @xmath434 and @xmath435 by zeros . then , the hyperplane @xmath436 contains @xmath20 and not @xmath35 . it remains to show proposition [ th - n2 ] when the equalities , hold . define the complete convex set @xmath437 as in . it follows from that @xmath438 . therefore , defining @xmath439 and @xmath440 as in , with @xmath20 replaced by @xmath437 , we get that the set @xmath161 of , where @xmath20 is replaced by @xmath437 , contains @xmath20 and not @xmath35 . by and , we have @xmath441 , so @xmath442 . also , by and , @xmath443 . moreover @xmath444 ( @xmath0 is conditionally complete and @xmath440 is the sup of a family of elements bounded from above by the unit ) . if @xmath445 we will have @xmath446 , and @xmath447 , and the set @xmath448 , where @xmath161 is as in , will be an affine hyperplane of @xmath364 . in order to show , take any @xmath122 . since the equalities and hold , we can find @xmath204 such that @xmath449 , and thanks to assumption ( a ) , @xmath450 , for some @xmath451 . hence , @xmath452 and @xmath453 which shows . when @xmath20 is a semimodule , the condition that @xmath20 is stable under filtered infs ( which is implied by the condition that @xmath20 is order closed ) can be dispensed with . [ th - n1 ] let @xmath20 be a subsemimodule of @xmath364 , and @xmath402 . assume that @xmath20 is stable under directed sups , and that assumption ( a ) is satisfied . then , there is a linear hyperplane of @xmath364 which contains @xmath20 and not @xmath35 . we reproduce the proof of proposition [ th - n2 ] , using directly , where @xmath454 , and noting that , by proposition [ prop - cbar2 ] , holds as soon as @xmath20 is stable under directed sups , when @xmath20 is a semimodule . in proposition [ th - n2 ] , we required the convex set to be order closed , but the separating sets , namely the affine hyperplanes of @xmath364 , where @xmath120 is infinite , need not be order closed , as shown by the following counter - example . let @xmath455 , @xmath36 , and let us separate @xmath456 from the convex set @xmath457 using proposition [ th - n2 ] . we obtain the affine hyperplane @xmath458 , where @xmath459 , and @xmath460 . consider the decreasing sequence @xmath461 , such that @xmath462 , for all @xmath463 , and @xmath464 , for @xmath465 , and @xmath466 , for @xmath467 , so that @xmath468 for all @xmath237 . we have @xmath469 , where @xmath470 for all @xmath463 , and @xmath471 , for all @xmath463 . since @xmath472 , @xmath161 is not stable under filtered infs . of course , this pathology vanishes in the finite dimensional case . [ affineareclosed ] affine hyperplanes of @xmath3 are closed in the order topology . this follows readily from lemma [ lem - birkhoff ] . the following example shows that the archimedean assumption is useful in proposition [ th - n1 ] . consider the semimodule @xmath473 , which is stable under directed sups , and consider the point @xmath474 , with @xmath475 , so that assumption ( a ) is satisfied . the proof of theorem [ th - n1 ] allows us to separate @xmath35 from @xmath20 by the linear hyperplane : @xmath476 however , consider now @xmath477 , which does not satisfy assumption ( a ) . we can not separate @xmath35 from @xmath20 by a linear ( or affine ) hyperplane , because such an hyperplane would be closed in the order ( = usual ) topology of @xmath382 ( by proposition [ affineareclosed ] ) whereas @xmath35 belongs to the closure of @xmath20 in this topology . thus assumption ( a ) can not be ommited in proposition [ th - n2 ] . in order to show that in the finite dimensional case , assumption ( a ) is not needed in proposition [ th - n2 ] , we establish some continuity property for projectors onto closed semimodules of @xmath3 . if @xmath117 is a subsemimodule of @xmath3 , we define @xmath479 as in ( the condition @xmath480 can be dispensed with , since @xmath117 is a semimodule ) , together with the projector @xmath481 @xmath482 since @xmath483 is a generating family of the complete semimodule @xmath484 , it follows from that @xmath485 if @xmath117 is a subsemimodule of @xmath3 , that is stable under directed sups , then the projector @xmath486 from @xmath487 onto @xmath484 admits a restriction @xmath148 from @xmath3 to @xmath117 . if @xmath488 , @xmath489 also belongs to @xmath3 , so that by , @xmath490 . we say that a map @xmath64 from @xmath42 to an ordered set @xmath70 ( resp . ) if @xmath491 ( resp . @xmath492 for all directed subsets @xmath493 bounded from above ( resp . for all filtered subsets @xmath494 bounded from below ) . [ theo - cont ] if @xmath117 is a subsemimodule of @xmath3 stable under directed sups and filtered infs , then @xmath148 preserves directed sups and filtered infs . let @xmath263 denote a filtered subset of @xmath117 , and @xmath495 . then , by @xmath496 , and since @xmath148 is isotone , we have @xmath497 furthermore , since @xmath148 is isotone , @xmath498 is filtered ( indeed , if @xmath499 is any finite subset of @xmath498 , we can write @xmath500 for some finite subset @xmath501 ; since @xmath263 is filtered , @xmath502 has a lower bound @xmath503 , and since @xmath148 is isotone , @xmath504 is a lower bound of @xmath500 , which shows that @xmath498 is filtered ) . hence , @xmath505 because @xmath117 is stable under filtered infs . since @xmath148 fixes @xmath117 , @xmath506 , and we get from , @xmath507 . the reverse inequality is an immediate consequence of the isotony of @xmath148 . consider now a directed subset @xmath508 bounded from above , and @xmath509 . we first show that for all @xmath510 , @xmath511 we shall assume that @xmath512 ( otherwise , the equality is trivial ) . since @xmath262 is directed , the net @xmath513 order converges to @xmath514 , and by corollary [ cor - order ] , this implies that @xmath515 order converges to @xmath516 . since @xmath517 is isotone , @xmath515 order converges to its sup . ( indeed , let @xmath518 denote an isotone map from @xmath262 to a conditionally complete ordered set , such that @xmath519 is bounded from above , and let us show more generally that @xmath519 order converges to its sup . observe that @xmath520 is independent of @xmath521 because @xmath262 is directed and @xmath518 is isotone . then , @xmath522 . also , since @xmath518 is isotone , @xmath523 , which shows that @xmath519 order converges to its sup . ) so , is proved . using , we get @xmath524 proposition [ theo - cont ] does not extend to semimodules of the form @xmath364 , where @xmath120 is an infinite set . indeed , take @xmath455 , @xmath36 , and let @xmath117 denote the semimodule spanned by the vector @xmath525 . for all @xmath526 , we have @xmath527 , where @xmath528 . consider now the sequence @xmath529 , such that @xmath530 if @xmath531 , and @xmath532 , otherwise . then , @xmath533 is a non - decreasing sequence with supremum @xmath372 . we have @xmath534 , but @xmath535 , which shows that @xmath148 does not preserve directed sups . the proof of theorem [ th-1 ] will rely on the following corollary of proposition [ theo - cont ] . [ cor - key ] if @xmath117 is a subsemimodule of @xmath3 stable under directed sups and filtered infs , and if @xmath536 , then , there is a vector @xmath537 with coordinates in @xmath538 , such that : @xmath539 let @xmath540 denote the set of vectors @xmath537 with coordinates in @xmath538 . let us assume by contradiction that @xmath541 since by proposition [ theo - cont ] , @xmath148 preserves filtered infs , we get from : @xmath542 since @xmath543 holds trivially , @xmath544 , hence , @xmath545 , a contradiction . the following finite dimensional separation theorem extends an earlier result of zimmermann @xcite . recall that when @xmath36 , the order topology on @xmath3 is the usual topology on @xmath37 . [ th-1 ] let @xmath20 denote a convex subset of @xmath3 that is closed for the order topology of @xmath3 , and let @xmath546 . then , there exists an affine hyperplane containing @xmath20 and not @xmath125 . we shall need the following lemma : [ support= ] if @xmath20 is a semimodule , and if @xmath547 , then @xmath548 . since @xmath549 , @xmath550 . conversely , pick any @xmath551 . since @xmath547 , we can find @xmath204 such that @xmath552 and @xmath553 . then , by , @xmath554 , and since @xmath555 , @xmath556 , which shows that @xmath557 . we showed in the first part of the proof of proposition [ th - n2 ] that we can always assume that @xmath558 we next prove theorem [ th-1 ] in the special case where @xmath20 is a semimodule , and then , we shall derive theorem [ th-1 ] , in general . the proof relies on a perturbation argument . pick a vector @xmath537 with coordinates in @xmath538 , ( hence , by , @xmath559 ) , and define , as in , : @xmath560 we will show that @xmath561 is a ( linear ) hyperplane , and that one can choose the above @xmath562 so that @xmath561 contains @xmath20 and not @xmath35 . it follows from @xmath563 and lemma [ support= ] that @xmath564 . since @xmath565 for all vectors @xmath566 of @xmath3 with coordinates different from @xmath75 , @xmath567 and @xmath568 belong to @xmath3 , which shows that @xmath561 is an hyperplane . by theorem [ th - separ ] , @xmath561 contains @xmath20 . let us check that : @xmath569 recall the classical residuation identity @xmath570 ( this can be shown by applying the first identity in to the map @xmath571 ) . if @xmath572 , we have @xmath573 , and by , @xmath574 . using @xmath575 and , we get @xmath576 , which shows . by corollary [ cor - key ] , there is a vector @xmath537 with entries in @xmath308 such that @xmath577 , and by , the hyperplane @xmath561 associated to this @xmath562 contains @xmath20 and not @xmath35 . associate to a convex set @xmath578 the semimodule : @xmath579 we denote by @xmath580 the closure of @xmath581 for the order topology of @xmath582 . we shall need the following : if @xmath20 is a convex subset of @xmath3 closed for the order topology , @xmath583 it suffices to show that @xmath584 is closed in @xmath582 for the order topology . take a net @xmath585 , with @xmath586 , order converging to some @xmath587 , with @xmath588 . we only need to show that if @xmath589 , @xmath590 . since @xmath589 , replacing @xmath232 by a set of the form @xmath591 , we may assume that @xmath592 , for all @xmath346 . then , @xmath593 , which implies that @xmath594 . since @xmath595 order converges to @xmath562 , and @xmath596 order converges to @xmath597 , by lemma [ lem - birkhoff ] , @xmath598 order - converges to @xmath599 . since , by our assumption , @xmath20 is closed for the order topology , @xmath600 , which shows that @xmath590 . so , @xmath601 is closed for the order topology . let us take @xmath602 . we note that by , @xmath603 . applying theorem [ th-1 ] , which is already proved in the case of closed semimodules , to @xmath580 , which is a semimodule thanks to lemma [ l-1 ] , we get a linear hyperplane @xmath604 , where @xmath605 , such that @xmath606 introducing @xmath607 , and @xmath608 , we see from that the affine hyperplane : @xmath609 contains @xmath20 and not @xmath35 . we needed to introduce the closure @xmath580 in the proof of theorem [ th-1 ] because @xmath581 need not be closed when @xmath20 is closed and convex . indeed , consider @xmath610 . we have @xmath611 and @xmath612 . [ ex - ncex ] when applied to example [ cex ] , the proof of theorem [ th-1 ] shows that for @xmath261 large enough , the hyperplane @xmath161 in separates the point @xmath243 from the convex set of figure [ fig - todo ] . the method of @xcite , which requires that the vector to separate from a convex set should have invertible entries in order to apply a normalization argument , does not apply to this case . we say that a map @xmath613 is if its epigraph is convex . by ( * ? ? ? * theorem 1 ) , @xmath64 is convex if , and only if , @xmath614 additionally , by ( * ? ? ? * theorem 2 ) , the ( lower ) level sets @xmath615 of @xmath64 are convex subsets of @xmath3 . when @xmath36 , we say that @xmath64 is . convex functions may of course be defined from an arbitrary @xmath0-semimodule @xmath14 to @xmath1 : we limit our attention to @xmath616 since the proof of the main result below relies on theorem [ th-1 ] which is stated for @xmath3 . we defined in the introduction functions and sets , when @xmath15 , see equations and . when more generally @xmath618 , we still define by , and extend by saying that a subset @xmath578 is if for all @xmath21 , we can find a map @xmath22 such that @xmath619 assume that @xmath20 is an intersection of ( lower ) level sets of maps in @xmath17 , that is , @xmath621 , where @xmath622 , @xmath623 , and @xmath232 is a possibly infinite set . if @xmath21 , @xmath624 , for some @xmath346 , so that @xmath625 . since @xmath626 , we have @xmath627 . we deduce that @xmath628 . hence , @xmath20 is @xmath17-convex . conversely , assume that @xmath20 is @xmath17-convex , and let @xmath629 denote the intersection of the sets @xmath630 , with @xmath22 and @xmath631 , in which @xmath20 is contained . trivially , @xmath632 . if @xmath21 , we can find @xmath22 satisfying . let @xmath633 . then , @xmath634 , and @xmath635 , so that @xmath636 . this shows that @xmath637 . thus , @xmath638 . we illustrate in table [ tab:1 ] below , and in figure [ fig : affine ] of [ sec - intro ] , the various shapes taken by differences of affine functions , when @xmath647 , and @xmath36 . for simplicity , a generic function in this class is denoted @xmath648 with @xmath649 . table [ tab:1 ] enumerates the four possible situations according to the comparisons of @xmath650 with @xmath651 and @xmath652 with @xmath653 . if @xmath566 is a difference of affine functions , by proposition [ prop - ls ] , the ( lower ) level sets of @xmath566 , are affine hyperplanes , which are closed by proposition [ affineareclosed ] , so @xmath566 is lower semi - continuous . as mentioned earlier , the function @xmath657 is convex if and only if its epigraph is convex . so , we consider two points @xmath658 and @xmath659 in the epigraph of @xmath566 , namely , @xmath660 which , by , is equivalent to @xmath661 let @xmath662 and @xmath663 in @xmath0 be such that @xmath664 . from the previous inequalities , we derive @xmath665 which , by , is equivalent to @xmath666 we have proved that @xmath667 belongs to the epigraph of @xmath566 . thus , @xmath566 is convex . 1 . @xmath20 is a convex subset of @xmath3 , and it is closed in the order topology ; 2 . @xmath20 is @xmath17-convex , where @xmath17 denotes the set of differences of affine functions @xmath668 , defined by . if @xmath20 is convex and closed , by theorem [ th-1 ] , for all @xmath33 , we can find an hyperplane containing @xmath20 and not @xmath35 . by remark [ rk - wpgeqws ] , we can choose @xmath669 in so that @xmath414 and @xmath415 . since @xmath670 , for all @xmath356 , @xmath671 for all @xmath356 , so that @xmath672 . since @xmath673 , and @xmath674 because @xmath414 and @xmath415 , we must have @xmath675 . then , @xmath676 , which shows that @xmath20 is @xmath17-convex . conversely , if @xmath20 is @xmath17-convex , proposition [ prop - inter ] shows that @xmath20 is an intersection of ( lower ) level sets of differences of affine functions . by proposition [ prop - ls ] , these ( lower ) level sets all are affine hyperplanes , and a fortiori , are convex sets . moreover , by proposition [ affineareclosed ] , affine hyperplanes are closed , so @xmath20 is closed and convex . [ tconverse ] a function @xmath677 is convex and lower semi - continuous if , and only if , it is a sup of differences of affine functions , i.e. , a @xmath17-convex function , where @xmath17 is the set of functions of the form . the proof relies on the following extension to the case of functions with values in a partially ordered set , of a well known characterization of abstract convexity of functions , in terms of `` separation '' ( see ( * ? ? ? 1.6i ) , or ( * ? ? 3.1 , eqn ( 3.31 ) ) ) . let @xmath683 . note first that @xmath64 is @xmath17-convex if , and only if , for all @xmath684 , @xmath685 , or equivalently , @xmath686 ( the other inequality always holds ) . recall that for all @xmath687 , @xmath688 denotes the upper set generated by @xmath689 . trivially : @xmath690 . applying this to @xmath691 and @xmath692 , we rewrite @xmath693 as @xmath694 since @xmath695 if , and only if , @xmath696 for some @xmath22 such that @xmath697 , and since it is enough to check the implication when @xmath698 ( if @xmath699 is the top element of @xmath1 , the implication trivially holds ) , the lemma is proved . as mentioned above , the epigraph of @xmath64 , @xmath703 , is a convex closed subset of @xmath704 . consider @xmath705 , so that @xmath706 . by theorem [ th-1 ] , there exist @xmath707 and @xmath708 in @xmath709 with @xmath710 such that @xmath711 @xmath712 that is , @xmath713 @xmath714 since the function identically equal to the top element of @xmath1 is trivially @xmath17-convex , we shall assume that @xmath715 , i.e. , @xmath716 . then , we claim that @xmath717 indeed , since @xmath718 , we may assume that @xmath719 . then , taking @xmath720 , with @xmath721 so large that @xmath722 , from we obtain @xmath723 . then , we can not have @xmath724 ( otherwise , @xmath725 would be bounded above independently of @xmath139 ) . therefore , for @xmath139 large enough @xmath726 , hence , @xmath727 , which , by @xmath728 , implies @xmath729 , and multiplying by @xmath597 , we get . let @xmath732 we claim that if @xmath733 , then @xmath719 . indeed , if @xmath734 , then and become @xmath735 @xmath736 but , , together with @xmath737 yields @xmath738 , for all @xmath720 , that is , @xmath739 when restricted to arguments lying in @xmath740 , in contradiction with , provided @xmath733 . this proves the claim @xmath741 in this case . equation implies that @xmath747 whence , by , @xmath748 and hence , defining @xmath749 and using that @xmath750 , from we see that @xmath751 for @xmath752 ; for @xmath753 , @xmath754 and this inequality is trivial , thus we have obtained the first half of . for the proof to be complete , we have to handle the case when @xmath758 which implies that @xmath759 . the previous arguments hold true up to a certain point when we can not claim that @xmath719 . either @xmath719 indeed , and the proof is completed as previously , or @xmath734 , and then boil down to @xmath760 @xmath761 without having to redefine the original @xmath762 . for any @xmath763 , define the functions @xmath764 which all belong to @xmath17 . because of , all those functions are identically equal to @xmath75 over @xmath740 , hence they are trivially less than or equal to @xmath64 over @xmath740 but also over the whole @xmath645 . on the other hand , because of , @xmath765 multiplying this strict inequality by a large enough @xmath662 , and using , we see that given @xmath699 , there exists @xmath662 for which @xmath766 , and a fortiori , @xmath767 . the proof is now complete . an illustration of theorem [ tconverse ] has been given in figure [ fig : fconvex ] in the introduction : the figure shows a convex function over @xmath4 , together with its supporting hyperplanes ( which are epigraphs of differences of affine functions , whose shapes were already shown in figure [ fig : affine ] ) . m. dubreil - jacotin , l. lesieur , and r. croisot . _ leons sur la thorie des treillis , des structures algbriques ordonnes , et des treillis gomtriques _ , volume xxi of _ cahiers scientifiques_. gauthier villars , paris , 1953 . i. singer . generalized convexity , functional hulls and applications to conjugate duality in optimization . in g. hammer and d. pallaschke , editors , _ selected topics in operations research and mathematical economics _ , number 226 in lecture notes econ . systems , pages 4979 . springer , 1984 .

we show that the probability that a finitely supported random walk on a non - elementary subgroup of the the mapping class group gives a non - pseudo - anosov element decays exponentially in the length of the random walk . more generally , we show that if is a set of mapping class group elements with an upper bound on their translation lengths on the complex of curves , then the probability that a random walk lies in decays exponentially in the length of the random walk .

1104.5543

let @xmath1 be a compact oriented surface of finite type , and let @xmath2 be the mapping class group of @xmath1 , i.e. the group of orientation preserving homeomorphisms of @xmath1 , up to isotopy . let @xmath3 be a probability distribution on @xmath2 with finite support . a random walk on @xmath2 is a markov chain with transition probabilities @xmath4 . we will always assume we start at the identity at time zero , and we will write @xmath5 for the location of the random walk at time @xmath6 . the probability distribution @xmath3 need not be symmetric , but we shall always assume that the semi - group generated by the support of @xmath3 is a non - elementary subgroup of the mapping class group . a subgroup of the mapping class group is _ non - elementary _ if it contains a pair of elements with distinct fixed points in @xmath7 , thurston s boundary for the mapping class group . rivin @xcite and kowalski @xcite showed that the probability that a random walk on the mapping class group gives rise to a pseudo - anosov element tends to one , as long as the group generated by the support of the mapping class group maps onto a sufficiently large subgroup of @xmath8 . furthermore , they showed that the probability that an element is _ not _ pseudo - anosov decays exponentially in the length of the random walk . malestein and souto @xcite and lubotzky and meiri @xcite extended this to the torelli subgroup , by considering the action of the torelli group on the homology of double covers of the surface . in @xcite it was shown that the probability that a random walk gives a element tends to one for all non - elementary subgroups of the mapping class , by considering the action of the mapping class group on the complex of curves , but no information was obtained about the rate of convergence . in this paper we show that the rate of convergence is exponential ; in fact , we show that for any constant @xmath9 , the probability that a random walk gives an element of translation length at most @xmath9 on the complex of curves decays exponentially in the length of the random walk ; the rate of decay depends on @xmath9 . we say the surface @xmath1 is _ sporadic _ if @xmath1 is a sphere with at most four punctures , or a torus with at most one puncture . the complex of curves @xmath10 is a simplicial complex , whose vertices consist of isotopy classes of simple closed curves , and whose simplices are spanned by disjoint simple closed curves . the mapping class group @xmath2 acts by simplicial isometries on the complex of curves , and masur and minsky @xcite showed that an element is if and only if its translation length on @xmath10 is positive . we will use landau s `` big @xmath11 '' notation , so @xmath12 denotes some function @xmath13 such that @xmath14 for some constant @xmath15 , and for all @xmath16 sufficiently large . [ theorem : main ] let @xmath2 be the mapping class group of a non - sporadic surface of finite type , and let @xmath5 be a random walk of length @xmath6 on @xmath2 generated by a finitely supported probability distribution @xmath3 , whose support generates a non - elementary subgroup of the mapping class group . then for any constant @xmath17 , there is a constant @xmath18 such that @xmath19 where @xmath20 is the translation length of @xmath5 acting on the complex of curves . if the surface is sporadic , then the mapping class group is either finite , or commensurable to @xmath21 , and in the latter case the result follows from the work of rivin @xcite or kowalski @xcite on random walks on matrix groups . theorem [ theorem : main ] does not apply to the torelli group of the genus two surface , as this group is not finitely generated , as shown by mccullough and miller @xcite . however , the results of @xcite hold in this case , but with no rate of convergence information . there are two main steps , both of which use the improper metric on @xmath2 arising from its action on the complex of curves , which we shall denote by @xmath22 , where @xmath23 is a basepoint in the complex of curves @xmath10 ; this is also known as a relative metric on @xmath2 . the first is to show that the random walk has a linear rate of escape in the relative metric , with exponential decay for the proportion of sample paths making progress at lower rate . the second is to consider the distribution of elements of bounded translation length on the complex of curves . if @xmath24 is an element of bounded translation length on the complex of curves , then @xmath24 is conjugate to an element @xmath25 , of bounded relative length . furthermore , if @xmath26 is chosen to be a shortest conjugating element , then the path @xmath27 is quasigeodesic , with quasigeodesic constants depending only on @xmath2 and the bound on translation length . this means that if a random walk @xmath5 is conjugate to an element of bounded translation length , then if the first half of a geodesic from @xmath28 to @xmath5 fellow travels with some geodesic from @xmath28 to @xmath26 , then the second half of the geodesic from @xmath28 to @xmath5 fellow travels with a translate of a geodesic from @xmath28 to @xmath29 . this fellow travelling condition is equivalent to the condition that the pair @xmath30 lies in a certain neighbourhood of the diagonal in @xmath31 , and we show that the probability that this occurs decays exponentially in the length of @xmath5 . we now give a brief summary of the organization of the paper . the remainder of this section is devoted to a detailed outline of the argument described in the previous paragraph . in section [ section : preliminaries ] we introduce some standard definitions and fix notation . in particular , we define subsets of @xmath2 , called shadows , and find upper bounds for the probability that a random walk lies in a shadow . in section [ section : linear progress ] we show the linear progress result , with an exponential decay bound for the proportion of paths making linear progress below some rate . finally , in section [ section : translation length ] we show that the fellow travelling property described above is equivalent to the condition that the pair @xmath30 lies in a certain neighbourhood of the diagonal in @xmath31 , consisting of unions of shadows , and we show that the probability that a random walk lies in one of these neighbourhoods decays exponentially in the length of @xmath5 . we will consider the action of the mapping class group on the complex of curves , see farb and margalit @xcite for an introduction to the mapping class group . the complex of curves @xmath10 is a simplicial complex , whose vertices are isotopy classes of simple closed curves , and whose simplices are spanned by disjoint simple closed curves . the complex of curves is finite dimensional , but not locally finite . we will only need to consider distances between vertices in the curve complex , and so we consider the @xmath28-skeleton of the complex of curves to be a metric space @xmath32 , by assigning every edge to have length @xmath28 . by abuse of notation , we will refer to this as a metric on the curve complex . the mapping class group acts on the complex of curves by simplicial isometries , and a choice of basepoint @xmath23 in the complex of curves determines a map from @xmath2 to @xmath10 defined by @xmath33 . we may therefore define an improper metric on the mapping class group by @xmath34 we emphasize that throughout this paper the metric @xmath35 will always refer to this improper metric induced from the action of the mapping class group on the complex of curves , and never a proper word metric on @xmath2 with respect to a finite generating set . however , the metric @xmath35 is quasi - isometric to a word metric on @xmath2 with respect to an infinite generating set , also known as a _ relative metric _ , formed by starting with a finite generating set and adding subgroups which stabilize vertices in the complex of curves which correspond to distinct orbits under the action of the mapping class group . masur and minsky @xcite showed that the curve complex is gromov hyperbolic , and we will write @xmath36 for the gromov boundary of @xmath2 , and @xmath37 for @xmath38 . a random walk of length @xmath6 on @xmath2 is a product of @xmath6 independent identically @xmath3-distributed random variables @xmath39 , which we shall call the _ steps _ of the random walk , so @xmath40 , and @xmath5 is distributed as @xmath41 , the @xmath6-fold convolution of @xmath3 with itself . a random walk converges to the gromov boundary almost surely , so this gives a hitting measure , known as _ harmonic measure _ on @xmath36 , and which we shall denote by @xmath42 . we will need to estimate the probability that a random walk lies in particular subsets of @xmath37 . we now define a family of subsets of @xmath37 which we shall call _ shadows_. recall that the gromov product of @xmath16 and @xmath43 with respect to @xmath28 is equal to the distance from @xmath28 to a geodesic from @xmath16 to @xmath43 , up to a bounded error which only depends on @xmath44 . given a real number @xmath45 , we can use the gromov product to define the shadow of a point @xmath16 in @xmath2 , which we shall denote by @xmath46 , @xmath47 we warn the reader that we use a different parameterization of shadows than that used by other authors , for example , blachre , hassinsky , and mathieu , @xcite define their shadows @xmath48 to be @xmath49 in our notation . we show that both the harmonic measures of shadows , and the @xmath41-measures of shadows , decay exponentially in @xmath45 , i.e. there are constants @xmath50 and @xmath18 such that @xmath51 and @xmath52 , for all @xmath16 , @xmath45 and @xmath6 . in @xcite , we showed that a random walk makes linear progress in the relative metric , almost surely , i.e. there is a constant @xmath53 such that @xmath54 we need a stronger version of this result , which gives an exponential decay bound for the rate of convergence . to be precise , we show : [ theorem : exponential linear progress ] let @xmath2 be the mapping class group of a non - sporadic surface of finite type , and let @xmath5 be a random walk of length @xmath6 on @xmath2 , generated by a finitely supported probability distribution @xmath3 , whose support generates a non - elementary subgroup of the mapping class group . then there are constants @xmath55 and @xmath18 , such that @xmath56 where @xmath35 is the non - proper metric on the mapping class group arising from its action on the complex of curves . we now give a brief overview of the proof of theorem [ theorem : exponential linear progress ] . consider taking the random walk @xmath57 steps at a time , i.e. consider @xmath58 instead of @xmath5 , which we shall refer to as the @xmath57-iterated random walk . we shall write @xmath59 for @xmath58 , and the steps of the @xmath57-iterated random walk are given by @xmath60 . the increments of the walk , @xmath61 , are all independent and identically distributed , with distribution @xmath62 . however , the distance travelled at time @xmath63 , given by @xmath64 is not the sum of the distances travelled at each step of the @xmath57-iterated walk , @xmath65 , as there may be some `` backtracking '' , illustrated schematically in figure [ pic : backtracks ] . the distance @xmath66 is the sum of the @xmath67 for @xmath68 , minus the total amount of backtracking . a key fact is that the distribution of the amount of backtracking at time @xmath69 is bounded above by an exponential function , and furthermore , the same upper bound holds for all @xmath69 , independent of the locations of the random walk , or the amount of backtracking at other times , and also independent of @xmath57 , the number of steps for each segment of the @xmath57-iterated random walk . we now explain why this is the case . the amount of backtracking can be estimated as follows . the size of the backtrack from @xmath70 to @xmath71 is roughly the distance from @xmath70 to the geodesic from @xmath28 to @xmath71 . after applying the isometry @xmath70 , this is the same as the distance from @xmath28 to the geodesic from @xmath72 to @xmath73 , as illustrated in figure [ pic : increment ] . the point @xmath73 is equal to @xmath74 , and the pair of random variables @xmath75 are independent , and distributed as @xmath76 , where @xmath77 is the @xmath6-fold convolution of the reflected distribution @xmath78 . if the backtrack is of length at least @xmath45 , then distance from @xmath28 to a geodesic from @xmath72 to @xmath74 is at least @xmath45 , up to bounded error , so in turn the gromov product @xmath79 is at least @xmath45 up to bounded error . therefore @xmath80 lies in @xmath81 , for some @xmath50 which only depends on @xmath44 . we show that both the harmonic measure @xmath42 , and the convolution measures @xmath82 , of shadows @xmath46 are bounded above by a function which decays exponentially in @xmath45 , and furthermore , the upper bound function is independent of both @xmath16 and @xmath6 . therefore , the probability that there is a backtrack of size @xmath45 decays exponentially in @xmath45 , independently of @xmath57 , and also independently of the locations of the random walk at other times . in particular , the expected size of a backtrack is bounded independently of @xmath57 , so by choosing @xmath57 sufficiently large , we can ensure that the expected value of each @xmath57-iterated step @xmath67 is larger than the expected value of a backtrack . furthermore , applying standard bernstein or chernoff - hoeffding estimates for concentration of measures , we obtain bounds for the probability that the sums of the first @xmath6 backtracks and @xmath57-iterated steps deviate from their expected values , and these bounds decay exponentially in @xmath6 . this implies that the distance away from the origin grows linearly at some rate , with exponential decay for the proportion of paths making progress below this rate . we now wish to show that the probability that @xmath5 is tends to @xmath28 exponentially quickly . the translation length of a group element on the complex of curves is @xmath83 masur and minsky @xcite showed that the elements are precisely those elements with non - zero translation length . the translation length of an element @xmath24 acting on the complex of curves is also coarsely equivalent to the shortest length of any conjugate of @xmath24 , measured in the relative metric on the mapping class group . in @xcite we showed that the has _ relative conjugacy bounds _ , i.e. there is a constant @xmath50 such that if two group elements @xmath84 and @xmath85 are conjugate , then @xmath86 for some element @xmath26 whose length is bounded in terms of the lengths of @xmath84 and @xmath85 , @xmath87 we emphasize that the distance @xmath35 here is the relative or curve complex distance on the mapping class group . every group element @xmath24 corresponds to a point in @xmath2 , but we may also think of @xmath24 as representing some choice of geodesic from @xmath28 to @xmath24 . we may therefore think of a product of group elements as representing a path in @xmath2 , composed of concatenating various translates of geodesics representing each element in the product . in particular , the word @xmath88 corresponds to a path consisting of three geodesic segments . as the curve complex is @xmath44-hyperbolic , one may show that if @xmath26 is chosen to be a conjugating word of shortest relative length , then the path @xmath88 is a quasigeodesic path in the curve complex , where the quasigeodesic constants depend on the relative conjugacy bound constant @xmath50 , and the length of @xmath85 . if we choose @xmath0 to be a collection of group elements of conjugacy length at most @xmath9 , then every element @xmath89 is equal to @xmath27 , where @xmath90 , and the paths @xmath27 are uniformly quasigeodesic over all elements of @xmath0 . this implies that the first half of the geodesic from @xmath28 to @xmath24 fellow travels with a translate of the inverse of the second half of the geodesic from @xmath28 to @xmath5 . in order to find an upper bound on the probability that this occurs , it is convenient to express this fellow travelling property in terms of the location of the pair @xmath91 in @xmath31 . the fact that @xmath27 is quasigeodesic implies that @xmath92 and @xmath93 , where @xmath45 is equal to @xmath94 , up to an additive error which only depends on @xmath44 and the quasigeodesic constants . we may extend the definitions of shadows to subsets @xmath95 by setting @xmath96 to be the union of all @xmath97 over all points @xmath98 . we may then extend the definition of shadows to subsets @xmath99 , by setting @xmath96 to be the union of all @xmath100 , over all @xmath101 . in particular , if a random walk @xmath5 lies in @xmath0 , then the pair @xmath30 lies in @xmath102 , a shadow of the diagonal @xmath103 in @xmath104 , where @xmath45 is roughly @xmath105 . by the linear progress result , we may assume that @xmath45 grows linearly in @xmath6 , up to a set of paths whose measure decays exponentially in @xmath6 . the distribution of pairs @xmath30 is obviously not independent , as @xmath5 determines @xmath106 , but they are asymptotically independent , and converge to @xmath107 . in fact , we may approximate the distribution of pairs @xmath108 by the distribution of pairs @xmath109 . this is because as sample paths converge to the boundary almost surely , it is probable that the the point @xmath5 looks close to the point @xmath110 , as viewed from the origin @xmath28 , as illustrated in figure [ pic : w2n ] below . similarly , standing at @xmath110 and looking back towards the origin @xmath28 , the point @xmath28 looks close to the midpoint of the path @xmath5 . if we apply the isometry @xmath111 , this implies that @xmath111 and @xmath112 look close together when viewed from @xmath28 . we can make this precise , and we show that the probability that @xmath110 lies in the shadow @xmath113 tends to one exponentially quickly , for some @xmath50 which only depends on the constant of hyperbolicity @xmath44 . the same argument shows that the probability @xmath111 lies in @xmath114 tends to one exponentially quickly . the pair @xmath115 is independent , and distributed as @xmath116 . we may then use the fact that the measure for shadows of points decays exponentially in @xmath45 to show that the @xmath116 measure of a shadow of the diagonal in @xmath104 also decays exponentially in @xmath45 . as @xmath45 grows linearly in @xmath6 , this shows that the probability @xmath5 has bounded translation length decays exponentially in @xmath6 . the author would like to thank g. margulis and d. thurston for useful conversations . the author was supported by nsf grant dms 0706764 . we now review some background on random walks on groups , see for example woess @xcite . let @xmath2 be the mapping class group of an orientable surface of finite type , which is not a sphere with three or fewer punctures , and let @xmath3 be a probability distribution on @xmath2 . we may use the probability distribution @xmath3 to generate a markov chain , or _ random walk _ on @xmath2 , with transition probabilities @xmath117 . we shall always assume that we start at time zero at the identity element of the group . the _ step space _ for the random walk is the product probability space @xmath118 , and we shall write @xmath119 for an element of the step space . the @xmath39 are a sequence of independent , identically @xmath3-distributed random variables , which we shall refer to as the _ increments _ of the random walk . the location of the random walk at time @xmath6 is given by @xmath120 , and so the distribution of random walks at time @xmath6 is given by the @xmath6-fold convolution of @xmath3 , which we shall write as @xmath41 . the _ path space _ for the random walk is the probability space @xmath121 , where @xmath122 is the set of all infinite sequences of elements @xmath2 , and the the measure @xmath123 is induced by the map @xmath124 . we shall always require that the group generated by the support of @xmath3 is _ non - elementary _ , which means that it contains a pair of elements with distinct fixed points in @xmath7 . we do not assume that the probability distribution @xmath3 is symmetric , so the group generated by the support of @xmath3 may be strictly larger than the semi - group generated by the support of @xmath3 . throughout this paper we will need to assume that the probability distribution @xmath3 has finite support . in @xcite , we showed that it followed from results of kaimanovich and masur @xcite and klarreich @xcite , that a sample path converges almost surely to a uniquely ergodic , and hence minimal , foliation in the gromov boundary of the relative space . this gives a measure @xmath42 on @xmath125 , known as _ harmonic measure_. the harmonic measure @xmath42 is @xmath3-stationary , i.e. @xmath126 @xcite [ theorem : converge ] consider a random walk on the mapping class group of an orientable surface of finite type , which is not a sphere with three or fewer punctures , determined by a probability distribution @xmath3 such that the group generated by the support of @xmath3 is non - elementary . then a sample path @xmath127 converges to a uniquely ergodic foliation in the gromov boundary @xmath125 of the relative space @xmath128 almost surely , and the distribution of limit points on the boundary is given by a unique @xmath3-stationary measure @xmath42 on @xmath125 . it will also be convenient to consider the _ reflected _ random walk , which is the random walk generated by the reflected measure @xmath129 , where @xmath78 . we will write @xmath130 for the corresponding @xmath129-stationary harmonic measure on @xmath125 . [ section : coarse ] we briefly recall some useful facts about gromov hyperbolic or @xmath44-hyperbolic spaces , and fix some notation . a @xmath44-hyperbolic space is a geodesic metric space which satisfies a _ @xmath44-slim triangles _ condition , i.e. there is a constant @xmath44 such that for every geodesic triangle , any side is contained in a @xmath44-neighbourhood of the other two . let @xmath131 be a @xmath44-hyperbolic space , which need not be proper . we shall write @xmath36 for the gromov boundary of @xmath2 , and let @xmath132 . given a subset @xmath133 , we shall write @xmath134 for the closure of @xmath135 in @xmath37 . given a point @xmath136 , the gromov product based at @xmath137 is defined to be @xmath138 we may extend the definition of the gromov product to points on the boundary by @xmath139 where the supremum is taken over all sequences @xmath140 and @xmath141 . this supremum is finite unless @xmath16 and @xmath43 are the same point in @xmath142 . we will make use of the following properties of the gromov product , see for example , bridson and haefliger @xcite*iii.h 3.17 . [ prop : gp ] + 1 . the gromov product @xmath143 is equal to the distance from @xmath137 to a geodesic from @xmath16 to @xmath43 , up to a bounded error of at most @xmath44 . [ prop : geodesic ] 2 . for any three points @xmath144 , @xmath145 [ prop : gpmin ] 3 . if @xmath146 , then there is a sequence @xmath147 with @xmath148 . [ prop : seqexist ] 4 . for any @xmath149 , and for any sequence @xmath150 , @xmath151 [ prop : seq ] we will also use the following stability property of quasi - geodesics in a @xmath44-hyperbolic space . let @xmath152 be a connected subset of @xmath153 . a _ quasi - geodesic _ is a map @xmath154 which coarsely preserves distance , i.e. there are constants @xmath50 and @xmath155 such that @xmath156 for every @xmath50 and @xmath155 there is a constant @xmath157 , which depends only on @xmath158 and @xmath44 , such that a finite @xmath159-quasigeodesic is hausdorff distance at most @xmath157 from a geodesic connecting its endpoints , see bridson and haefliger @xcite*iii.h theorem 1.7 . finally , we will also use the fact that nearest point projection onto a geodesic @xmath160 is coarsely well defined , i.e. there is a constant @xmath50 , which only depends on @xmath44 , such that if @xmath161 and @xmath162 are nearest points on @xmath160 to @xmath16 , then @xmath163 . furthermore , if @xmath43 is a point on a geodesic @xmath164 , and @xmath16 is a point with nearest point projection @xmath161 on @xmath164 , then the path consisting of a geodesic from @xmath16 to @xmath161 , and then from @xmath161 to @xmath43 is a bounded hausdorff distance @xmath165 from a geodesic from @xmath16 to @xmath43 , where @xmath165 only depends on @xmath44 , see for example @xcite*proposition 3.1 . [ section : shadows ] given a point @xmath149 and a real number @xmath45 , we define the _ shadow _ of @xmath16 based at @xmath28 , written as @xmath46 , to be @xmath47 if @xmath166 , and @xmath167 , then @xmath46 is empty . if @xmath168 , then @xmath46 consists of all of @xmath37 . we warn the reader again that this definition of a shadow differs slightly from that of other authors , for example blachre , hassinsky , and mathieu @xcite , who define their shadows @xmath169 to be @xmath170 , in our notation . we also remark that it is possible to use the gromov product to define a metric on the gromov boundary , where roughly speaking the distance between two boundary points is @xmath171 , where @xmath35 is the gromov product of the two points based at @xmath28 . in this case , the intersection of a shadow with the boundary is a small metric neighbourhood of the boundary point . however , we wish our neighbourhoods to include points in @xmath2 , for which the boundary metric is not defined , so we find our definition of shadows more convenient . we may extend the definition of shadows from points to arbitrary subsets of @xmath37 . given a subset @xmath172 , we define the _ shadow _ of @xmath173 based at @xmath28 , written @xmath96 , to be the union of the shadows of all points of @xmath173 , i.e. @xmath174 note that if @xmath173 contains points in @xmath2 , then in general @xmath175 . however , it is not hard to show that @xmath176 , though we will not use this fact directly . there is a lower bound on the gromov product of any two points in the shadow of a single point , which we now state as a proposition . this is a direct consequence of property [ prop : gp].[prop : gpmin ] above . [ prop : product bound ] for any @xmath177 , the gromov product @xmath178 . shadows are closed subsets of @xmath37 , and we now show that a shadow of a point is the closure of its intersection with @xmath2 . [ prop : closure ] @xmath179 . suppose @xmath180 is a sequence of points in @xmath181 , and @xmath182 . then , by the definition of a shadow , @xmath183 for all @xmath69 . this implies that @xmath184 , and so @xmath185 . therefore , by the definition of the gromov product for points in the boundary , @xmath186 , and so @xmath187 . conversely , if @xmath188 , then by property [ prop : gp].[prop : seqexist ] , there is a sequence @xmath147 with @xmath148 , and as the gromov product takes values in @xmath189 , we may pass to a subsequence such that @xmath190 , and so this gives a sequence @xmath180 contained in @xmath46 , which converges to @xmath43 . we shall write @xmath191 for all points which lie in a metric @xmath192-neighbourhood of @xmath193 , i.e. @xmath194 we now show that all points in a metric @xmath192-neighbourhood of a shadow of @xmath195 , are contained in a slightly larger shadow of @xmath195 . [ prop : metric nest ] for any @xmath196 , @xmath197 if @xmath198 , then there is a point @xmath199 with @xmath200 , and a point @xmath201 with @xmath202 . by the definition of the gromov product , @xmath203 which in turn is at least @xmath204 , so @xmath205 , as required . we will use the following properties of shadows of points , which follow from elementary arguments , see calegari and maher @xcite for detailed proofs . we state the results using the current notation of this paper . [ lemma : nested shadows ] @xcite*lemma 4.5 there is a constant @xmath206 , which only depends on @xmath44 , such that for all positive constants @xmath207 and @xmath45 , and any @xmath208 with @xmath209 , the shadow @xmath210 is disjoint from the complement of the shadow @xmath211 . furthermore for any pair of points @xmath212 such that @xmath213 and @xmath214 , the distance between @xmath84 and @xmath85 is at least @xmath207 . [ lemma : change basepoint ] @xcite*lemma 4.7 there are constants @xmath215 and @xmath216 , which only depend on @xmath44 , such that for any @xmath45 , and any three points @xmath217 with @xmath218 , there is an inclusion of shadows , @xmath219 where @xmath220 . [ lemma : shadow complement ] @xcite*lemma 4.6 there is a constant @xmath221 , which only depends on @xmath44 , such that for all constants @xmath222 , and all @xmath208 with @xmath223 , @xmath224 we may further extend the definition of a shadow to subsets of @xmath225 . let @xmath99 , and define the _ shadow _ @xmath96 to be @xmath226 we shall continue to write @xmath96 for the shadow in this case . hopefully this will not cause confusion , as it should be clear from context whether @xmath195 is a subset of @xmath37 or @xmath227 . finally , we remark that the lower bound for the gromov product in a shadow , proposition [ prop : product bound ] , immediately implies that the @xmath45-shadow of an @xmath25-shadow is contained in the shadow @xmath228 . [ prop : nested neighbourhoods ] let @xmath195 be a subset of either @xmath37 or @xmath104 . then @xmath229 for all @xmath45 and @xmath25 . in this section we show the following upper bounds for measures of shadows . [ lemma : exponential shadow ] let @xmath3 be a finitely supported probability distribution on @xmath2 whose support generates a non - elementary subgroup , and let @xmath42 be the corresponding harmonic measure . then there are constants @xmath230 , @xmath231 and @xmath18 , such that for any @xmath16 with @xmath232 and for any @xmath45 , @xmath233 and @xmath234 the constants @xmath230 , @xmath231 and @xmath155 depend on @xmath3 and @xmath44 , but not on @xmath45 or @xmath16 , as long as @xmath235 . here we write @xmath236 instead of @xmath237 , as it will be convenient to know explicitly the dependence of the implicit constants in @xmath237 . this result also applies to the reflected random walk generated by the probability distribution @xmath78 , and we may choose the constants to be the same for both random walks . the proof of this result is essentially the same as the proof of exponential decay of measures of halfspaces from @xcite . shadows are slightly more general sets than halfspaces , so the shadow result is not an immediate consequence of the halfspace result , although the halfspace result does follow from the version for shadows . although the shadow version could be deduced from the halfspace version , this still requires extra work , so we choose to give an argument here purely in terms of shadows . we start by giving some conditions on a family of nested subsets @xmath238 of @xmath37 , which guarantee that their measures decay exponentially in the index of the sets . we then show that a shadow @xmath46 is contained in a nested family of shadows satisfying the conditions , and furthermore , the number of sets in the nested family is linear in @xmath45 . if @xmath207 and @xmath9 are subsets of @xmath37 , then we define @xmath239 , the distance between @xmath207 and @xmath9 , to be the smallest distance between any pair of points in @xmath240 and @xmath241 . if either of these sets is empty , the distance is undefined . [ lemma : nested sets ] let @xmath3 be a probability distribution of finite support of diameter @xmath192 . let @xmath242 be a sequence of nested closed subsets of @xmath37 with the following properties : @xmath243 then there are constants @xmath18 and @xmath50 , which only depend on @xmath244 , such that @xmath245 and @xmath246 . by properties , and proposition [ prop : closure ] , any sequence of points which converges into the limit set of @xmath247 must contain points in @xmath248 . as the diameter of the support of @xmath3 is @xmath192 , property implies that any sample path which converges into @xmath247 must contain at least one point in @xmath249 . therefore , in order to find an upper bound for the probability a sample converges into @xmath247 , given that it converges into @xmath248 , we can condition on the location at which the sample path first hits @xmath250 . let @xmath251 be the distribution of first hitting times in @xmath250 for those sample paths which converge into @xmath247 , then , @xmath252 for all @xmath253 , there is an upper bound @xmath254 , by property . as @xmath248 is contained in @xmath255 , there is a lower bound @xmath256 , and by property , @xmath257 . therefore @xmath258 as @xmath259 . therefore @xmath245 , where we may choose @xmath260 . the measure @xmath42 is @xmath3-stationary , and so @xmath41-stationary for all @xmath6 , i.e. @xmath261 as all terms in the sum are positive , we may discard some of the terms and the sum will still be bounded above by the upper bound for @xmath262 , i.e. @xmath263 the measure @xmath264 is at least @xmath265 by , which implies @xmath266 and we may rewrite this as @xmath267 as @xmath268 this implies @xmath269 so @xmath246 , where @xmath270 . the constant @xmath50 only depends on @xmath244 , as @xmath155 only depends on @xmath244 . we wish to apply this lemma to shadows of points . we start by showing that as the harmonic measure @xmath42 is non - atomic , the harmonic measure of the shadows of points @xmath46 tends to zero as @xmath45 tends to infinity , uniformly in @xmath16 . [ prop : epsilon ] for any @xmath271 there is a constant @xmath272 , which depends on @xmath244 and @xmath3 , such that if @xmath273 then @xmath274 . suppose not , then there is an @xmath271 , and a sequence of shadows @xmath275 , with @xmath276 such that @xmath277 . let @xmath278 , and let @xmath279 , so @xmath173 consists of all points which lie in infinitely many @xmath45-shadows . the sets @xmath280 are decreasing , i.e. @xmath281 , and @xmath282 for all @xmath6 , so @xmath283 , and so in particular @xmath173 is non - empty . given @xmath284 , pass to a subsequence , which by abuse of notation we shall still refer to as @xmath285 , such that @xmath286 for all @xmath69 . let @xmath180 be any sequence of points with @xmath287 . by proposition [ prop : product bound ] , @xmath288 which tends to infinity as @xmath289 , which implies that @xmath290 . but this implies @xmath291 , which must have measure zero , as the measure @xmath42 is non - atomic , which contradicts the fact that @xmath292 . it will be convenient to choose @xmath259 , so from now on we will fix a value of @xmath272 which ensures that proposition [ prop : epsilon ] holds for some @xmath244 with @xmath259 . we now complete the proof of lemma [ lemma : exponential shadow ] by showing that a shadow @xmath46 has a nested family of sets @xmath293 satisfying lemma [ lemma : nested sets ] , where the number of sets is linear in @xmath45 . the constant @xmath157 in lemma [ lemma : nesting ] depends only on @xmath3 and @xmath44 , as does the choice of constant @xmath244 from proposition [ prop : epsilon ] , so the constants arising from lemma [ lemma : nested sets ] will depend only on @xmath3 and @xmath44 . [ lemma : nesting ] for any constant @xmath192 , there is constant @xmath157 , which depends on @xmath3 and @xmath44 , with the following properties . for any shadow @xmath294 , with @xmath295 , let @xmath296 be the largest integer such that @xmath297 . then the sets @xmath298 , for @xmath299 , form a sequence of nested sets , which contain @xmath46 , and which satisfy properties ( 15 ) from lemma [ lemma : nested sets ] above . let @xmath300 , where @xmath192 is the diameter of the support of @xmath3 , and the constants @xmath301 are the constants from lemmas [ lemma : nested shadows ] , [ lemma : change basepoint ] , [ lemma : shadow complement ] and proposition [ prop : epsilon ] respectively . we may assume that @xmath55 . the sets @xmath302 are nested , i.e. @xmath303 , by the definition of shadows , and @xmath304 for @xmath305 . we now check properties ( 15 ) from lemma [ lemma : nested sets ] . the gromov product @xmath306 . for all @xmath307 the gromov product @xmath308 , so @xmath309 . by property [ prop : gp].[prop : seq ] of the gromov product , for any sequence @xmath150 , @xmath310 . therefore , if @xmath311 , then for any sequence @xmath147 , all but finitely many points lie in @xmath298 , as @xmath312 . therefore @xmath313 , as required . two shadows which are sufficiently nested in terms of their shadow parameters , are also metrically nested in terms of the distance in @xmath2 , by lemma [ lemma : nested shadows ] . we shall apply lemma [ lemma : nested shadows ] , choosing the constant @xmath207 to be @xmath192 and the constant @xmath45 to be @xmath314 . recall that @xmath315 , where @xmath192 is the diameter of the support of @xmath3 , and @xmath206 is the constant from lemma [ lemma : nested shadows ] . this implies that @xmath316 for all @xmath317 , by our choice of @xmath296 . therefore lemma [ lemma : nested shadows ] implies that @xmath318 , so @xmath319 , as required . suppose that @xmath320 . we wish to show that that @xmath321 is contained in a shadow with basepoint @xmath43 , with a lower bound on the size of its @xmath45-parameter . this in turn will give an upper bound on the harmonic measure of the shadow . we may change the basepoint for the shadows using lemma [ lemma : change basepoint ] , so as long as @xmath322 , lemma [ lemma : change basepoint ] implies that the shadow @xmath46 is contained in @xmath323 , where @xmath324 as @xmath320 , this implies that @xmath325 . therefore choosing @xmath326 implies that @xmath327 , and as we have chosen @xmath328 , the conditions of lemma [ lemma : change basepoint ] are satisfied . therefore @xmath329 is an upper bound for @xmath330 . the harmonic measure @xmath329 is equal to @xmath331 , and this is at most @xmath259 as long as @xmath332 , by proposition [ prop : epsilon ] . we now verify this last inequality . by the definition of the gromov product , @xmath333 as @xmath334 , and @xmath335 this implies that @xmath336 . as we have chosen @xmath337 , this implies that @xmath338 , as required . suppose that @xmath339 . we wish to show that @xmath340 is contained in a shadow with basepoint @xmath43 , with a lower bound on the size of its @xmath45-parameter , which gives an upper bound on the harmonic measure of the shadow . we have chosen @xmath157 such that @xmath341 , and @xmath342 , so by lemma [ lemma : shadow complement ] , @xmath343 where @xmath344 . the argument is now essentially the same as in case ( 4 ) , except with @xmath28 and @xmath16 interchanged . let @xmath345 , so @xmath346 . as @xmath347 , we may apply lemma [ lemma : change basepoint ] , which implies that @xmath348 , where @xmath349 we now wish to use proposition [ prop : epsilon ] to find an upper bound for @xmath350 which is equal to @xmath351 . by thin triangles and the definition of the gromov product , @xmath352 , so @xmath353 which we may rewrite as @xmath354 where the left hand side is equal to @xmath25 . as we have chosen @xmath355 , this shows that @xmath356 . therefore proposition [ prop : epsilon ] implies that @xmath357 , so @xmath358 , as required . in this section we prove theorem [ theorem : exponential linear progress ] , i.e. we show that sample paths make linear progress at some rate @xmath157 , and furthermore , the proportion of sample paths at time @xmath6 which are distance at most @xmath359 from @xmath28 decays exponentially in @xmath6 . as @xmath360 is equal to @xmath361 the reflected random walk also makes linear progress at the same rate @xmath157 , and with the same exponential decay constant for the proportion of sample paths distance less than @xmath359 from the origin @xmath28 . a random walk of length @xmath63 , determined by a probability distribution @xmath3 , may be thought of as a random walk of length @xmath6 , determined by the probability distribution @xmath362 . we shall write @xmath59 for @xmath363 , and we shall call this the _ @xmath57-iterated random walk_. the steps of the @xmath57-iterated random walk are @xmath364 , and so for each @xmath69 , the segment of the random walk from @xmath365 to @xmath366 is independently and identically distributed according to the probability distribution @xmath362 , the @xmath57-fold convolution of @xmath3 . however , the distance from @xmath28 to @xmath366 is at most @xmath367 , but may be smaller , as the random walk may have `` backtracking , '' i.e. the geodesic from @xmath365 to @xmath366 may fellow travel with a terminal segment of the geodesic from @xmath28 to @xmath365 . this is illustrated schematically in figure [ pic : backtracks ] , for the first few steps of the @xmath57-iterated random walk . set @xmath368 to be the random variable corresponding to the change in distance from the basepoint @xmath28 from time @xmath369 to time @xmath69 of the @xmath57-iterated random walk , i.e. @xmath370 which may be negative . the sum of the first @xmath6 random variables @xmath368 is equal to the distance travelled at the @xmath6-th step of the @xmath57-iterated walk , i.e. @xmath371 we may write @xmath372 , where @xmath373 is the distance the @xmath57-iterated random walk travels between steps @xmath369 and @xmath69 , i.e. @xmath374 and @xmath375 . the @xmath373 form an independent collection of random variables , but the @xmath376 do not . by the definition of the gromov product , @xmath377 and we may think of @xmath376 as the amount of backtracking the iterated random walk @xmath59 does from step @xmath369 to step @xmath69 . in particular , the @xmath376 are non - negative . in order to find lower bound estimates for the sums of the @xmath378 , it suffices to find lower bound estimates for the sums of the @xmath379 , and upper bound estimates for the sums of the @xmath380 , and we now show how to do this , using standard results from the theory of concentration of measures . the distances @xmath381 form a sequence of independent , identically distributed random variables , so estimates on the behaviour of the sums of these random variables are well known . let @xmath382 be the expected value of @xmath373 , which depends on @xmath57 , but not on @xmath69 . as the trajectories of the random walk converge to the boundary almost surely , @xmath383 as @xmath384 . we will use the following bernstein or chernoff - hoeffding estimate , see for example dubhashi and panconesi @xcite*theorem 1.1 which says that the probability that the sum of @xmath6 copies of @xmath373 deviates from the expected mean @xmath385 by at least @xmath386 decays exponentially in @xmath6 . [ theorem : bernstein ] let @xmath379 be a sequence of bounded independent identically distributed random variables with mean @xmath387 . then for any @xmath271 there is a constant @xmath388 such that @xmath389 we now show a similar bound for the sums of the @xmath376 . we start by showing that the distribution functions of the @xmath376 are bounded above by the same exponential function , for all @xmath57 and @xmath69 . furthermore , the upper bound for @xmath376 holds independently of the values of @xmath390 for @xmath391 . as @xmath376 is a function of @xmath392 for @xmath393 , it suffices to show that the upper bound is independent of the values of @xmath392 for @xmath391 . [ prop : gp estimate ] there are constants @xmath50 and @xmath18 , which do not depend on @xmath57 or @xmath69 , such that @xmath394 by the definition of @xmath376 , if @xmath395 then @xmath396 . by the definition of shadows , this condition is equivalent to the condition @xmath397 , therefore @xmath398 we may apply the isometry @xmath72 , and use the fact that @xmath399 , to obtain , @xmath400 as @xmath80 is distributed as @xmath62 , and is independent of the @xmath392 for @xmath391 , this implies , @xmath401 now using lemma [ lemma : exponential shadow ] , there are constants @xmath231 and @xmath18 such that the bound @xmath402 , is independent of @xmath24 and @xmath57 , so this implies @xmath403 as required . in particular , this gives an upper bound for the expected value of @xmath376 which is independent of @xmath57 . therefore , by choosing @xmath57 to be large , we can make the expected value of @xmath373 much larger than the expected value of @xmath376 . we now show that there is a constant @xmath404 , which is independent of @xmath57 , such that the probability that the sum @xmath405 is larger than @xmath359 decays exponentially in @xmath6 . [ lemma : z estimate ] let @xmath59 be the @xmath57-iterated random walk of length @xmath6 , generated by a finitely supported probability distribution @xmath3 which generated a non - elementary subgroup of the mapping class group @xmath2 , and let @xmath406 . then there are constants @xmath407 and @xmath18 , which depend on @xmath3 but are independent of @xmath57 , such that @xmath408 for all @xmath6 . we have shown that the probability that @xmath395 decays exponentially in @xmath45 , with exponential decay constants which do not depend on either @xmath57 or @xmath69 , or the values of any other @xmath390 for @xmath409 . as @xmath376 is also independent of @xmath390 for @xmath410 , this implies that the exponential bounds for @xmath376 hold independently of the vales of @xmath390 for all @xmath411 . therefore , the probability distribution of the sum @xmath405 will be bounded above by a multiple @xmath412 of the @xmath6-fold convolution of the exponential distribution function with itself . we will use the following chernoff - hoeffding bound for sums of exponential random variables . the version stated below is an exercise from dubhashi and panconesi @xcite , but we provide a proof in appendix [ appendix ] for completeness . @xcite*problem 1.10 . let @xmath413 be independent identically distributed exponential random variables , with expected value @xmath207 . then for any @xmath414 , @xmath415 the upper bound for the sum of the @xmath376 will therefore be @xmath412 times the upper bound above , i.e. @xmath416 the expected value @xmath417 is bounded above for all @xmath57 , so by choosing @xmath418 sufficiently large , we may ensure that the base of the exponent on the right is strictly less that @xmath28 . this completes the proof of lemma [ lemma : z estimate ] . we now complete the proof of theorem [ theorem : exponential linear progress ] . recall that @xmath419 , and @xmath372 . so if @xmath420 then @xmath421 , or @xmath422 , though of course both conditions may be satisfied . furthermore , we may choose @xmath57 sufficiently large such that @xmath423 is positive . the probability that at least one of the events occurs is at most the sum of the probability that either occurs , so @xmath424 for some constants @xmath425 from theorem [ theorem : bernstein ] and @xmath426 from lemma [ lemma : z estimate ] , and this decays exponentially in @xmath6 , as required . this completes the proof of theorem [ theorem : exponential linear progress ] . in this section we prove theorem [ theorem : main ] . we start by showing that translation length of @xmath24 is coarsely equivalent to the length of the ( relative ) shortest element in the conjugacy class of @xmath24 , which we shall denote @xmath427 $ ] , i.e. @xmath428 = \inf_{h \in g } d(1 , hgh^{-1}).\ ] ] let @xmath2 be the mapping class group of a non - sporadic surface . there is a constant @xmath50 such that @xmath429 \right| } \le k$ ] . let @xmath24 be a conjugate of minimal relative length @xmath427 $ ] . by the definition of translation length , @xmath430 $ ] . we now show the bound in the other direction . there is a constant @xmath431 , which depends only on the surface , such that every non - element is conjugate to an element of relative length at most @xmath431 , see for example @xcite*lemma 5.5 so we shall choose @xmath432 and then we may assume that @xmath24 is . let @xmath433 be a quasi - axis for @xmath24 , i.e. a bi - infinite quasigeodesic such that @xmath433 and @xmath434 are @xmath435-fellow travellers for all @xmath6 . let @xmath436 be a closest point on @xmath433 to @xmath28 , then @xmath437 is distance at most @xmath438 from @xmath433 . this implies that the distance from @xmath436 to @xmath437 is at most @xmath439 , so @xmath440 is at most @xmath438 . therefore @xmath427 \le \t(g ) + k$ ] , as required . the mapping class group has _ relative conjugacy bounds _ , @xcite*theorem 3.1 , i.e. there is a constant @xmath50 , which only depends on the surface , such that if @xmath84 and @xmath85 are conjugate , then @xmath86 for some element @xmath26 with @xmath441 if @xmath24 is a group element , then we may think of @xmath24 as a point in the metric space @xmath131 . however , we can also represent @xmath24 by a choice of geodesic in @xmath2 from @xmath28 to @xmath24 . geodesics need not be unique , but any two distinct choices of geodesics are hausdorff distance at most @xmath435 apart . this gives two ways of representing a product @xmath437 of two group elements @xmath24 and @xmath436 . we may choose a single geodesic from @xmath28 to @xmath437 , or alternatively choose a path from @xmath28 to @xmath437 consisting of two geodesic segments , the first consisting of a geodesic from @xmath28 to @xmath24 , and the second consisting of a geodesic from @xmath24 to @xmath437 , which is the translate of a geodesic from @xmath28 to @xmath436 . therefore if @xmath24 is equal to @xmath27 , we can represent @xmath24 by a path composed of three geodesic segments , each consisting of a translate of @xmath26 , @xmath25 and @xmath29 respectively , and this is what we mean when we refer to the _ path _ @xmath27 . the fact that @xmath2 has relative conjugacy bounds implies that if an element @xmath24 is conjugate to a short element @xmath25 , and @xmath26 is a conjugating word of shortest possible relative length , then the path @xmath27 is quasigeodesic , where the quasigeodesic constants depend on @xmath442 , the constant of hyperbolicity @xmath44 , and the relative conjugacy bounds constant @xmath50 . @xcite*lemma 4.2 let @xmath2 be a weakly relatively hyperbolic group with relative conjugacy bounds . let @xmath24 be an element of @xmath2 which is conjugate to an element @xmath25 , i.e. @xmath443 , for some @xmath444 . if we choose @xmath26 to be a conjugating word of shortest relative length , then the word @xmath27 is quasi - geodesic in the relative metric , with quasi - geodesic constants which depend only on the relative length of @xmath25 , and the group constants @xmath44 and @xmath50 . [ prop : kpath ] for any constant @xmath195 there is a constant @xmath445 , which only depends on @xmath195 , the constant of hyperbolicity @xmath44 , and the relative conjugacy bounds constant , such that if @xmath24 is conjugate to an element @xmath25 of relative length at most @xmath195 , then @xmath446 for some @xmath26 with the following properties : 1 . @xmath447 2 . @xmath448 3 . @xmath449 let @xmath446 , where @xmath450 and @xmath26 is a conjugating element of shortest ( relative ) length . the first inequality follows from the triangle inequality , which implies that @xmath451 . as the path @xmath27 is a quasigeodesic , there is a constant @xmath157 , which only depends on @xmath195 , the constant of hyperbolicity @xmath44 , and the conjugacy bounds constant , such that the distance from @xmath26 to a geodesic from @xmath28 to @xmath24 is at most @xmath157 . this implies that if @xmath161 is the nearest point projection of @xmath26 to a geodesic from @xmath28 to @xmath24 , then @xmath452 . as any geodesic from @xmath26 to @xmath24 is contained in a @xmath165-neighbourhood of the nearest point projection path , consisting of a geodesic from @xmath26 to @xmath161 , and then from @xmath161 to @xmath24 , where @xmath165 only depends on @xmath44 . this implies that the distance from @xmath28 to any geodesic from @xmath26 to @xmath24 is at least @xmath453 . finally , as the gromov product @xmath454 is equal to the distance from @xmath28 to a geodesic from @xmath26 to @xmath24 , up to bounded additive error @xmath435 , this implies that @xmath455 , where @xmath456 , which only depends on the constant of hyperbolicity @xmath44 . this means that @xmath457 , and the same argument applied to the points @xmath458 and @xmath459 implies that @xmath460 , as @xmath461 , for the same constant @xmath50 . we may therefore choose @xmath445 to be the maximum of @xmath50 and @xmath462 . proposition [ prop : kpath ] above shows that the probability that a random walk @xmath5 is conjugate to an element of relative length at most @xmath195 , is bounded above by the probability that there is an element @xmath26 , with @xmath463 , such that @xmath464 , and @xmath465 . we shall write @xmath293 for the measure corresponding to the distribution of pairs @xmath30 on @xmath104 , i.e. @xmath466 for any subset @xmath99 . as @xmath467 decays exponentially , by theorem [ theorem : exponential linear progress ] , this gives the following upper bound for that the probability that @xmath5 is conjugate to an element of length at most @xmath195 , @xmath468 for some constant @xmath18 , and where @xmath103 is the diagonal in @xmath104 . therefore , in order to complete the proof of theorem [ theorem : main ] , it suffices to show : [ lemma : xdiagonal ] let @xmath157 be a constant such that @xmath469 decays exponentially in @xmath6 . then for any @xmath50 , there is a constant @xmath18 , which depends on @xmath50 and @xmath3 , such that @xmath470 the rest of this section is devoted to the proof of lemma [ lemma : xdiagonal ] . in fact , it will be convenient to obtain upper bounds for @xmath471 rather than @xmath293 . this suffices to obtain upper bounds for @xmath293 for all @xmath6 , as if @xmath192 is the diameter of the support of @xmath3 , then @xmath472 , by proposition [ prop : metric nest ] . we start by showing that it is very likely that a random walk @xmath110 lies in the shadow @xmath473 . [ lemma : walk in halfspace ] the probability that @xmath110 lies in @xmath473 tends to one exponentially quickly as @xmath6 tends to infinity , i.e. @xmath474 for some @xmath18 . we shall find an upper bound for the probability that @xmath110 does not lie in the shadow @xmath475 . conditioning on @xmath476 , and using the fact that the complement of the shadow @xmath475 is contained in @xmath477 , lemma [ lemma : shadow complement ] , gives @xmath478 which decays exponentially in @xmath6 , as required . applying this result to the reflected random walk implies that the probability that @xmath111 does not lie in @xmath479 also decays exponentially . we now use this to find an upper bound for @xmath471 in terms of @xmath116 . [ lemma : xmu estimate ] let @xmath195 be a subset of @xmath104 . there are constants @xmath55 and @xmath18 such that @xmath480 we have shown that the probability that each of the following four events occurs tends to one exponentially quickly . @xmath481 therefore the probability that all four of them occur tends to one exponentially quickly . if all four events occur , then @xmath482 , and similarly , @xmath483 . furthermore , if @xmath108 lies in @xmath484 , then there is a point @xmath485 such that @xmath486 and @xmath487 . this implies that @xmath488 , and @xmath489 , and so @xmath490 , as required . finally , we now show that the @xmath116-measure of a shadow of the diagonal @xmath491 decays exponentially in @xmath45 . [ lemma : diagonal ] there are constants @xmath425 and @xmath426 such that @xmath492 for all @xmath6 and @xmath45 . let @xmath493 and @xmath5 be random walks determined by @xmath3 and @xmath129 respectively . if @xmath494 , then there is a point @xmath16 such that @xmath495 and @xmath496 . therefore @xmath497 , and so @xmath498 . by the upper bound for measures of shadows , lemma [ lemma : exponential shadow ] , for any @xmath5 with @xmath499 , the probability that @xmath500 is at most @xmath501 , for some @xmath425 . furthermore , by theorem [ theorem : exponential linear progress ] , there is a @xmath502 such that the probability that @xmath503 is at most @xmath504 , for @xmath505 . therefore @xmath506 , as required . combining propositions [ lemma : xmu estimate ] and [ lemma : diagonal ] establishes lemma [ lemma : xdiagonal ] , and so completes the proof of theorem [ theorem : main ] . [ appendix ] in this section we provide the details for the following chernoff - hoeffding bound for exponential random variables . this proof is the solution given by dubhashi and panconesi to @xcite*problem 1.10 , which appeared in the initial draft version , but not in the final published version . let @xmath380 have probability density function @xmath509 , so the expected value of @xmath380 is @xmath510 , and set @xmath511 . consider the moment generating function @xmath512 for @xmath513 . therefore @xmath514 it now follows from markov s inequality that @xmath515 the right hand side above is minimized by choosing @xmath516 , which gives @xmath517 setting @xmath518 , and using the fact that @xmath510 , yields , @xmath519 as required .

we consider a system of partial differential equations describing the steady flow of a compressible heat conducting newtonian fluid in a three - dimensional channel with inflow and outflow part . we show the existence of a strong solution provided the data are close to a constant , but nontrivial flow with sufficiently large dissipation in the energy equation .

1211.7303

we investigate the stationary flow of a heat conducting compressible fluid in a cylindrical domain . the fluid is assumed to be newtonian . then the flow is described by the stationary navier - stokes - fourier ( nsf ) system : @xmath0 where @xmath1 $ ] with @xmath2 smooth , @xmath3 is the velocity field of the fluid , @xmath4 is the density , @xmath5 is the absolute temperature . we assume that the pressure @xmath6 is a twice continuously differentiable function on @xmath7 such that @xmath8 we consider the newtonian compressible fluid , i.e. the stress tensor has the form @xmath9 where the constant viscosities @xmath10 and @xmath11 fulfill @xmath12 and @xmath13 . further , the friction coefficient @xmath14 . finally , @xmath15 is the heat flux . note that we could treat the situation @xmath16 , @xmath17 and @xmath18 with suitable assumptions on these functions . it would only lead to further complications , therefore we take them rather constant not to hide the main ideas by too many technicalities . the boundary of the domain is divided in a natural way into the inflow part @xmath19 , the outflow part @xmath20 and the impermeable wall @xmath21 . more precisely , we define @xmath22 we consider a channel - like flow , i.e. we further assume that @xmath23 , @xmath24 and @xmath25 $ ] . we set @xmath26 on @xmath27 and @xmath28 to be a given positive constant on the wall @xmath21 . we set @xmath29 on @xmath21 but we admit @xmath30 on @xmath27 ( see fig.[rys1 ] ) . under such choice , on the inflow and outflow part the boundary condition for the temperature reduces to the inhomogeneous neumann condition . from the point of view of modeling these parts of the boundary are artificial and we can measure the parameters of the flow , what can be reflected in different values of @xmath31 . the condition on @xmath21 means that the flux of the temperature through the wall is proportional to the difference between the outside temperature @xmath32 and the temperature of the fluid . the level of thermal insulation of the wall is given by @xmath28 . existence and uniqueness of the special solution @xmath33 under our choice of boundary conditions is discussed in the next chapter . the total energy is given by @xmath34 , where the internal energy @xmath35 is given by @xmath36 with @xmath37 i.e. the internal energy fulfills the maxwell relation . without loss of generality we set in @xmath38 . equivalently we can consider system ( [ main_system ] ) with total energy balance ( [ main_system])@xmath39 replaced with the internal energy balance @xmath40 we also set @xmath41 * known results * the existence theory for the steady compressible navier stokes equations in the framework of strong solutions was intensively studied in eighties , see e.g. @xcite , @xcite and many other papers . in the beginning of this century , inspired by the results of p.l . lions ( see @xcite ) , the attention turned rather towards the weak solutions , i.e. towards the existence theory without any smallness assumption on the data . the best results in this direction so far can be found in @xcite ( slip boundary conditions ) and @xcite ( space periodic situation ) . note , however , that the theory is not able to treat the situation with nonzero inflow / outflow conditions . strong solutions with inhomogeneous boundary data , which are in our scope of interest in this paper , have been considered for the first time in @xcite in hilbert spaces and later in @xmath42 framework ( @xcite , @xcite , @xcite ) . in @xcite a constant flow in the direction of the axis of the cylinder is investigated and existence of a strong solutions in its vicinity is shown in the barotropic case . concerning the nsf system , the existence of weak solutions was studied much later , see @xcite , @xcite . the best result in this direction can be found in @xcite . note , however , that similarly to the barotropic case , all these results do not allow to treat the inflow / outflow problems . the theory of strong solutions for the nsf system describing the thermal effects is much less developed , to our knowledge the only result in @xmath42 framework is the paper by beirao da veiga @xcite where existence of strong solutions in a vicinity of a constant flow with zero velocity is shown . * special solution . * here we are interested in showing existence of strong solutions , i.e. we are looking for @xmath43 , solving system ( [ main_system ] ) so that the solution is close to a given special solution . hence we arrive at a problem with some smallness of the data . even though smallness obviously facilitates the analysis , the results of this kind are still missing in the theory of compressible flow , especially for heat - conducting fluids ( see the overview above ) . this is mostly due to high complexity of the system ( [ main_system])@xmath44 , in particular due to its mixed , elliptic ( or parabolic in time dependent case ) and hyperbolic character . investigation of stationary solutions close to given laminar flows is of particular importance if we want to investigate stability of special solutions , which is quite well developed for weak solutions but so far almost not investigated in strong solutions framework ( see @xcite for an overview of results ) . the simplest example of a special solution to perturb is a solution with zero velocity and constant density and temperature . such flow has been investigated in @xcite with homogeneous dirichlet boundary conditions . from the point of view of applications it is important to investigate flows with large velocity , which leads in a natural way to inhomogeneous boundary data . in a cylindrical domain a natural example of such solution is a constant flow in the direction of the axis of the cylinder : @xmath45 , \quad \bar \rho = 1.\ ] ] in other words , we can state that such a flow is natural to be investigated as a special solution to the system ( [ main_system ] ) in our domain @xmath46 as for the temperature of the basic solution the simplest choice would be to assume it constant . it seems however quite artificial with the flow @xmath47 defined above . it is more interesting from physical point of view to consider a flow with nontrivial temperature function @xmath33 which corresponds to the solution @xmath47 as a solution to corresponding energy balance equation . on the boundary we assume the temperature of the form @xmath48 with constant , possibly large @xmath49 and small variation @xmath50 . substituting @xmath47 to the internal energy balance equation we see that @xmath33 should satisfy the system @xmath51 now we write @xmath52 , where @xmath53 solves @xmath54 \kappa \frac{\partial \overline { \theta}_0}{\partial n}|_{\gamma } = g & \mbox { on } \gamma,\\[2pt ] \frac{1}{|\omega|}\int_{\omega } \overline { \theta}_0 \,dx = t_0 , \end{array}\ ] ] and @xmath55 the existence of a solution @xmath56 to ( [ eqn_tilde_theta ] ) is a classical elliptic result , we only need to apply symmetry to deal with corner singularities . the details are given in lemma [ lem_neumann ] . in particular @xmath53 satisfies @xmath57 \leq c \|g\|_{w^{1 - 1/p}_p(\gamma)}.\ ] ] now we see that @xmath58 satisfies @xmath59 the existence and uniqueness of solution to the above system is discussed in subsection [ sec_theta1 ] . our goal is to show the existence of a solution to ( [ main_system ] ) close to @xmath60 . hence it is convenient to introduce the following quantity to measure the distance of the data from the special solution : @xmath61 we are now in a position to formulate our main result . [ main_thm ] assume that @xmath62 defined in ( [ d0 ] ) is small enough , @xmath63 is large enough , @xmath28 is large enough on @xmath21 , @xmath64 large enough on @xmath19 and @xmath65 . then there exists a solution @xmath66 to system ( [ main_system ] ) such that @xmath67 this solution is unique in the class of solutions satisfying ( [ est_main ] ) . * notation . * we use standard notation for the sobolev spaces , i.e. @xmath68 . for the spaces defined on @xmath46 we will skip the domain ; for example we write @xmath69 instead of @xmath70 . a generic constant that is controlled ( but not necessarily small ) will be denoted by @xmath71 , whereas @xmath72 will denote a small constant ( i.e. which can be arbitrarily small for the data small enough ) . we will also use the space @xmath73 , for simplicity we denote it by @xmath74 . in this section we collect some standard results which we use throughout the paper . we start with the sobolev imbedding theorem , let us state here the particular case which is used throughout the paper ( the general version with the proof can be found in ( * ? ? ? * theorem 5.4 ) ) let @xmath46 be defined as above and let @xmath75 with @xmath65 . then @xmath76 and there exist @xmath77 s.t . @xmath78 next we recall the poincar inequality . among its different versions we will need the following two . let @xmath79 and @xmath80 denote two nontrivial parts of the boundary @xmath81 and @xmath82 . then @xmath83\ ] ] and @xmath84,\ ] ] where @xmath85 . ( interpolation inequality ) : [ lem_int ] + let @xmath86 . then for all @xmath87 such that @xmath88 : @xmath89 the interpolation inequality in the lebesgue spaces and the imbedding @xmath90 with @xmath91 for @xmath65 , @xmath92 , arbitrary for @xmath93 and @xmath94 for @xmath95 yield @xmath96 now application of the young inequality leads to ( [ int1 ] ) . another result we use is the following version of the korn inequality [ lem_korn ] let @xmath97 , @xmath12 and if @xmath98 then @xmath46 is not axially symmetric . then @xmath99 it is a straightforward generalization of ( * ? ? ? * lemma 4 ) . we also apply the following well - known result in finite dimensional hilbert spaces ( the proof can be found in @xcite ) : [ lem_p ] let @xmath100 be a finite dimensional hilbert space and let @xmath101 : @xmath102 be a continuous operator satisfying @xmath103 then @xmath104 our proofs will be based on the @xmath42 regularity of elliptic problems with neumann and slip boundary conditions . these are classical results , in case of cylindrical domain we only have to deal with the singularities of the boundary , for the sake of completeness we give the proofs . the first result concerns the neumann problem . [ lem_neumann ] let @xmath105 , @xmath106 , @xmath107 and @xmath108 . let @xmath109 . then there exists a solution @xmath110 to the problem @xmath111 \frac{\partial u}{\partial n } = d & \mbox { on } & \gamma , \\[2pt ] \frac{1}{|\omega|}\int_\omega u \ , dx & = c_0 , \end{array}\ ] ] and @xmath112.\ ] ] moreover , this solution is unique in the given regularity class . in fact , the only difficulty we have to overcome are the singularities of the boundary . if @xmath113 , then we can simply extend the weak solution using symmetric reflection that preserves the homogeneous neumann condition . in the neighbourhood of @xmath19 it will be defined as @xmath114 where @xmath115 and @xmath116 . for @xmath117 we obtain a neumann problem in a domain with smooth boundary that we solve using classical elliptic theory . if @xmath118 does not vanish on @xmath27 , we reduce the problem to the previous case . let us consider @xmath19 , on @xmath20 we proceed the same way . introducing an extension @xmath119 such that @xmath120 and @xmath121 , we consider @xmath122 that satisfies @xmath123 where @xmath124 $ ] . the second result concerns the lam system with slip boundary conditions : @xmath125 the following lemma gives existence of a solution and the maximal regularity elliptic estimate that we apply in our proofs . [ lem_lame ] under the assumptions of theorem [ main_thm ] there exists a unique @xmath110 solving ( [ lam ] ) with @xmath126.\ ] ] under the assumptions on @xmath10 and @xmath11 ( [ lam ] ) is elliptic so we easily get a weak solution . the only problem we encounter showing regularity of the weak solution are the singularities of the boundary on the junctions of @xmath21 with @xmath19 and @xmath20 . these singularities can again be dealt using the symmetry arguments . hence we define the operator @xmath127 extending a vector field defined for @xmath128 on the whole space as @xmath129 where @xmath130 is as above and @xmath131 $ ] . then we have @xmath132 and on the plane @xmath133 the extension @xmath127 preserves the slip boundary conditions , i.e. @xmath134 and @xmath135 . now we can show higher regularity of @xmath136 , extension of the weak solution to ( [ lam ] ) , and since @xmath127 preserves the boundary conditions we conclude that @xmath137 is a strong solution to ( [ lam ] ) satisfying the boundary conditions on @xmath19 . application of analogous antisymmetric extension on @xmath20 completes the proof . our next aim is to construct solutions to . recall that @xmath138 where @xmath139 solves . therefore by lemma [ lem_neumann ] we have @xmath140 as well as by the lax - milgram theorem @xmath141 next for @xmath142 we have problem ( [ eqn_theta1 ] ) . our aim is to show existence of a weak solution to this problem as well as its regularity . as the problem is linear , the existence and uniqueness of a weak solution is direct consequence of corresponding a priori estimates . first we show estimates in @xmath143 , then in @xmath144 . since we will require @xmath63 and @xmath28 large , we have to control how all constants depend on them . for the sake of simplicity we assume that @xmath63 and @xmath28 are approximately of the same size . multiplying ( [ eqn_theta1 ] ) by @xmath145 and integrating over @xmath46 we get @xmath146 all terms on the l.h.s . are nonnegative . applying the poincar inequality to the first term on the r.h.s the young inequality to the other two ones we get @xmath147 therefore we get , for @xmath63 and @xmath28 sufficiently large with respect to @xmath148 , @xmath149 hence , using also @xmath150 where @xmath71 is a fixed constant independent of @xmath63 and @xmath28 , provided @xmath151 . uniqueness is a direct consequence of ; if the data are zero , then the solution is zero as well . finally , proceeding similarly as above we will deduce the existence of strong solutions together with the estimates of the corresponding norms . first we rewrite ( [ eqn_theta1 ] ) to the form @xmath152 \kappa \frac{\partial \overline{\theta}_1}{\partial n } = -l(\overline{\theta}_1 + \overline{\theta}_0 - t_0 -t_1 ) & \mbox { on } \gamma , \end{array}\ ] ] and using lemma [ lem_neumann ] @xmath153 now , applying the poincar inequality in the form @xmath154 and using we end up with [ l7 ] let @xmath155 , @xmath156 , @xmath65 and let @xmath63 , @xmath28 be sufficiently large . then there exist a unique weak solution @xmath145 to ( [ eqn_theta1 ] ) such that @xmath157.\ ] ] moreover , the solution is strong , i.e. @xmath158 and @xmath159.\ ] ] here , the constants @xmath71 are independent of @xmath63 and @xmath28 . as we are interested in small perturbations of @xmath60 , it is natural to introduce the perturbations as unknown functions . for technical reasons it is better to get rid of inhomogeneity on the boundary . hence we construct @xmath160 such that @xmath161 . more precisely , we set @xmath162 , where @xmath163 solves @xmath164 next we introduce the perturbations @xmath165 replacing the total energy balance ( [ main_system])@xmath39 with the internal energy balance ( [ int_en_bal ] ) we can rewrite ( [ main_system ] ) as @xmath166 where @xmath167 g(u,\sigma ) = -(\sigma + 1 ) \ , { \rm div } \ , u_0 - \sigma \ , { \rm div } \,u , \\ [ 5pt ] h(u,\sigma,\eta ) = { \bf s}(\nabla(u+u_0 ) ) \cdot \nabla ( u+u_0 ) - t_0 \partial_\rho \pi(1+\sigma , \eta + \overline{\theta}_0 + \overline{\theta}_1 ) { \rm div}\,u_0 \\ -(1+\sigma ) ( u+u_0 ) \cdot \nabla ( \overline{\theta}_0 + \overline{\theta}_1 + \eta ) + t_0 { \rm div } \,u \big(p_2- \partial_\theta \pi ( 1+\sigma , \overline{\theta}_0 + \overline{\theta_1 } + \eta)\big ) \\ -(\eta + ( \overline{\theta}_0 -t_0 ) + \overline{\theta}_1 ) \partial_\theta \pi(1+\sigma , \overline{\theta}_0 + \overline{\theta_1 } + \eta ) { \rm div}\ , ( u+u_0 ) \\ + \partial_{x_1 } ( \overline{\theta}_0 + \overline{\theta}_1 + \eta ) \big(e_2- \partial_\theta e ( 1+\sigma , \overline{\theta}_0 + \overline{\theta_1 } + \eta)\big ) , \\[5pt ] b_i = b_i - \alpha \tau_i^{(1 ) } - ( { \bf s}(\nabla u_0 ) n ) \cdot \tau_i -\alpha u_0 \cdot \tau_i . \end{array}\ ] ] the above expressions result directly from substitution of ( [ pert ] ) to ( [ main_system ] ) . it is only worth to notice that to derive the expression for @xmath168 we have applied the continuity equation which yields @xmath169 now in ( [ main_system])@xmath39 the first term on the r.h.s . of ( [ div_id ] ) can be combined with the @xmath170 , what finally gives the above form of @xmath171 . let us state at this early stage the following estimates on the quantities @xmath172 defined in ( [ fgh ] ) : @xmath173 , \\ \\ \smallskip \|g(u,\sigma)\|_{w^1_p } & \leq c \big[(\|u\|_{w^2_p } + \|\sigma\|_{w^1_p})^2 + d_0(1+d_0)\big],\\ \\ \smallskip \|h(u,\sigma,\eta)\|_{l_p } & \leq c \big [ ( \|u\|_{w^2_p } + \|\sigma\|_{w^1_p } + \|\eta\|_{w^2_p})^2 \\ & + ( \|u\|_{w^2_p } + \|\sigma\|_{w^1_p}+ \|\eta\|_{w^2_p})^3 + d_0(1+d_0)\big ] , \\ \end{array}\ ] ] where @xmath62 is defined in ( [ d0 ] ) . the bounds ( [ est_fgh ] ) are straightforward , it is enough to observe that @xmath174 and @xmath168 are composed of terms which are either quadratic or cubic with respect to the functions @xmath175 or linear with respect to them multiplied by something which can be bounded by @xmath62 . note that in the estimate of terms contained differences of the derivatives of @xmath176 or @xmath35 we use the fact fact that these functions are twice continuously differentiable and hence we may use the taylor expansion theorem . we sketch briefly the idea of the proof of theorem 1 , and hence the structure of the remainder of the paper . in subsection 2.4 we explain how to deal with the linearization of the continuity equation , i.e. equation @xmath177 . to show existence of solutions we can apply a method of successive approximations and look for the solution as a limit of a sequence @xmath178 in section 3 we deal with the linear system ( [ system_lin ] ) showing the a priori estimates . the first step is the energy estimate which is derived in quite a standard way . the energy estimate is then applied to show the maximal regularity @xmath179 estimates , hence estimates in the space where we look for the solution ( lemma [ lem_est_lin ] ) . it can be considered as the main result of section 3 . in the end of section 3 we solve the linear system , hence we show that the sequence ( [ system_seq ] ) is well defined . in section 4 we deal with the sequence of approximations . using the estimates for the linear system combined with ( [ fgh ] ) we show it is bounded in @xmath180 and has a contraction property in a weaker space @xmath181 . an analogous estimate gives uniqueness in the class of solutions satisfying ( [ est_main ] ) . in other words we apply a generalization of the banach fixed point theorem to show the convergence of the sequence of approximations to the solution of ( [ main_system ] ) . in this section we show the well posedness of the operator @xmath182 defined as @xmath183 where @xmath184 is a weak solution to in the sense as introduced in . the solvability of the above problem , to which we refer as a steady transport equation , will be needed to show the energy estimate for the density in section 3 and to solve the continuity equation in section 4 . the result is given in the following [ lem_s ] let @xmath185 and @xmath186 . let @xmath187 be such that @xmath188 is small enough and @xmath189 . then there exists a unique solution @xmath190 and @xmath191.\ ] ] the proof follows @xcite . for the sake of completeness we recall here the most important steps , referring to @xcite for the missing details . the idea is to get rid of the term @xmath192 introducing a change of variables @xmath193 satisfying the identity @xmath194 we construct the mapping @xmath195 in the following [ lem_change ] let @xmath188 be small enough . then there exists a set @xmath196 and a diffeomorphism @xmath197 defined on @xmath198 such that @xmath199 and ( [ change_id ] ) holds . moreover , if @xmath200 and @xmath201 then @xmath202 , where @xmath203 is the outward normal to @xmath198 . before we sketch the proof let us make one remark . the last condition states that the first component of the normal to @xmath204 vanishes , but since @xmath195 is defined only on @xmath198 we formulate this condition using the limits . it means simply that the image @xmath205 is also a cylinder with a flat wall . _ sketch of the proof of lemma [ lem_change ] . _ the identity ( [ change_id ] ) means that @xmath195 must satisfy @xmath206 a natural condition is that @xmath207 . thus we can search for @xmath208 , where for all @xmath209 such that @xmath210 the function @xmath211 is a solution to a system of ode : @xmath212 using the regularity and smallness of @xmath213 we show existence for ( [ ode ] ) , hence we have @xmath195 defined on some @xmath198 such that @xmath199 . the smallness of @xmath213 implies also that @xmath214 , where @xmath215 is small and we conclude that @xmath195 is a diffeomorphism . let us denote @xmath216 . now it is natural to define the subsets of @xmath217 as @xmath218 where @xmath219 , @xmath220 and @xmath221 . to show that @xmath222 for @xmath223 it is enough to observe that @xmath224 ) = [ 1+u^1(x),u^2(x),u^3(x)],\ ] ] where @xmath197 . but for @xmath225 the vector on the r.h.s is tangent to @xmath21 since @xmath226 . we can conclude that on @xmath227 the image in @xmath195 of a straight line @xmath228 is a curve tangent to @xmath21 , and thus @xmath227 is a sum of such lines and so we have @xmath222 . the proof of lemma [ lem_change ] is completed . @xmath229 _ proof of lemma [ lem_s ] . _ first we define @xmath230 for a continuous function @xmath231 as @xmath232 the condition @xmath233 on @xmath234 guarantees that a straight line @xmath235 has a picture in @xmath46 and thus we integrate along a curve contained in @xmath46 . it means that @xmath236 is well defined for continuous functions defined on @xmath46 and the construction of @xmath195 clearly ensures that @xmath236 satisfies ( [ def_s ] ) . next we have to extend @xmath236 on @xmath70 . to this end we show the estimate ( [ est_s ] ) for continuous @xmath231 . let @xmath237 denote an @xmath238-cut of @xmath46 and let @xmath239 . then by ( [ s ] ) we have @xmath240 ^ 2 \ , d \bar x \\ \leq 2 \|w_{in}\|_{l_2(\gamma_{in})}^2 + c \int_{\omega_{x_1 } } \int_0^{\phi_1(x ) } h^2(\psi(s,\phi_2(x),\phi_3(x ) ) ) \,ds \,d\bar x \leq c \ , \big [ \|w_{in}\|_{l_2(\gamma_{in})}^2 + \|h\|_{l_2(\omega)}^2 \big ] . \end{array}\ ] ] the above estimate holds for every @xmath241 what implies ( [ est_s ] ) . now we can define @xmath242 for @xmath243 using a standard density argument , the details are given in ( @xcite ) . to show the uniqueness of the solution we can rewrite the r.h.s of ( [ def_s ] ) as @xmath244 where the equation is to be understood in the sense of distribution , and , treating @xmath238 as a `` time '' variable , adapt di perna - lions theory of transport equation ( @xcite ) that implies the uniqueness of solution to ( [ def_s_1a ] ) in the class @xmath245 . the proof is thus complete . in this section we deal with the linear system @xmath246 with given functions @xmath247 such that @xmath248 note that this system actually represents , where we only redefined a few constants . the main difficulty lies in deriving appropriate estimates for ( [ system_lin ] ) . we start with the energy estimates , then applying the properties of slip boundary conditions , helmholtz decomposition of the velocity and classical elliptic theory we derive the @xmath42 estimates . apart from solving the system ( [ system_lin ] ) , the estimates derived in this section are used later , combined with ( [ est_fgh ] ) , to show the convergence of the sequence of approximations ( [ system_seq ] ) . [ l 8 ] let @xmath249 solve system ( [ system_lin ] ) with @xmath250 with @xmath188 small enough , @xmath251 on @xmath252 fulfilling conditions above , where @xmath253 is the dual space to @xmath254 assume further that @xmath255 and @xmath64 fulfill the assumptions of theorem [ main_thm ] . then the following estimate holds : @xmath256 . \end{array}\ ] ] we multiply ( [ system_lin])@xmath257 by @xmath137 and integrate applying the following identity : @xmath258 with application of the korn inequality and the boundary condition ( [ system_lin])@xmath259 we get @xmath260 next we divide ( [ system_lin])@xmath39 by @xmath261 ( recall it is constant ) , multiply by @xmath262 and integrate . it yields @xmath263 hence @xmath264 to get rid of the boundary term on the r.h.s we can apply the poincar inequality ( [ poin1 ] ) and rewrite ( [ ene2 ] ) as @xmath265 where @xmath266 is the constant from ( [ poin1 ] ) . for @xmath63 large enough and @xmath28 large enough on @xmath21 the first two terms on the l.h.s . will be positive . now we can combine ( [ ene1 ] ) and ( [ ene2 ] ) obtaining @xmath267 \leq \int_{\omega } h\eta \,dx + \int_{\omega } f \cdot u \,dx \\ + \int_{\omega } g\sigma \,dx + \frac{1}{2 } \int_{\omega } \sigma^2 { \rm div } \ , u \,dx + \frac{1}{2 } \int_{\gamma_{in } } \sigma_{in}^2 ( 1-u\cdot n ) \,ds + \sum_{i=1}^2 \int_\gamma b_i u \cdot \tau_i \ , ds . \end{array}\ ] ] using ( [ poin2 ] ) , the imbedding @xmath268 and hlder s inequality we derive @xmath269 recall that e denotes a small constant and so to complete the proof we have to find the bound on @xmath270 . but this follows from the existence result in the previous subsection , see . hence we have @xmath271 combining ( [ ene3 ] ) and ( [ ene4 ] ) we conclude ( [ ene ] ) . the main result of this section is [ lem_est_lin ] let @xmath272 solve system ( [ system_lin ] ) with @xmath273 , where @xmath188 is small enough and @xmath274 satisfy the assumptions of theorem [ main_thm ] . then @xmath275 : = d_{lin}. \end{array}\ ] ] the proof of ( [ est_lin ] ) will be performed in several consecutive lemmas . we show the bound on @xmath276 . then we proceed with the bound on @xmath277 which is the most demanding . first we show bound on the vorticity which , together with helmholtz decomposition of the velocity , makes possible to eliminate the term @xmath278 from the continuity equation , leading to ( [ trans ] ) . using this equation we show the bound on the density . then ( [ est_lin ] ) easily follows from the classical elliptic estimate on @xmath279 . we start with the bound on @xmath262 ; to show it let us rewrite ( [ system_lin])@xmath280 as @xmath281 the classical elliptic estimate ( @xcite , @xcite ) for the above system yields @xmath282.\ ] ] applying the trace theorem to the boundary term and then the interpolation inequality ( [ int1 ] ) we get @xmath283 + \epsilon \|u\|_{w^2_p}.\ ] ] by the energy estimate ( [ ene ] ) we conclude @xmath284 now we proceed towards the bound on @xmath277 . as mentioned before , we start with the estimate on the vorticity . taking the curl of ( [ system_lin])@xmath257 we get @xmath285 & \mbox{in } & \omega , \\ \alpha \cdot \tau_2 = ( 2 \chi_1 - \frac{\alpha}{\mu } ) u \cdot \tau_1 + \frac{b_1}{\mu } & \mbox{on } & \gamma , \\ \alpha \cdot \tau_1 = ( \frac{\alpha}{\mu } - 2 \chi_2 ) u \cdot \tau_2 - \frac{b_2}{\mu } & \mbox{on } & \gamma , \\ { \rm div}\,\alpha = 0 & \mbox{on } & \gamma , \end{array}\ ] ] where @xmath286 denote the curvatures of the curves generated by tangent vectors @xmath287 . in order to show the boundary relations ( [ system_rot])@xmath288 it is enough to differentiate ( [ system_lin])@xmath259 with respect to the tangential directions and apply ( [ system_lin])@xmath39 . a rigorous proof is given in @xcite or @xcite . the condition @xmath289 in @xmath46 results simply from the fact that @xmath290 . we introduce this relation as a boundary condition ( [ system_rot])@xmath259 , that completes the conditions on the tangential parts of the vorticity . note that the boundary conditions ( [ system_rot])@xmath288 give the tangential parts of the vorticity on the boundary the regularity of the velocity itself and the data . we will now use this feature of slip boundary conditions to show the higher estimate on the vorticity ( see @xcite,@xcite , @xcite ) . for the above system we have ( see @xcite , theorem 10.4 ) : @xmath291.\ ] ] applying the interpolation inequality ( [ int1 ] ) and then ( [ ene ] ) we arrive at @xmath292 for any @xmath293 , where @xmath294 is defined in ( [ est_lin ] ) . now we introduce the helmholtz decomposition of the velocity ( see e.g. @xcite ) @xmath295 where @xmath296 and @xmath297 . we see that the field @xmath298 satisfies the following system @xmath299 for this system we have ( see @xcite ) : @xmath300 what by ( [ alpha_w1p_1 ] ) can be rewritten as @xmath301 + \epsilon \|u\|_{w^2_p}\ ] ] for any @xmath293 . now we substitute the helmholtz decomposition to ( [ system_lin])@xmath257 . we get @xmath302 = f - \partial_{x_1 } a + \mu \delta a - \partial_{x_1 } \nabla \phi - p_2\nabla \eta,\ ] ] but @xmath303 . we denote @xmath304 combining the last equation with ( [ system_lin])@xmath305 we arrive at @xmath306 where @xmath307 and @xmath308 equation ( [ trans ] ) makes possible to estimate the @xmath309-norm of the density in terms of @xmath309 - norm of @xmath310 , which in turn will be controlled by ( [ nablak ] ) using interpolation and the energy estimate ( [ ene1 ] ) . first we estimate @xmath277 in terms of @xmath310 . the result is stated in the following lemma assume that @xmath311 satisfies equation ( [ trans ] ) with @xmath312 . then @xmath313.\ ] ] in order to find a bound on @xmath314 we multiply ( [ trans ] ) by @xmath315 and integrate over @xmath46 . integrating by parts and next using the boundary conditions we get @xmath316 with the smallness of @xmath317 , the above implies @xmath318 and so @xmath319 .\ ] ] now we estimate the derivatives . in order to find a bound on @xmath320 we differentiate ( [ trans ] ) with respect to @xmath321 . note that @xmath322 is not defined in general , however , for @xmath323\ ] ] and @xmath324 we may set @xmath325 . hence we can differentiate ( [ trans ] ) with respect to @xmath321 , multiply by @xmath326 and integrate . since @xmath327 , we have @xmath328 next , since @xmath329 , we can write @xmath330 = - \frac{1}{p } \int_{\omega } |\partial_{x_i}\sigma|^p \ , { \rm div}\ , \tilde u \,dx + \frac{1}{p } \int_{\gamma } |\partial_{x_i}\sigma|^p \ , \tilde u \cdot n \ , ds \\[5pt ] = - \frac{1}{p } \int_{\omega } |\partial_{x_i}\sigma|^p \ , { \rm div}\ , u \,dx - \frac{1}{p } \int_{\gamma_{in } } |\partial_{x_i}\sigma_{in}|^p \ , ( 1+u^1 ) \,ds + \frac{1}{p } \int_{\gamma_{out } } |\partial_{x_i}\sigma|^p \ , ( 1+u^1 ) \,ds . \end{array}\ ] ] for @xmath331 we have @xmath332 and hence the above defines the trace of @xmath333 on @xmath20 . we arrive at @xmath334 for @xmath331 ( [ wxi ] ) gives straightforward bound on @xmath335 . in order to estimate @xmath336 we also differentiate ( [ trans ] ) with respect to @xmath238 and multiply by @xmath337 . the difference is that @xmath336 is not given on @xmath19 . to overcome this difficulty we can observe that on @xmath19 equation ( [ trans ] ) reduces to @xmath338 \,\partial_{x_1}\sigma = k,\ ] ] what can be rewritten as @xmath339,\ ] ] where the lower index @xmath340 denotes the tangential component . thus we have @xmath341.\ ] ] using this bound in ( [ wxi ] ) , @xmath342 , we arrive at the estimate @xmath343.\ ] ] applying the trace theorem to the term @xmath344 and then combining ( [ wxi ] ) ( for @xmath345 and @xmath346 ) with ( [ wx1 ] ) we get @xmath347.\ ] ] the term @xmath348 can be put on the l.h.s . due to the smallness assumption and thus by young s inequality @xmath349,\ ] ] what combined with ( [ w ] ) yields @xmath350.\ ] ] the lemma is proved . now we estimate @xmath310 in terms of the data . the result is [ lemk ] let @xmath310 be defined in ( [ k ] ) . then @xmath351 we have @xmath352 where @xmath294 is defined in ( [ est_lin ] ) . applying first ( [ int1 ] ) and then ( [ ene1 ] ) we get @xmath353.\ ] ] hence , it is enough to find the bound on @xmath354 . by ( [ nablak ] ) we have @xmath355,\ ] ] where @xmath356 is the helmholtz decomposition of the velocity . applying ( [ aw2p ] ) , ( [ eta_w2p ] ) , trace theorem and ( [ int1 ] ) we arrive at @xmath357 \\ + \delta_1 \|u\|_{w^2_p } + c(\delta_1 ) \big[\|f\|_{l_2 } + \|g\|_{l_2 } + \|b\|_{l_2(\gamma ) } \big ] . \end{array}\ ] ] combining ( [ klp ] ) and ( [ nablahlp ] ) we get ( [ lemk_teza ] ) . the bound on @xmath311 now follows directly . substituting ( [ lemk_teza ] ) to ( [ w_w1p ] ) we get @xmath358 where @xmath294 is given by ( [ est_lin ] ) . now the only missing piece to complete the proof of ( [ est_lin ] ) is the bound on @xmath279 . to show it note that the velocity satisfies the lam system @xmath359 the classical theory of elliptic equations ( @xcite,@xcite ) yields @xmath360.\ ] ] applying ( [ int1 ] ) to the term @xmath361 and then ( [ ene1 ] ) we get @xmath362 _ proof of lemma [ lem_est_lin ] . _ we combine ( [ eta_w2p ] ) , ( [ u_w2p ] ) and ( [ w_w1p_ost ] ) . in ( [ w_w1p_ost ] ) we choose for example @xmath363 where @xmath71 is the constant from ( [ u_w2p ] ) . now ( [ est_lin ] ) follows immediately . in this section we solve the linear system ( [ system_lin ] ) and hence show that the sequence ( [ system_seq ] ) is well defined . we start with showing the existence of an appropriately defined weak solution applying the galerkin method modified in a way to deal with the continuity equation . then we show the regularity of the solutions for the given regularity of the data . for simplicity let us denote @xmath364.\ ] ] in order to define a weak solution to ( [ system_lin ] ) we recall the definition of the space @xmath365 ( [ def_v ] ) . a natural definition of a weak solution to the system ( [ system_lin ] ) is @xmath366 such that * 1 . * the identity @xmath367 is satisfied @xmath368 ; * 2 . * ( [ system_lin])@xmath305 is satisfied in the following sense : @xmath369 @xmath370 , where @xmath371 is defined in ( [ tilde_u ] ) , and * 3 . * the equation @xmath372 holds for all @xmath373 . we introduce an orthonormal basis of @xmath365 formed by @xmath374 . we consider finite dimensional spaces : . the sequence of approximations to the velocity will be searched for as @xmath375 due to the equation ( [ system_lin])@xmath305 we have to define the approximations to the density in an appropriate way . namely , we set @xmath376 , where @xmath377 is defined in ( [ def_s ] ) , with @xmath378 . finally , to define the approximations of the temperature we introduce @xmath379 as a weak solution operator to equation ( [ system_lin])@xmath280 , i.e. we set @xmath380 the well - posedness of the operator @xmath381 is direct as it is just a solution operator for a standard elliptic equation . to show it is well defined it is enough to recall the estimate from the proof of ( [ ene ] ) . in the same way we show the estimate for ( [ weak3 ] ) which gives existence of @xmath262 satisfying ( [ weak3 ] ) for given @xmath382 , hence @xmath381 is well defined . in particular , we have @xmath383.\ ] ] with the operators @xmath236 and @xmath381 well defined we can proceed with the galerkin method . taking @xmath384 , @xmath385 , @xmath386 and @xmath387 in ( [ weak1 ] ) , @xmath388 , we arrive at a system of @xmath389 equations @xmath390 where @xmath391 is defined as @xmath392 \ , ( v^n \cdot \tau_j ) \,ds - \int_{\omega } f \cdot v^n \,dx . \end{array}\ ] ] now , if @xmath393 satisfies ( [ system_aprox ] ) for @xmath394 and @xmath395 and @xmath396 , defined as above , then the triple @xmath397 satisfies ( [ weak1])([weak3 ] ) for @xmath398 , @xmath399 . we will call such a triple an approximate solution to ( [ weak1])([weak3 ] ) . the following lemma gives existence of a solution to system ( [ system_aprox ] ) : let @xmath400 , @xmath401 , @xmath402 , @xmath403 . assume that @xmath188 and @xmath63 , @xmath28 , @xmath64 fulfill the assumptions of theorem [ main_thm ] . then there exists @xmath404 satisfying ( [ system_aprox ] ) for @xmath386 . moreover , @xmath405 we will apply a well - known tool of the galerkin method , lemma [ lem_p ] . thus we define @xmath406 as @xmath407 we have to show that @xmath408 on some sphere in @xmath409 . as @xmath410 is linear in the second variable , we have @xmath411 by the korn inequality we have @xmath412 for @xmath64 large enough . to deal with @xmath413 let us denote for a moment @xmath414 . then setting @xmath415 in ( [ weak3 ] ) ( with @xmath416 instead of @xmath137 ) we have @xmath417\ ] ] and repeating the reasoning from the proof of ( [ ene ] ) we infer @xmath418 we have to find a bound on @xmath419 . denoting @xmath420 we have @xmath421 using ( [ est_s ] ) we get @xmath422 the remaining part of ( [ i3_1 ] ) is also not very difficult . with the first integral on the r.h.s we have @xmath423 \,dz . % \end{array } = \frac 12 \int_\omega \big(\partial_{x_1 } |\sigma^n|^2 + u \cdot \nabla |\sigma^n|^2\big ) \ , dx \\[2pt ] = -\frac12 \int_{\gamma_{in } } |\sigma^n|^2 \ , ds + \frac 12 \int_{\gamma_{out } } |\sigma^n|^2 \ , ds + \frac 12 \int_{\gamma } u\cdot n |\sigma^n|^2 \ , ds - \frac 12 \int_{\omega } |\sigma^n|^2 { \rm div}\ , u \ , dx \\[2pt ] \geq - \frac 12 \int_{\gamma_{in } } \sigma_{in}^2 \ , ds + \frac 12 \int_{\gamma } u\cdot n |\sigma_{in}|^2 \ , ds - \frac 12 \|{\rm div}\ , u\|_{l_\infty } \|\sigma^n\|_{l_2}^2 \\[2pt ] \geq -c \|\sigma_{in}\|_{l_2}^2 -e \|\sigma^n\|_{l_2}^2 \geq -c -e \big(\|g\|_{l_2}^2 + \|\sigma^n\|_{w^1_2}^2 + \|\sigma_{in}\|_{l_2}^2\big ) . \end{array}\ ] ] hence we have @xmath424 combining ( [ est_p_1]),([est_p_2 ] ) and ( [ est_p_3 ] ) we conclude @xmath425,\ ] ] where @xmath426 . thus there exists @xmath427 such that @xmath428 , and applying lemma [ lem_p ] we conclude that @xmath429 . by the definition of @xmath430 , @xmath431 is a solution to ( [ system_aprox ] ) . now showing the existence of weak solution is straightforward . the result is [ lem_weak ] assume that @xmath432 , @xmath401 , @xmath402 , @xmath403 . let @xmath188 be small enough and assume @xmath433 satisfy the assumptions of theorem [ main_thm ] . then there exists @xmath366 , that is a weak solution to system ( [ system_lin ] ) . moreover , the weak solution satisfies estimate ( [ ene ] ) . let us set @xmath434 and @xmath435 , where @xmath393 is the solution to ( [ system_aprox ] ) . estimates ( [ est_s ] ) , ( [ est_r ] ) and ( [ est_un ] ) imply that @xmath436 . thus , at least for a chosen subsequence ( denoted however in the same way ) @xmath437 for some @xmath438 . passing to the limit in ( [ weak1])([weak3 ] ) for @xmath397 we conclude that @xmath272 satisfies ( [ weak1])([weak3 ] ) , thus we have the weak solution . to show the boundary condition on the density we can rewrite the r.h.s of ( [ system_lin])@xmath439 as @xmath440 and , treating @xmath238 as a `` time '' variable , adapt di perna - lions theory of transport equation ( @xcite ) that implies the uniqueness of solution to ( [ def_s_1 ] ) in the class @xmath245 . the proof is thus complete . with the estimate the only problem is to tackle the singularities at the junction of @xmath21 with @xmath19 and @xmath20 . to this end we reflect the weak solution in a way which preserves the boundary conditions and then apply the classical elliptic theory to the extended solution of the lam system . the details are given in lemma [ lem_lame ] . to show the regularity of the temperature we can apply lemma [ lem_neumann ] since on @xmath27 the boundary condition on the density reduces to the neumann condition . in this section we will show the bounds on the sequence ( [ system_seq ] ) . due to the term @xmath441 in the continuity equation we are not able to show directly the convergence of this sequence in @xmath180 to the strong solution of ( [ system ] ) . we can show however its boundedness in @xmath180 . next , using this bound we derive the cauchy condition in @xmath181 , and thus show the convergence in this space to some @xmath272 . on the other hand , the boundedness implies weak convergence in @xmath180 , and the limit must be @xmath272 ; hence the solution is strong . the following lemma gives the boundedness of @xmath442 in @xmath180 . [ lem_seq_bound ] let @xmath443 be a sequence of solutions to ( [ system_seq ] ) starting from @xmath444,0,0)$ ] . then @xmath445 where @xmath446 can be arbitrarily small provided that @xmath62 defined in , quantities @xmath447 , @xmath403 , @xmath448 and @xmath188 are small enough and @xmath64 , @xmath28 , @xmath63 fulfill the assumptions of theorem [ main_thm ] . estimate ( [ est_lin ] ) for system ( [ system_seq ] ) reads @xmath449 . \end{array}\ ] ] denoting @xmath450 , from ( [ est_fgh ] ) and ( [ unwn_1 ] ) we get @xmath451 we aim at showing that @xmath452 thus @xmath452 is bounded by a constant that can be arbitrarily small provided that @xmath453 and @xmath62 are small enough . indeed , let us fix @xmath454 . ( note that the constant @xmath71 can be without loss of generality taken larger than @xmath455 . ) assume that @xmath456 . then ( [ an1 ] ) entails an implication @xmath457 and we can conclude that @xmath458 the next lemma almost completes the proof of the cauchy condition in @xmath181 for the iterating scheme . [ lem_cauchy1 ] let the assumptions of lemma [ lem_seq_bound ] hold . then we have @xmath459 where @xmath446 is the constant from ( [ est_seq_bound ] ) and @xmath460 can be taken arbitrarily small . subtracting ( [ system_seq])@xmath461 from ( [ system_seq])@xmath462 we arrive at @xmath463 @xmath464 @xmath465 @xmath466 estimate ( [ ene ] ) applied to this system yields @xmath467 in order to derive ( [ est_cauchy1 ] ) from the above inequality we have to examine the r.h.s . the differences in @xmath468 and @xmath168 are bounded in a straightforward way : using the imbedding @xmath469 and hlder s inequality we derive @xmath470 the difference in @xmath471 must be investigated more carefully . a direct calculation yields @xmath472 where in @xmath473 we include all the terms which do not contain @xmath474 as well as other terms containing the gradient of density or its difference . direct calculation using the imbedding @xmath475 , @xmath469 and hlder s inequality yields @xmath476 next , @xmath477 we have to compute @xmath253 norm of @xmath478 , hence we multiply by @xmath479 and integrate . with the first term we have @xmath480 hence @xmath481 we have used the fact that @xmath482 and the imbedding @xmath483 . the other term @xmath484 can be estimated similarly , only without integration by parts . combining the estimates on @xmath473,@xmath485 and @xmath484 we conclude @xmath486 the only term that remains to estimate is @xmath487 . we emphasize that this is the term which makes it impossible to show the convergence in @xmath488 directly . namely , if we would like to apply the estimate ( [ est_lin ] ) to the system for the difference then we would have to estimate @xmath489 what can not be done as we do not have any knowledge about @xmath490 . fortunately we only need the @xmath69-norm of this awkward term , which can be bounded in a direct way as @xmath491 since @xmath492 for @xmath95 . we have thus completed the proof of ( [ est_cauchy1 ] ) . now , lemma [ lem_seq_bound ] implies that the constant @xmath493 provided that the data is small enough and the starting point @xmath494,0,0)$ ] . it completes the proof of the cauchy condition in @xmath495 for the sequence @xmath442 . * lemmas [ lem_seq_bound ] and [ lem_cauchy1 ] hold for any starting point @xmath496 small enough in @xmath180 , but we can start the iteration from @xmath497,0,0)$ ] without loss of generality . in this section we prove our main result , theorem [ main_thm ] . first we show existence of the solution passing to the limit with the sequence @xmath442 and next we show that this solution is unique in the class of solutions satisfying ( [ est_main ] ) . * existence of the solution*. since we have the cauchy condition on the sequence @xmath442 only in the space @xmath181 , first we have to show the convergence in the weak formulation of the problem ( [ system ] ) . the sequence @xmath442 satisfies in particular the following weak formulation of ( [ system_seq ] ) : now using the convergence in @xmath181 combined with the bound ( [ est_seq_bound ] ) in @xmath180 we can pass to the limit in ( [ weak1_seq])([weak3_seq ] ) . the convergence of the l.h.s . of ( [ weak1_seq])([weak3_seq ] ) is obvious . recalling the definition ( [ fgh ] ) of @xmath504 and @xmath171 we verify easily that @xmath505 and @xmath506 and the only step in showing the convergence of @xmath471 which requires more attention is to show that @xmath507 however , since @xmath508 and @xmath509 in @xmath510 as well as @xmath511 in @xmath512 , we easily verify that holds true . we conclude that @xmath272 satisfies ( [ weak1_seq])([weak3_seq ] ) . now we need to show that this implies the strong formulation ( [ system ] ) , what can be done in a standard way , just integrating by parts in the weak formulation . now we set @xmath513 and we see that @xmath66 solves ( [ main_system ] ) . we clearly have @xmath514 where @xmath62 is defined in ( [ d0 ] ) , hence estimate ( [ est_main ] ) holds . * uniqueness . * the uniqueness in the class of solutions satisfying ( [ est_main ] ) actually results directly from the method of the proof , more precisely from the proof of ( [ est_cauchy1 ] ) . namely , we can show uniqueness on the level of perturbations , i.e. uniqueness for ( [ system ] ) . consider two solutions with the same data , denote it by @xmath515 and @xmath516 . their difference then satisfy @xmath517 @xmath518 @xmath519 @xmath520 now recall that to show ( [ est_cauchy1 ] ) we applied only ( [ dif_seq1])([dif_seq4 ] ) and ( [ est_seq_bound ] ) , hence from the above equations we conclude @xmath521 provided the data are small enough we have @xmath522 and so @xmath523 which completes the proof of uniqueness and hence of theorem [ main_thm ] . 99 r.adams , _ sobolev spaces _ , pure and applied mathematics , vol . 65 . academic press , new york - london , 1975 s.agmon , a.douglis , l.nirenberg , _ estimates near the boundary for solutions of elliptic partial differential equations satisfying general boundary conditions i _ , comm.pure appl.math . 12 ( 1959 ) , 623727 . s.agmon , a.douglis , l.nirenberg , _ estimates near the boundary for solutions of elliptic partial differential equations satisfying general boundary conditions ii _ , comm.pure appl.math . 17 ( 1964 ) , 3592 . da veiga , _ an @xmath179-theory for the n - dimensional , stationary , compressible navier - stokes equations , and the incompressible limit for compressible fluids . the equilibrium solutions _ , comm.math.phys . 109 ( 1987 ) , 229248 . r.j.diperna , p.l.lions , _ ordinary differential equations , transport theory and sobolev spaces _ , invent.math . 98 ( 1989 ) , 511547 . e.feireisl , _ dynamics of viscous compressible fluids _ , oxford lecture series in mathematics and its applications , 26 . oxford university press , oxford , 2004 . g.p.galdi , _ an introduction to the mathematical theory of the navier - stokes equations _ , vol.i , springer - verlag , new york , 1994 . d.jessl , a.novotn , _ existence of renormalized weak solutions to the steady equations describing compressible fluids in barotropic regimes _ , to appear in annales ihp analyse nonlinaire . d.jessl , a.novotn , m.pokorn , _ steady navier stokes fourier system with slip boundary conditions _ , submitted . s.jiang , c.zhou , _ existence of weak solutions to the three dimensional steady compressible navier stokes equations _ , annales ihp analyse nonlinaire 28 ( 2011 ) , 485498 . r.b.kellogg , j.r.kweon , _ compressible navier - stokes equations in a bounded domain with inflow boundary condition _ , siam j.math.anal . 28,1(1997 ) , 94108 . r.b.kellogg , j.r.kweon , _ smooth solution of the compressible navier - stokes equations in an unbounded domain with inflow boundary condition _ , j.math.anal . and app . 220 ( 1998 ) , 657675 . p.l.lions , _ mathematical topics in fluid mechanics . 2 . compressible models _ , oxford lecture series in mathematics and its applications , 10 . oxford science publications . the clarendon press , oxford university press , new york , 1998 . p.b.mucha , _ on cylindrical symmetric flows through pipe - like domains _ , j.differential equations 201 ( 2004 ) , 304323 . p.b.mucha , _ on navier - stokes equations with slip boundary conditions in an infinite pipe _ , acta applicandae mathematicae 76 ( 2003 ) , 115 . p.b.mucha , m.pokorn , _ on a new approach to the issue of existence and regularity for the steady compressible navier - stokes equations _ , nonlinearity 19 ( 2006 ) , 17471768 . p.b.mucha , m.pokorn , _ on the steady compressible navier stokes fourier system _ , comm . ( 2009 ) , 349377 . p.b.mucha , m.pokorn , _ weak solutions to equations of steady compressible heat conducting fluids _ , mathematical models and methods in applied sciences 20,5 ( 2010 ) , 785813 . p.b.mucha , r.rautmann , _ convergence of rothe s scheme for the navier - stokes equations with slip conditions in 2d domains _ , zamm z. angew . 86,9 ( 2006 ) , 691701 . a.novotn , m. padula , _ @xmath179-approach to steady flows of viscous compressible fluids in exterior domains _ , arch . rational mech . 126,3 ( 1994 ) , 243297 . a.novotn , k.pileckas _ steady compressible navier - stokes equations with large potential forces via a method of decomposition _ , math . methods appl . 21,8 ( 1998 ) , 665684 . a.novotn , i.straskraba , _ an introduction to the mathematical theory of compressible flows _ , oxford science publications , oxford 2004 . t.piasecki , _ on an inhomogeneous slip - inflow boundary value problem for a steady flow of a viscous compressible fluid in a cylindrical domain _ , journal of differential equations 248 ( 2010 ) , 21712198 . p.i.plotnikov , e.v.ruban , j.sokolowski , _ inhomogeneous boundary value problems for compressible navier - stokes equations : well - posedness and sensitivity analysis _ , siam j.math.anal . 40,3 ( 2008 ) , 11521200 . p.i.plotnikov , j.sokolowski , _ on compactness , domain dependence and existence of steady state solutions to compressible isothermal navier - stokes equations _ , j.math.fluid.mech . 7 ( 2005 ) , 529573 . m.pokorn , p.b.mucha , _ 3d steady compressible navier - stokes equations _ , discrete and continuous dynamical systems s , 1(1 ) ( 2008 ) , 151163 . v.a.solonnikov , _ overdetermined elliptic boundary value problems _ , zap.nauch.sem.lomi 21 ( 1971 ) , 112158 . r.temam , _ navier stokes equations _ , north - holland , amsterdam , 1977 . a.valli , w.m.zajaczkowski , _ navier stokes equations for compressible fluids : global existence and qualitative properties of the solutions in the general case _ , comm . math . phys . 103,2 ( 1986 ) , 259296 . w.m.zajaczkowski , _ existence and regularity of solutions of some elliptic systems in domains with edges _ , dissertationes math . , 274(1989 )

the fermion representation for spins is generalized to spins with arbitrary magnitudes . the symmetry properties of the representation is analyzed where we find that the particle - hole symmetry in the spinon hilbert space of/2 fermion representation is absent for . as a result , different path integral representations and mean field theories can be formulated for spin models . in particular , we construct a lagrangian with restored particle - hole symmetry , and apply the corresponding mean field theory to one dimensional ( 1d ) and antiferromagnetic heisenberg models , with results that agree with haldane s conjecture . for a open chain , we show that majorana fermion edge states exist in our mean field theory . the generalization to spins with arbitrary magnitude is discussed . our approach can be applied to higher dimensional spin systems . as an example , we study the geometrically frustrated afm on triangular lattice . two spin liquids with different pairing symmetries are discussed : the gapped-wave spin liquid and the gapless-wave spin liquid . we compare our mean field result with the experiment on nigas , which remains disordered at low temperature and was proposed to be in a spin liquid state . our fermionic mean field theory provide a framework to study spin liquids with fermionic spinon excitations .

1005.5626

quantum magnetism is one of the oldest central problems in condensed matter and many body physics . few exact results are known except at one dimension or when long range spin order is established at low temperature . in the later case , landau s symmetry breaking paradigm is applicable , and the long range spin order determines the low energy physics of the system . the low energy spin excitations are spin waves or magnons.@xcite systematic corrections to spin wave theory can be obtained through semiclassical approach ( @xmath10 expansion@xcite ) and/or nonlinear @xmath11 model.@xcite however , it was suggested that for some anti - ferromagnetic materials , strong quantum fluctuations due to small spin magnitude and low dimensionality combined with geometric frustration may lead to quantum coherent , spin disordered ground states.@xcite in such cases , a quantum fluid approach@xcite is probably a better starting point to describe the low energy physics instead of usual spin wave or semiclassical approach . there exist at present two common quantum fluid approaches based on different particle " representations of spins . in the bosonic " approach@xcite spins are represented by spin up and down schwinger bosons with the constraint that total boson number at each site is @xmath12 . a spin - disordered state is obtained in a mean field theory as long as bose condensation does not occur . the low lying excitations coming from this theory are bosonic . the other possibility is to represent spins by fermions . up to now , the fermionic approach is mainly restricted to @xmath0 spin systems@xcite . because of the pauli exclusion principle , we can not put more than two spin up or down fermions on a site to form @xmath2 objects as in bosonic approach . one way out is to introduce more species of fermions . it was proposed in ref . , that @xmath13-species of boson / fermions can be used to construct the spin-@xmath5 swapping operators . later , it is shown that the fermionic representation can be used to describe a @xmath3 valence bond solid state.@xcite in this paper , we systematically study the fermionic representation for arbitrary spins . to construct a spin operator with spin magnitude @xmath5 , we follow ref . and introduce different fermion operators for different @xmath14 states . we find that for half - odd - integer spins , the spin operator thus constructed is invariant under a @xmath15 transformations,@xcite i.e. the symmetry group is @xmath15 group whereas for integer spins the symmetry group is @xmath16 . fermionic path integral formulations for spin systems with arbitrary spin-@xmath5 can then be formulated and the corresponding mean field theories can be studied . we shall see that one of the mean field theories applied to 1d antiferromagnetic heisenberg model produces mean field excitation spectrums which agrees with haldane s conjecture.@xcite moreover , there exist zero energy majorana edge modes for open integer spin chains . we also apply our approach to two - dimensional ( 2d ) antiferromagnetic heisenberg models on triangular lattices where we obtain two translational and rotational invariant spin liquid solutions , one ( with @xmath6-wave paring symmetry ) is gapped and the other one ( with @xmath7-wave pairing symmetry ) has nodes at the spinon fermi level . comparing with experiments on @xmath17 , we argue that the @xmath7-wave pairing spin liquid is a plausible ground state . it should be noted that rather different fermionic / bosonic spinon approaches have also been applied to @xmath18 spin models where the dynamic symmetry is generalized from @xmath15 to @xmath18 , and the hamiltonian is composed of @xmath18 generators .. in these approaches , @xmath19-species of particles are introduced to construct the @xmath18-spin operator and hamiltonians . the major difference between our approach and the @xmath18 approaches is that @xmath13 species of fermions are introduced to form a @xmath15-spin ( or irreducible representation of @xmath15 group ) in our approach . unlike the @xmath18 models , our approach is not limited to bipartite lattices . the paper is organized as follows . in section ii , we introduce the general fermionic spinon representation for arbitrary spin @xmath5 and discuss the symmetry properties associated with the representation . in section iii , we introduce the path integral formalism where different ways of imposing particle number constraints are presented . a particle - hole symmetric way of imposing constraint is discussed . in section iv.a and iv.b , we focus on 1d heisenberg models for spin-1 and spin-3/2 , respectively , and study two versions of the mean field theories . the cases for general spin @xmath5 are discussed in section iv.c . in section v we study 2d afm models on triangular lattice . our paper is summarized and concluded in section vi . we begin with the fermionic representation for spins . for spin @xmath0 , two species of fermions @xmath20 and @xmath21 are introduced to construct the three spin operators @xmath22 . to generalize this fermionic representation to arbitrary spin-@xmath5 , we introduce @xmath23 species of fermionic operators @xmath24 satisfying anti - commutation relations , @xmath25 where @xmath26 . the spin operator can be expressed in terms of @xmath24 and @xmath27 s , @xmath28 where @xmath29 and @xmath30 @xmath31 is a @xmath32 matrix whose matrix elements are given by @xmath33 it is easy to see that the operators @xmath34 satisfy the @xmath15 angular momentum algebra , @xmath35=i\epsilon^{abc}\hat s^c$ ] . under a rotational operation , @xmath36 is a spin-@xmath5 spinor " transforming as @xmath37 and @xmath38 is a vector transforming as @xmath39 , here @xmath40 is the @xmath23-dimensional irreducible representation of @xmath15 group generated by @xmath41 and @xmath42 is the adjoint representation . as in the @xmath0 case , a constraint that there is only one fermion per site is needed to project the fermionic system into the proper hilbert space representing spins , i.e. @xmath43 where @xmath44 is the site index and @xmath45 ( the particle representation , one fermion per site ) . alternatively , it is straightforward to show that the constraint @xmath46 ( the hole representation , a single hole per site ) represents a spin equally . the @xmath45 representation can be mapped to the @xmath46 representation _ via _ a _ particle - hole transformation_. for @xmath0 , the particle picture and the hole picture are identical , reflecting an intrinsic particle - hole symmetry of the underlying hilbert space which is absent for @xmath47 . following affleck _ et al . _ , @xcite we introduce another spinor " @xmath48 , whose components can be written as @xmath49 , where the index @xmath50 runs from @xmath5 to @xmath51 as in @xmath36 . to examine whether @xmath52 is really a spinor " , we construct a spin singlet state for two spins at site @xmath44 and @xmath53 , @xmath54 . therefore @xmath55 is a scalar operator as @xmath56 , meaning that @xmath52 and @xmath36 must behave identically under spin rotation . consequently , the spin operators can also be written in terms of @xmath52 , @xmath57 combining @xmath36 and @xmath52 into a @xmath58 matrix @xmath59 , @xcite we can reexpress the spin operators as @xmath60 and the constraint can be expressed as @xmath61 where the @xmath62 sign implies @xmath45 and @xmath63 sign implies @xmath46 . we are interested in two kinds of operations ( or groups ) acting on @xmath64 . the first kind belongs to a @xmath13-dimensional irreducible representation of @xmath15 group , acting on the left of @xmath64 . suppose @xmath65 is an element of this irreducible representation , then @xmath66 ( @xmath42 belongs to the adjoint irreducible representation of @xmath15 group , which is a @xmath67 matrix ) . under transformation @xmath68 , the spin operator @xmath69 becomes @xmath70 , which means _ a rotation of the spin_. it is obvious that the particle number constraint eq . ( [ contr2 ] ) remains unchanged under the action of @xmath65 . the other kind belongs to a @xmath71 unitary group acting on the right of @xmath64 which keeps the spin operator eq . ( [ spin ] ) and the fermionic statistics eq . ( [ fermi ] ) invariant . this group reflects the symmetry properties of the underlying hilbert space structure in the fermionic representation . we call it an _ internal symmetry group_. the internal symmetry group is different for integer and half - integer spins . it is @xmath72 for the former and @xmath15 for the latter . we leave the rigorous proof in appendix a , and shall explain qualitatively the reason behind here . notice that @xmath36 and @xmath52 are not independent . the operators in the internal symmetry group will mix " the two fermion operators in the same row of @xmath36 and @xmath73 , i.e. @xmath74 and @xmath75 , @xmath76 and @xmath77 , etc . for integer spins , @xmath78 and @xmath79 will be mixed " . to keep the relation @xmath80 invariant , there are only two methods of mixing " : one is an @xmath81 transformation , the other is interchanging the two operators . these operations form the @xmath16 group . for half - odd - integer spins , the pair @xmath82 do not exist , and the symmetry group is the maximum group @xmath15 . the difference between integer and half - integer spins is a fundamental property of the fermionic representation as we shall see more later . now let us see how the constraints eq . ( [ newcontr1 ] ) transform under the symmetry group . for @xmath0 , the constraint eq . ( [ newcontr1 ] ) is invariant under the transformation @xmath83 because the right hand side vanishes ( due to the particle - hole symmetry of the hilbert space ) . for integer spins , if @xmath84 , then @xmath85 , and eq . ( [ newcontr1 ] ) is invariant . if @xmath86 , then @xmath87 , meaning that the particle " picture(@xmath62 sign in eq . ( [ contr2 ] ) ) and the hole " picture ( @xmath63 sign in eq . ( [ contr2 ] ) ) are transformed to each other . for a half - odd - integer spin with @xmath88 , @xmath89 is a rotation and we need to extend the constraint into a vector form similar to @xmath0 case,@xcite so that eq . ( [ contr2 ] ) becomes @xmath90 under the group transformation @xmath83 , @xmath91 where @xmath92 , @xmath93 , i.e. @xmath42 is a 3 by 3 matrix representing a 3d rotation in the internal hilbert space . the transformed constraint represents a new hilbert subspace which is still a @xmath94-dimensional irreducible representation of the spin @xmath15 algebra . any measurable physical quantity such as the spin @xmath95 remains unchanged in the new hilbert space . therefore , for half - odd - integer spins ( @xmath88 ) , there exists infinitely many ways of imposing the constraint that gives rise to a hilbert subspace representing a spin . however , for integer spins , there exists only two possible constraint representations . the different representations of constraints for @xmath2 systems result in different path integral representations and different mean field theories . these mean field theories are equivalent in the sense that they can be transformed to each other by the internal symmetry group . in the next section , we shall study the heisenberg model in different representations and shall construct a mixed " path integral representation which restored particle - hole symmetry , and the internal symmetry group becomes almost " a gauge symmetry.@xcite the mixed " representation is studied in section iv where a new mean field theory is proposed which recovers haldane conjecture at 1d for the heisenberg model . we shall focus on the antiferromagnetic heisenberg model @xmath96 with @xmath97 in the rest of the paper . we start by presenting some useful algebraic manipulations of the hamiltonian , and then discuss the general path integral formalism . the different ways of handling constraints and how they affect the symmetry of the lagrangian will be discussed in the process . it is known that for spin-1/2 the heisenberg interaction can be written as @xcite @xmath98 where @xmath99 are spin - singlet operators . @xmath100 and @xmath101 will be extended to arbitrary spin magnitudes with the above definition and will be used in the following discussions . interestingly , an expression almost the same as eq . ( [ s1/2 ] ) holds for @xmath3 , but this is no longer true for larger spins . we shall present precise formula for @xmath4 and @xmath102 , and provide a general discussion for higher spins . for @xmath3 , the three spin matrices are @xmath103 where @xmath104 . the matrix operator @xmath64 is given by @xmath105 the spin operator can be written in form of eq . ( [ spin ] ) , and it can be shown after some straightforward algebra that the hamiltonian can be written as @xmath106 for @xmath4 , the three spin matrices are @xmath107 and the matrix operator @xmath64 is given by @xmath108 where @xmath109 and @xmath110 . in this case the singlet operators @xmath100 and @xmath101 alone are not enough to represent the heisenberg hamiltonian , and triplet hoping and pairing terms are necessary . after some straightforward but tedious algebra , we find that the hamiltonian for spin @xmath4 heisenberg model can be written as @xmath111:\nonumber\\\end{aligned}\ ] ] it is interesting to note that the above form holds also for @xmath3 @xcite . however , for @xmath3 , the two terms in the square bracket can be transformed from one to another , which does not hold for @xmath4 . this suggests that @xmath112 heisenberg models have a higher hidden symmetry " compared with @xmath4 . actually , there is another hidden symmetry for the spin @xmath0 systems which is absent for spin-1 , @xmath113 where @xmath114 is the particle number on site @xmath44 . our analysis suggests that smaller spins have higher symmetry when the spin interaction is expressed in the fermionic representation . for @xmath5=2 , the heisenberg hamiltonian can be written as @xmath115 as in the case of lower spins , the terms @xmath116 and @xmath117 have finite overlaps and are not completely independent of each other . for @xmath118 , it is not possible to represent the hamiltonian in the above two terms alone . quintet and higher multipolar hoping and pairing operators are needed to represent the heisenberg hamiltonian and we are not able to obtain any general expression . summarizing , the heisenberg model can be written in the fermionic representation as @xmath119\end{aligned}\ ] ] where @xmath120 are parameters dependent on the spin magnitude @xmath5 and @xmath121 represents quintet and higher multiplet terms . in the following , we shall study in detail a hamiltonian which keeps the first two terms only . the hamiltonian can be considered as an effective hamiltonian for constructing trial ground state wavefunctions in a variational calculation . notice that the effective hamiltonian is in fact exact " up to @xmath102 if @xmath122 and @xmath123 are chosen properly . in imaginary time path integral formalism , the partition function is given by @xmath124+h\nonumber\\\end{aligned}\ ] ] where @xmath125 is the lagrange multiplier field introduced to impose the constraint eq . ( [ contr2 ] ) . the @xmath126 sign corresponds to the particle ( @xmath127 ) and hole ( @xmath128 ) representations . the grassmann number @xmath64 satisfy the antiperiodic boundary condition @xmath129 . notice that the lagrangian eq . ( [ u ] ) is not invariant under the action of the internal symmetry group because of the non - invariant particle number constraint eq . ( [ contr2 ] ) , i.e. the internal symmetry group of the spin operator is not a symmetry group of the lagrangian . in the following we will employ a trick to restore the symmetry of the lagrangian . the integer- and half - odd - integer- spin models will be discussed separately because of the intrinsic difference in their internal symmetry groups . our trick is to consider an average of all the partition functions with different particle - number constraint representations . in the integer spin case , the average is given by @xmath130+h)d\tau}\times\nonumber\\ & & \prod_{\{i,\tau\}}\frac{1}{2}\left[\delta(\hat n_i-1)+\delta(\hat n_i-2s)\right ] \nonumber\\ & = & \int d\psi d\psi^\dagger d\lambda e^{-\int_0^\beta ( \sum_i\frac{1}{2}\mathrm{tr}[\psi_i\partial_\tau\psi^\dagger_i]+h)d\tau}\times\nonumber\\ & & \prod_{\{i,\tau\}}\frac{1}{2}\left(e^{i\frac{\lambda_i}{2}[\mathrm{tr}(\psi_i\sigma_z \psi_i^\dagger)-2s+1]}\right.\nonumber\\&&\left . + e^{i\frac{\lambda_i}{2}[\mathrm{tr}(\psi_i\sigma_z\psi_i^\dagger)+2s-1]}\right).\end{aligned}\ ] ] notice that we do not make any approximations in deriving eq . ( [ path1 ] ) from eq . ( [ u ] ) and the new " partition function is a faithful representation of heisenberg model except that it averages over all possible particle " and hole " representations of constraints locally . similar ideas have been applied to generate the supersymmetric representation and the @xmath15 representation of the @xmath131-@xmath132 model @xcite . the fermion particle - hole symmetry is restored in this representation . the lagrangian corresponding to eq . ( [ path1 ] ) is @xmath133 -2\ln\cos\frac{(2s-1)\lambda_i}{2}\right]\nonumber\\ & & + h.\nonumber\end{aligned}\ ] ] this form of lagrangian is invariant under the internal symmetry group @xmath16 and the symmetry group becomes almost a gauge symmetry " of the new lagrangian as we shall see in the following . we note that the lagrangian is complex because of the multiplier @xmath134 . this problem can be solved by lifting the contour of integration over @xmath134 into complex plane via analytic continuation to find a saddle point in the imaginary axis @xcite . for convenience , we write @xmath135 , and integrate @xmath136 from @xmath137 to @xmath138 ( the saddle point of @xmath136 is real ) . then the lagrangian becomes @xmath139 -2\ln\cosh\frac{(2s-1)\tilde\lambda_i}{2}\right]\nonumber\\ & & + h,\end{aligned}\ ] ] we shall now consider the path integral representation in terms of the effective hamiltonian eq . ( [ htr ] ) keeping only the first two terms . the effective hamiltonian can be decoupled by a hubbard - stratonovich matrix field ( see appendix b ) @xmath140 where @xmath141 , and @xmath142 . note that @xmath143 and @xmath144 form two sets of spin triplet operators , respectively . the hamiltonian becomes @xmath145\right.\nonumber\\ & & \left.+b(s)\mathrm{tr}[\hat { \mathbf v}_{ij}^\dagger\cdot\hat { \mathbf v}_{ij } -(\hat{\mathbf v}_{ij}\cdot \psi_j^\dagger\mathbf i\psi_i+h.c.)]\right\}.\end{aligned}\ ] ] the @xmath146 and @xmath136 fields define @xmath16 lattice gauge fields coupling to the fermionic particles which we shall call spinons in the following . under the local gauge transformation @xmath147 , the temporal component and the spacial component of the gauge fields transform as @xmath148 notice that @xmath125 changes sign under an uniform @xmath149 gauge transformation , and @xmath100 , @xmath101 exchange their roles under a staggered " @xmath149 transformation where we have @xmath150 on one sublattice and @xmath151 on the other sublattice . notice also that the lagrangian is not invariant under a time - dependent gauge transformation , because the term @xmath152 is not invariant . a similar situation occurs in the supersymmetric representation of heisenberg model . because of this restriction the internal symmetry group does not generate a complete gauge symmetry " in the path integral formalism . integrating out the fermion fields , we get an effective action in terms of the @xmath153 , @xmath154 and @xmath125 fields . the mean field values of the these fields are given by the saddle point of the effective action determined by the self - consistent equations @xmath155 , @xmath156 with the mean field hamiltonian @xmath157+b(s)[-(\mathbf v^*\cdot c_i^\dagger\mathbf i c_j+\mathbf u^*\cdot\bar c_i^\dagger\mathbf i c_j+h.c.)\right.\nonumber\\ & & \left.+|\mathbf u|^2+|\mathbf v|^2]\right ] + \sum_i[\tilde\lambda(\hat n_i-\frac{2s+1}{2})\nonumber\\ & & -\ln\cosh\frac{(2s-1)\tilde\lambda}{2}],\end{aligned}\ ] ] where @xmath158 , @xmath159 , @xmath160 , @xmath161 , and @xmath162 is determined by the condition @xmath163 notice that the averaged particle number satisfies @xmath164 because we are mixing the particle and hole pictures in the constraint . if the ground state does nt break the particle - hole symmetry , then @xmath165 and we get the half - filling condition @xmath166 . we first examine how the two particle number constrains @xmath167 are transformed into each other by the @xmath15 internal symmetry group . we consider the vector constraint eq . ( [ veccontr ] ) for @xmath168 @xmath169 where @xmath170 the @xmath171 and @xmath172 components indicate that the one - site pairings @xmath173 @xmath174 are equal to 0 , and the @xmath175 component imposes the constraint that there is exactly one fermion per site . notice that the first two components of the constraint is automatically satisfied if the third component is satisfied rigorously . under the @xmath15 transformation @xmath176 , the constraint becomes @xmath177 , where @xmath178 , @xmath42 is an @xmath67 rotation . since the constraint is invariant under an @xmath81 gauge transformation generated by @xmath179 , the continuum space formed by different constrains is the surface of a sphere @xmath180 . in particular , for @xmath181 , the north pole of the sphere which corresponds to the positive sign in eq . ( [ contr2 ] ) is transformed to the south pole , @xmath182 , corresponding to the negative sign in eq . ( [ contr2 ] ) . now we construct a lagrangian which is invariant under the @xmath15 internal symmetry group . in the path integral formalism , the particle number constraint @xmath183 can be realized by introducing a vector langrange multiplier field @xmath184 @xmath185\}.\end{aligned}\ ] ] the vector @xmath186 changes into @xmath187 under @xmath15 gauge transformation , where @xmath188 is an unit vector on the surface of the sphere @xmath189 . as in the integer spin case , a @xmath15 invariant path integral formalism can be obtained by averaging over all possible particle number constraints : @xmath190 where @xmath191 . introducing @xmath192 , we obtain a lagrangian with almost " @xmath15 gauge symmetry : @xmath193 -(2s-1)\ln\frac{\sinh\tilde\lambda_i}{\tilde\lambda_i}]\nonumber\\&&+h.\end{aligned}\ ] ] as in the case of integer spins , we replace @xmath194 by the effective hamiltonian eq . ( [ htr ] ) with only the first two terms . introducing the hubbard stratonovich field as in the integer spin case ( see appendix b for details ) , @xmath195 then @xmath196 and @xmath197 form an @xmath15 lattice gauge field coupling to the spinons . ( notice the sign difference in the matrix elements of @xmath198 and @xmath199 between integer and half - odd - integer spins . ) under gauge transformation @xmath147 ( @xmath200 ) the gauge fields transform as , @xmath201 and the lagrangian eq . ( [ lsu2 ] ) is invariant only under time - independent gauge transformations as in the integer spin case because of the non - invariant term @xmath202 . this is different from the @xmath0 case , where the lagrangian is invariant under time - dependent gauge transformations . the mean field effective hamiltonian corresponding to eq . ( [ lsu2 ] ) is @xmath203+b(s)[-(\mathbf v^*\cdot c_i^\dagger\mathbf i c_j+\mathbf u^*\cdot\bar c_i^\dagger\mathbf i c_j+h.c.)\right.\nonumber\\ & & \left.+|\mathbf u|^2+|\mathbf v|^2]\right ] -\sum_i{2s-1\over2}(\pmb{\tilde\lambda}\cdot\hat { \mathbf m}_i + \ln\frac{\sinh\tilde\lambda}{\tilde\lambda})\nonumber\\\end{aligned}\ ] ] where the parameters @xmath204 are solved self - consistently by the mean field equations @xmath205 , @xmath206 and @xmath207 is determined by @xmath208 the averaged particle number per site is again given by @xmath166 if the ground state respects particle - hole symmetry @xmath209 . it can be proved ( see appendix b ) that for integer spins , @xmath210 , @xmath211 , @xmath212 and @xmath213 , whereas for half - odd - integer spins , @xmath210 , @xmath214 , @xmath215 and @xmath213 . notice that the parity of the paring terms @xmath101 and @xmath143 are different for integer and half - odd - integer spins . these operators are central to the mean field theory , because the ground states are completely determined by their expectation values . in particular , for a mean field theory with @xmath216 and @xmath217 , the mean field ground state is a bcs spin - singlet pairing state where the order parameter @xmath218 has even parity for half - odd - integer spins , but has odd parity for integer spins . we shall see how this important difference leads to different excitation spectrums between integer and half - odd - integer spin systems in the fermionic mean field theory . similar result exists for states with @xmath219 and @xmath220 . in this case the mean field ground state is a bcs spin - triplet pairing state and the parity of the order parameters are reversed . we note that as in the @xmath0 case , the mean field hamiltonian should be viewed as a trial hamiltonian for the ground state wavefunction of the spin systems after gutzwiller projection . for the mean field theory with particle - hole symmetry the gutzwiller projection is rather non - trivial . the state is a coherent superposition of states which allows sites with both one fermion and with @xmath12 fermions . the numerical analysis of such a state is complicated and we shall not go into details in this paper . in this section we apply our mean field theory to the antiferromagnetic heisenberg model in one dimension . we first consider the cases of spin @xmath3 and @xmath4 where two versions of mean field theories based on different methods of implementing the constraints will be discussed . the mean field results are summarized in section iv.c where the case of general spin @xmath5 and haldane conjecture will be discussed . to simplify our analysis we make use of the wagner - mermin theorem which asserts that a continuous symmetry can not spontaneously break in one dimension@xcite . therefore we assume in our mean - field theory that the ground state is a spin singlet ( spin liquid state ) and the expectation values of @xmath221 and @xmath222 are zero , since the rotational symmetry will be broken otherwise . as a result , we keep only the @xmath122 term in the trial hamiltonian . for simplicity we shall also restrict ourselves to translational invariant solutions of the mean - field theory in this paper . we choose @xmath223 to be the same as eq . ( [ h1 ] ) , i.e. , @xmath224 . for convenience of discussion , we introduce two new fermions @xmath225 which are the symmetric and antisymmetric combinations of @xmath226 and @xmath227 , @xmath228 the mean field hamiltonian eq . ( [ heff1 ] ) becomes completely decoupled in terms of @xmath225 and @xmath78 . in fourier space , it can be written as @xmath229\nonumber\\ & & + jn(|\chi|^2+|\delta|^2)-n(\frac{3}{2}\tilde\lambda+\ln\cosh\frac{\tilde\lambda}{2}),\end{aligned}\ ] ] where @xmath230 , @xmath231 ( since the phases of @xmath232 and @xmath233 are unimportant in 1d , we choose @xmath232 and @xmath233 to be real numbers ) and @xmath19 is the length of the chain . the mean field hamiltonian can be diagonalized by a bogoliubov transformation : @xmath234 the coefficients @xmath235 satisfy the bogoliubov - de gennes ( bdg ) equations , @xmath236 solving the equations we obtain @xmath237 where @xmath238 is given by @xmath239 and the diagonalized hamiltonian is @xmath240 where @xmath241 is the ground state energy . minimizing @xmath242 , we obtain the self - consistent mean field equations at zero temperature , [ mfeq ] @xmath243 with @xmath3 . the above equations are solved numerically where we obtain @xmath244 and @xmath165 as shown in the first row of table [ tab : spin1 ] . the averaged particle number of the spinons per site is @xmath245 , indicating that the ground state respects particle - hole symmetry . the ground state energy is @xmath246 , and the excitation spectrum is flat ( i.e. @xmath247-independent ) with a finite energy gap @xmath248 . the spin - spin correlation function is @xmath249\\\end{aligned}\ ] ] and is nonzero only for @xmath250 and @xmath251 , with @xmath252 and @xmath253 . the correlation function is zero for @xmath254 , indicating that the mean field theory describes a short - ranged valence - bond - solid ( vbs ) state . .solutions of two versions of mean fields for spin-1 af heisenberg chain . @xmath255 is the ground state energy per site and @xmath256 is the excitation gap . [ cols="^,^,^,^,^,^,^",options="header " , ] comparing the mean field energy dispersion eq . ( [ ex1 ] ) and eq . ( [ ex3/2 ] ) , we find that the excitation spectrum of spin-1 heisenberg model is gapped whereas the excitation spectrum for the spin-3/2 heisenberg model is gapless in the mean field formulation with particle - hole symmetry . this difference between integer spin and half - odd - integer spin persists for any spin @xmath5 , if we consider the trial hamiltonian eq . ( [ heff1 ] ) or eq . ( [ heff2 ] ) with @xmath257 , i.e. if we consider bcs spin - singlet ground state wavefunctions . the mean field equations for arbitrary @xmath5 are the same as eq . ( [ mfeq ] ) or eq . ( [ mf3/2 ] ) for arbitrary @xmath5 , and the dispersion is qualitatively the same as eq . ( [ ex1 ] ) or eq . ( [ ex3/2 ] ) , as long as we consider particle - hole symmetric mean field solutions which do not break translational invariance . the main effect of changing @xmath5 in mean field theory is to change the number of fermionic spinon species . this result is consistent with the haldane conjecture , @xcite which asserts that the integer spin heisenberg chains have singlet ground state with finite excitation gaps and exponentially decaying spin - spin correlation functions , while half - odd - integer spin chains have singlet ground states with gapless excitation spectrums and power law decaying correlations . our particle - hole symmetric mean field results agree well with haldane conjecture . haldane noticed that the time reversal operator satisfies @xmath258 for integer spin and @xmath259 for half - odd - integer spin , and this results in different berry phase contributions from the topological excitations ( skymion or instanton ) in the path integral formulation given by @xmath260 , where @xmath261 is the skyrmion number . for @xmath262 odd integer , the berry phase @xmath263 is @xmath264 for integer spins and @xmath265 for half - odd - integer spins . the difference in the @xmath266 factor between integer spin and half - odd - integer spin is the origin of haldane s conjecture . the situation is quite similar in our mean field theory . noticing that @xmath267 is a singlet formed by two spin-@xmath5 particles . the clebsch - gordan coefficients implies that @xmath211 for integer @xmath5 and @xmath214 for half - odd - integer @xmath5 ( see appendix b ) . this sign or parity difference results in appearance of @xmath268 in eq . ( [ ex1 ] ) for @xmath3 and @xmath269 in eq . ( [ ex3/2 ] ) for @xmath4 , which leads to different symmetries of ground state wavefunctions and different excitation spectrums between integer and half - odd - integer spin chains . and @xmath270 are two reciprocal vectors . the brillouin zone is divided into three regular pieces by the three new bases @xmath271 , @xmath272 and @xmath273 . , width=316 ] the topological term eq . ( [ tplg ] ) leads to exponentially localized edge states for open integer spin chains which is absent for half - odd - integer spin chains . @xcite this important difference between integer spin and half - odd - integer spin chains is also reflected in our mean field theory where zero energy majorana edge fermions exist at the open boundaries for integer spin chains which are absent for half - odd - integer spin chains . the existence of majorana edge fermions for integer spin chains is a direct consequence of the odd - pairing symmetry for integer spin chains . @xcite the details of the majorana edge fermions is discussed in appendix c. we show in the previous section that our mean field theory is able to capture the essential physics of spin - liquid states in 1d antiferromagnetic quantum spin chains . in this section we shall apply our mean field theory to the 2d @xmath274-@xmath275 heisenberg model on triangular lattice . we shall show that our mean field theory admits new spin liquid solutions not explored before . some results of this model have been reported in a previous paper@xcite , where particle number constraint is treated in the particle representation " ( @xmath276 ) . in this paper , we shall revisit this model with the particle - hole symmetric constraint . the hamiltonian of the @xmath274-@xmath275 model is @xmath277}\mathbf s_i\cdot \mathbf s_{j},\end{aligned}\ ] ] where @xmath278 denotes nearest neighbor ( nn ) and @xmath279 $ ] the next next nearest neighbors ( nnnn ) . for @xmath3 , there are two channels of decoupling the spin interaction , namely , @xmath280 or @xmath281 . since there is no mermin - wigner theorem to protect us at zero temperature at 2d , we have to keep both terms in the effective hamiltonian eq . ( [ heff1 ] ) . we shall choose @xmath282 , with the weight @xmath283 determined by minimizing the ground state energy . . we indicate the relative phase of the nearest neighbor pairing @xmath284 . ( a ) , @xmath7-wave pairing . the dished lines are the zero lines of the pairing function ( in lattice space or momentum space ) . ( b ) , @xmath6-wave pairing . there are no zero lines , the pairing function vanishes only at the @xmath285 point.,width=288 ] next we introduce the mean field parameters @xmath286 , @xmath287 , @xmath288 , @xmath289(for nn ) and @xmath290 , @xmath291 , @xmath292 , @xmath293 ( for nnnn ) to decouple the hamiltonian . we assume that @xmath294 and @xmath295 are parallel , so we only keep the @xmath296 component of these vectors , and note them simply as @xmath297 and @xmath298 ( which are set as real numbers ) . as in 1-d case we consider here only spin liquid solutions that respect translation and rotational symmetries . since the pairings have odd parity @xmath299 , only @xmath300-wave and @xmath7-wave like pairings are allowed . solving the mean field equations we find two kinds of solutions ( see fig . [ fig : pairing ] ) , which have @xmath7-wave and @xmath6-wave pairing symmetries , respectively . we shall concentrate on these two kinds of solutions in the following . in momentum space ( fig . [ fig : bz ] ) , the mean field hamiltonian eq . ( [ heff1 ] ) can be written as @xmath301\nonumber\\&&+\sum_{\mathbf{k } } v_k ( c^\dagger_{1k}c_{1k}-c^\dagger_{-1k}c_{-1k } ) \nonumber\\&&-\sum_{\mathbf{k } } u_{k}^*(c_{-1-k}c_{1k}+h.c.),\ ] ] with @xmath302 where @xmath162 is the lagrange multiplier and @xmath303,\\ & \psi_k^{f}={1\over3}[\sin k_x+\sin(-{k_x\over2}+{\sqrt3 k_y\over2 } ) + \sin(-{k_x\over2}-{\sqrt3k_y\over2})],\\ & \psi_k^{p+ip}={1\over3}[\sin k_x+e^{i{\pi\over3}}\sin({k_x\over2}+ { \sqrt3 k_y\over2})\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \\ & \ \ \ \ + e^{i{2\pi\over3 } } \sin(-{k_x\over2}+{\sqrt3 k_y\over2})].\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \\end{aligned}\ ] ] the pairing symmetries can be verified easily by expanding the pairing terms at small @xmath247 : @xmath304 , and @xmath305 . there are three lines of zeros for the @xmath7-wave pairing function as shown in fig . [ fig : pairing ] . introducing a vector @xmath306 , the hamiltonian eq . ( [ mf_hb ] ) can be written in a matrix form : @xmath307 where @xmath308 @xmath309 can be diagonalized by the bogoliubov transformation , @xmath310 where @xmath311 and @xmath312 . the corresponding eigenvalues are given by @xmath313 and @xmath314 , where @xmath315 and the self - consistent mean field equations are @xmath316 , \nonumber\\ & & \bar \delta_{1,3}=\frac{1}{n}\sum_{\mathbf{k}}\psi_{k,2k}[\frac { |\delta_k|}{e_{1k}}+ \frac{|\delta_k|}{2e_{0k } } ] , \nonumber\\ & & \bar u_{1,3}=\frac{1}{n}\sum_{\mathbf{k}}\gamma_{k,2k}\frac { u_k}{e_{1k } } , \nonumber\\ & & \bar v_{1,3}=0 , \nonumber\\ & & \langle \hat n_i-\frac{3}{2}\rangle=\frac{1}{2}\tanh \frac{\tilde\lambda}{2 } , \ ] ] where we have adopted the particle - hole symmetric constraint in writing down the mean field equations . we have also checked the mean field solutions with constraint @xmath317@xcite and found that the excitation spectrums are qualitatively the same as those in the particle - hole symmetric theory . the ground state energy is lowest for vanishing @xmath318 , i.e. @xmath319 , @xmath320 and the ground state is a spin - singlet . the excitations are characterized by three branches of fermionic spinons with @xmath321 and identical dispersion @xmath322 . -@xmath275 heisenberg model with @xmath323 . we use the constraint @xmath324 . the red dot line indicate energy per site for the @xmath6 state and the black square line is for the f - wave state . dirac nodes exist for the f - wave state . the number of nodes is 6 when @xmath274 is dominating and 24 when @xmath275 is dominating.,width=268 ] our mean field theory for the @xmath274-@xmath275 model contains two regimes of spin liquid states for both @xmath7 and @xmath6 pairing symmetries as a function of @xmath325 . a first order phase transition occurs between the two regimes at @xmath326 ( see fig . [ fig : hsb ] ) . when @xmath274 dominates , the spin liquid state is characterized by @xmath327 and @xmath328 ( consequently @xmath329 and @xmath330 ) ; while when @xmath275 dominates , @xmath331 and @xmath332 , @xmath333 ( consequently @xmath334 , @xmath335 ) . the @xmath6 states remain lower in energy in both regimes . the @xmath7-wave pairing solution respects particle - hole symmetry and has @xmath336 . the excitation is gapless with several dirac cones in the first brillouin zone . a cut of the spinon dispersion is shown in fig . [ fig : f_band ] where the particle - hole symmetry is obvious . the mean field solution with @xmath337 has similar properties , except that the particle - hole symmetry is lost and the position of the dirac nodes are shifted.@xcite for the @xmath6-wave pairing , we find that the ground state breaks particle - hole symmetry and @xmath338 . the excitation spectrum is fully gapped . similar to integer spin chains , we find that the @xmath6-wave state is topologically nontrivial ( see also appendix c ) . the solution has @xmath339 , so that @xmath340 can be either positive or negative , depending on @xmath341 . therefore the bogoliubov spinor space described by the vector @xmath342 ( which is @xmath189 ) covers the brillouin zone ( @xmath341 space , which is also @xmath189 ) at least once . in other words , the topological ( skyrmion ) number of mapping from the @xmath341 space to the spinor space is nonzero @xmath343 . however in the vacuum where the spinon density is zero , @xmath344 is very big and @xmath340 can only take positive values , which gives a zero topological number . @xcite since the bulk and vacuum belong to different topological sectors , the boundary defines a domain wall between the two phases , and there should exists gapless ( chiral ) majorana edge states in the @xmath6 state following the analysis of read and green . @xcite in this sense , the @xmath6-state describes a time - reversal symmetry breaking topological spin liquid . for @xmath345 , @xmath346 . the figure on the right shows the dispersion along the line linking two @xmath347 points of the brillouin zone ( see the blue lines on the left ) . the negative energy part is filled and the excitation is gapless with 6 dirac nodes in the first brillouin zone . the dispersion for the mean field solution with @xmath337 is similar to the above one except that the locations of the nodes are shifted a little bit.,width=316 ] next we consider the specific heat and spin susceptibility for the @xmath7- and @xmath6 states . for the @xmath6-wave pairing states , since the excitations are fully gapped as in bcs superconductors , the specific heat and spin susceptibility will show exponential behavior @xcite at low temperature . for the @xmath7-wave pairing states , both the specific heat and the spin susceptibility show power law behavior due to the dirac cone structure ( see fig . [ fig : f_band ] ) . the energy for the system is @xmath348 in mean field theory . at low temperature , the specific heat is dominate by the low energy excitations near the nodes of the fermi surface , which is given approximately by @xmath349 , where @xmath350 is the fermi velocity ( the anisotropy of the dispersion can be removed by a re - scaling of momentum and will not affect the result qualitatively ) . the magnetic specific heat is given by @xmath351 where @xmath352 is the number of dirac cones . the magnetic susceptibility can be calculated from the linear response theory with @xmath353\nonumber\\\end{aligned}\ ] ] the static susceptibility is given by @xmath354 . the detailed calculation is given in appendix d. recently , a spin-1 material @xmath355 was discovered , where the @xmath3 @xmath356 ions form a triangular lattice with antiferromagnetic ( afm ) interaction . the system has a curie - weiss temperature @xmath357k , and is found to exhibit no conventional magnetic order down to @xmath358k . several plausible ground states have been proposed for this system , @xcite including an antiferro - nematic ( afn ) state @xcite , a ferro - nematic state , @xcite and a kt phase driven by vortices . @xcite spin liquid is also proposed to be a plausible ground state . in our mean field theory , the @xmath7-wave pairing spin liquid has coherent gapless excitation with specific heat scaling as @xmath359 at low temperature which is consistent with the experimental result . @xcite furthermore , the knight shift data in ref . shows that the intrinsic susceptibility decreases with decreasing temperature ( but the very low temperature data are missing . @xcite ) this matches qualitatively with our mean field result eq . ( [ kubo ] ) . we therefore proposed that the @xmath7-wave pairing spin liquid state is a plausible ground state for the material . summarizing , we show in this paper how the fermionic representation for spin @xmath0 systems can be generalized to spin systems with @xmath2 . the symmetry group of the spin operator is @xmath15 group for integer spin and @xmath16 group for half - odd - integer spin . different path integral formulations and mean field theories are developed corresponding to different ways of handling the constraints . in 1d , we show that the particle - hole symmetric mean field theory for @xmath3 and @xmath4 spin chains are consistent with haldane s conjecture , and we argue that the difference reflects a fundamental difference between integer and half - odd - integer spin chains . we also study 2d spin-1 afm on triangular lattice where we find two spin liquid states , one is a gapless @xmath7-wave spin liquid and the other is a topological @xmath6 spin liquid state . we propose that the gapless @xmath7-wave spin liquid is a plausible ground state for the material niga@xmath8s@xmath9 . our approach can be applied to any other spin models and provides a new approach to spin liquid states for @xmath2 spin systems . we note that because of limitation in scope we have addressed only a very limited number of issues in the study of spin - liquid states in this paper . within the mean - field theory we have restricted ourselves to spin liquids solutions of hamiltonian ( [ hamiltonian ] ) with spin rotational symmetry ( 1d ) and lattice translational symmetry ( 1d and 2d ) . we note that solutions which break spin rotational symmetry exist in our theory which may serve as ground states of hamiltonian ( [ hamiltonian])@xcite . for example , we show in ref . that state with long - ranged magnetic order exists as lowest energy state of the mean - field theory of the @xmath274-@xmath275 model and anti - ferro nematic order may exist when the bi - quadratic ( @xmath360 ) spin - spin interaction exists in the hamiltonian . for simplicity we only considered the mean field states without breaking the translational symmetry . we note that states that break translational symmetry ( such as dimerized states ) are believed to be ground states of some 1d or 2d spin models.@xcite another very important issue is whether the spin liquid states we find are stable against gauge fluctuations . in 1-d , the gauge fluctuations can be removed by a time - dependent read - newns gauge transformation and thus have no effect to the low energy properties@xcite . the situation is very different in two or higher dimensions where the stability of the mean - field state depends on dimensionality and the ( gauge ) structure of the gauge field fluctuations . for instance , @xmath149 spin liquid is believed to be stable at 2d@xcite and gapless @xmath81 dirac fermionic spin liquids is stable at 2d in the large-@xmath19 limit@xcite . in our case of spin liquid solutions for the @xmath3 heisenberg model , the @xmath81 gauge fluctuation is gapped by the spinon - pairing term via the anderson - higgs mechanism . the ground state is stable against the low energy gauge fluctuations which is described by an effective @xmath149 gauge theory . we note that the emergence of gauge - field structure is a direct consequences of particle number constraints and a reliable answer to the question of stability of spin liquid states can be obtained only if we can handle the particle number constraint reliably . gauge field theory can only handle long distance , low energy gauge fluctuations and a more satisfactory answer to the question of stability of the spin liquid states can be obtained only after the gutzwiller projection wavefunctions are studied carefully . we thank prof . michael ma for the discussion about 1d models . we also thank prof . a. lee and naoto nagaosa for suggesting the @xmath6-ansatz for afm on triangular lattice , and thank mr . cheung chan and dr . xiao - yong feng for helpful discussions . zxl and tkn are supported by rgc grant of hksar and hkust3/crf/09 . yz is supported by national basic research program of china ( 973 program , no.2011cb605903 ) , the national natural science foundation of china(grant no.11074218 ) and the fundamental research funds for the central universities in china . we shall look for allowed transformations @xmath361 , which keep the spin operator eq . ( [ spin ] ) invariant . at first glance , this condition is satisfied as long as @xmath362 . however , it is _ not true _ because @xmath24 s and @xmath363 s are not independent . by checking each row of @xmath64 directly , it is easy to see that @xmath364 transforms to @xmath365 , and @xmath366 . noticing that @xmath367 , we obtain @xmath368 or @xmath369 on the other hand , @xmath370 transforms to @xmath371 . noticing that @xmath372 and @xmath373 , we obtain @xmath374 or @xmath375 eq . ( [ w1 ] ) and eq . ( [ w2 ] ) impose the condition @xmath376 for integer spin , eq . ( [ w3 ] ) becomes @xmath377 , which results in @xmath378 or @xmath379 , i.e. @xmath380 , where @xmath381 is the usual gauge symmetry group and @xmath382 is the particle - hole symmetry group . notice that the two groups do nt commute , thus the total symmetry group is a semidirect product of the @xmath81 group and the @xmath149 group @xmath383 . for half - odd - integer spin , eq . ( [ w3 ] ) implies @xmath384 , which is equivalent to @xmath385 since @xmath386 . this indicate that @xmath89 . thus the symmetry group is @xmath15 group . first we introduce a @xmath32 matrix @xmath387 , which is essential in our proof : @xmath388 the matrix is the cg coefficients combining two spin - s to a spin - singlet state , i.e. @xmath389 . notice that for integer spin @xmath390 while for half - odd - integer spin @xmath391 . the following are some common properties of the @xmath387 matrix . suppose @xmath392 is an rotation operation , and @xmath393 is its @xmath32 dimensional matrix representation , then we have @xmath394 , and @xmath395 on the other hand , @xmath396 , so @xmath397 comparing these two , we have @xmath398 , or @xmath399 _ this means that @xmath387 is the transformation matrix which transfers the representation matrix of a rotational operator to its complex conjugate . _ in general , @xmath393 can be written as @xmath400 , where @xmath401 are the @xmath15 group parameters ( real numbers ) for the operator @xmath392 , and consequently @xmath402 . from eq . ( [ bdb ] ) we have @xmath403 so we have @xmath404 noticing @xmath405 , above equation is equivalent to @xmath406 now let us consider the spin algebra . the spin interaction is decomposed by the following two operators @xmath407 we define @xmath408 , @xmath409 and @xmath410 , @xmath411 . the following discussions distinguish integer- and half - odd - integer- spins and prove the results in section iii.c . * integer spins * ( @xmath412 ) the second term on the first row of eq . ( [ cd ] ) can be written as @xmath413 in second step , we have used the fact that the transverse of a scalar operator is equivalent to reversing the order of the constituting components . commuting two fermionic operators we obtain a minus sign . on the other hand , @xmath414 , so @xmath211 . the second term on the second row of eq . ( [ cd ] ) is @xmath415 it is obvious that @xmath210 . to simplify eq . ( [ uv ] ) , we will use the following relations : @xmath416 , @xmath417 and @xmath418 . the second term on the first row of eq . ( [ uv ] ) is @xmath419 on the other hand , above operator can be written into another form , @xmath420 . comparing with eq . ( [ uij ] ) we get @xmath421 . the second term on the second row of eq . ( [ uv ] ) is @xmath422 and the relation @xmath423 manifests itself . * half - odd - integer spins * ( @xmath424 ) repeating the above procedures , it is straightforward to show that @xmath425 with @xmath426 . it is known that integer spin antiferromagnetic heisenberg model has free edge states with spin magnitude - s/2 . @xcite majorana fermion edge states also exist in our mean field theory under open boundary condition . our discussion will follow the argument of ref . for 2d pairing fermions . firstly , we point out the existence of two topologically distinct phases under periodic boundary condition and then we discuss the zero energy edge state solution at an open boundary . in our mean field theory , all the properties of the ground state are completely determined by @xmath427 and @xmath428 ( or @xmath340 and @xmath429 ) . noticing that @xmath427 and @xmath428 obey @xmath430 , so they can be viewed as a spinor . this spionor is equivalent to a vector @xmath431 where we have used eq . ( [ ukvk ] ) . obviously , the spinor @xmath432 spans a @xmath433 space . recalling that the first brillouin zone is also @xmath433 , the functions @xmath434 describe a mapping from @xmath433(k space ) to @xmath433(spinor space ) . the mapping degree is characterized by the first homotopy group @xmath435 . when @xmath436 ( which is the case for our mean field theory ) , the map is nontrivial because every point of the spinor @xmath433 is covered once , in other words , the mapping degree is @xmath437 . this topological number defines a phase , we call it a phase . when @xmath438 , the mapping is topologically trivial since @xmath340 is always positive(or negative ) and the lower half circle(or upper circle ) of the spinor @xmath433 is never covered , so the mapping degree is @xmath439 . we call this region b phase . since the mapping can not be smoothly deformed from @xmath439 to @xmath437 , a topological phase transition occurs at @xmath440 . we will see that the existence of zero energy edge states is tied with a phase boundary between the bulk ( a phase ) and the vacuum ( b phase ) . next we consider an infinite chain with a single edge . we assume that the edge is located at @xmath441 , and @xmath442 is the vacuum where the spinon density is zero . the boundary can be described by a potential @xmath443 that is large and positive at @xmath442 so that the wavefunction of the spinon vanishes exponentially at @xmath442 . equivalently , we can assume a position dependent lagrange multiplier term @xmath444 which becomes very large and positive(so that the condition @xmath438 is satisfied ) at @xmath442 . this implies that the vacuum belongs to the b phase and the boundary is a domain wall between a phase and b phase . we now study the edge states in the continuum approximation . it is convenient to introduce @xmath445 such that @xmath446 is positive in the bulk , becomes zero near the edge and turns negative outside the edge . expanding @xmath340 and @xmath429 to first order in @xmath247 near @xmath447 , we obtain @xmath448 and @xmath449 . to see the effects of the domain wall , we consider the bdg equation eq . ( [ bdg - k ] ) , which can be written in position space as @xmath450 where we have replaced @xmath255 and @xmath247 by @xmath451 and @xmath452 , respectively . the bdg equation is compatible with @xmath453 , the spinor @xmath432 satisfying this relation describes majorana fermions . there is a normalizeable zero energy solution for the above equations . putting @xmath454 , then @xmath455 and the solution is @xmath456 notice that the bdg equation eq . ( [ bdg ] ) @xmath457 at @xmath458 admits _ no _ normalizable bound solutions . this shows that the zero mode is the only plausible bound state solution . in the case of a pair of boundaries at @xmath441 and @xmath459 , we can replace the two first order equations by a second order equation , @xmath460(u\pm v),\ ] ] and the energy @xmath255 is not zero for finite @xmath461 . however two bound state solutions with energy going to zero exponentially with increasing @xmath461 exist . the solution @xmath462 is centered at @xmath441 and @xmath463 is centered at @xmath459 . notice that there is only one solution for the above bdg equations , and the two modes @xmath464 are majorana fermions . the above discussion is applicable to our mean field theory for @xmath3 open antiferromagnetic chain . the discussion can be generalized to larger integer spins @xmath465 if we adopt the effective hamiltonian eq . ( [ hm1 ] ) . unlike the @xmath3 case , our @xmath4 mean field theory produces a mapping from @xmath247-space @xmath433 to the @xmath432 space @xmath433 which is topologically trivial ( essentially because of the even parity of the pairing term ) , and the majorana edge states do nt exist anymore . this is in agreement with the result that no exponentially localized edge state exists for open half - odd - integer spin chains . the existence of power - law localized edge states in half - odd - integer spin chains @xcite is more subtle and is not produced by our mean field theory . substituting the bogoliubov transformation eq . ( [ boglbv ] ) into the kubo formula eq . ( [ kubo ] ) , we obtain @xmath466 [ -\mathcal c_{q+k}^2 g_a(q+k,\tau)+\mathcal s_{q+k}^2 g_b(-q - k,-\tau)]\right.\nonumber\\&&\left . -\mathcal c_q\mathcal s_q [ g_a(q,-\tau ) + g_b(-q,\tau)]\times\mathcal c_{q+k}\mathcal s_{q+k } [ g_a(q+k,\tau)+ g_b(-q - k,-\tau)]\right]\nonumber\\ & = & \frac{2}{\beta v}\sum_{q , i\omega } \left[[\mathcal c_q^2 g_a(q , i\omega)-\mathcal s_q^2 g_b(-q ,- i\omega ) ] [ -\mathcal c_{q+k}^2 g_a(q+k , i\omega+i\omega)+\mathcal s_{q+k}^2 g_b(-q - k ,- i\omega - i\omega)]\right.\nonumber\\ & & \left.-\mathcal c_q\mathcal s_q\mathcal c_{q+k}\mathcal s_{q+k } [ g_a(q , i\omega ) + g_b(-q ,- i\omega ) ] [ g_a(q+k , i\omega+i\omega)+ g_b(-q - k ,- i\omega - i\omega)]\right]\nonumber\end{aligned}\ ] ] with the notations @xmath467 and @xmath468 . using the results @xmath469 and @xmath470 , we get the static susceptibility @xmath471\nonumber\\ & = & \lim_{k\to0}\frac{2}{v}\sum_{q}\left[-(\mathcal c_q\mathcal c_{q+k}+\mathcal s_q\mathcal s_{q+k})^2 \frac{n_f(\varepsilon_{q+k})-n_f(\varepsilon_q)}{\varepsilon_{q+k } -\varepsilon_q}\right]\nonumber\\ & = & -2\int d^2q\frac{\partial n_f(\varepsilon_q)}{\partial \varepsilon_q}\end{aligned}\ ] ] which is the pauli susceptibility for free fermions . at low temperature , the excitation spectrum can be approximated by dirac cones , so we have @xmath472 actually , a similar expression like eq . ( [ h3/2 ] ) also holds for @xmath0 , @xmath473:$ ] . notice that the first term in the square bracket has different sign comparing to the expression for @xmath3 and 3/2 . the two terms in the square bracket are also identical . we note that another method to deal with disordered spin states is modified spin - wave theory where zero magnetization is enforced by a magnon number constraint . minoru takahashi [ prog . 87 , 233 ( 1986 ) ; phys . rev . b 40 , 2494 ( 1989 ) ] applied this method to study 1d ferromagnets and 2d antiferromagnets on square lattice . the results agree well with that of bethe ansatz and schwinger boson mean field theory , respectively . this is because that the elementary excitations in both situations are spin-1 magnons which can be viewed as bound states of two spin-1/2 spinons . however , for 1d half - odd - integer antiferromagnetic spin chains , elementary exciations are spin-1/2 spinons . so far we do not know how this modified spin wave - theory can be generalized to deal with this situation . ian affleck , nucl . b 265 , 409 ( 1986 ) ; s. k. yip , phys . rev . 90 , 250402(2003 ) ; j. j. garca - ripoll , m. a. martin - delgado , and j. i. cirac , phys . 93 , 250405 ( 2004 ) ; rajiv r. p. singh , zheng weihong , c. j. hamer , and j. oitmaa , phys . b 60 , 7278 ( 1999 ) .

the proximity profile in the spectra of quasars , where fluxes extend blueward of the he ii ly wavelength , is one of the most important spectral features in the study of the intergalactic medium . based on the _ hst _ spectra of 24 quasars , we find that the majority of them display a proximity profile , corresponding to an ionization radius as large as 20 mpc in the source s rest frame . in comparison with those in the spectra of the quasars at , the proximity effect is more prominent and is observed over a considerably longer period of reionization . the proximity zone sizes decrease at higher redshifts , particularly at . this trend is similar to that for , signaling an onset of reionization at . for quasar sdss1253 + 6817 ( ) , the he ii absorption trough displays a gradual decline and serves as a good case for modeling the reionization . to model such a broad profile requires a quasar radiation field whose energy distribution between 4 and 1 rydberg is considerably harder than normally assumed . the uv continuum of this quasar is indeed exceptionally steep , and the ionization level in the quasar vicinity is higher than the average level in the intergalactic medium . these results are evidence that a very hard euv continuum from this quasar produces a large ionized zone around it . distinct exceptions are the two brightest quasars at , for which no significant proximity profile is present , probably implying that they are very young .

1503.06741

the epoch at redshift @xmath8 marks the peak of quasar formation @xcite as well as the reionization of the intergalactic helium ( * ? ? ? * m09 hereafter ) . as the reionization of singly ionized helium requires high - energy photons above 4 rydberg , the hard uv background radiation field in the vast intergalactic space , commonly referred to as `` metagalactic '' , at @xmath0 is believed to be photons originating from quasars instead of hot stars @xcite . because helium is difficult to ionize and readily recombines , its opacity at this redshift range is considerably higher than that of hydrogen . most of our knowledge of the intergalactic helium at these high redshifts is based upon a few bright quasars with clear lines of sight : qso0302@xmath9003 @xcite , hs1700 + 6416 @xcite and he2347@xmath94342 @xcite . thanks to the sloan digital sky survey ( sdss , * ? ? ? * ) and galaxy evolution explorer @xcite , the number of known quasars with unobscured sightlines to their ly @xmath1 wavelengths has increased dramatically from only three in mid-1990s to more than 50 @xcite . of these quasars the highest redshift is 3.93 @xcite . the cosmic origins spectrograph ( cos , * ? ? ? * ) instrument aboard the _ hst _ has greatly enhanced our ability to probe multiple lines of sight and investigate the absorption features at higher signal - to - noise ( s / n ) ratios . in the five years since the cos installation , more than two dozen quasars have been confirmed with the ly @xmath1 features . the ensemble of spectra of these quasars is now sufficiently large that we can begin to overcome the cosmic variance and investigate the reionization process over a significant range of redshift . it is now well established that the gunn - peterson effect @xcite in quasar spectra becomes significant only at @xmath10 for @xcite and at @xmath11 for @xcite , signaling the end of the respective reionization epochs . the intergalactic - medium ( igm ) reionization process is lengthy , and little is known about the igm evolution at much higher redshifts , _ i.e. _ , @xmath12 for helium , @xmath13 for hydrogen . the ionization properties of the igm change significantly near luminous quasars , whose radiation enhances the ionization level in their vicinity . this `` proximity effect '' was first found in quasars at @xmath0 where the number of ly @xmath1 forest lines in their optical spectra declines in the quasar vicinity @xcite . at a high igm opacity , the proximity effect results in an expected residual flux blueward of the ly @xmath1 wavelength @xcite . such an absorption profile has been observed in the spectra of most quasars at @xmath14 @xcite . for helium , the presence of proximity profiles was ambiguous , as there were only a handful of known quasars , and some of them do not display a proximity profile . the ly @xmath1 absorption troughs may be complex @xcite , as their shapes are affected by several factors : a proximity profile blueward of the ly @xmath1 wavelength , a damped absorption profile that extends to the wavelengths redward of it , and possibly another damped absorption profile from an associated absorber . in this paper , we report our analysis of the proximity effect using a large sample of quasars with confirmed ly @xmath1 features . the distances quoted in this paper are proper distances in the quasar s rest frame , calculated using the codes of @xcite , with the standard cosmological parameter values : @xmath15 , @xmath16 and @xmath17 . we investigate the _ hst _ spectra of 24 quasars in which the ly @xmath1 absorption feature is present at the anticipated wavelength and is not severely contaminated by geocoronal emission . our database consists of three parts : ( 1 ) cos spectra of four quasars at @xmath18 from our program ( go 12249 : pi zheng ) , see table [ tbl-1 ] ; ( 2 ) archival cos spectra of seven quasars ( go 11742 , 13013 : pi worseck ) ; and ( 3 ) published spectra of 13 quasars , taken with cos and stis . six other quasars are excluded because their break is contaminated by geocoronal ly @xmath1 or @xmath19 emission . table [ tbl-2 ] lists the eleven sources in parts 1 and 2 ; and table [ tbl-3 ] lists the 13 sources in part 3 . optical spectra of these quasars are useful as they provide information of the absorption counterparts . we retrieved 15 sdss spectra of the quasars , the vlt / uves spectra of quasars pks1935@xmath9692 and he2347 - 4342 , and the keck / hires spectra of qso0302@xmath9003 and hs1700 + 6416 . in total , we have the optical spectra of 17 quasars , among which four are at high resolution . we carried out _ hst_/cos spectroscopy of four quasars : sdss1253 + 6817 , sdss1319 + 5202 , sdss1711 + 6052 , sdss2346@xmath90016 ( go program 12249 ) . they were selected as a bright sample ( the uv continuum level at @xmath20 @xmath21 ) at @xmath18 with confirmed ly @xmath1 absorption breaks with the acs prism data @xcite . the new cos data provide considerably higher s / n ratios and spectral resolution . the _ hst_/cos observations of these four quasars were performed between 2010 november and 2011 december . the g140l spectra cover a wavelength range of approximately 1100 - 2150 with a pixel scale of 0.08 and resolution @xmath22 . the source fluxes were extracted using a small slit width of 25 pixels , which is considerably smaller than the default size of 57 pixels ( 059 , * ? ? ? * ; * ? ? ? * ) . to determine the level of geocoronal - line contamination and scattered light , we reduced portions of the orbital night separately and compared with the full set . only in one case ( sdss2346@xmath90016 ) the airglow is significant even during the orbital night @xcite , and the dataset taken in 2011 december has to be excluded . the spectrum of sdss1319 + 5202 reveals a lyman - limit system ( lls ) at @xmath23 ( @xmath24 in the observed frame ) , and the source flux below 1550 is considerably suppressed . based on low - resolution prism data ( @xmath25 at 1460 ) , we suggested an associated damped absorption system whose effect extends redward of the ly @xmath1 wavelength @xcite . with the improved wavelength accuracy of our higher - resolution cos spectrum , the absorption profile is confirmed to straddle the ly @xmath1 wavelength and can not be explained by a simple proximity profile . the cos spectra of these four quasars are shown in figure [ fig - cos ] . to study the absorption profiles , it is essential to accurately determine the quasar systemic redshifts . we retrieved the optical spectra of these quasars from the sdss and used the iraf task specfit @xcite to fit both the optical and uv spectra with multiple components . for the cos spectra , a power - law continuum , and a ly @xmath1 emission line were used to fit the data in wavelength redward of the ly @xmath1 break . for the optical spectra , we used a power - law continuum , multiple emission lines , including ly @xmath1 ( narrow + broad ) , , , , , @xcite , and a set of absorption templates that simulate the accumulated igm absorption and the proximity profiles . the systemic redshifts were verified using the @xmath26 emission line as low - ionization lines are believed to be less affected by the local motion of broad - line regions . for our sample , the redshifts are generally consistent with those in the sdss data release seven quasars catalog ( dr7 , * ? ? ? * ) , except for sdss2346@xmath90016 . we obtained a spectrum of sdss2346@xmath90016 using the triplespec instrument @xcite on the arc ( the astrophysical research consortium ) 3.5-m telescope . the observations were made on 2011 november 10 and 16 , under an ambient temperature of 1 - 3 c , and the total integration time was 10 hours . as shown in figure [ fig - nir ] , the [ ] @xmath275007 emission is not visible . the and h@xmath28 emission lines , while unusually weak , yield a redshift consistent with the line . the cos spectra are of sufficient spectral resolution to confirm that that our measurements of the proximity profile are not affected noticeably by the small difference in quasar redshifts . we retrieved the archival cos / g140l spectra of 15 other quasars ( go 11742 and 13013 ) via the mikulski archive for space telescopes and processed with the standard pipeline calcos ( * ? ? ? * v2.19 ) . the difference between the pipeline extraction and our narrow - window extraction ( see [ sub_cos ] ) is mainly in the residual flux level at wavelengths far away from the quasar , but this difference has no effect to our study . for qso0233@xmath90149 , the geocoronal @xmath19 line is close to the ly @xmath1 wavelength , therefore we used the orbital - night portion of the data . we also used the night portion of the qso0916 + 2405 data as geocoronal @xmath29 emission is near its proximity profile . four of these quasars were excluded because of severe contamination by geocoronal emission . four more were excluded as the flux breaks in their spectra are observed at wavelengths considerably longer than that for ly @xmath1 . most likely , these are attributed to partial llss along the lines of sight . the details of the seven archival spectra used are listed in the lower parts of tables [ tbl-1 ] and [ tbl-2 ] , below the details for our prime cos spectra listed at the top of the tables . figure [ fig - cos2 ] plots the spectra of seven quasars that we include in our sample , with contaminating lines flagged with the earth symbols . a number of the cos spectra of quasars have been published @xcite , most of which were observed with the g140l grating . we estimated the proximity sizes in the literature . for most of them , their systemic redshifts are from the sdss catalog . the references for their redshifts are listed in tables 3 and 4 . one stis spectrum of quasar pks1935@xmath9692 is used . the study of proximity profiles is more challenging than that of at @xmath30 as they are more complex . we measured a proximity zone from the ly @xmath1 wavelength to a point where the flux drops to below propagated errors as calculated in a bin of approximately 1.5 . we used these nominal bin sizes and combined the redshift errors to estimate the uncertainties in zone sizes . a difference of 1.5 corresponds to a redshift difference of 0.005 for ly @xmath1 . this definition of proximity zones is based on an assumption of zero continuum flux beyond a proximity zone , as implied by the high opacity at @xmath31 @xcite . an exception is the quasar hs1700 + 6416 ( @xmath32 ) , where the continuum flux level blueward of the ly @xmath1 wavelength is known to be nonzero . in figure 8 of @xcite , the ly @xmath1 absorption region is plotted along with a keck / hires spectrum . we estimated the flux level in two regions where no strong absorption is present : @xmath33 and @xmath34 . the average ly @xmath1opacity there was estimated as @xmath35 . extrapolating this value to @xmath36 using a theoretical scaling relation @xmath37 @xcite , the normalized flux level there should be 0.36 ( @xmath38 ) . this is approximately the flux level around @xmath39 and 2.739 , where no strong absorption features are present , and a potential flux excess near these two redshifts is marginal at best . it is likely that the proximity zone in hs1700 + 6416 ends at the absorption feature of @xmath40 . a generous estimate would mark the zone down to @xmath41 , where there is no significant absorption line . we therefore list the zone size as @xmath42 mpc in table [ tbl-3 ] , to reflect the two possible values of 2 and 6 mpc and other uncertainties . all broad proximity profiles are embedded with strong absorption features . at a relatively high flux , these discrete absorption lines can be easily identified . however , at a low - flux end , they may cut off the residual flux and cause an underestimate of the proximity zone size . when an optical spectrum is available , we checked whether a strong absorption line is present near the wavelength that corresponds to the proximity zone s endpoint . in six cases in tables [ tbl-2 ] and [ tbl-3 ] , no such a absorber is identified within 6 of the endpoint s wavelength in optical spectra ( @xmath43 ) . in the other 11 cases such absorption features are found near the proximity - zone endpoint , and for the other seven quasars no optical spectra are available . therefore the possibility is real that their proximity zones may be underestimated . in these 18 cases we increased the measurement errors by adding a term of 3 mpc ( @xmath44 at @xmath45 ) , approximately the nominal width of a strong absorption feature in g140l spectra . the quasar he2347@xmath94342 does not show a proximity zone , and there is no absorption feature in the uves spectrum near the systemic redshift that can account for a potential lls . given the sample size , it would be informative to study the evolution of proximity zones with redshift and luminosity , which has been reported for at @xmath30 ( [ sub_comp ] ) . we plot the distribution of proximity - zone sizes over redshift and luminosity , respectively , in figures [ fig - z ] and [ fig - lum ] . the redshift scales in the top and bottom panels of figure [ fig - z ] are set to reflect the same ratio of cosmic time . our results demonstrate that the proximity effect is common among quasars , and their proximity - zone sizes can be considerably larger than those in the quasars at @xmath3 . the most significant trend seems to be at @xmath4 , where the zone size decreases towards higher redshift . this trend may be understood in terms of an increasing fraction and signals the onset of the intergalactic reionization . the two most luminous quasars , hs1700 + 6416 and he2347@xmath94342 , are located at lower left in the top panel of figure [ fig - lum ] and clearly disconnected from the rest of the sample . recently a bright quasar at @xmath46 was discovered @xcite : with a similar luminosity and a moderate zone size , this quasar is also at odds with the model prediction that luminous quasars display a larger proximity zone . we ran statistical tests to check potential correlations . the tool is the iraf astronomical survival analysis package , which allows upper or lower limits as inputs @xcite . a moderate anti - correlation between zone sizes and quasar redshifts is confirmed : the spearman correlation coefficient @xmath47 , rejecting a no - correlation hypothesis at 99% level . proximity - zone sizes do not appear to be correlated with quasar luminosities : the spearman correlation coefficient between them is @xmath48 , suggesting a 65% probability for no correlation . even without the two data points at the high - luminosity end , the correlation is still poor . a major source of uncertainties in our measurements of proximity zones is the systemic redshifts of quasars . the sdss quasars have redshift uncertainties on the order of @xmath49 @xcite , which are derived mainly at @xmath50 when narrow emission lines are present . we verified the redshifts of sdss quasars in our sample with @xmath51emission in the optical spectra . emission is a weak feature and consists of multiple components , which carry a statistically weighted average of 1303.49 @xcite . furthermore , @xmath51 emission may be blended with @xmath52 @xcite . assuming a solar abundance , the weighted average for + is 1304.22 . we adopted this value with an uncertainty of 1 , which is comparable to the statistical errors of 5 ( in the observer s frame ) in our fitting to the sdss spectra . combining these terms and converting into the observer s frame for ly @xmath1 , we estimated uncertainties of approximately 2 in the uv spectra of our sample . this term corresponds to @xmath53 , 450 @xmath54 in velocity dispersion , and 1.2 mpc in proper distance at @xmath45 . we verified the redshifts of sdss quasars using fits to @xmath51 emission when possible . in a number of cases , @xmath51 emission is too weak to be useful in redshift estimates , we then used the redshift values from dr7 . for the non - sdss quasars in tables [ tbl-2 ] and [ tbl-3 ] , we assumed a redshift error of @xmath55 @xmath54 in velocity space , unless quoted explicitly . another source of uncertainties in our estimates of the proximity - zone sizes arises from their ending points in the low - flux regions . our data are mainly moderate - resolution spectra with limited s / n levels . for the cos g140l data , this nominal uncertainty is 1.5 , which corresponds to approximately 0.9 mpc in proper distance at @xmath45 and 1.1 mpc at @xmath56 . as discussed in [ sub_res ] , the frequent occurrence of strong absorption features ( column density @xmath57 @xmath58 ) adds a term of uncertainty in many cases . other rare effects , such as an infalling absorber could affect the redshifts , but are unlikely to be thick enough to account for the lack of a proximity effect @xcite . quasar spectra display the signature of numerous igm components , and density fluctuations have a significant impact on the ionization process . llss of column density @xmath59 @xmath58 , both in and low redshift and at high , produce random flux cutoffs . at @xmath60 , a lls may mimic a absorption edge . the probability of encountering such a system per unit redshift is @xmath61 @xcite , which yields a rate that is consistent with other results @xcite . over a wavelength range of 15 ( @xmath62 mpc in proper distance ) near ly @xmath1 , there is less than 2% probability of intercepting a lls of optical depth @xmath63 . this value would be even smaller if we consider this effect only in the region near the proximity - zone s endpoint . for , the igm components have considerably higher column densities and a filtering power . if the ratio of column densities @xmath64 , even a moderate absorber may become a potential lls greatly reduce the number of -ionizing photons from the quasar , reducing both the growth of a ionization zone and the subsequent ionization rate . a simple estimate @xcite suggests that a system becomes opaque ( @xmath65 ) to the ionizing photons at 4 rydberg if its column density @xmath66 @xmath58 . based on the high - resolution data at @xmath67 @xcite , the mean value of @xmath68 is around 80 , which is derived from a majority of weak components ( column density @xmath69 @xmath58 ) . at @xmath70 , the mean @xmath68 value may be around 200 @xcite . however , very high @xmath68 values are mostly associated weak lines and very difficult to confirm , particularly at @xmath71 when the opacity is high . evidence suggests that strong components display lower @xmath68 values : @xcite found @xmath72 at @xmath73 @xmath58 , which are supported by the data in another bright quasar ( @xmath74 at @xmath75 @xmath58 , * ? ? ? adopting a fiducial value of @xmath76 , only the systems of column density @xmath77 @xmath58 may filter out the quasar -ionizing radiation . from table 3 of @xcite the distribution can be expressed as @xmath78 . the probability of encountering systems with @xmath79 @xmath58 within 1.5 to the endpoint of a proximity zone ( see [ sub_res ] ) is estimated as @xmath80 . theoretical models , however , predict high @xmath68 around 100 , based on our knowledge of the quasar s euv ( extreme uv ) continuum , and a re - analysis of the average @xmath68 values at @xmath81 @xcite does not find evidence for the extremely low @xmath68 values noted in previous studies . this recent analysis is based on g140l data , while the previous studies used high - resolution fuse data . to check the reality of low @xmath68 at @xmath82 , we used the published results of qso0302@xmath9003 at a higher redshift . @xcite used a high - resolution keck spectrum and cos / g130 m spectrum to derive @xmath68 values with a bin size of @xmath83 . in the range of @xmath84 there are 21 absorption lines at @xmath85 @xmath58 @xcite . at @xmath86 and @xmath87 there are additional eight absorption lines with restframe equivalent widths greater than 0.3 . from figures 10 - 13 of @xcite we estimated 29 @xmath68 values : twelve of them are lower limits and 17 are measurements . for the twelve lower limits , their mean value is @xmath88 . for the 17 measured @xmath68 values , their mean value is 13 . if lls are common , no broad proximity profiles should have been observed . we analyzed the sdss spectra of our sample and identified 40 strong absorption features in the proximity zones : nearly all see a significant flux at their blueward wavelengths . if the @xmath68 values are around 100 for these absorbers , they should display a sharp flux cutoff . furthermore , in the sdss spectra of five quasars , no strong absorbers are found near the wavelengths that correspond to a proximity - zone endpoint ( tables [ tbl-2 ] and [ tbl-3 ] ) . the upper limit to the column density would be approximately @xmath89 @xmath58 , assuming a minimum equivalent width of 0.3 in the restframe and a nominal velocity dispersion of 30 @xmath54 . if an endpoint of the proximity profiles in these quasars is attributed to an assumed lls , the @xmath68 value may be 8000 or higher , which is not supported by observations . we therefore conclude that , while such a possibility can not be ruled out , it is unlikely that proximity profiles are significantly altered by dense lls . the strmgren radius is generally calculated from a balance of the ionizing photon rate with the total recombination rate @xcite . in a simple model , we assume that the igm is homogeneous around a quasar , and the quasar radiation is isotropic . the total quasar ionizing luminosity in the restframe , in units of photon rate , is calculated as @xmath90 where @xmath91 is the bolometric luminosity distance , @xmath92 is the flux density in the observer s frame and @xmath93 is the observed frequency corresponding to 4 rydberg . the strmgren radius is then : @xmath94 where @xmath1 is the case - a recombination coefficient of ions , @xmath95 the electron density , and @xmath96 the local baryon overdensity in units of the mean igm density , and @xmath97 the proper distance . the mean density of @xmath98 @xmath99 at @xmath100 is derived from its value at @xmath101 @xcite . the value given in equation ( 1 ) is an upper limit ; real ionized zones are considerably smaller for at least two reasons . first , this hypothetical limiting radius grows faster than the ionization front expands due to the decreasing igm density as the universe expands @xcite . secondly , for intergalactic helium , the recombination time scale is so long , @xmath102 gyr , that it exceeds the nominal quasar ages . as a result , the igm around quasars would not reach a balance to form a strmgren zone . if the gas is denser than average ( @xmath103 ) or clumpy , and if the quasar age is old , the recombination rate may become comparable to the ionization rate . a proximity zone does not mean that a strmgren sphere has been established , in the sense of numbers of recombinations balancing numbers of ionizations . this calculation could just indicate that the gas within the sphere is over - ionized by the presence of the quasar radiation compared with that in the general igm . the timing criterion for the ionization fraction to stay in ionization balance is only that the age be longer than ( ionization rate)@xmath104 for highly ionized gas ( m09 ) . assuming an average igm density , the opacity is estimated to be on the order of @xmath105^{3/2}$ ] for hydrogen and @xmath106^{3/2}$ ] for @xcite . @xcite provide a more accurate formula that can be written for the chosen cosmological parameters : @xmath107\ ] ] where @xmath108 is the fraction and @xmath109 is the helium mass fraction . in the vicinity of a luminous quasar , the ionizing radiation field is enhanced over any metagalactic ionizing background radiation , producing a cosmic `` bubble '' that may be observed in the quasar spectrum as a proximity effect . this effect should apply to @xcite as well as @xcite absorption . however , the overall optical depth could be as high as @xmath110 if the intergalactic helium is in a form of singly ionized state and remains poorly constrained at @xmath111 . in the wavelength region 1325 - 1340 ( @xmath112 ) of the spectrum of sdss1253 + 6817 , the effective optical depth is @xmath113 at 95% confidence level . if we used a wider slit width as the calcos pipeline default , @xmath114 at 68% confidence level , or @xmath115 at 95% level . it is therefore possible that the flux extends beyond 12 mpc from the quasar , suggesting that the metagalactic -ionizing radiation field is extremely weak but present , or the quasar has a -ionizing luminosity considerably stronger than the extrapolation from a nominal power law , or the ionization level has not yet recovered after an earlier episode of an enhanced quasar luminosity . the proximity profile is subject to the quasar age . a large zone size requires sufficient ionizing photons from the quasar , even after a long period , if the surrounding intergalactic helium was not fully ionized , and a large number of photons are needed to ionize it . for a young quasar , the proximity profile would display a sharp cutoff . several factors affect the proximity profile : ( 1 ) the igm density and its fluctuations ; ( 2 ) the quasar luminosity ; ( 3 ) the quasar age ; and ( 4 ) the recombination process . for the case when the quasar age is considerably smaller then the recombination time scale , the total number of -ionizing photons over a lifetime @xmath116 is equal to that of the ionized helium atoms in the quasar vicinity : @xmath117 we used the proximity profile in sdss1253 + 6817 to model the evolution of intergalactic helium . while other cases of significant proximity effect exist , the profiles are often shelf - like , possibly due to the duty cycle of the quasar or the transverse proximity effect by other quasars @xcite . the g140l spectrum , after a reddening correction of @xmath118 , @xmath119 , and the extinction curve of @xcite , displays a threefold increase in flux from 1950 to 1400 , which is steeper than that in qso0302@xmath9003 . since no lls break is obvious in this wavelength range , we fitted the dereddened continuum between 1400 and 1900 with a simple power law and a potential partial lls between 2000 and 3000 . the best fit suggests @xmath120 ( @xmath121 ) and a lls of @xmath122 at @xmath123 . if no lls is assumed , the best - fit power law has @xmath124 . both fitting results suggest a similar continuum level of @xmath125 @xmath21at the lyman - limit wavelength of @xmath126 . the estimated total photon rate at 4 rydberg is @xmath127 sec@xmath104 . as shown in the next section , this hard continuum is not strong enough to produce the observed broad proximity profile if the intergalactic helium is not fully ionized . the ionized zones represented by equations ( 1 ) and ( 3 ) are extreme cases ; the real ionization by a quasar is a gradual and slower process , which can be described by an analytical expression derived from a simplified differential equation @xcite for a fixed igm temperature . a more accurate evaluation of the ionization fraction requires simultaneously solving for the gas temperature and the ionization . we computed the time dependence of ionization around a quasar , as outlined in m09 . specifically , we solved the time - dependent spherically symmetric set of coupled ionization rate equations for uniformly distributed hydrogen and helium ( equations 41 and 46 of m09 ) on a radial grid , including the photoelectric attenuation by , he i and of the radiation from the quasar . we solved these equations simultaneously with the photoionization heating and cooling equations as described in m09 ( sec.iii.b ) for a pure hydrogen / helium gas of cosmic abundances . this calculation used the atomic rates in m09 , except for adopting the electron excitation and ionization cooling rate of from @xcite . our calculations assumed case a recombination and included compton cooling off the cosmic microwave background and adiabatic expansion losses in the gas . we considered a sequence of quasar activation redshifts corresponding to -region ages at @xmath128 of 10 , 20 , 30 , 50 , and 100 myr . to match the observed proximity profile , the -ionizing continuum must be very strong : @xmath129 at 228 in the restframe , which is approximately three times of the estimated level from observations . it would imply an euv continuum that is even steeper than qso0302@xmath9003 @xcite . lower - luminosity models , such as a -ionizing continuum extrapolated from the flux at ly @xmath1 wavelength with a nominal power - law index of @xmath130 @xcite were not able to match the size of the ionized region and the flux simultaneously . we emphasize that the measured absorption signal is co - temporaneous with the quasar observed . since the characteristic time to reionize the gas and maintain ionization equilibrium is on the order of millions to tens of millions of years , depending on the distance from the quasar , the luminosity of the quasar that produced the proximity zone may have been larger than the observed value ; we have few constraints on quasar variability on million year timescales . we estimated the black - hole mass as @xmath131 , using the line width , the underlying continuum flux in the sdss spectrum and the formula in @xcite . following the work of @xcite , we derived the rest - frame luminosity near the lyman limit as @xmath132 @xmath133 , which is below the eddington limit of @xmath134 @xmath133 . a -ionizing metagalactic background was not assumed . in our simulations of the ionization zone , a hydrogen - ionizing and he i - ionizing metagalactic background was turned on at @xmath135 . figure [ fig - time ] shows the ionization level , the igm temperature and the normalized observed flux at different epochs after the quasar s birth . shortly after the quasar s birth , a sharp ionization front expands over time , as described by equation ( 3 ) . at each radius and each time interval , the temperature is computed from the energy equation and used to evaluate the atomic rates . helium within the sphere is highly ionized , as the ionizing flux is strong and recombination is not effective . at later epochs , when the ionization front reaches large distances , ionization balance between the geometrically - diluted quasar ionization flux and recombinations produces a more gradual rise in the fraction with distance . as shown in figure [ fig - time ] , a smooth decline of flux may be observed after @xmath136 myr . the high luminosity assumed in this simulation and that in equation ( 3 ) probably represents an upper limit as high - energy photons from the quasar need to ionize a large volume of singly ionized helium . this requirement would be eased if a -ionizing background field existed prior to @xmath137 . as equation ( [ eq : time ] ) shows , the ionizing luminosity may be reduced by a factor of three to @xmath138 sec@xmath104 if the fraction @xmath108 is 30% . in either case , the ionization front is likely moving nearly at the speed of light , as shown in figure [ fig - time ] . @xcite studied the time - retardation effects on the observed proximity - zone size . adapting their equations to , the speed - of - light corrections are important to the observed size until times large compared with @xmath139 myr for an assumed boosted quasar ionizing photon luminosity of @xmath140 , where @xmath141 . the observed ionization zone sizes then increase slowly with time , as @xmath142 . after an expansion time of 50 myr , the ionization front in figure [ fig - time ] is restricted by the speed of light , but the ionization front velocity slows to 0.5c by 100 myr . according to equation ( 3 ) , the age of the observed quasar producing the ionization zone after an expansion time of 50 myr is @xmath143 myr , which provides a minimal age to the quasar . if the igm were highly clumped , the region would expand more slowly , as recombinations would slow it down . the high quasar luminosity used in figure [ fig - time ] is required to match both the position of the ionization front and the ly @xmath1 flux level behind the front assuming a homogeneous igm . if instead the igm opacity is dominated by line - blanketing by the ly @xmath1 forest , the quasar luminosity need not be so high to match the flux level while still matching the position of the front . this would , however , require a somewhat older quasar . the model in figure [ fig - time ] demonstrates the extreme assumptions that must be made for the quasar spectrum to match both the flux level and the position of the front assuming a homogeneous igm . line - blanketing by the ly @xmath1 forest is a more plausible explanation for the flux level @xcite . in principle , a proximity profile should be present in every quasar spectrum , as a cosmic `` bubble '' would exist even without a metagalactic radiation field . at the beginning , ionizing photons do not travel very far , as the mean free path @xmath144 is less than 0.1 mpc , where @xmath145 @xmath146 is the photoionization cross section at 4 rydberg and @xmath147 @xmath148 the number density of singly ionized helium at @xmath149 , is less than 0.1 mpc . high - energy photons from the quasars , however , penetrate considerably deeper into the igm , raising its temperature @xcite . these ionized spheres expanded and eventually overlapped , gradually completing the reionization of intergalactic helium . while it is anticipated that luminous quasars exhibit a broader proximity profile ( equations ( 1 ) and ( 3 ) ) , it is not the case for the two brightest quasars at @xmath150 . the uv spectroscopic properties of hs1700 + 6416 ( @xmath32 , v=16.2 ) have been extensively studied @xcite . its proximity profile is insignificant , extending at most 6 mpc ( see [ sub_res ] ) . the other well - known quasar , he2347@xmath94342 ( @xmath151 , v=16.1 , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) has no proximity effect observed . the redshifts from @xmath271304 and @xmath275007 agree @xcite . there may be a strong infalling absorption system observed in the uv spectrum @xcite . @xcite studied this infalling system and concluded that no satisfactory explanation exists for the absence of a proximity effect around this luminous quasar other than the possibility that it has only recently turned on ( within the last myr ) @xcite and @xcite discovered a proximity zone in the quasar qso0302@xmath9003 at @xmath152 ( v=17.5 ) . blueward of the break , a shelf of residual flux corresponding to an opacity @xmath153 extends out about 15 . the recent cos observations @xcite provide details in the proximity zone . the broad proximity profile , extending up to 15 mpc , may be attributed to a line - of - sight effect from the quasar itself , or to a transverse effect by another quasar q0301@xmath900 ( @xmath154 ) near the line of sight . in the latter case , the age of the ionization zone around q0301@xmath900 should be at least 34 myr . it is interesting to note that the -ionizing continuum is considerably harder ( power - law index @xmath155 ) than previous work suggested : broad proximity profiles appear to be related to a hard quasar continuum . similarly , in pks1935@xmath9692 ( @xmath156 , * ? ? ? * ) , a proximity shelf of flux extends blueward of the break by at least 20 . this absorption feature displays a strong recovery void at 1246.5 , likely produced by the radiation of a foreground source . the reionization of intergalactic hydrogen ended by redshift @xmath30 , as characterized by a sharp increase in opacity at @xmath14 @xcite . this trend is similar to what is observed for helium at @xmath0 , and it is instructive to compare these two major cosmic processes . the number of -ionizing photons of a quasar is considerably higher than that of -ionizing photons . on the other hand , the igm at @xmath30 is denser than that at @xmath157 . the recombination time scale of hydrogen is lower than helium . as a result of these factors , our estimate based on equation ( 1 ) suggests that the proximity profile of intergalactic hydrogen at @xmath3 is somewhat weaker than that for at @xmath0 . @xcite reported proximity profiles of hydrogen in the spectra of 27 quasars ; their `` near zones '' ( proximity ) lie in the range of @xmath158 mpc . they found a significant trend for a decrease in the near - zone size with increasing redshift , as evidence for the evolution of an increasing neutral fraction of intergalactic hydrogen toward higher redshifts . they also found that the near - zone sizes increase with the quasar uv luminosity , as expected for photoionization dominated by quasar radiation . we add eight more data points from recent literature . one quasar is at the highest known redshift @xmath159 @xcite and another is luminous ( @xmath160 , * ? ? ? these two data points carry a significant weight in the correlation analysis . the paper of @xcite does not list the near - zone sizes , and we made estimates from their figure 4 . a spearman test over these 35 data points finds a correlation coefficient @xmath161 between zone sizes and redshifts , rejecting a no - correlation hypothesis at 99.95% level . another test between zone sizes and absolute magnitude finds @xmath162 , rejecting a no - correlation hypothesis at 99.4% level . the luminous quasar j0100 + 2802 displays a proximity zone of 8 mpc . while this is a significant size , it does not scale well to luminosity ( see the lower panel of figure [ fig - lum ] ) . its normalized zone size , @xmath163 at the restframe wavelength of 1450 , is 4 mpc , below the average at the absolute magnitude @xmath164 . this result is actually consistent with the two luminous quasars that they do not show a significant proximity profile . maybe some other factors , such as the density and ionization level in the quasar s vicinity or a young quasar age , play an important role . we utilize a sample of 24 quasars to investigate the interplay between these quasars and the surrounding igm . the ionization zone around quasars is often more prominent than that for at @xmath3 . the large redshift range for these quasars allows us to gain insight into the igm reionization over a long epoch before intergalactic helium became fully ionized . the proximity - zone sizes decline significantly at @xmath4 , and it is likely that helium reionization started well before @xmath165 . in the quasar sdss1253 + 6817 , the source flux extends considerably blueward of the ly @xmath1 wavelength , suggesting a quasar age of @xmath166 myr . the uv flux rises dramatically from 1950 to 1400 , suggesting an exceptionally hard euv continuum . the @xmath68 value in the proximity zone is lower , consistent with such strong -ionizing radiation from the quasar that produces a broad proximity zone . the two brightest quasars do not display a significant proximity profile . while this is at odds with model expectations , we notice that a luminous quasar does not display the largest proximity profile either . it is possible that these hyperluminous quasars are young , or they are surrounded by a dense igm . it should be stressed that the observed proximity - zone size is not a direct measurement of the quasar lifetime , as the structure of the zone is affected by many possible factors , including the igm density fluctuations and quasar variability over igm ionization timescales . a quasar with small proximity zone may be considerably younger compared with the light - travel time across its proximity zone allowing for retardation effects . further understanding of the quasar lifetime awaits improved igm simulations that take these factors into account . support for this research was provided by nasa through grant go-12249 from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 . this work is based on observations made with the nasa / esa hubble space telescope , obtained at the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 and observations obtained with the apache point observatory 3.5-meter telescope , which is owned and operated by the astrophysical research consortium . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the sdss web site is http://www.sdss.org/. this research has made use of the keck observatory archive ( koa ) , which is operated by the w. m. keck observatory and the nasa exoplanet science institute ( nexsci ) , and the data obtained from the eso science archive facility under request number 154374 . anderson , s. f. , hogan , c. j. , williams , b. f. & carswell , r. f. 1999 , , 117 , 56 bajtlik , s. , duncan , r. c. & ostriker , j. p. 1988 , , 327 , 570 becker , g. d. , hewett , p. c. , worseck , g. , & prochaska , j. x. 2013 , mnras , 430 , 2067 becker , r. h. , fan , x. , white , r. l. et al . 2001 , , 122 , 2850 carilli , c. l. , wang , r. , fan , x. et al . 2010 , , 714 , 834 carswell , r. f. , webb , j. k. , baldwin , j. a. & atwood , b. 1987 , , 319 , 709 cen , r. & haiman , z. 2000 , , 542 , 75 dallaglio , a. , wisotzki , l. , & worseck , g. 2008a , , 480 , 359 , 2008b , , 491 , 465 davidsen , a. f. , kriss , g. a. & zheng , w. 1996 , nature 380 , 47 donahue , m. & shull , j. m. 1987 , , 323 , l13 fan , x. , carilli , c. l. , & keating , b. , 2006 , araa , 44 , 415 fardal , m. a. , giroux , m. l. , & shull , j. m. , 1998 , , 115 , 2206 fechner , c. , baade , r. , & reimers , d. 2004 , , 418 , 857 fechner , c. , reimers , d. , kriss , g , a. et al . 2006 , , 455 , 91 feldman , p. d. , davidsen , a. f. , blair , w. p. et al . 1992 , geophy . lett . , 19 , 453 fitzpatrick , e. l. 1999 , pasp , 111 , 63 giroux , m. l. , fardal , a. a. , & shull , j. m. 1995 , , 451 , 477 green , j. c. , froning , c. s. , osterman , s. et al . 2012 , , 744 , 60 gunn , j. & peterson , b. 1965 , , 142 , 1633 haardt , f. & madau , p. 1996 , , 461 , 20 heap , s. r. , williger , g. m. , smette , a. et al . 2000 , , 534 , 69 hewett , p. c. & wild , v. 2010 , , 405 , 2302 hodge , p. e. 2011 , in _ astronomical data analysis software and systems xx _ , eds . i. n. evans , a. accomazzi , d. j. mink , & a. h. rots , ( _ a . s. p. conf . series 442 _ , asp , san francisco ) , 391 hogan , c. j. , anderson , s. f. & rugers , m. h. 1997 , , 113 , 1495 hogg , d. w. 1999 , astro - ph/9905116 isobe , t. , feigelson , e. d. & nelson , p. i. 1986 , apj , 306 , 490 jakobsen , p. , boksenberg , a. , deharveng , j. m. , greenfield , p. , jedrzejewski , r. , & paresce , f. 1994 , nature , 370 , 35 janknecht , e. , reimers , d. , lopez , s. & tytler , d. 2006 , , 458 , 427 kim , t .- s . , carswell , r. f. , cristiani , s. , dodorico , s. , & giallongo , e. 2002 , , 335 , 555 kim , t .- s . , partl , a. m. , carswell , r. f. , & mller , v. 2013 , , 552 , 77 kirkman , d. , tytler , d. , lubin , d. , & charlton , i. 2007 , mnras , 376 , 1227 komatsu , e. , smith , k. m. , dunkley , j. et al . 2011 , , 192 , 18 kriss , g. a. 1994 , in _ astronomical data analysis software and systems iii _ , eds . d. r. crabtree , r. j. hanisch , & j. barnes , ( _ a . s. p. conf . series 61 _ , asp , san francisco ) , 437 kriss , g. a. , shull , j. m. , oegerle , w. et al . 2001 , science , 293 , 1112 madau , p. & meiksin , a. 1994 , , 433 , l53 madau , p. & rees , m. j. 2000 , , 542 , 69 martin , d. c. , fanson , j. , schiminovich , d. et al . 2005 , , 619 , l1 mcquinn , m. , lidz , a. , zaldarriaga , m. , et al . 2009 , , 694 , 842 mcquinn , m. & worseck , g. 2014 , , 440 , 2406 meiksin , a. 2005 , mnras , 356 , 596 2006 , mnras , 365 , 807 2009 , rev . modern phys . , 841 , 1405 ( m09 ) miralda - escud , j. 1998 , , 501 , 15 mortlock , d. j. , warren , s. j. , venemans , b. p. et al . 2011 , nature , 474 , 616 morton , d. c , 1991 , , 77 , 119 murdoch , h. s. , hunstead , r. w. , pettini , m. & blades , j. c. 1986 , , 309 , 19 osterbrock , d. e. & ferland g. j. 2006 astrophysics of gaseous nebulae and active galactic nuclei ( university science books , sausalito , ca ) reimers , d. , khler , s. , wisotzki , l. , groote , d. , rodriguez - pascual , p. & wamsteker , w. 1997 , , 327 , 890 ribaudo , j. , lehner , n. & howk , j. c. 2011 , , 736 , 42 richards , g. t. , strauss , m. a. , fan , x. et al . 2006 , , 131 , 2766 schneider , d. p. , richards , g. t. , hall , p. b. et al . 2010 , , 139 , 2360 scholz , t. t. & walters , h. r. j. 1991 , , 380 , 302 shapiro , p. r. 1986 , , 98 , 1014 shull , j. m. , france , k. , danforth , c. w. et al . 2010 , , 722 , 1312 shull , j. m. , tumlinson , j. giroux , m. l. et al . 2004 , , 600 , 570 smette , a. , heap , s. r. , willinger , g. m. , tripp , t. m. , jenkins , e. b. , & songalia , a. 2002 , , 564 , 542 steinhardt , c. l. & elvis , m. 2011 , , 410 , 201 stengler - larrea , e. a. boksenberg , a. , steidel , c. c. et al . 1995 , apj , 444 , 64 storrie - lombardi , l. j. , mcmahon , r. g. , irwin , m. j. , & hazard , c. 1994 , apj , 427 , l13 strmgren , b. 1939 , , 89 , 526 syphers , d. 2010 , phd dissertation , university of washington syphers , d. , anderson , s. f. , zheng , w. et al . 2009 , , 185 , 20 2009 , , 690 , 1181 2011 , , 726 , 111 2011 , , 742 , 99 2012 , , 143 , 100 syphers , d. & shull , j. m. 2013 , , 765 , 119 2014 , , 784 , 42 telfer , r. , zheng , w. , kriss , g. a. , & davidsen , a. f. 2002 , , 565 , 733 tittley , e. r. , & meiksin , a. 2007 , , 380 , 1369 trainor , r. f. & steidel , c. c. 2012 , , 752 , 39 tytler , d. 1987a , , 321 , 49 1987b , , 321 , 69 vanden berk , d. e. , richards , g. t. , bauer , a. et al . 2001 , , 122 , 549 vron - cetty , m .- p . & vron , p. 2010 , , 518a , 10 venemans , b. p. , baados , e. , decarli , r. et al . 2015 , arxiv 15020.1927 venemans , b. p. , findlay , j. r. , sutherland , w. j. et al . 2013 , , 779 , 24 vestergaard , m. , & peterson , b. 2006 , , 641 , 689 white , r. l. , becker , r. , fan , x. , & strauss , m. a. 2003 , , 126 , 1 wilson , j. c. , henderson , c. p. , herter , t. l. et al . 2004 , in ground - based instrumentation for astronomy , proc . spie 5492 , eds . a. f. m. moorwood & m. iye , 1295 worseck , g. , prochaska , j. x. , hennawi , j. f. & mcquinn , m. 2014 , arxiv 1405.7405 worseck , g. , prochaska , j. x. , mcquinn , m. et al . 2011 , , 733 , 24 wu , x. , wang , f. , fan , x. et al . 2015 , nature , 518 , 512 york , d. g. , adelman , j. , anderson , j. e. jr . 2000 , , 120 , 1579 zheng , w. & davidsen , a. f. 1995 , , 440 , l53 zheng , w. , kriss , g. a. , deharveng , j .- m . et al . 2004 , , 605 , 631 zheng , w. , kriss , g. a. , telfer , r. c. , grimes , j. p. & davidsen , a. d. 1997 , , 475 , 469 zheng , w. , meiksin , a. , pifko , k. et al . 2008 , , 686 , 195 approximately half of all the known quasars were discovered in three acs prism surveys ( go 10907 , 11215 , 11982 : pi anderson ; * ? ? ? * ; * ? ? ? the acs pr130l prism offers a low spectral resolution , @xmath167 between 1360 and 1250 . while the potential merit of these prism spectra is limited , their sample size is significant . it is therefore interesting to explore the proximity profiles in the `` other half '' of the quasar sample . we estimated the proximity profiles in the published prism spectra of 15 quasars at @xmath168 . the spectra of six other quasars at higher redshifts were not used because the spectral resolution degrades significantly redward of 1400 . to understand the effect of spectral deconvolution , we simulated thousands of spectra with different proximity profiles ( @xmath169 ) at @xmath45 and a pixel scale of 0.1 , then convolved with a gaussian kernel appropriate to the wavelength - dependent acs resolution . the input spectra were of simple constant flux , with a proximity profile that starts at 1367 and declines linearly towards zero at the end of the proximity zone . the flux at shorter wavelengths of the proximity zone was set to zero . the noise level was set for an exposure time of 4500 sec , which is typical for these prism observations . as the s / n level in prism spectra is quite high ( @xmath170 ) , our simulation results are not sensitive to noise . each of these simulated spectra was smoothed to a resolution of the pr130l prism and then binned with small wavelength offsets . these offsets reflect the possibility that a spectrum may shift within a prism - data pixel and subsequently affect the deconvolution result . these simulated prism data were deconvolved using the idl task max_likelihood with a point - spread - function kernel of two pixels and then compared with the input data . our measurements of proximity zones in approximately 90% of the simulated spectra were within an error of one prism pixel ( @xmath171 ) , and the others recovered within two pixels . a conservative estimate for the errors in prism data as 7 ( @xmath172 mpc at @xmath100 ) , which are considerably higher . we compared the cos / g140l spectra of quasars sdss1253 + 6817 and sdss2346@xmath90016 with their counterparts in the acs / pr130l prism data . as shown in figure [ fig - prm ] , prism spectra display an extended wing at a level @xmath173 of the unattenuated flux . this may be due to the extended wing of the instrumental line - spread function or potential misalignment of a few individual prism images . we measured the proximity zone to a flux level of 10% level . the proximity profiles in the prism data are consistent with that in the cos / g140l data : sdss1253 + 6817 displays a broad profile and sdss2346@xmath90016 does not . most of the acs prism data yielded no detection of proximity profiles ( @xmath174 mpc ) . in table [ tbl-4 ] we list three quasars whose proximity profiles in the acs prism spectra are larger than 7 . they are at relatively high redshifts , and the data suggest that the decline of proximity - zone size may start at @xmath175 . the study of the proximity effect has been based on high - resolution quasar spectra @xcite , where the numbers of forest lines are compared in the quasar s vicinity and at large distances . the effect of enhanced radiation in a quasar s vicinity may be seen even at a low spectra resolution . @xcite used a flux transmission technique on the vlt / fors2 spectra ( @xmath176 ) of 17 bright quasars at @xmath177 to measure the effective optical depth along the lines of sight and detect a proximity effect . the effective optical depth accumulated from most igm components is estimated as @xmath178 @xcite . approximately half of this term is from those at column density @xmath85 @xmath58 , most of which cases can be identified at the sdss spectral resolution and s / n ( [ sub_res ] ) . from a power - law distribution of @xmath179 where @xmath180 is the column density and @xmath181 the number of individual absorbers @xcite , we estimated approximately six strong absorbers over a range of 10 mpc ( @xmath182 on the sdss wavelength scale ) . these lines affect roughly 60% of the pixels in this bin , assuming that each line affects five sdss spectral pixels . the fluxes in other pixels in this bin are believed to bear the signature of weak igm components . since these lines are unsaturated , their intensity is sensitive to the proximity effect . by measuring the flux decrements in these pixels , we tried to detect the signature of proximity zones . we generated a set of absorption templates @xmath183 , where r is the distance to the quasar and @xmath184 the proximity size . the range of @xmath184 is 5@xmath935 mpc . the fitting task specfit allows such a `` userabs '' component in the form of opacity vs. wavelength . we ran specfit on sdss spectra with the following components : an underlying power law , gaussian emission components of ly @xmath1 ( narrow + broad ) , , , , , , and a set of absorption templates . the fitting windows consists of two parts at the following restframe wavelengths : ( 1 ) 1216@xmath91680 , which cover the red wing of ly @xmath1 and five other emission lines ( [ sub_cos ] ) and ( 2 ) a number of small windows between 1200 and 1216 , where no strong forest line is detected . we ran tests on the sdss spectra two quasars with different proximity zones . for sdss1253 + 6817 , our fitting results suggest a symmetrical ly @xmath1 profile , implying a strong proximity profile . the average opacity @xmath185 over a bin @xmath186 blueward of the ly @xmath1 wavelength . the calculation includes 22 pixels that are not in absorption windows . for sdssj1711 + 6052 , the fitting result suggests that the blue ly @xmath1 wing is weaker than the red wing , as expected from igm absorption . the average optical depth in a similar wavelength range is @xmath187 . the s / n in the second quasar is @xmath188 per pixel in the proximity zone , and the errors in our measurements are dominated by pixel - to - pixel variations . a higher average optical depth with noticeable pixel - to - pixel fluctuations is consistent with a significant structure in the forest lines and a weak proximity effect . these fitting results suggest that the signature of a broad proximity zone may be detected in sdss spectra . lcccccc + sdss1253 + 6817 & @xmath189 & 3.4727 & 19 & @xmath190 & 20 & 12 + sdss2346@xmath90016 & @xmath191 & 3.4895 & 19 & @xmath192 & 6 & 3.5 + sdss1711 + 6052 & @xmath193 & 3.8269 & 8 & @xmath194 & 5 & 2.5 + sdss1319 + 5202 & @xmath195 & 3.8991 & 2 & @xmath196 & @xmath197 & 4.7 + + sdss0936 + 2927 & @xmath198 & 2.9239 & 15 & @xmath199 & 20 & 16 + qso2157 + 2330 & & 15 & @xmath200 & 26 & 19 + sdss1237 + 0126 & & 15 & @xmath201 & 11 & 8 + qso2149@xmath90859 & & 6 & @xmath202 & 4 & 2.7 + qso0233@xmath90149 & & 11 & @xmath203 & 11 & 7.2 + qso0916 + 2405 & & 18 & @xmath204 & 7 & 4 + qso1630 + 0435 & & 23 & @xmath205 & 10 & 5.1 + llcccl hs1700 + 6416 & @xmath206 & @xmath207 & 4 & 3.7 & @xcite + q1216 + 1656 & 2.818 & @xmath190 & 15 & 13 & @xcite + hs1024 + 1849 & 2.8475 & @xmath204 & 21 & 18 & @xcite + 4c57.27 & 2.858 & @xmath208 & 11 & 9 & @xcite + he2347@xmath94342 & @xmath209 & @xmath210 & 0 & 0 & @xcite + sdss1508 + 1654 & 3.1716 & @xmath211 & 18 & 12 & @xcite + pks1935@xmath9692 & 3.185 & @xmath205 & 25 & 17 & @xcite + sdss0856 + 1234 & 3.1948 & @xmath212 & 20 & 14 & @xcite + sdss0955 + 432 & 3.2388 & @xmath213 & 18 & 12 & @xcite + qso0302@xmath9003 & 3.2860 & @xmath205 & 17 & 11 & @xcite + sdss0915 + 4756 & 3.3369 & @xmath214 & 6 & 3.9 & @xcite + sdss2345 + 1512 & 3.5880 & @xmath194 & 5 & 2.8 & @xcite + sdss2257 + 0016 & 3.7721 & @xmath203 & 5 & 2.6 & @xcite +

in a series of papers the authors associated to an-acyclic group an invariant that is a formal difference of polytopes in the vector space . this invariant is in particular defined for most 3-manifold groups , for most 2-generator 1-relator groups and for all free - by - cyclic groups . in most of the above cases the invariant can be viewed as an actual polytope . in this survey paper we will recall the definition of the polytope invariant and we state some of the main properties . we conclude with a list of open problems .

1611.01857

a _ polytope in a finite dimensional real vector space @xmath4 _ is defined as the convex hull of a finite non - empty subset of @xmath4 . given a polytope @xmath5 we denote by @xmath6 the mirror image of @xmath5 in the origin . in the origin is often denoted by @xmath7 . in our paper @xmath7 will have a very different meaning . ] we say that two polytopes @xmath5 and @xmath8 are _ translation - equivalent _ if there exists a vector @xmath9 with @xmath10 . we denote by @xmath11 the set of all translation - equivalence classes of polytopes in @xmath4 . the _ minkowski sum _ of two polytopes @xmath5 and @xmath8 in @xmath4 is defined as the polytope @xmath12 this turns @xmath11 into an abelian monoid , where the identity element @xmath13 is represented by any polytope consisting of a single point . it is straightforward to show , see e.g. ( * ? ? ? * lemma 3.1.8 ) , that @xmath14 has the cancellation property , i.e. for any @xmath15 with @xmath16 we have @xmath17 . we denote by @xmath18 the set of all equivalence classes of pairs @xmath19 where we say that @xmath20 if @xmath21 . note that @xmath18 is an abelian group , and since @xmath14 has the cancellation property it follows that the map @xmath22 given by @xmath23 is a monomorphism . we will use this monomorphism to identify @xmath14 with its image in @xmath18 . given @xmath5 and @xmath24 we may write @xmath25 . we refer to @xmath18 as the _ grothendieck group of polytopes_. now let @xmath1 be a group . we write @xmath26 and we say a polytope in @xmath3 is _ integral _ if all the vertices lie in the image of @xmath27 . similarly we write @xmath28 and we say an element in @xmath29 is _ integral _ if it can be represented by the difference of two integral polytopes . we will define later on will lie in @xmath30 . in fact they will lie in a subgroup given by polytopes with integral vertices . therefore it makes sense to study the polytope group of integral polytopes , which has also been studied in its own sake by funke @xcite . ] a group homomorphism @xmath31 induces natural homomorphisms @xmath32 and @xmath33 . we say that a group @xmath1 is of _ type @xmath34 _ if it admits a finite model for @xmath35 . ( a group of type @xmath34 is well - known to be torsion - free , see e.g. ( * ? ? ? * corollary viii.2.5 ) for a proof . ) we say @xmath1 is _ @xmath0-acyclic _ if all its @xmath0-betti numbers @xmath36 vanish . the following gives examples of @xmath0-acyclic groups of type @xmath34 . fundamental groups of admissible 3-manifolds ( here a @xmath37-manifold is _ admissible _ if it is connected , orientable , irreducible , its boundary is empty or a disjoint union of tori , and its fundamental group is infinite ) . free - by - cyclic groups , i.e. groups that can be written as a semidirect product @xmath38 with @xmath34 a free group . torsion - free groups with two generators and one non - trivial relator . we refer to ( * ? ? ? * theorem 0.1 ) , ( * ? ? ? * section 4.2 ) and @xcite for proofs that these groups are indeed @xmath0-acyclic of type @xmath34 . in this paper we are mostly interested in groups for which the whitehead group is trivial . it is conjectured that the whitehead group is trivial for any torsion - free group . this conjecture has now been proven for large classes of groups . for example the whitehead group is known to be trivial for all torsion - free hyperbolic groups @xcite , all virtually solvable @xcite and the aforementioned three types of groups , see e.g. ( * ? ? ? * ( c.36 ) ) and ( * ? ? ? * and p. 250 ) for details . we refer to ( * ? ? ? * theorem 2 ) for a detailed list of all groups for which it is currently known that the whitehead group is trivial . furthermore , for the most part we want to restrict ourselves to groups that satisfy the atiyah conjecture . a torsion - free group @xmath1 satisfies the atiyah conjecture if given any @xmath39-matrix @xmath40 over @xmath41 $ ] the @xmath0-dimension of the kernel of the map @xmath42 defined by @xmath43 is a natural number . the class of torsion - free groups that are known to satisfy the atiyah conjecture is considerably smaller than the class of torsion - free groups for which it is known that the whitehead group is trivial . we refer to ( * ? ? ? * theorem 2.3 ) for a comprehensive summary of what is known about the atiyah conjecture . for us it is of interest that the atiyah conjecture is known for fundamental groups of most admissible 3-manifolds and for free - by - cyclic groups , see e.g. ( * ? ? ? * ( h.21 ) ) and * theorem 1.5 ) . in the following we refer to a group with trivial whitehead group and which satisfies the atiyah conjecture as a _ @xmath44-group_. it is an open question whether all torsion - free groups are @xmath44-groups . in section [ section : definition - g - pi ] we will use reidemeister torsion over an appropriate skewfield to associate to an @xmath0-acyclic @xmath44-group @xmath1 of type @xmath34 an integral element @xmath2 of @xmath29 . we refer to @xmath2 as the _ polytope invariant of @xmath1_. in section [ section : elementary - examples ] we will see that there exists an @xmath0-acyclic @xmath44-group @xmath45 of type @xmath34 such that neither @xmath46 nor @xmath47 is represented by an actual polytope . the following theorem summarizes some of the structural properties of this invariant . we refer to section [ section : properties - of - p ] for details . [ thm : properties - of - p ] suppose @xmath48 and @xmath49 are @xmath0-acyclic @xmath44-groups of type @xmath34 . let @xmath50 and @xmath51 be monomorphisms . we consider the corresponding amalgamated product @xmath52 and denote by @xmath53 and @xmath54 the monomorphisms from @xmath48 and @xmath49 into @xmath52 . then @xmath52 is also @xmath0-acyclic of type @xmath34 . furthermore , if @xmath52 is a @xmath44-group , then @xmath55 let @xmath56 be an exact sequence of groups of type @xmath34 . suppose @xmath57 is a @xmath44-group and @xmath58 is @xmath0-acyclic . then @xmath59 if @xmath1 is the fundamental group of an aspherical @xmath60-dimensional manifold @xmath61 and if @xmath1 is an @xmath0-acyclic has zero @xmath0-betti numbers if and only if @xmath62 . ] @xmath44-group of type @xmath34 , then @xmath63 the polytope invariant is in general rather difficult to calculate . the following theorem summarizes what is known for special classes of groups . in the interest of readability we keep the language somewhat informal . we refer to section [ section : examples ] for more carefully formulated statements . [ thm : examples ] if @xmath64 is an admissible 3-manifold that is not a closed graph manifold , then @xmath65 equals the dual of the thurston norm ball . in particular @xmath65 is an integral polytope and it determines and is determined by the thurston norm . if @xmath45 is a group that admits a presentation @xmath66 such that @xmath67 is non - empty , reduced and cyclically reduced , then @xmath46 can be easily read off the word @xmath67 . if @xmath68 is a closed admissible 3-manifold that is not a graph manifold and if @xmath68 admits a cw - structure with one 0-cell , two 1-cells , two 2-cells and one 3-cell , then @xmath65 can be determined immediately from the corresponding chain complex of the universal cover . it follows in particular from theorem [ thm : examples ] that for 3-manifolds to which ( 2 ) or ( 3 ) applies one can easily obtain the thurston norm from the fundamental group or the chain complex . next we turn to the question of what information the polytope invariant contains . the following theorem gives a partial answer to that question . it is again formulated in a slightly informal way , the precise statements will be given later in sections [ section : marked - polytopes ] and [ section : polytope - meaning ] . [ thm : polytope - meaning ] if @xmath69 is an admissible 3-manifold that is not a closed graph manifold , then we can mark some of the vertices of the polytope @xmath70 such that a class @xmath71 pairs maximally with a marked vertex if and only if it represents an element in the bns - invariant @xcite @xmath72 . let @xmath45 be a group with @xmath73 that admits a presentation @xmath66 such that @xmath67 is non - empty , reduced and cyclically reduced . then the following hold : the `` thickness '' of @xmath46 is determined and determines the minimal complexity of `` hnn - splittings '' of @xmath45 along groups . we can mark some of the vertices of the polytope @xmath46 such that a class @xmath74 pairs maximally with a marked vertex if and only if it represents an element in the bns - invariant @xmath75 . _ [ funke kielak ] _ if @xmath1 is an descending hnn - extension of a free group on two generators , then the geometry of @xmath2 is closely related to @xmath75 . most of the results in this paper are already explicit or at least implicit in our previous papers @xcite . only theorem [ thm : examples ] ( 3 ) is a completely new statement . + the paper is organized as follows . in section [ section : definition - g - pi ] we will use reidemeister torsion to introduce the polytope invariant @xmath46 for any @xmath0-acyclic @xmath44-group @xmath1 of type @xmath34 . furthermore we prove several statements regarding polytopes and reidemeister torsions that will in particular imply theorem [ thm : properties - of - p ] . in section [ section : examples ] we give a more detailed discussion of the various statements of theorem [ thm : examples ] . in sections [ section : marked - polytopes ] and [ section : polytope - meaning ] we will explain in more detail the statement and the references for theorem [ thm : polytope - meaning ] . we conclude this paper with a long list of questions in section [ section : questions ] . we wish to thank the many people who shared their ideas and insight with us as we developed the subject . we are especially grateful to nathan dunfield , florian funke , dawid kielak and kevin schreve for their contributions . the first author is supported by the sfb 1085 `` higher invariants '' at the university of regensburg , funded by the deutsche forschungsgemeinschaft ( dfg ) . the second author is financially supported by the leibniz - preis granted by the dfg and the erc advanced grant `` kl2mg - interactions '' ( no . 662400 ) granted by the european research council . the third author is partially supported under the australian research council s discovery funding scheme ( project number dp140100158 ) . let @xmath76 be a subring of a ring @xmath77 . the _ division closure _ @xmath78 is the smallest subring of @xmath77 which contains @xmath76 and is division closed , i.e. , any element @xmath79 which is invertible in @xmath77 is already invertible in @xmath80 . the _ rational closure _ @xmath81 is the smallest subring of @xmath77 which contains @xmath76 and is rationally closed , i.e. , for any natural number @xmath60 and matrix @xmath82 if @xmath40 is invertible over @xmath83 then @xmath40 is already invertible over @xmath84 . the division closure and the rational closure always exist . obviously @xmath85 . consider a group @xmath1 . let @xmath86 be the group von neumann algebra which can be identified with the algebra @xmath87 of bounded left @xmath1-equivariant operators @xmath88 . denote by @xmath89 the algebra of operators that are affiliated to the group von neumann algebra , see ( * ? ? ? * section 8) for details . this is the same as the ore localization of @xmath86 with respect to the multiplicatively closed subset of non - zero divisors in @xmath86 , see ( * ? ? ? * theorem 8.22 ( 1 ) ) . by the right regular representation we can embed @xmath90 and hence also @xmath91 as a subring in @xmath86 . we will denote by @xmath92 and @xmath93 the division and the rational closure of @xmath91 in @xmath89 . summarizing we get a commutative diagram of inclusions of rings @xmath94 \ar[d ] & \mathcal{n}(\gamma ) \ar[dd ] \\ \mathcal{d}(\gamma ) \ar[d ] & \\ \mathcal{r}(\gamma ) \ar[r ] & \mathcal{u}(\gamma).}\ ] ] we will use these inclusions to identify each ring with its monomorphic images . the following lemma is well - known to the experts , full references are given in ( * ? ? ? * lemma 1.21 ) . [ lem : l2-acyclic - and - zpi - contractible ] let @xmath95 be a finite chain complex of free left-@xmath91-modules . then the following assertions are equivalent : 1 . the @xmath0-betti numbers of @xmath95 are all zero the @xmath92-chain complex @xmath96 is contractible . [ lem : rational - equals - division - closure ] let @xmath1 be a group . if @xmath1 is torsion - free and if @xmath1 satisfies the atiyah conjecture , then the rational closure @xmath92 agrees with the division closure @xmath93 of @xmath97 . furthermore , @xmath93 is a skew field . this lemma is proved in ( * ? ? ? * lemma 10.39 ) for the atiyah conjecture over @xmath98 with the division closure of @xmath99 $ ] instead of the division closure of @xmath41 $ ] , but the proof verbatim implies lemma [ lem : rational - equals - division - closure ] . throughout this section let @xmath1 be a torsion - free group that satisfies the atiyah conjecture . by lemma [ lem : rational - equals - division - closure ] @xmath100 is a skew field . our goal is to associate to any element in the multiplicative group of units @xmath101 an integral element of @xmath29 . we outline the main steps in the construction and refer to @xcite , where this map was first introduced , for more details . write @xmath102 and view @xmath103 as a multiplicative group . let @xmath104 be the canonical projection and @xmath58 be the kernel of @xmath105 . choose a map of sets @xmath106 with @xmath107 . we denote by @xmath108*_s h$ ] the ring , often referred to as the crossed product of @xmath108 $ ] and @xmath103 , whose underlying abelian group is given by finite formal sums @xmath109 with each @xmath110 $ ] and for which multiplication is extended from the multiplication on @xmath108 $ ] by the rule @xmath111 for @xmath112 and by the rule @xmath113 for @xmath114 and @xmath115 . it is straightforward to see that @xmath108*_s h$ ] is a ring and that @xmath116 & \to & \z[k]*_s h\\ { \mbox{$\textstyle \sum\limits_{g\in \g}^{}$ } } a_g g&\mapsto & { \mbox{$\textstyle \sum\limits_{g\in \g}^{}$ } } a_g gs(\op{pr}(g))^{-1}\cdot \op{pr}(g ) \ea\ ] ] is a ring isomorphism , which we will sometimes use to identify these two rings . given a non - zero element @xmath117*_s h$ ] we consider the integral polytope @xmath118 defining @xmath119 analogously , given any non - zero element @xmath120 we obtain the corresponding polytope @xmath121 . since @xmath122 is a skew field , in particular a domain , we obtain from elementary arguments that for any non - zero @xmath123 we have [ equ : p - multiplicative ] ( fg)= ( f)+(g ) . we denote by @xmath124 the set of all non - zero elements of the domain @xmath119 . we can form the ore localization @xmath125a proof of this fact is for example given in ( * ? ? ? * theorem 6.4 ) or ( * ? ? ? * example 8.16 ) . there exists a canonical isomorphism @xmath126 which we will use to identify these rings , see e.g. ( * ? ? ? * lemma 10.69 ) . ( again the lemma in @xcite is stated for the division closure of @xmath127 $ ] but it also holds with the same proof for the division closure of @xmath128 $ ] . ) now let @xmath129 be non - zero . we define @xmath130 it follows from ( [ equ : p - multiplicative ] ) that this is well - defined and that this map defines a group homomorphism @xmath131 . since the target is abelian this descends to a group homomorphism @xmath132\,\ , \to\,\ , \mfg(\gamma).\ ] ] one easily checks that this homomorphism is independent of the choice of @xmath133 . [ section : definition - of - p(gamma ) ] given a ring @xmath76 we denote by @xmath134 the usual @xmath135-group , as defined in @xcite . if @xmath136 is a skew field , then the dieudonn determinant , see ( * ? ? ? * corollary 4.3 on page 133 ) and @xcite , gives rise to an isomorphism @xmath137.\ ] ] in the following , given a torsion - free group @xmath1 that satisfies the atiyah conjecture we will use this canonical isomorphism to identify @xmath138 with @xmath139 . @xmath0-acyclic pair _ @xmath140 consists of a finite connected cw - complex and a homomorphism @xmath141 such that @xmath142 for all @xmath143 . here @xmath144 denotes the @xmath0-betti numbers of the covering space of @xmath145 corresponding to @xmath146 , viewed as a @xmath1-cw - complex . suppose that _ @xmath147 is an @xmath0-acyclic pair _ and suppose that @xmath1 is torsion - free and that it satisfies the atiyah conjecture . we denote by @xmath148 the universal cover of @xmath145 . by picking orientations of the cells of @xmath145 and by picking lifts of the cells of @xmath145 to @xmath148 we can view @xmath149 as a chain complex of based free @xmath150$]-left modules . similarly we can view @xmath151 } c_*(\wti{x})$ ] as a chain complex of based free @xmath100-left modules . we had assumed that @xmath140 is @xmath0-acyclic . together with lemmas [ lem : l2-acyclic - and - zpi - contractible ] and [ lem : rational - equals - division - closure ] this implies that the chain complex @xmath151 } c_*(\wti{x})$ ] is acyclic . we denote by @xmath152 the corresponding torsion as defined in @xcite . furthermore we define the _ polytope invariant of @xmath140 _ as @xmath153 the minus sign in the definition of @xmath154 might initially be a little surprising . this choice of sign ensures that in many situations of interest , the polytope invariant @xmath154 lies in @xmath155 , i.e. it indeed can be represented by a polytope . the following proposition summarizes some of the key properties of the polytope invariant of an @xmath0-acyclic pair . the proof of the proposition follows from standard properties of reidemeister torsion and the fact that the map @xmath156 is a homomorphism . we refer to ( * ? ? ? * theorem 2.5 ) for details . [ prop : properties - of - p ] let @xmath147 be an @xmath0-acyclic pair and suppose that @xmath1 is a torsion - free group that satisfies the atiyah conjecture . _ ( induction ) _ if @xmath157 is a monomorphism to a torsion - free group @xmath158 that satisfies the atiyah conjecture , then @xmath159 , where @xmath160 is the induced map . _ ( simple homotopy invariance ) _ if @xmath161 is a simple homotopy equivalence of cw - complexes , i.e. if @xmath162 , then @xmath163 . _ ( homeomorphism invariance ) _ if @xmath161 is a homeomorphism , then we have @xmath163 . now we can make the following two definitions . we say that a group @xmath1 is _ @xmath0-acyclic of type @xmath34 _ if it admits a finite @xmath35 and if all its @xmath0-betti numbers vanish . if @xmath1 is a @xmath44-group , then we define @xmath164 where @xmath145 is any finite @xmath35 . it follows from proposition [ prop : properties - of - p ] ( 2 ) that this definition does not depend on the choice of @xmath145 . we refer to @xmath2 as the _ polytope invariant _ of @xmath1 . let @xmath61 be a compact manifold and let @xmath165 be a homomorphism to a group @xmath1 that satisfies the atiyah conjecture . as above we say that @xmath166 is @xmath0-acyclic if @xmath167 for all @xmath168 . now suppose that @xmath166 is @xmath0-acyclic . we pick a cw - structure @xmath145 for @xmath61 and we define @xmath169 . it follows from proposition [ prop : properties - of - p ] ( 3 ) that this definition does not depend on the choice of @xmath145 . sometimes we write @xmath170 . the following proposition collects a few more structural properties of the polytope invariant . these are again a consequence of ( * ? ? ? * theorem 2.5 ) . [ prop : properties - of - p-2 ] let @xmath171 be a decomposition of a finite cw - complex @xmath145 into two connected cw - complexes @xmath40 and @xmath172 such that @xmath173 is also connected . let @xmath174 be a homomorphism to a torsion - free group @xmath1 that satisfies the atiyah conjecture . we denote by @xmath175 , @xmath176 and @xmath177 the inclusion maps . if @xmath178 , @xmath179 and @xmath180 are @xmath0-acyclic , then @xmath140 is also @xmath0-acyclic and @xmath181 let @xmath182 be a fibration of finite cw - complexes . let @xmath183 be a homomorphism to a torsion - free group @xmath1 that satisfies the atiyah conjecture . if @xmath184 is @xmath0-acyclic , then @xmath185 let @xmath61 be an @xmath60-dimensional closed orientable manifold and let @xmath165 be a homomorphism to a group @xmath1 that satisfies the atiyah conjecture . if @xmath166 is @xmath0-acyclic , then @xmath186 it is clear that theorem [ thm : properties - of - p ] is an immediate consequence of propositions [ prop : properties - of - p ] and [ prop : properties - of - p-2 ] and the definitions . we first consider the infinite cyclic group @xmath187 . in this case we take @xmath188 . the cellular chain complex of the universal cover of @xmath189 is then isomorphic to @xmath190 it follows that @xmath191 and therefore @xmath192\in \mfg(\ll t\rr)=\mfg(\r)$ ] . in this case @xmath193 is therefore not represented by a polytope . now consider @xmath194 , where @xmath195 denotes , as usual , the free group on three generators . it follows from the above calculation and from theorem [ thm : properties - of - p ] ( 2 ) that @xmath196 is represented by an interval of length @xmath197 . we also consider @xmath198 . ( strictly speaking we do not know of a proof that @xmath172 satisfies the atiyah conjecture , the following discussion implicitly assumes that this is the case so that we can define @xmath199 . ) similar to the above we see that @xmath199 is represented by minus an interval of length @xmath200 . now let @xmath201 and pick a monomorphism @xmath50 that factors through a monomorphism @xmath202 and similarly pick a monomorphism @xmath203 that factors through a monomorphism @xmath204 . we denote by @xmath1 the corresponding amalgamated product . it is a consequence of theorem [ thm : properties - of - p ] ( 2 ) that @xmath2 is a difference of two intervals and it is straightforward to see that @xmath2 is neither represented by a polytope nor by minus a polytope . in general it is very hard to compute the polytope invariant , the main difficulty lies in determining the dieudonn determinant of a square matrix . there is only one situation in which the calculation of the dieudonn determinant is straightforward , namely when the matrix is a @xmath205-matrix . as we will see , this observation makes it straightforward to determine the polytope invariant for groups with two generators and one relator . we say that a presentation @xmath66 is _ nice _ if it satisfies the following conditions : @xmath67 is a non - empty , reduced and cyclically reduced word , and @xmath206 , where @xmath207 denotes the group defined by the presentation @xmath208 . following @xcite we will associate to a nice presentation @xmath66 a polytope @xmath209 . the definition is illustrated in figure [ fig : brown - intro ] . first we identify @xmath210 with @xmath211 such that @xmath212 corresponds to @xmath213 and @xmath214 corresponds to @xmath215 . then the relator @xmath67 determines a discrete walk on the integer lattice in @xmath216 and the polytope @xmath209 is obtained from the convex hull of the trace of this walk in the following way : start at the origin and walk across @xmath211 reading the word @xmath67 from the left . take the convex hull @xmath217 of the set of all lattice points reached by the walk . an elementary argument , using the fact that @xmath67 is reduced and cyclically reduced , shows that one can take the minkowski difference with the square @xmath218 of length one , i.e. there exists a polytope @xmath209 with @xmath219 . figure [ fig : brown - intro ] illustrates the construction of the polytope for the nice presentation @xmath220 for the final minkowski difference see also figure [ fig : minkowski - sum ] . we now view @xmath209 as an element in @xmath221 , where we identify @xmath211 with @xmath222 by sending @xmath215 to @xmath212 and @xmath215 to @xmath214 . a priori @xmath223 is an invariant of the presentation and not of the underlying group @xmath224 . the following theorem gives another strong indication that the polytope is an invariant of the group . [ thm : polytopes - for-2 - 1-presentations ] let @xmath225 be a nice presentation . if @xmath224 is torsion - free is torsion - free if and only if the word @xmath67 is not a proper power of another word , see e.g. ( * ? ? ? * proposition ii.5.17 ) for details . ] and if @xmath224 is a @xmath44-group , then @xmath224 is @xmath0-acyclic of type @xmath34 and @xmath226 in ( * ? ? ? * theorem 1.3 ) it was shown @xmath209 is an invariant of the underlying group if @xmath224 is residually a group that is elementary - amenable and torsion - free ( without the assumption that @xmath224 satisfies the atiyah conjecture ) . in ( * ? ? ? * conjecture 1.9 ) wise ( see also @xcite ) conjectured that most hyperbolic groups @xmath227 act properly and cocompactly on a @xmath228 cube complex . by agol s theorem @xcite a proof of this conjecture would imply that such groups are virtually special . by ( * theorem 3.8 ) together with ( * ? ? ? * ( h.26 ) ) this would imply that any word hyperbolic torsion - free group with a nice presentation @xmath227 is residually a group that is elementary - amenable and torsion - free . summarizing , a proof of wise s conjecture would show that @xmath209 is an invariant of the underlying group for torsion - free hyperbolic groups with a nice @xmath229-presentation . by ( * ? ? ? * proposition iii.11.1 ) the 2-complex @xmath145 corresponding to the given presentation is aspherical . we denote the universal of @xmath145 by @xmath148 and we write @xmath230 . as usual we view @xmath231 $ ] as a subring of @xmath232 . the chain complex @xmath233 } c_*(\wti{x})$ ] is isomorphic to @xmath234 where @xmath235 and @xmath235 denote the fox derivatives @xcite of the word @xmath67 with respect to the generators @xmath212 and @xmath214 . by @xcite and lemmas [ lem : l2-acyclic - and - zpi - contractible ] and [ lem : rational - equals - division - closure ] this chain complex is acyclic . since @xmath236 it follows from standard arguments , see e.g. ( * ? ? ? * theorem 2.2 ) or ( * ? ? ? * lemma 3.1 ) that the torsion of the chain complex equals @xmath237 . so we obtain @xmath238 } c_*(\wti{x})\big)\big)\,\,=\,\,\pp(r_x\cdot ( y-1)^{-1})\,\,=\,\,\pp(r_x)-\pp(y-1).\ ] ] but by ( * ? ? ? * proposition 3.5 ) the polytope @xmath209 agrees with @xmath239 in @xmath240 . in ( * ? ? ? * section 8) we also deal with presentations of the form @xmath241 where @xmath67 is non - empty , reduced and cyclically reduced , but where @xmath242 . in this case we can also ` naively ' define a polytope ( i.e. an interval ) @xmath209 in @xmath243 and the obvious analogue of theorem [ thm : polytopes - for-2 - 1-presentations ] holds . let @xmath68 be a 3manifold . for each @xmath244 there is a properly embedded oriented surface @xmath245 , such that @xmath246\in h_2(n , \partial n ; \mathbb{z})$ ] is the poincar dual to @xmath247 . letting @xmath248 , where @xmath249 are the connected components of @xmath245 , the _ thurston norm _ of @xmath244 is defined as @xmath250 ) = \phi$ } \big\}.\ ] ] thurston @xcite showed that @xmath251 is a seminorm on @xmath252 and elementary arguments show that @xmath251 extends to a seminorm @xmath251 on @xmath253 . we denote by @xmath254 the dual of the unit norm ball of @xmath251 . thurston @xcite showed that @xmath255 is a polytope with integral vertices . the following theorem is ( * ? ? ? * theorem 3.35 ) . the proof relies on the recent work of agol @xcite , przytycki - wise @xcite and wise @xcite . [ thm : polytopes - agree - for-3-manifold ] for any admissible 3-manifold @xmath69 that is not a closed graph manifold we have @xmath256 the combination of theorems [ thm : polytopes - for-2 - 1-presentations ] and [ thm : polytopes - agree - for-3-manifold ] gives us the following corollary . this corollary was first proved in @xcite using a different approach . [ cor : thurstonnorm-2 - 1 ] let @xmath68 be an admissible 3-manifold such that @xmath257 admits a nice presentation @xmath66 . then @xmath258 the following theorem says that for `` small '' closed 3-manifolds one can easily obtain the thurston norm from the chain complex of the universal cover . this theorem can be viewed as a version of corollary [ cor : thurstonnorm-2 - 1 ] for closed 3-manifolds . it applies , in particular , to all closed admissible 3manifolds of heegaard genus equal to two . it is known through work of kobayashi @xcite and hempel @xcite in combination with perelman s solution of the geometrisation conjecture that if the splitting distance of a genus two heegaard splitting of a 3-manifold is larger than two , then the 3-manifold is hyperbolic ( whence admissible ) . maher @xcite used this fact to show that _ most _ 3manifolds of heegaard genus two are hyperbolic . [ thm:3-manifold - with - two - generators ] let @xmath68 be a closed admissible 3-manifold that is not a graph manifold . we write @xmath259 . suppose @xmath68 admits a cw - structure with one 0-cell , two 1-cells , two 2-cells and one 3-cell . the corresponding cellular chain complex of the universal cover @xmath260 is of the form @xmath261\xrightarrow{\scriptsize{\bp c_{1}&c_2\ep } } \z[\pi]^2\xrightarrow{\scriptsize{\bp b_{11}&b_{12}\\ b_{21}&b_{22}\ep } } \z[\pi]^2\xrightarrow{\scriptsize{\bp a_1\\ a_2\ep}}\z[\pi]\to 0.\ ] ] then there exist @xmath262 with @xmath263 and @xmath264 , and @xmath265 as we pointed out earlier , the chain complex @xmath266 } c_*(\wti{n})$ ] is acyclic . it thus follows that there exist @xmath262 with @xmath263 and @xmath264 . since @xmath267 is a skew field one can show , similar to ( * ? ? ? * theorem 2.2 ) , that @xmath268 this implies that @xmath269 . the theorem now follows from theorem [ thm : polytopes - agree - for-3-manifold ] . a _ marked polytope _ is a pair @xmath270 , where @xmath5 is a polytope and @xmath271 is a ( possibly empty ) subset of the set of vertices of @xmath5 . we refer to the vertices in @xmath271 as the _ marked vertices_. if @xmath272 and @xmath273 are two marked polytopes , then we define the _ @xmath274marked@xmath275 minkowski sum @xmath276 of @xmath272 and @xmath273 _ as the minkowski sum of the corresponding polytopes , where the marked vertices of the minkowski sum are precisely those that are the sum of a marked vertex of @xmath272 and a marked vertex of @xmath273 . an example is given in figure [ fig : summp ] , where the marked vertices are indicated by a dot . marked polytopes appear naturally in many contexts : let @xmath68 be a 3-manifold . a class @xmath277 is called _ fibered _ if it can be represented by a non - degenerate closed 1-form . by @xcite an integral class @xmath278 is fibered if and only if there exists a locally trivial fiber bundle @xmath279 such that @xmath280 . thurston @xcite showed that we can turn @xmath255 into a marked polytope @xmath281 which has the property that @xmath282 is fibered if and only if it pairs maximally with a marked vertex . this means that there exists a marked vertex @xmath283 of @xmath281 , such that @xmath284 for any vertex @xmath285 in the underlying polytope @xmath255 . let @xmath1 be a torsion - free group . we will now see that we can associate a marked polytope to a non - zero element in @xmath231 $ ] . ( in our subsequent discussion of this assignment we will use the notation introduced in section [ section : the_polytope_homomorphism ] . ) we recall that we have an identification @xmath41= \z[k]*_s h$ ] , where @xmath286 and @xmath287 . given a non - zero element @xmath288 we consider the marked polytope @xmath289 which is given by the polytope @xmath290 where we mark a vertex @xmath291 of @xmath121 if @xmath292 for some @xmath115 . if @xmath231 $ ] is a domain , then for any non - zero @xmath293 $ ] we have @xmath294 . let @xmath66 be a nice presentation such that @xmath224 is torsion - free . we showed in ( * ? ? ? * proposition 3.5 ) that there exists a unique marked polytope @xmath295 with @xmath296 . the set of marked polytopes in a vector space forms a monoid under minkowski sum . as illustrated in figure [ fig : summp - not - cancel ] this monoid does _ not _ have the cancellation property . this implies that the monoid of marked polytopes does _ not _ inject into the corresponding grothendieck group . therefore we can not associate a meaningful notion of a difference of marked polytopes to an @xmath0-acyclic group @xmath208 of type @xmath34 at this point of time . in section [ section : definition - of - p(gamma ) ] we associated to an @xmath0-acyclic @xmath44-group @xmath1 of type @xmath34 its polytope invariant @xmath297 . in this section we want to determine information encoded in this invariant . given a polytope @xmath5 in a real vector space @xmath4 and given @xmath298 we refer to @xmath299 as the _ thickness of @xmath5 in the @xmath247-direction_. since thickness is translation invariant we can also define @xmath300 for any @xmath301 and since thickness is additive under minkowski sum we can also define @xmath302 for any @xmath303 . given a polytope @xmath5 in @xmath4 we refer to @xmath304 as the _ symmetrization of @xmath5_. this definition extends to a symmetrization map on @xmath14 and @xmath18 . it is clear that the thickness of @xmath5 depends only on the symmetrization of @xmath5 . the following theorem is now a straightforward consequence of theorem [ thm : polytopes - agree - for-3-manifold ] . [ thm : thickness - equals - thurston - norm ] for any admissible 3-manifold @xmath69 that is not a closed graph manifold and any @xmath277 we have @xmath305 ideally one would like to generalize the statement of theorem [ thm : thickness - equals - thurston - norm ] to larger classes of groups . the first problem that arises is that there is no satisfactory purely group theoretic definition of the thurston norm . we refer to @xcite for several ideas and approaches . in an attempt to generalize theorem [ thm : thickness - equals - thurston - norm ] we will work with the notion of a splitting of a group . let @xmath1 be a finitely presented group and let @xmath306 be an epimorphism . let @xmath172 be a finitely generated group . a _ splitting of @xmath307 over @xmath172 _ is an isomorphism @xmath308 such that the following hold : @xmath40 is finitely generated , @xmath172 is a subgroup of @xmath40 and @xmath309 is a monomorphism , @xmath310 for @xmath311 and @xmath312 . it is well - known , see e.g. @xcite or ( * ? ? ? * theorem b * ) , that any such pair @xmath307 admits a splitting over a finitely generated group . we define the _ splitting complexity of @xmath307 _ as @xmath313 where @xmath314 is defined as the minimal number of generators of @xmath172 . [ thm : splittings ] let @xmath1 be a torsion - free group that admits a nice presentation @xmath227 . furthermore suppose that @xmath1 satisfies the atiyah conjecture . then for any epimorphism @xmath306 we have @xmath315 in @xcite it was shown that the statement of the theorem also holds if @xmath208 is the fundamental group of a knot complement . but the equality of theorem [ thm : splittings ] does not hold for the fundamental groups of all 3-manifolds . for example , if @xmath68 is the 3-torus , then @xmath316 and it follows from theorem [ thm : properties - of - p ] and the discussion in section [ section : elementary - examples ] that @xmath70 is a point . in particular the thickness for each @xmath247 is zero . on the other hand it is straightforward to see that for any epimorphism @xmath317 we have @xmath318 . the inequality @xmath319 was shown in ( * ? ? ? * proposition 7.6 ) if @xmath1 is residually a group that is elementary - amenable and torsion - free ( without the assumption that @xmath1 satisfies the atiyah conjecture ) . as we remarked after theorem [ thm : polytopes - for-2 - 1-presentations ] , potentially most torsion - free groups with a nice presentation satisfy this condition . the inequality @xmath320 was shown in ( * ? ? ? * proposition 7.3 ) . as was suggested to us by nathan dunfield , the proof of this inequality follows from a careful reading of the first step in the proof of the freiheitssatz . the proof of the reverse inequality follows along the same lines as the proof ( * ? ? ? * proposition 7.6 ) but one needs to replace the ore localizations of the elementary - amenable and torsion - free groups by @xmath100 . more precisely , it follows easily from the combination of theorem 3.6(4 ) and lemmas 4.3 , 6.12 and 6.13 of @xcite together with a slight variation on theorem 4.1 of @xcite . we leave the details to the reader . let @xmath1 be a finitely generated group . the bieri neumann strebel @xcite invariant @xmath321 of @xmath1 is by definition a subset of @xmath322 . we refer to @xcite for the precise definition , but in order to give a flavour of the invariant we recall three properties : an epimorphism @xmath323 represents an element in @xmath321 if and only if it corresponds to an _ ascending hnn - extension_. to be precise : if and only if there exists an isomorphism @xmath324 where @xmath40 is a finitely generated group and @xmath325 is a monomorphism , such that @xmath247 corresponds under @xmath326 to the epimorphism given by @xmath327 and @xmath328 for all @xmath311 . a non - trivial homomorphism @xmath323 has the property that @xmath247 and @xmath329 represent elements in @xmath321 if and only if @xmath330 is finitely generated . @xmath75 is an open subset of @xmath331 . the first two properties follow from ( * ? ? ? * proposition 4.3 ) ( see also ( * ? ? ? * corollary 3.2 ) ) and the third one is ( * ? ? ? * theorem a ) . it is shown in ( * ? ? ? * theorem e ) that given a 3-manifold @xmath68 the fibered classes of @xmath332 correspond precisely to the classes that lie in @xmath72 . the following is now a reformulation of the statement of the first example in section [ section : marked - polytopes ] . let @xmath69 be an admissible 3-manifold that is not a closed graph manifold . then the polytope @xmath65 admits a marking with the property that for any non - trivial @xmath333 we have @xmath334\in \sigma(\pi_1(n))\quad \longleftrightarrow \quad \phi\mbox { pairs maximally with a marked vertex of $ \pp(n)$.}\ ] ] for groups with a nice presentation @xmath227 we obtain a similar theorem . [ mainthm ] let @xmath225 be a nice presentation . a non - trivial class @xmath335 represents an element in @xmath336 if and only if @xmath247 pairs maximally with a marked vertex of @xmath295 . the theorem is stated as ( * ? ? ? * theorem 1.1 ) . in that paper the theorem is proved using the generalised novikov rings of sikorav @xcite . it can also be viewed as a reformulation of brown s algorithm ( * ? ? ? * theorem 4.3 ) . finally a relationship between the polytope invariant and the bns - invariant was also found for a different class of groups by funke kielak @xcite . in order to state their theorem we need to introduce a few more definitions . let @xmath337 be a monomorphism of a group @xmath208 . we denote by @xmath338 the corresponding descending hnn - extension . we refer to the epimorphism @xmath339 given by @xmath340 for all @xmath341 and @xmath342 as the _ canonical epimorphism_. let @xmath5 be a polytope in a vector space @xmath4 and let @xmath343 . following @xcite we refer to @xmath344 as the _ @xmath247-minimal face of @xmath5_. let @xmath4 be a vector space and let @xmath345-[\qq]\in \mfg(v)$ ] . following @xcite we say that @xmath346 are _ @xmath347-equivalent _ if @xmath348 and @xmath349 . note that this definition is independent of the choice of @xmath5 and @xmath8 . the following is the main result of @xcite . [ thm : funke - kielak ] let @xmath350 be a monomorphism of the free group on two generators . let @xmath351 be a homomorphism such that the class @xmath352\in s(\gamma)$ ] is not represented by the canonical epimorphism . then there exists an open neighborhood @xmath353 of @xmath352 $ ] in @xmath331 such that for every non - trivial @xmath354 with @xmath355\in u$ ] that is @xmath2-equivalent to @xmath247 we have @xmath356\in \sigma(\gamma)\quad \longleftrightarrow \quad [ -\psi]\in \sigma(\gamma).\ ] ] we conclude this survey with a long list of questions and conjectures . before we state our next question we recall that the thurston norm of an admissible hyperbolic 3-manifold @xmath68 is a norm . ( this is a direct consequence of the fact that hyperbolic admissible 3-manifolds are atoroidal , i.e. any embedded torus is boundary parallel , see e.g. ( * ? ? ? * proposition d.3.2.8 ) . ) this implies that if @xmath68 is an admissible hyperbolic 3-manifold with @xmath358 , then @xmath359 is not a point . [ qu:2-dimensional - k(pi,1)-hyperbolic ] let @xmath357 be an @xmath0-acyclic @xmath44-group of type @xmath34 that admits a 2-dimensional @xmath35 . furthermore suppose that @xmath1 is a non - elementary hyperbolic group and that @xmath360 . does it follow that @xmath361 ? it is known that the bns invariants of metabelian and 3manifold groups are polyhedral . an affirmative ( even partial ) answer to question [ question : polytope - bns ] may establish polyhedrality of bns invariants for new families of groups , as well as whether the polyhedra are rational or not . we can therefore also ask the following question . let @xmath1 be a hyperbolic @xmath0-acyclic group of type @xmath34 with a 2-dimensional @xmath35 . then there exists an integral marked polytope @xmath364 such that a non - trivial @xmath365 represents an element in @xmath75 if and only if @xmath247 pairs maximally with a marked vertex of @xmath272 . the combination of theorem 3.6(4 ) and lemmas 4.3 , 6.12 and 6.13 of @xcite shows that if @xmath1 is an @xmath0-acyclic @xmath44-group of type @xmath34 , then for any epimorphism @xmath366 we have @xmath367 in section [ section : thickness ] we saw that for 3-manifold groups and many two - generator one - relator groups the thickness of the polytope invariant is related to the complexity of splittings of the underlying group . one partial answer is given by the discussion of the bns - invariant . this invariant is in general not symmetric and if the polytope invariant is related to the bns - invariant it can not be symmetric in general . nonetheless , there are many groups with empty bns - invariant and non - symmetric polytope invariant . finally we want to discuss which elements of @xmath369 can be realized as the polytope invariant of a manifold . here we say that @xmath370 _ can be realized by a @xmath371-dimensional manifold _ if there exists a pair @xmath372 where @xmath68 is a closed @xmath0-acyclic @xmath371-dimensional manifold @xmath68 such that @xmath257 is a @xmath44-group and where @xmath373 is an isomorphism , such that @xmath374 . from our above results we know that not all @xmath370 can be realized by manifolds . more precisely , in proposition [ prop : properties - of - p-2 ] we showed that polytopes realized by closed orientable manifolds have a symmetry . furthermore in theorem [ thm : polytopes - agree - for-3-manifold ] we showed that the polytope invariant of an admissible 3-manifold @xmath69 that is not a closed graph manifold can be represented by a polytope . so we can only hope to realize elements of @xmath375 by 3-dimensional manifolds . let @xmath34 be a free group on @xmath380 generators . we set @xmath381 . then @xmath35 is 2-dimensional and it is @xmath0-acyclic by ( * ? ? ? * theorem 1.35 ( 4 ) ) . furthermore @xmath1 is a @xmath44-group by ( * ? ? ? * theorem 2 ) and ( * ? ? ? * theorem 2.3 ) . we pick an isomorphism @xmath382 and we pick a set - theoretic section @xmath383 of the projection map @xmath384 . now let @xmath370 . we can find @xmath385 $ ] with @xmath386 . we write @xmath387 it follows immediately from the definitions that @xmath388 . now let @xmath378 . it follows easily from a slight generalization of ( * ? ? ? * lemma 2.9 ) that there exists a closed orientable @xmath371-dimensional manifold @xmath68 with @xmath389 such that @xmath390 it follows immediately that @xmath391 . the first question seems to be very hard since by theorem [ thm : polytopes - agree - for-3-manifold ] it is a reformulation of the question of which thurston norm balls are realized by aspherical 3-manifolds . this question has been open since the 1970 s . the second question might be more accessible , especially in dimensions 5 and 6 . it is conceivable that in dimension 4 the answer depends on whether one studies topological or smooth manifolds . 10 i. agol , _ criteria for virtual fibering _ , j. topol . 1 ( 2008 ) 2 , 269284 . i. agol , _ the virtual haken conjecture _ , with an appendix by i. agol , d. groves and j. manning , documenta math . 18 ( 2013 ) , 10451087 . m. aschenbrenner , s. friedl and h. wilton , _ 3-manifold groups _ , ems series of lectures in mathematics ( 2015 ) a. bartels , w. lck and h. reich , _ equivariant covers for hyperbolic groups _ , 12 ( 2008 ) , 17991882 . r. benedetti and c. petronio , _ lectures on hyperbolic geometry _ , universitext , springer , berlin , 1992 . r. bieri , w. d. neumann and r. strebel , _ a geometric invariant of discrete groups _ , invent . ( 1987 ) , 451 - 477 . r. bieri and r. strebel , _ almost finitely presented soluble groups _ , comment . helv . 53 ( 1978 ) , 258278 . k. s. brown , _ cohomology of groups _ , graduate texts in mathematics 87 , springer , 1982 . k. s. brown , _ trees , valuations , and the bieri - neumann - strebel invariant _ , invent . ( 1987 ) , 479 - 504 . j. button and r. kropholler , _ non hyperbolic free - by - cyclic and one - relator groups _ , preprint ( 2015 ) w. dicks and p. linnell , _ @xmath0-betti numbers of one - relator groups _ , math . 337 , no . 4 ( 2007 ) , 855874 . j. dodziuk , p. linnell , v. mathai , t. schick and s. yates , _ approximating @xmath0-invariants , and the atiyah conjecture _ , preprint series sfb 478 mnster , germany . communications on pure and applied mathematics , vol . 56 , no . 7 ( 2003 ) , 839 - 873 . j. dubois , s. friedl and w. lck , _ the @xmath0-alexander torsion of 3-manifolds _ , j. topol . 9 , no . 3 ( 2016 ) , 889926 . r. h. fox , _ free differential calculus i , derivation in the free group ring _ , ann . math . 57 ( 1953 ) , 547560 . s. friedl , _ reidemeister torsion , the thurston norm and harvey s invariants _ , pac . j. math . 230 ( 2007 ) , 271296 . s. friedl , _ novikov homology and noncommutative alexander polynomials _ , preprint ( 2015 ) , to be published the tim cochran memorial volume of the j. knot . thy . ram . s. friedl and w. lck , _ @xmath0-euler characteristics and the thurston norm _ , preprint ( 2016 ) , arxiv:1609.07805 . s. friedl and w. lck , _ universal @xmath0-torsion , polytopes and applications to @xmath37-manifolds _ , preprint ( 2016 ) , arxiv:1609.07809 . s. friedl , k. schreve and s. tillmann , _ thurston norm via fox calculus _ , preprint ( 2015 ) , arxiv:1507.05660 . s. friedl , d. silver and s. williams , _ splittings of knot groups _ , mathematische annalen 362 ( 2015 ) , 401424 . s. friedl , d. silver and s. williams , _ the turaev and thurston norms _ , pac . j. math . 284 ( 2016 ) , 365382 . s. friedl and s. tillmann , _ two - generator one - relator groups and marked polytopes _ , preprint ( 2015 ) , arxiv:1501.03489 . f. funke , _ the integral polytope group _ , preprint ( 2016 ) , arxiv:1605.01217 . f. funke and d. kielak , _ alexander and thurston norms , and the bieri - neumann - strebel invariants for free - by - cyclic groups _ , preprint ( 2016 ) , arxiv:1605.09067 . j. hempel , _ 3-manifolds as viewed from the curve complex _ , topology 40 ( 2001 ) , no . 3 , 631657 . h. kammeyer , w. lck and h. rping , _ the farrell - jones conjecture for arbitrary lattices in virtually connected lie groups _ , geometry and topology 20 ( 2016 ) , 12751287 . t. kobayashi , _ casson - gordon s rectangle condition of heegaard diagrams and incompressible tori in 3-manifolds _ , osaka j. math . 25 ( 1988 ) , no . 3 , 553573 . p. a. linnell , _ division rings and group von neumann algebras _ , forum math . 5 ( 1993 ) , vol . 6 , 561576 . p. linnell and w. lck , _ localization , whitehead groups , and the atiyah conjecture _ , preprint ( 2016 ) j. lott and w. lck , _ @xmath0-topological invariants of 3-manifolds _ , invent . ( 1995 ) , no . 1 , 1560 . w. lck , _ @xmath0-invariants : theory and applications to geometry and k - theory _ , ergebnisse der mathematik und ihrer grenzgebiete , 3 . folge , vol . 44 , springer , berlin , 2002 . r. lyndon and p. schupp , _ combinatorial group theory _ , ergebnisse der mathematik und ihrer grenzgebiete , band 89 . springer - verlag , berlin - new york , 1977 . j. maher , _ random heegaard splittings _ , journal of topology ( 2010 ) 3 ( 4 ) , 9971025 . j. milnor , _ whitehead torsion _ , 72 ( 1966 ) , 358426 . p. przytycki and d. wise , _ mixed 3manifolds are virtually special _ , preprint ( 2012 ) . j. rosenberg , _ algebraic k -theory and its applications _ , graduate texts in mathematics 147 , springer , new york , 1994 . r. schneider , _ convex bodies : the brunn - minkowski theory _ , cambridge univ . press ( 1993 ) j.c . sikorav , _ homologie de novikov associe une classe de cohomologie relle de degr un _ , thse orsay , 1987 . j. silvester , _ introduction to algebraic @xmath395-theory _ , chapman and hall mathematics series ( 1981 ) i. m. singer , _ some remarks on operator theory and index theory _ , in k - theory and operator algebras ( proc . georgia , athens , ga . , 1975 ) , pages 128138 . lecture notes in math . , vol . 575 . springer - verlag , berlin , 1977 . r. strebel , _ finitely presented soluble groups _ , in group theory : essays for philip hall , academic press , london , 1984 . thurston , _ a norm for the homology of 3manifolds _ , mem . 59 , no . 339 ( 1986 ) , d. tischler , _ on fibering certain foliated manifolds over @xmath396 _ , topology 9 ( 1970 ) , 153154 . v. turaev , _ introduction to combinatorial torsions _ , lectures in mathematics , eth zrich ( 2001 ) f. waldhausen , _ algebraic k - theory of generalized free products ii _ , ann . of math . 108 ( 1978 ) , 135256 . c. wegner , _ the farrell - jones conjecture for virtually solvable groups _ , j. topol . 8 ( 2015 ) , 9751016 . d. wise , _ the structure of groups with a quasi - convex hierarchy _ , electronic res . ( 2009 ) , 4455 . d. wise , _ the structure of groups with a quasi - convex hierarchy _ , 189 pages , preprint ( 2012 ) , downloaded on october 29 , 2012 from + ` http://www.math.mcgill.ca/wise/papers.html ` d. wise , _ from riches to raags : 3manifolds , right angled artin groups , and cubical geometry _ , cbms regional conference series in mathematics , 2012 . d. wise , _ cubular tubular groups _ , trans . 366 ( 2014 ) , 55035521 .

with advances in fabrication technologies it is now possible to create precisely controlled geometries and pinning landscapes for vortex matter in type - ii superconductors . here we use numerical simulations to examine vortex states and dynamics in periodic funnel geometries where a drive is applied in the easy flow direction . we show that this system exhibits a number of different commensurability effects when the vortex configurations match to both the periodicity of the array and the geometry of the funnels . the vortex configurations in this system are generally different from those observed for single isolated triangular superconducting samples due to the coupling of vortices in adjacent funnels . at certain matching fields , peaks in the critical current are absent due to the particular vortex configurations that occur at these fields . we find that the overall depinning force increases with increasing vortex density as a result of the enhanced vortex - vortex interactions caused by a crowding effect at the funnel tips . when a system becomes less mobile as a result of increased particle interactions , it is said to exhibit a jamming behavior . under an applied drive we observe a series of elastic and plastic vortex flow phases which produce pronounced features such as jumps or dips in the transport curves . in all of the flow phases , only one vortex can pass through the funnel tip at a time due to the vortex - vortex repulsion forces . as a consequence of this constraint , we observe the remarkable result that the sum of the vortex velocities at a fixed drive remains nearly constant with increasing magnetic field rather than increasing linearly . this result is similar to the behavior of sand in an hourglass . we also show how noise fluctuations can be used to distinguish the different flow phases . our results should be readily generalizable to other systems of particles flowing in periodic funnel geometries , such as colloids or wigner crystals .

1003.0932

advanced nanostructuring techniques permit the creation of specific superconducting structures for controlling vortex motion in type - ii superconductors . experiments and simulations for square and triangular artificial pinning arrays demonstrate that pronounced commensurability effects occur when the number of vortices is an integer multiple of the number of pinning sites , and that these effects can be observed as peaks in the critical current.@xcite above the first matching field , multiple vortices may occupy individual pinning sites,@xcite or only a portion of the vortices may be captured by the pinning sites with the remaining vortices located in interstitial sites where they can still be pinned by the repulsive interactions with vortices at the pinning sites.@xcite vortex matter in these periodic pinning arrays can also exhibit a remarkably rich variety of distinct dynamical phases originally predicted in simulations.@xcite these include one - dimensional flow of interstitial vortices between immobile vortices in the pinning sites , disordered or turbulent phases where the number of moving vortices fluctuates strongly , and laminar states where the vortex flow can organize along the rows of pinning sites . the transitions between these different states appear as specific features in the voltage - current curves , including negative differential conductivity where the number of moving vortices or the average vortex velocity decreases with increasing drive . in very recent experiments , these dynamical phases were observed in both low temperature @xcite and high temperature @xcite superconductors with periodic pinning arrays . it is also possible to create periodic pinning arrays that contain intrinsic asymmetry , such as with asymmetric thickness modulation,@xcite funnel geometries,@xcite composite pinning sites,@xcite or arrays of triangular traps.@xcite this asymmetry can produce a diode effect when the depinning force is higher in one direction , and can give a ratchet effect in which a net dc vortex flow occurs upon application of an ac drive.@xcite reversals of the ratchet flow from the easy asymmetry direction to the hard asymmetry direction can occur as a function of magnetic field and other parameters @xcite due to various collective interactions of the vortices . one of the earliest proposals for a vortex ratchet involved a periodic asymmetric channel or funnel,@xcite and such geometries have now been experimentally fabricated.@xcite the first experiments on this system verified the existence of ratchet effects , but also detected boundary effects caused by the disordered injection of vortices at the edge of the sample.@xcite the edge effect can be overcome by using sample geometries where the vortices flow in annular asymmetric channels so that vortices need not enter or exit the sample.@xcite such a technique may also provide the resolution required to investigate individual vortex motion through the channels.@xcite in this work we study the dynamics of vortices in a periodic asymmetric channel geometry or funnel array and show that a variety of new types of vortex dynamics and behaviors can arise , including a jamming effect and flow patterns that are organized such that only one vortex at a time passes through the funnel tip . such a geometry could be realized by etching the edge of a single superconducting strip into a periodic funnel shape so that the vortices would flow in a true single channel . funnel geometries created with periodic channels in two - dimensional superconducting samples should also exhibit many of the same properties we observe ; however , in these systems vortices located between the asymmetric channels could be important and under high drives could depin . the presence of vortices in the regions between the channels can be deduced from the onset of hysteresis in the critical current curves @xmath1 at fields higher than the fields at which many of the ratchet effects are observed.@xcite although our study focuses on interacting vortices in type - ii superconductors , we expect that the same dynamics will occur for other systems of repulsively interacting particles in funnel geometries . these include charged colloidal assemblies,@xcite magnetic colloids,@xcite charged metallic dots,@xcite and classical electron crystals.@xcite additionally , in recent experiments of ion flow through a single funnel , effects such as negative differential conductivity were observed.@xcite our results suggest that the same type of clogging effect we observe could be occurring in these artificial ion channels . an important feature that differentiates funnel geometries from the other vortex ratchet geometries is that in the funnel the vortices are forced to move through a narrow bottleneck . at this constricted point , the repulsive vortex - vortex interactions are very important and favor the organization of the vortex flow into a pattern that permits only a single vortex to pass through the bottleneck at a time . the bottleneck has many features in common with granular hopper geometries where grains flow through a funnel . in the granular case , it is known that decreasing the width of the hopper aperture can cause the flow of grains to be impeded or jammed . @xcite systems that become immobile due to particle - particle interactions are often referred to as jammed . the physics of jamming has attracted growing interest as a way to understand many types of loose particle assemblies such as grains , colloids , and emulsions in situations where these systems exhibit a sudden onset of resistance to shear , with possible connections to the glass transition.@xcite in a granular hopper where grains flow through a thin funnel , the jamming effect occurs when some grains block the motion of other grains . the interaction between the grains is short ranged and has a sharp cutoff . vortices also experience a repulsive interaction ; however , it is significantly longer in range and smoother than the interaction among grains . understanding how systems with intermediate range or longer range interactions can develop a jammed state is an open question . generally , in a vortex system , increasing the effective vortex - vortex interaction strength reduces the effectiveness of the pinning and causes the depinning force to decrease . if vortex matter can exhibit a jamming effect , the opposite behavior would occur and the system would become more immobile with increasing vortex - vortex interaction strength . we note that in the peak effect phenomenon , the effective pinning force increases with increasing vortex density or temperature . many explanations of this effect involve the reduction of the effective vortex - vortex interaction force with increasing density or temperature due to changes in the penetration depth or softening of the vortex lattice near @xmath2 or @xmath3 @xcite in order to match the normal expectation of increased pinning force caused by decreased vortex interaction strength . superficially , however , the increased pinning force generated by increased vortex density resembles a jamming effect . in simulations of vortex systems with more vortices than pinning sites,@xcite gradually increasing the strength of the vortex - vortex interactions initially increased the depinning force since the vortices at the pinning sites blocked the free vortices from moving , similar to a jamming effect . on the other hand , when there are more pinning sites than vortices , the depinning force monotonically decreases as the vortex - vortex interaction strength increases.@xcite this shows that if vortices are to exhibit jamming behavior , it must arise from the collective interactions of the vortices rather than from vortex - pin interactions alone . in our system we fix the bare vortex - vortex interaction strength and study the appearance of jamming and clogging effects due to density - induced changes in the effective importance of the vortex - vortex interactions . following previous techniques for simulating vortices in periodic pinning geometries , we employ langevin dynamics in a two - dimensional system.@xcite we consider a sample containing a single channel composed of @xmath4 funnels . each funnel has a small aperture size of @xmath5 , a wide aperture size of @xmath6 , and a length of @xmath7 , where lengths are measured in units of the london penetration depth @xmath8 . the funnels are aligned in the @xmath9 direction and the sample has periodic boundary conditions along the @xmath9 direction only . a total of @xmath10 vortices are placed only inside the funnel channel and the region outside the channel is empty . the motion of the vortices is calculated by integrating the following overdamped equation of motion : @xmath11 here the damping constant is @xmath12 in a crystal of thickness @xmath13 , where @xmath14 is the elementary flux quantum , @xmath15 is the superconducting coherence length , and @xmath16 is the normal state resistivity . the vortex - vortex force is repulsive with a potential @xmath17 , where @xmath18 , @xmath19 is a modified bessel function , @xmath20 is the position of vortex @xmath21 , and @xmath22 . if a vortex approaches one of the channel walls sufficiently closely , it experiences a wall force @xmath23 . the channel walls are constructed out of @xmath24 repulsive elongated potential barriers which are inverted versions of the potential wells employed in ref . each barrier has a central rectangular region which repels vortices in the direction transverse to the long direction of the rectangle , along with two half - parabolic cap regions which repel vortices from the ends of the barrier . we have @xmath25 here @xmath26 , @xmath27 is the position of the center point of barrier @xmath28 , @xmath29 is the barrier radius or half width , @xmath30 is the barrier strength , @xmath31 is half the length of the central rectangular region of barrier @xmath28 , and @xmath32 ( @xmath33 ) is a unit vector parallel ( perpendicular ) to the axis of barrier @xmath28 . the individual barriers are connected together to form a pair of sawtooth shapes as illustrated in the inset of fig . [ fig:1 ] . the barrier lengths are @xmath34 for the vertical walls and @xmath35 for the slanted funnel walls . the central portion of the channel is featureless and the vortices experience confining forces only from the barrier walls and not from the channel itself . the barriers are sufficiently strong that vortices can never cross them under the conditions considered in this work . the initial vortex positions are obtained by placing the vortices evenly throughout the funnel and performing simulated annealing . temperature is modeled as langevin kicks @xmath36 with the following properties : @xmath37 and @xmath38 , where @xmath39 is the boltzmann constant . after the vortex positions are initialized , we apply an external drive @xmath40 representing the lorentz force from an applied current in the positive @xmath9 or easy - flow direction and measure the average vortex velocity @xmath41 . the drive is slowly increased in small increments with a fixed waiting time between each increment . the waiting time is taken sufficiently long that the system always reaches a steady state velocity at each drive before we make our measurements . the depinning force @xmath42 is defined as the drive at which @xmath43 . in this work we examine only the dynamics in the easy flow direction and note that in general @xmath44 is higher for driving in the hard or negative @xmath9 direction so that a diode effect is possible , in agreement with experiments.@xcite for driving in the hard direction , clogging type dynamics occur which will be explored elsewhere.@xcite in our system the penetration depth @xmath8 is less than the funnel size , so many of our results should carry over to typical colloidal systems with optical trap arrays . in the vortex ratchet experiments of ref . @xcite , the size of the triangular traps is less than @xmath8 so we are working in a different regime from the experiment , although we expect that many of the same effects can appear in both systems . experiments with thicker films or larger funnel length scales should also be possible which would be much closer to the regime we consider . vs @xmath45 for vortices in a periodic funnel array for driving in the @xmath46 , easy - flow direction . a series of peaks appear at matching fields of @xmath47 , 5 , 7 , 8 , 11 , 12 and 14 . weaker peaks occur at @xmath48 , 10 , and 13 , while peaks are absent at @xmath49 , 6 , and 9 . inset : a portion of the sample showing the funnel array geometry . , width=336 ] in fig . [ fig:1 ] we plot the the depinning force @xmath44 for driving in the easy @xmath46 direction vs @xmath45 , where @xmath50 denotes the field at which there is one vortex per funnel . in our geometry , @xmath51 at @xmath52 since each vortex can move unimpeded through the bottleneck and then flow freely along the center axis of the funnel . fig . [ fig:1 ] shows peaks in @xmath44 at @xmath53 , 5.0 , 7.0 , 8.0 , 11.0 , 12.0 , and 14.0 . smaller peaks in @xmath42 appear at @xmath54 , 10.0 , and 13.0 , while peaks are absent at @xmath55 , 6.0 , and 9.0 . the varying sizes and shapes of the commensuration peaks are consistent with the existence of distinct arrangements of singly quantized vortices within the funnel plaquettes at different matching fields . commensurability effects in two - dimensional square or triangular periodic pinning arrays can take two different forms depending on whether multiple vortices are trapped at each pinning site or whether interstitial vortices are present . if multiple vortex pinning occurs , then the vortex configuration at each matching field @xmath56 , with integer @xmath57 , is the same as the configuration at @xmath52 , but with @xmath57-quantized vortices trapped at each pinning site . the result is a peak in the critical current at every matching field.@xcite similarly , the vortex configurations at fractional fields such as @xmath58 are repeated at all fields @xmath59.@xcite such multi - quanta commensuration effects are very similar to the commensuration effects observed in superconducting wire networks.@xcite if the pinning sites can capture a maximum of one vortex , then for fields above @xmath60 , some vortices will be located in the interstitial regions and the vortex lattice structures can be different at each matching field.@xcite at matching fields where the interstitial vortex structure is disordered , there is no peak in the critical current @xcite . the magnitude and shape of the commensuration peaks show striking variations when different types of vortex crystals form at different matching fields . for example , in square pinning arrays , a square vortex lattice forms at @xmath55 , a less stable dimer lattice with a smaller critical current peak appears at @xmath54 , a very stable triangular lattice with a strong critical current peak is present at @xmath53 , and the partially disordered vortex structures at @xmath61 and @xmath62 produce no peaks in the critical current.@xcite for the asymmetric funnel array , the quasi - one - dimensional nature of the system might be expected to produce identical commensuration effects at each matching field ; however , it is possible for the vortices within each funnel to distort in both the @xmath9 and @xmath63 directions in order to try to form triangular ordering on a local scale , and thus the response of the system differs at different matching fields . simulations and experiments on single mesoscopic triangular superconducting samples have shown that the vortices can form triangular or partially ordered configurations at magic fillings such as @xmath64 , 6.0 , and @xmath65.@xcite at these matching fields , we find weak or missing commensuration peaks in the funnel geometry , as indicated in fig . [ fig:1 ] . the high symmetry of these well ordered states results in poor pinning of the vortices in the easy - flow direction since it allows some vortices to simultaneously align along the @xmath9 axis of the channel while closely approaching the narrow aperture of the funnel . the alignment of the vortices produces an additional @xmath9-direction force on the vortex closest to the funnel tip , permitting it to flow out of the funnel at a relatively low driving force . in contrast , at commensurate fillings where the vortices adopt partially disordered configurations or have degenerate ground states , the critical current is high . this is likely due to the lack of a well - defined easy shear direction for the disordered vortices . figure [ fig:1 ] shows a vanishing critical current at low fields which results when the vortices are able to sit along the center of the funnel channel and can flow along the channel unimpeded by the funnel geometry . the experiments of ref . @xcite revealed a finite critical depinning force at low fields . since the experimental channels are less than @xmath8 wide , this finite critical force could be the result of a bowing effect in the center of the funnel channel that creates some effective pinning even for very low fields . the edge barrier may also be playing a role at low fields , and it is likely that some intrinsic random pinning exists throughout the entire sample which would create a finite @xmath44 at all fields . if the intrinsic pinning is weak , it should be possible to observe the commensuration effects shown in fig . [ fig:1 ] . the intrinsic pinning effects can be strongly suppressed near @xmath66 , and in a later section we show that the commensuration effects in fig . [ fig:1 ] are robust at finite temperature . we also note that for colloidal particles in an optical funnel trap array , such intrinsic pinning effects would not be present . at incommensurate fields @xmath67 , due to the discreteness of the vortices the funnels are occupied by a mixture of @xmath57 and @xmath68 vortices . the depinning force is reduced at these fields as a result of the asymmetric repulsive force experienced by a vortex at the boundary between two funnels of different occupancy . [ fig:1 ] indicates that peaks or enhancements of the depinning force can arise at non - matching fields such as at @xmath69 and @xmath70 ; however , unlike the matching peaks , these non - matching peaks are not robust against thermal fluctuations . , an aligned dimer state forms . ( b ) at @xmath71 , a triangular state forms with the triangle orientation alternating in every other funnel . ( c ) @xmath72 . ( d ) @xmath73 . ( e ) at @xmath74 , a triangular structure forms that fits well with the triangular funnel geometry and that matches the vortex configuration found in an individual mesoscopic triangular superconductor . this field also corresponds to a missing commensuration peak in fig . [ fig:1 ] . ( f ) @xmath75 . , width=336 ] in fig . [ fig:2](a - f ) we illustrate the vortex configurations at zero applied drive for @xmath55 , 3.0 , 4.0 , 5.0 , 6.0 , and 7.0 . for @xmath55 , fig . [ fig:2](a ) shows that each funnel captures two vortices which form a dimer state . all of the dimers are aligned in the same direction with one vortex in each dimer located at the upper left corner of the funnel and the other vortex located along the lower wall . we note that this is not the same configuration that would arise for a single isolated triangular superconductor , where the two vortices would maximize their spacing by sitting with one vortex at the tip of the triangle and the other vortex in the center of the opposing triangle wall . interactions between vortices in adjacent funnels in our system would make such an arrangement energetically unfavorable since the vortex at the tip of the triangle would be too close to the leftmost vortex in the adjacent funnel . the tilted dimer configuration in fig . [ fig:2](a ) minimizes the vortex - vortex interactions both within a single funnel and in adjacent funnels . the vortex dimer state in fig . [ fig:2](a ) is two - fold degenerate ; in the other possible orientation , a vortex is located at the lower left corner of each funnel . in an infinitely long system at finite temperature or in the presence of quenched disorder , it is possible that domain wall excitations could form where the dimer orientation flips from one ground state to the other . in this case , it may be possible to map the dimer state to a one - dimensional ising model which is known to have a long - range ordered ground state only at @xmath76 . the mapping of effective dimer and trimer states of particles in periodic substrates to ising and other spin models has been proposed previously for colloids on two - dimensional periodic substrates @xcite and vortices in honeycomb pinning arrays.@xcite for @xmath54 , fig . [ fig:2](b ) shows that the three vortices in each funnel form a triangle with one vortex located in the corner of the funnel . the orientation of the triangle alternates in every other plaquette from having the upper funnel corner occupied by a vortex to having the lower funnel corner occupied by a vortex . this ground state has similarities to a one - dimensional antiferromagnetic ordering . for @xmath53 , shown in fig . [ fig:2](c ) , both corners of each funnel are occupied by vortices and the ground state is non - degenerate . this is the first filling at which a pronounced peak in @xmath42 emerges , as seen in fig . [ fig:1 ] . the ground state at @xmath77 , illustrated in fig . [ fig:2](d ) , is very similar to the configuration at @xmath53 with the addition of one vortex in the center of the channel near the tip of the funnel , while at @xmath61 , shown in fig . [ fig:2](e ) , there are two vortices in the center of the channel . at @xmath78 , fig . [ fig:2](f ) indicates that the configuration changes from states with four vortices along the walls and the remaining vortices in the center of the channel to a state with six vortices lining the walls and only one vortex in the center of the channel . figure [ fig:1 ] indicates that there is a pronounced peak in @xmath42 at @xmath78 but not at @xmath61 . the vortex configurations at @xmath55 , 3.0 , 4.0 , 5.0 , and @xmath62 differ from the configurations found in isolated mesoscopic triangular superconductors;@xcite however , the configuration at @xmath61 is almost the same as that in an isolated triangle since the vortices can form an almost perfect triangular ordering within the funnel at this field . since the accommodation of the triangular ordering to the boundaries is so energetically favorable , it overcomes the energy cost of placing two vortices close together near the funnel aperture with one vortex shifted slightly in the positive @xmath9-direction . since this vortex experiences an extra force from the other vortices , the effective pinning potential at the tip of the funnel for this vortex is depressed , lowering @xmath42 . at @xmath78 this condition no longer holds since the extra vortex that had been near the funnel tip shifts to a new location along the wall , where it no longer exerts an extra force on the vortex at the center of the channel . . ( b ) @xmath79 , where there is a missing peak in @xmath42 . ( c ) @xmath80 . ( d ) @xmath81 . ( e ) @xmath82 . ( f ) @xmath83 . , width=336 ] in fig . [ fig:3](a ) , the ordered vortex configuration at @xmath84 has four vortices on the upper wall of each funnel and three vortices on the bottom wall , along with one vortex in the center of the channel near the wide end of the funnel . this configuration has a two - fold degenerate ground state since either the upper or the lower funnel wall could be occupied with the four vortices . at @xmath85 , fig . [ fig:3](b ) shows that there are now four vortices lining both the top and bottom funnel walls . the vortices on one wall are slightly more compressed than the vortices on the other wall so that the vortices near the funnel tip are not aligned in the @xmath63-direction . the wall with the stronger compression alternates from top to bottom in adjacent funnels . figure [ fig:1 ] indicates that there is no peak in @xmath42 at @xmath85 . the relatively low depinning threshold at this field is a result of the close proximity of a vortex near each funnel tip to the vortex in the center of the channel combined with the asymmetry of the vortex compression along the walls ; the extra force experienced by the channel vortex from the less compressed wall of the funnel causes it to depin more readily . at @xmath86 , shown in fig . [ fig:3](c ) , two vortices occupy the center of the channel and the remaining eight vortices line the upper and lower funnel walls in a symmetric configuration . in fig . [ fig:3](d ) we plot the configurations at @xmath87 , where four vortices line each funnel wall and a triangular vortex structure forms near the wide end of the funnel . at @xmath88 , fig . [ fig:3](e ) shows that an asymmetric configuration of five vortices on one funnel wall and four vortices on the other forms along with a skewed triangle of vortices in the open portion of the channel . at @xmath89 , the symmetric configuration illustrated in fig . [ fig:3](f ) forms with five vortices along each funnel wall and a triangle of vortices in the center of the channel . at @xmath90 vs @xmath45 . the line indicates a fit to @xmath91 . , width=336 ] figure [ fig:1 ] shows that the depinning force tends to increase overall with increasing @xmath0 . this is the result of a crowding effect that makes it more difficult for the vortex structure to distort in order to permit individual vortices to pass through the bottleneck . we can show that the crowding is geometrically induced by measuring the average closest neighbor distance @xmath92 for the vortex configuration at the depinning force @xmath90 . to determine @xmath92 , we solve the all - nearest - neighbors problem with a simple algorithm . the distance from each vortex @xmath93 to its closest neighbor at @xmath94 is @xmath95 . then , @xmath96 . the value of @xmath92 at the depinning threshold has a convincing @xmath91 form , as shown in fig . [ fig : dmin ] . as expected in a system where the pinning originates from vortex - vortex interaction forces , at a given field @xmath0 a vortex can depin when the driving force pushes it closer to its neighboring vortex than the average spacing between vortices . as @xmath0 increases , the average vortex spacing should drop as @xmath91 , consistent with the behavior of @xmath92 . the critical force curve shown in fig . [ fig:1 ] is very far from being a smooth function of @xmath0 , unlike @xmath92 , and this simply indicates that the particular geometric arrangements of the vortices at different fields may make it easier or more difficult for two vortices to approach each other closely enough to depin . for the most commonly studied types of pinning , such as random pinning or arrays of individual pinning sites , when the number of vortices exceeds the number of pins , the depinning force tends to decrease with increasing @xmath0 due to the relative increase in the strength of the vortex - vortex interactions compared to the pinning energy . in this case , as the vortex lattice becomes stiffer , some vortices are forced to shift out of the pinning sites and occupy interstitial regions . in other words , a stiff vortex lattice can not adjust to the pinning site configuration as well as a soft vortex lattice can . one proposed mechanism for the peak effect observed in superconductors with random pinning is a softening of the vortex lattice due either to thermal fluctuations or to changes in @xmath8 which increase the effectiveness of the pinning . in the funnel geometry we consider here , depinning does not require the vortices to overcome the pinning strength of individual pinning sites . instead , the vortices depin once they are able to overcome the vortex - vortex interactions blocking the passage of individual vortices through the funnel tips . a system that becomes less mobile due to increased interactions among the particles is said to be jammed.@xcite studies on systems with short range interactions where a single probe particle is pushed through a collection of other particles in the absence of pinning have shown that the probe particle motion at a finite and constant driving force @xmath97 becomes slower for increasing particle density , and at a critical density the probe particle becomes stuck or jammed.@xcite in this jammed state , a critical driving force @xmath44 must be applied to unjam the probe particle , and this critical force monotonically increases with increasing density.@xcite the critical force is analogous to the depinning force needed to move the vortices through the funnels in our system , which also shifts to higher values as the vortex density increases . there are many similarities between fig . [ fig:1 ] and the behavior of jamming in colloidal systems . at very low densities , the critical depinning force drops to zero since the vortices are so far apart that they interact only extremely weakly . previous studies of probe particles in systems with longer range particle - particle interactions found that a finite depinning force , similar to a jamming effect , exists even at low densities and increases monotonically with particle density.@xcite a key ingredient for jamming in our system is that a portion of the vortices must remain immobile . the vortices lining the walls of the funnel are held in place due to the strength of the repulsive vortex - vortex interactions ; when one of these vortices moves toward the tip of the funnel , it experiences a barrier due to the compression of the vortices at the funnel tip , providing a finite depinning force . vortices away from the funnel walls are also pinned by means of this compressive repulsion . if the funnel walls did not converge , but instead remained a fixed distance apart , the depinning force would be absent . in experiments and simulations on straight channels where there are vortices both inside and outside the channels , the vortices outside of the channel are strongly pinned and create a periodic potential modulation for the vortices within the channel which allow the channel vortices to be pinned.@xcite for the funnel geometry we consider , additional immobile vortices outside of the channel are not needed to create an effective periodic pinning potential . in ratchet channel geometries @xcite there are strongly pinned vortices outside of the channels ; however , we do not expect the presence of such vortices to qualitatively affect the results we report here . for colloids moving through an asymmetric optical trap array , it would be possible to remove all colloids outside of the channel so that particles are present only inside the ratchet channel . vs @xmath45 at @xmath98 , 0.025 , 0.02 , 0.015 , 0.01 , and @xmath99 ( from top to bottom ) . , width=336 ] to quantify the greater difficulty with which vortices flow at higher vortex densities , we examine the average normalized vortex velocity @xmath100 versus @xmath45 for different values of the driving force @xmath101 . in fig . [ fig:1 ] the initial depinning force @xmath44 was determined by the initial onset of vortex flow ; however , the normalized velocity can be taken at any value of @xmath101 and in fig . [ fig:4 ] we plot @xmath100 versus @xmath45 for @xmath98 , 0.025 , 0.02 , 0.015 , 0.01 , and 0.005 . in all cases , @xmath100 decreases with increasing @xmath45 expect for some small oscillations caused by commensuration effects . at @xmath102 , @xmath103 for those fields at which @xmath104 . the reason that the overall flow velocity decreases for increasing @xmath0 is that the flow patterns organize in such a way that only one vortex at a time is able to pass through the funnel tip , as will be shown later . this is a result of the very large energy cost that would be associated with the passage of two or more vortices through the tip simultaneously . the single passage constraint causes the flow rate of vortices through the system to be roughly constant for fixed @xmath105 , such that for a number of moving vortices @xmath106 , we find @xmath107 . for fixed @xmath106 , we obtain @xmath108 , which is approximately the behavior shown in fig . [ fig:4 ] . the results in figs . [ fig:1 ] and [ fig:4 ] show that jamming phenomena can be realized in superconducting systems or other systems of particles with intermediate to long range interactions . further , since stronger pinning is often a desirable feature for many applications of superconductors , some of the concepts from studies of jammed systems could be employed to increase the effective pinning strength in a device . vs @xmath101 for @xmath55 , 2.25 , 2.5 , and @xmath109 ( from top to bottom ) . here the average velocity decreases with increasing @xmath0 . , width=336 ] we next examine the different vortex dynamical phases that can arise in the funnel geometry . in fig . [ fig:5 ] we plot @xmath100 versus @xmath101 for @xmath55 , 2.25 , 2.5 , and @xmath109 . the depinning force increases with increasing @xmath110 over this field range , as was seen in fig . [ fig:1 ] , and above depinning @xmath100 for a particular value of @xmath101 decreases as more vortices are added to the system , as was shown in fig . [ fig:4 ] . for @xmath111 ( a ) 2.0 , ( b ) @xmath112 , ( c ) @xmath113 , ( d ) @xmath109 , ( e ) @xmath114 , and ( f ) @xmath115 . here the flow at the commensurate fields is highly ordered . , width=336 ] in fig . [ fig:6 ] we plot the vortex trajectories at fixed @xmath116 for @xmath55 , 2.25 , 2.5 , 2.75 , 3.0 , and @xmath115 . the flow is ordered at @xmath117 and @xmath71 in fig . [ fig:6](a ) and fig . [ fig:6](e ) . the vortices move in fixed trajectories and continuously maintain the same neighbors , indicating that all of the vortices are moving . for @xmath118 , 2.5 , and @xmath109 in fig . [ fig:6](b , c , d ) , the trajectories become increasingly disordered and the vortices no longer keep the same neighbors over time , indicative of plastic flow . the disordered nature of the flow permits the vortices to explore larger regions of phase space , including regimes in which some of the vortices move much more slowly than others or become temporarily pinned . vs @xmath101 for @xmath54 , 3.25 , 3.5 , and @xmath119 ( from top right to bottom right ) , showing the same trends as in fig . 5 . , width=336 ] fig . [ fig:6](f ) indicates that at @xmath120 , the vortex trajectories are more disordered than at @xmath54 . in fig . [ fig:7 ] we plot @xmath100 vs @xmath101 for @xmath54 , 3.25 , 3.5 , and @xmath119 , which show the same trend observed at and above the second matching field . the value of @xmath100 at a fixed @xmath101 above the depinning threshold decreases with increasing @xmath45 and the vortex flow becomes partially disordered at the incommensurate fields . vs @xmath101 for ( a ) @xmath53 . ( b ) @xmath121 . ( c ) @xmath122 showing a two step depinning process with a cusp near @xmath123 where @xmath100 decreases with increasing @xmath101 . ( d ) @xmath124 showing some small steplike features . , width=336 ] at @xmath116 where an ordered flow occurs . ( b ) the initial ordered flow at @xmath125 and @xmath126 , below the cusp in @xmath100 in fig . [ fig:8](c ) , where only the vortices along the top wall are moving . ( c ) the disordered flow for @xmath122 at @xmath127 , above the cusp in @xmath100 in fig . [ fig:8](c ) , where all of the vortices are moving . ( d ) the high drive phase at @xmath128 for @xmath125 . ( e ) the initial disordered flow at @xmath126 for @xmath129 . ( f ) the flow at @xmath116 for @xmath124 , below the step in @xmath100 which occurs near @xmath130 in fig . [ fig:8](d ) . for drives above the step in @xmath100 at this field , the vortex flow resembles the flow shown in panel ( a ) . , width=336 ] for @xmath131 , the transport becomes more complicated and transitions between distinct dynamical phases begin to occur . the moving phases for @xmath131 are generally characterized as plastic since a portion of the vortices can remain immobile . we find both ordered plastic motion , where the trajectories of the mobile vortices follow a fixed path , and disordered plastic flow phases where the vortex motion is more chaotic and the vortices follow many different paths . in fig . [ fig:8](a - d ) we plot @xmath100 versus @xmath101 for @xmath53 , 4.25 , 4.5 , and 4.75 . at @xmath53 , fig . [ fig:8](a ) indicates that there is a single well defined depinning transition where the vortices flow elastically , as illustrated in fig . [ fig:9](a ) . for @xmath132 , the value of @xmath42 is depressed as seen in fig . [ fig:8](b ) . here there are no sharp jumps in the transport curve , and the vortex motion resembles that observed at @xmath53 but with some additional fluctuations in the vortex trajectories . at @xmath125 , a two step depinning process occurs as shown in fig . [ fig:8](c ) . initially , only one ordered channel of moving vortices forms along one of the funnel walls while the remaining vortices are immobile . this is illustrated in fig . [ fig:9](b ) for a simulation in which the vortices along the upper funnel wall depinned first ; depending upon the initial random fluctuations , it is also possible for the vortices along the lower wall to depin first . the remaining vortices depin near @xmath126 , where a cusp feature appears in @xmath100 . at the cusp , the average vortex velocity decreases with increasing @xmath101 , creating a region of negative differential conductivity where @xmath133 . the negative differential conductivity at @xmath125 is not as pronounced as that observed in simulations and experiments with square pinning arrays ; @xcite however , we find that the type of negative differential conductivity illustrated in fig . [ fig:8](c ) is a common feature at a number of the higher order incommensurate fillings we have examined in the funnel geometry . in contrast , for the square pinning array the negative differential conductivity occurs only near the first matching field . this indicates that it may actually be easier to observe negative differential conductivity in a funnel geometry than in a square pinning array . above the cusp at @xmath126 , fig . [ fig:8](c ) shows that @xmath100 smoothly increases with increasing @xmath101 . in fig . [ fig:9](b ) we show that the vortex flow below the cusp in @xmath100 at @xmath125 and @xmath126 is ordered with only a single channel of vortex flow . figure [ fig:9](c ) indicates that for the same field at @xmath127 , above the cusp in @xmath100 , the vortices flow in disordered paths , each of which predominantly runs along either the upper or lower funnel wall . the transition from only a single flowing channel to effectively two flowing channels of vortices at the second depinning transition associated with the cusp in @xmath100 causes a drop in the mobility of the vortices due to the competition between the two channels for passing a vortex through the bottleneck of the funnel . only a single vortex can fit through the bottleneck at a time , but vortices in the upper and lower flowing channels do not arrive at the bottleneck at synchronized times due to the unequal distribution of vortices between the two effective channels which causes the two channels to flow at different average speeds . as a result , a vortex in one channel may reach the bottleneck too soon while a vortex from the other channel is still moving through the bottleneck , forcing the vortices in the first channel to move more slowly until the vortex in the second channel has exited the bottleneck and freed it for passage of a vortex in the first channel . as @xmath105 increases , the difference in flow speed for the two unequally populated channels of moving vortices decreases until it is small enough that the flow of the two channels becomes synchronized on average and a much more orderly passage of vortices through the bottleneck occurs , alternating between the two channels . in this case , the disordered trajectories become smoother , as shown for @xmath128 in fig . [ fig:9](d ) . for @xmath129 , fig . [ fig:8](d ) shows that there are several small steps in the @xmath100 versus @xmath101 curve which correspond to changes in the flow . in fig . [ fig:9](e ) , the initial flow for this field at @xmath126 is partially disordered . by @xmath116 , shown in fig . [ fig:9](f ) , a transition to a more ordered flow state has occurred where there are two possible paths for vortices to follow in the lower channel . at higher drives , one of these two lower paths closes , corresponding to the jump in @xmath100 near @xmath130 , and the flow at higher drives resembles that shown in fig . [ fig:9](a ) . vs @xmath101 for ( a ) @xmath77 . ( b ) @xmath134 . ( c ) @xmath135 . ( d ) @xmath136 . , width=336 ] shown in a small portion of the sample . dots : vortex positions ; heavy lines : funnel geometry ; light lines : vortex trajectories . ( a ) right before depinning , at @xmath137 , there are four vortices lining the walls and one vortex located in the center of the channel near the funnel tip . ( b ) the first rearrangement just below @xmath138 occurs when the center vortices move through the aperture in the positive @xmath9 direction and pass into the adjacent funnels . ( c ) at depinning , which occurs at @xmath138 , these center vortices move against the funnel walls in an alternating pattern associated with a transient flow . ( d ) above depinning at @xmath139 , the vortices flow strictly along the funnel walls . , width=336 ] at @xmath77 there is a single depinning transition , as illustrated in the plot of @xmath100 versus @xmath105 in fig . [ fig:10](a ) . right at the depinning transition , the vortices undergo a structural transition which is highlighted in fig . [ fig:11 ] . the vortex configuration for drives well below depinning consists of four vortices arranged symmetrically along the funnel walls with a fifth vortex in the center of the channel near the funnel tip , as shown in fig . [ fig:11](a ) . at the onset of depinning , illustrated in fig . [ fig:11](b ) just below @xmath138 , the vortices at the funnel tips move in the positive @xmath9 direction and shift into the adjacent funnel . at depinning , which occurs at @xmath138 , fig . [ fig:11](c ) indicates the rearrangement that occurs when the center vortices move against one of the funnel walls in an alternating pattern . above depinning , as shown in fig . [ fig:11](d ) for @xmath139 , the vortices flow strictly along the funnel walls in two channels that never mix . for @xmath140 . ( b ) the initial depinning at @xmath139 for @xmath141 . ( c ) the trajectories in the fluctuating phase for @xmath142 at @xmath141 . ( d ) the high drive ordered phase at @xmath143 for @xmath141 . , width=336 ] at @xmath140 , 5.5 , and 5.75 , shown in fig . [ fig:10](b ) , [ fig:10](c ) , and [ fig:10](d ) , respectively , there can be multiple dynamical transitions between ordered and disordered flow phases . at @xmath140 , the flow at large @xmath101 is very similar to that shown for @xmath77 in fig . [ fig:11](d ) ; however , the flow initiates at a much lower drive and takes the form of an ordered winding channel in which about 40% of the vortices are moving , as shown in fig . [ fig:12](a ) . as the vortices in the winding channel move through the system , the vortices pinned along the funnel walls undergo an oscillatory motion which is most easily seen in fig . [ fig:12](a ) for the pinned vortex near the center of each long funnel wall . this oscillation is a response to the passage of an individual vortex through the flowing channel ; the pinned vortex shifts in the positive @xmath9 direction as the moving vortex approaches , and shifts back again after the moving vortex has passed . the vortices at the corners of the funnel undergo little to no shift . in fig . [ fig:10](d ) at @xmath141 the transitions between different dynamical phases are associated with changes in the fluctuations of the velocity signal . at this filling , the sharp depinning transition takes the system into the ordered braiding flow phase illustrated in fig . [ fig:12](b ) , a state with low levels of velocity fluctuations . a transition to a highly fluctuating phase occurs near @xmath144 and corresponds to the onset of the disordered flow phase illustrated in fig . [ fig:12](c ) . for higher drives , a transition to an ordered phase occurs near @xmath145 when the trajectories become partially ordered . a velocity jump near @xmath146 marks the transition to the completely ordered flow phase shown in fig . [ fig:12](d ) . vs @xmath101 for ( a ) @xmath61 . ( b ) @xmath147 . , width=336 ] for @xmath61 . ( b ) the flow above depinning at @xmath126 for @xmath61 . ( c ) the flow at @xmath142 for @xmath61 . ( d ) the flow at @xmath148 for @xmath61 . ( e ) the ordered vortex flow at @xmath146 for @xmath149 , below the cusp in @xmath100 vs @xmath101 seen in fig . [ fig:13](b ) . ( f ) the disordered flow at @xmath148 for @xmath149 , just after the cusp in @xmath100 . , width=336 ] at @xmath150 , where a commensuration peak in @xmath42 was missing in fig . [ fig:1 ] , the velocity - force curve shows a single depinning transition into a random flow phase , as shown in fig . [ fig:13](a ) . no sharp transitions between different phases appear ; however , the general form of the vortex flow changes gradually as @xmath105 increases . figure [ fig:14](a ) shows an initial shift in vortex positions occurring at a drive just below the depinning transition . the vortex at the tip of the leftmost funnel has pushed the vortex at the base of the adjacent funnel into a position along the lower wall in a pattern that repeats every two funnel plaquettes , indicating that the vortices in the center of the channel were not well pinned , resulting in lack of a peak in @xmath44 at this field . just above depinning , shown at @xmath126 in fig . [ fig:14](b ) , a disordered flow occurs in which the two vortices at the corners of each funnel do not participate . although the flow is disordered , there are clearly defined regions inside the funnels which the vortices completely avoid , as shown by the lack of trajectories passing through large areas of the funnels . as @xmath101 is further increased , the trajectories become increasingly disordered and the vortices in the corners of the funnels begin to take part in the motion , as shown in figs . [ fig:14](c , d ) . more clearly defined transitions between ordered and disordered flow states occur at @xmath151 , 6.5 , and 6.75 , as illustrated in fig . [ fig:13](b ) for @xmath149 . the transitions are characterized by cusp structures in @xmath100 associated with negative differential conductivity . [ fig:14](e ) shows the initial ordered flow at @xmath148 for @xmath149 , below a pronounced cusp in @xmath100 . here the two vortices in the corners of each funnel remain immobile so that the flow is actually plastic . these vortices become mobile at the transition to the disordered phase , illustrated at @xmath148 in fig . [ fig:14](e ) . a similar set of dynamics appears at @xmath151 and @xmath152 . for driving forces higher than those we have examined , it is possible that further dynamical transitions could occur at which different ordered phases arise . vs @xmath101 for ( a ) @xmath78 and ( b ) @xmath153 , where three distinct moving phases appear . , width=336 ] for @xmath78 . ( b ) the braided flow phase at @xmath126 for @xmath153 . ( c ) the second flow phase at @xmath154 for @xmath153 . ( d ) the third flow phase at @xmath155 for @xmath153 . , width=336 ] depinning at @xmath156 occurs in a single step , as illustrated in fig . [ fig:15](a ) where we plot @xmath100 versus @xmath105 . at this field , [ fig:1 ] indicates that there is a peak in @xmath42 . above depinning , the vortices flow in the ordered pattern shown in fig . [ fig:16](a ) , where two vortices remain pinned in the corners of each funnel . for drives @xmath157 higher than those we consider here , a transition to a disordered flow state may occur once the drive is large enough to cause the two immobile vortices to depin . at @xmath158 , 7.5 , and @xmath159 , multiple steps occur in the velocity - force curves , as illustrated in fig . [ fig:15](b ) for @xmath153 . each step is associated with a distinct type of ordered flow . at low drives , we find the braided flow shown in fig . [ fig:16](b ) at @xmath126 . above the first low step in @xmath100 near @xmath116 , an alternating braided flow occurs in which every other plaquette contains two possible flow paths along the lower funnel wall , as illustrated in fig . [ fig:16](c ) at @xmath154 . above the second larger step in @xmath100 near @xmath139 , the completely ordered flow phase shown in fig . [ fig:16](d ) at @xmath155 occurs . vs @xmath101 for ( a ) @xmath84 , ( b ) @xmath160 , ( c ) @xmath161 , and ( d ) @xmath162 . , width=336 ] for @xmath84 . ( b ) the disordered flow at @xmath127 for @xmath163 . ( c ) the initial partially ordered flow phase at @xmath126 for @xmath164 . ( d ) the disordered flow phase with two immobile vortices in the corners of each funnel at @xmath146 for @xmath164 . , width=336 ] for higher fields @xmath166 we observe the same trends found for fields just below @xmath84 . the integer matching fields typically display a single ordered flow phase , while at the incommensurate fields , multiple flow phases occur with ordered - ordered or ordered - disordered flow transitions . another trend is that as the field increases , the number of immobile vortices appearing in the ordered flow phases increases . in fig . [ fig:17 ] we plot @xmath100 versus @xmath101 for @xmath84 , 8.25 , 8.5 , and 8.75 , where the trend outlined above can be seen . there is a single depinning step in fig . [ fig:17](a ) at @xmath84 into the ordered flow phase shown in fig . [ fig:18](a ) at @xmath167 where two vortices remain pinned at the corners of each funnel . figure [ fig:18](b ) shows the disordered flow at the higher drive @xmath127 for @xmath163 . at this field , there is a transition near @xmath146 to an ordered phase with vortex trajectories that are very similar to those shown in fig . [ fig:18](a ) for @xmath84 . the initial flow at @xmath164 is partially ordered with four immobile vortices in each funnel plaquette , as shown at @xmath126 in fig . [ fig:18](c ) . at higher drives for this field , the flow becomes disordered but only two of the immobile vortices begin to move , leaving two vortices immobile in the corners of each funnel , as shown in fig . [ fig:18](d ) for @xmath146 . vs @xmath101 for ( a ) @xmath85 and ( b ) @xmath168 . , width=336 ] showing braided flows . ( a ) first flow regime shown at @xmath116 . ( b ) second flow regime shown at @xmath167 . ( c ) third flow regime shown at @xmath169 . , width=336 ] for @xmath168 . ( b ) the second flow phase at @xmath139 for @xmath168 . ( c ) the higher drive ordered flow phase at @xmath148 for @xmath168 . ( d ) the flow at @xmath127 for @xmath87 . ( e ) a complex partially ordered flow pattern at @xmath127 for @xmath170 . ( f ) a higher drive ordered alternating flow pattern at @xmath146 for @xmath171 . , width=336 ] we find a distinctive integer matching field at which there is not merely a single dynamic flow phase state . this field , @xmath172 , also fails to produce a depinning peak as shown in fig . [ fig:1 ] . we plot @xmath100 versus @xmath105 for @xmath172 in fig . [ fig:19](a ) , which indicates that there are three distinct moving phases . all three phases involve braided vortex flows , but the details of the braiding and the number of vortices participating in the flow varies as a function of drive , as shown in fig . [ fig:20 ] where the three different braided flows are illustrated . the suppression of the depinning peak at @xmath172 is related to the suppression of the peak at @xmath173 . at each of these fields , the system exhibits an instability in which the number of vortices present in the center of the channel for zero applied drive is different than the number of vortices that can be stabilized along the center of the channel just above depinning . at @xmath174 , in each funnel plaquette there are two vortices positioned along the center line of the channel at zero drive , but under finite drive one of these vortices moves against the funnel wall , leaving only one vortex in the center of the channel . this transition can be seen by comparing fig . 2(e ) with fig . 15(a , b ) . at this filling , there is only a small energy difference between placing two vortices at the center of the channel and placing only one vortex at the center of the channel with the other vortex against the funnel wall . this produces an instability which contributes to the disordered flow we find at @xmath174 . an opposite transition occurs for @xmath175 . as seen in fig . 3(b ) and fig . 21(a ) , there is only one vortex per plaquette in the center of the channel at zero applied drive , but above depinning a second vortex moves into the center of the channel and is stabilized there . the existence of this change in the occupancy of the center of the channel at depinning from one to two or two to one is associated with the suppression of commensuration peaks in the critical depinning force at the fields @xmath174 and @xmath175 seen in fig . 1 . a series of flow transitions also occurs at @xmath176 in fig . [ fig:19](b ) , corresponding to the dynamical flows illustrated in fig . [ fig:21](a , b , c ) . in the initial flow phase at low drives , fig . [ fig:21](a ) indicates that only a portion of the vortices are moving in a path that passes through the bottom half of each funnel , forming a single flowing channel . at higher drives a transition to the symmetric flow state illustrated in fig . [ fig:21](b ) occurs . at still higher drives , an intermediate disordered phase appears before the flow reorders into the state shown for @xmath148 in fig . [ fig:21](c ) , where there are two immobile vortices on the bottom half of each funnel and one immobile vortex on the top half of the funnel , with flow occurring through two well - defined channels . at higher fields , similar dynamical regimes occur and larger numbers of vortices can become immobile along the funnel walls , such as at @xmath86 where there is a single step depinning transition into a state with four immobile vortices in each funnel . it is also possible for vortices to be immobilized in the center of the channel rather than along the funnel walls , as illustrated for @xmath177 in fig . [ fig:21](d ) . at this field , a single step depinning transition occurs into the state shown where a total of three vortices are immobilized : two in the corners of the funnel and one right in the funnel center . as @xmath45 increases , additional partially ordered phases occur with increasingly intricate flow patterns such as the one shown in fig . [ fig:21](e ) at @xmath170 . we also find phases with asymmetric flow where the asymmetry alternates from one funnel to the next , such as the state illustrated in fig . [ fig:21](f ) for @xmath171 . here the leftmost funnel contains two flowing channels of vortices interacting with the upper funnel wall and one channel interacting with the lower funnel wall , while in the middle funnel this flow pattern is reversed . we expect the trends described above to continue for higher matching fields . eventually , however , there may be a crossover to a state where the vortex lattice constant is smaller than the width of the funnel aperture , at which point it would be possible to move two vortices through the funnel tip at the same time without much of an additional energy cost . in this case , the average velocity might sharply increase since the flow in the aperture would transition from one - dimensional to quasi - two - dimensional , and it may be possible that further oscillations in the average velocity would arise at even higher fields as larger groups of vortices could pass through the aperture simultaneously for increasing magnetic field values . versus frequency @xmath178 in inverse simulation time steps . lower curve : the ordered phase at @xmath179 for @xmath77 , where a series of peaks occurs due to the periodic velocity signal of the vortices . upper curve : the disordered flow phase at @xmath179 for @xmath61 , where a broadband noise signal arises with a @xmath180 form . the same types of spectra appear for ordered and disordered flow phases at other fields . , width=336 ] with current experimental noise measurement techniques , it should be possible to characterize the difference between ordered and disordered flow phases . the vortex velocities we measure in our simulations are proportional to the experimentally measured vortex voltage signal , which can be analyzed using the power spectrum defined as @xmath181 in fig . [ fig:22 ] we plot @xmath182 for @xmath77 at @xmath179 in a regime of ordered flow . we find narrow band noise peaks at the characteristic frequencies of the ordered oscillatory motion of the vortices . also plotted in fig . [ fig:22 ] is @xmath182 for @xmath61 at @xmath179 , where the vortex flow is disordered . here the power spectrum has a broad band or @xmath183 noise characteristic with @xmath184 . in general , for other values of @xmath45 , the ordered phases produce narrow band spectra while the disordered phases produce broad band noise signals . the appearance of the narrow band noise in the ordered phase implies that phase locking phenomena could be induced by adding an ac drive component to the dc driving force . when the frequency of the ac drive matches the intrinsic or higher harmonic frequencies of the moving vortices in the ordered phase , a series of steps should appear in the velocity force curves . we next consider the effect of temperature . in periodic pinning arrays , the most pronounced matching field peaks observed in experiments occur at temperatures near @xmath3 . this has been attributed to the thermal suppression of the intrinsic pinning in the sample that competes with the periodic pinning . another factor that could play a role is that the vacancies or interstitials present in the vortex lattice away from commensurate fields can be relatively mobile under thermal activation , and so increasing the temperature could depress the depinning force to a larger degree at incommensurate fields than at commensurate fields where vacancies or interstitials are not present . we find that for a temperature of @xmath185 , which is well below the melting temperature of the vortex lattice , the depinning force peaks at the commensurate fields are robust while the depinning force at the incommensurate fields is depressed . the peaks in @xmath42 that appear at @xmath125 and @xmath135 for @xmath186 are absent at @xmath185 , indicating that for finite temperatures , pronounced peaks are only observable at the matching fields whereas submatching field peaks are less robust . the general trend of increasing @xmath44 with increasing @xmath45 also remains robust at finite temperature . although our work is focused on a superconducting vortex system , we expect our results to be general for other systems of particles with repulsive interactions . for example , in colloidal systems where the volume density is sufficiently low that steric contact of the colloidal particles does not occur , many of the same results should still apply . one aspect of the vortex system that we do not take into account is the possibility of multi - quanta vortex formation under certain conditions . in static systems , the formation of giant vortices has been observed in cases where the giant vortices produce a higher symmetry of the vortex configuration , such as in quasiperiodic pinning arrays.@xcite it is possible that dynamical multi - quanta vortices could form in a funnel geometry . for example , if two vortices are forced into close proximity near the funnel tip , they could merge to form a two - quanta vortex which would then pass through the funnel tip . the formation of such dynamical multi - quanta vortices could produce interesting signatures in the transport curves . the experimental system closest to the system studied here is periodic asymmetric channel geometries @xcite where small oscillations in the critical current due to commensuration effects were observed . since these commensuration effects were very weak , it is possible that they were due to edge effects . more recent experiments with geometries that avoid the effect of edges have now revealed much more prominent commensuration effects,@xcite indicating that some of the phases we observe may be occurring in this system . in summary , we have used numerical simulations to examine the vortex configurations and dynamics in a periodic funnel array . the vortex configurations we observe are generally different from those found for a single isolated triangular sample due to the coupling between vortices in adjacent funnels . as a function of field we find a series of depinning threshold peaks at the matching fields where the vortex configurations are ordered . in some cases , matching peaks are missing due to the fact that the vortex configuration contains pairs of vortices that are located close together near the funnel apertures . we also observe a general increase in the depinning threshold with increasing vortex density , which is opposite from the normal trend for vortices in two - dimensional pinning arrays where the vortices are directly trapped by pinning sites . in the funnel geometry , the pinning is a result of the repulsive vortex - vortex interaction forces , and as the vortex density increases it becomes more difficult for the vortices to overcome these repulsive forces and flow through the funnel tip . we also find a rich variety of dynamical phases , including ordered elastic and ordered plastic phases where the vortices follow fixed trajectories and disordered phases where vortices mix chaotically . the phases generally organize in such a way that only one vortex passes through the tip of the funnel at a time . due to this constraint , the average velocity of an individual vortex decreases with increasing field such that the sum of the velocities of all of the vortices at fixed drive remains close to constant for increasing field rather than increasing with increasing field . this behavior is similar to the response of grains in an hourglass . transitions between the different dynamical phases appear as jumps or cusps in the velocity - force curves , and there are even regimes where the average vortex motion decreases with increasing drive , producing a negative differential conductivity . at higher fields , moving states can form in which a single vortex remains immobile at the center of a funnel while other vortices flow around it . in general , ordered flow phases occur at the matching fields , while at non - matching fields the flow is disordered for at least some regime of driving forces . the ordered phases are associated with sharp narrow band or washboard velocity noise signals while the disordered phases have @xmath187 velocity noise spectra . our results should also be generalizable to other systems of repulsively interacting particles moving through a funnel geometry , such as colloids or wigner crystals . 99 m. baert , v.v . metlushko , r. jonckheere , v.v . moshchalkov , and y. bruynseraede , phys . 74 * , 3269 ( 1995 ) ; k. harada , o. kamimura , h. kasai , t. matsuda , a. tonomura , and v.v . moshchalkov , science * 274 * , 1167 ( 1996 ) . mart ' in , m. v ' elez , j. nogu ' es , and i.k . schuller , phys . lett . * 79 * , 1929 ( 1997 ) ; d.j . morgan and j.b . ketterson , phys . 80 * , 3614 ( 1998 ) ; j.e . villegas , m.i . montero , c .- p . li , and i.k . schuller , phys . lett . * 97 * , 027002 ( 2006 ) . field , s.s . james , j. barentine , v. metlushko , g. crabtree , h. shtrikman , b. ilic , and s.r.j . brueck , phys . lett . * 88 * , 067003 ( 2002 ) ; a.n . grigorenko , g.d . howells , s.j . bending , j. bekaert , m.j . van bael , l. van look , v.v . moshchalkov , y. bruynseraede , g. borghs , i.i . kaya , and r.a . stradling , phys . b * 63 * , 052504 ( 2001 ) . siwy , m.r . powell , e. kalman , r.d . astumian and r.s . eisenberg , nano lett . * 6 * , 473 ( 2006 ) ; z.s . siwy , m.r . powell , a. petrov , e. kalman , c. trautmann , and r.s . eisenberg , nano lett . * 6 * , 1729 ( 2006 ) .

we study one - dimensional disordered fermions that either undergo metal - insulator transitions or topological phase transitions to become trivial anderson insulators . we focus on using entanglement to elucidate how the spatial , momentum , and internal degrees of freedom of fermions are affected by the presence of disorder in such cases . we develop entanglement tools that reveal the existence of metallic states in the presence of disorder and further show clear signatures of the corresponding localization transition even in the presence of interactions . in systems where the internal degrees of freedom are coupled with the motion of the electrons , topological phases develop . we subject a topological insulator model to different types of disorder and discuss how the topological aspects of the system can be captured through entanglement , even at strong disorder .

1403.6129

the physics of anderson localization transitions has been intensively studied for almost half a century , yet new aspects of this phenomenon continue to emerge @xcite . among many recent developments , the theory of anderson transitions has been used to classify topological insulators and analyze their robust properties @xcite , and to study thermalization behavior in many - body interacting systems @xcite . furthermore , new systems have been proposed that exhibit anderson localization physics , for example in photonic systems @xcite and ultra - cold atoms in optical lattices @xcite . one approach that has recently been applied to study anderson transitions has been the use of quantum information and entanglement measures to probe the correlations of many - body states . for example , the spatial entanglement of quantum states reveals the non - local properties of many - body wave functions @xcite which naturally suggests its usefulness for identifying localized and delocalized states . additionally , entanglement is particularly useful in capturing universal properties that can not be obtained through the expectation value of local operators @xcite . in particular , it has been used in condensed matter systems to gain insight into the nature of different phases of matter and the critical points that separate them @xcite . over the past few years there have been a number of articles that focus on applying entanglement methods to disordered fermion systems . in [ ] , for example , the scaling of the entanglement entropy was studied to understand a disordered system at criticality . in [ ] , the von neumann entanglement entropy was analyzed using the multi - fractal spectrum of critical wave functions at localization transitions , and in [ ] , the multi - fractal spectrum of disordered systems was shown to have connections with the rnyi entropy of the single - particle states . in [ ] and [ ] , relations were found between the level spacing statistics of the hamiltonians of disordered topological insulators and of their corresponding spatial entanglement spectrum from which critical delocalized states were identified . recent work has also considered the entanglement properties of low - dimensional systems with correlated disorder@xcite . in this article , we complement these studies by analyzing the utility of several different entanglement methods for determining signatures of anderson localization transitions . spatial entanglement methods , such as those used in refs . [ ] have given insight into metal - insulator transitions and topological - to - anderson - insulator transitions . recently , in ref . [ ] , the momentum entanglement spectrum was proposed and used to elucidate the nature of metal - insulator transitions in one - dimensional disordered fermion systems . this was followed by a recent article which uses momentum - entanglement to study correlated disorder in low - dimensional models@xcite . in our work we will focus solely on one - dimensional fermionic models . we study a particular set of models with anderson transitions that can be precisely determined either analytically or numerically in order to calibrate our entanglement techniques . although most of the focus is on free - fermion models , we also perform calculations on an interacting model , and discuss a generalization of the single - particle entanglement signatures to the many - body case . the outline of this work is as follows . we begin in section [ sec : one ] by reviewing the notion of bipartite quantum entanglement for generic systems and the simpler case of free - fermion models . we discuss the different types of entanglement partitions that we will apply in our analysis , and the dual nature of spatial and momentum entanglement for disordered systems . after developing this intuition we move on to some explicit calculations . in section [ sec : entanglement1 ] we discuss two types of one - dimensional lattice models ( i ) the random n - mer models which have correlated disorder and ( ii ) the quasi - crystal aubry - andr model . the localization transitions in these models have been previously studied and identified , and we evaluate several different types of entanglement measures to study the critical behavior . after calibrating with these two models we add interactions to the aubry - andr model and study its entanglement properties . finally in section [ sec : entanglement2 ] we study a disordered topological insulator model and develop a hybrid momentum - orbital entanglement partition that clearly illustrates the localization transition . let us briefly review how bipartite entanglement of a given quantum state @xmath0 is quantified . we start by dividing the hilbert space of many - body states into two subspaces , say @xmath1 and @xmath2 , such that : ( 1 ) no states are shared between both subspaces ; and ( 2 ) their union reproduces the original hilbert space of the system . given this partition , bipartite entanglement can be quantified using the so - called von - neumann entropy , which is defined as @xmath3 where the reduced density matrix @xmath4 is obtained by first constructing the density matrix @xmath5 of the full system , and then tracing out the degrees of freedom that correspond to the subspace @xmath2 . to see why eq . [ sa ] is connected with the concept of entanglement , we can make use of the so - called schmidt decomposition of @xmath0 . given @xmath0 , we can always find its schmidt decomposition , which reads @xmath6 where @xmath7 and @xmath8 are sets of many - body basis states that span the subspaces @xmath1 and @xmath2 , respectively . from this expression , we see that the set of eigenvalues @xmath9 controls the degree to which @xmath0 can be factorized into states from the subspaces @xmath1 and @xmath2 . in other words , the set @xmath9 controls the degree of quantum entanglement of the state @xmath10 for example , we can use eq . [ schmidt ] to explicitly compute the reduced density matrix of subspace @xmath1 . this leads to the expression @xmath11=\sum_n \lambda_n^2 { | \phi_n^a \rangle}{\langle \phi_n^a |},\ ] ] from which we conclude , using eq.[sa ] , that @xmath12 the von - neumann entanglement entropy is a single number one can calculate from the @xmath13 but the full set , which is the so - called entanglement - spectrum " represents the full information about the wavefunction under the @xmath1-@xmath2 partitioning@xcite . the calculation of the entanglement spectrum and @xmath14 can be simplified for non - interacting fermions@xcite . in the absence of interactions , the quantum state of the full system @xmath15 can always be written as a slater determinant . a consequence of this is that any correlation function can be expressed in terms of the correlation matrix @xmath16_{ij}={\langle \omega |}c^{\dagger}_i c_j{| \omega \rangle}=\text{tr}\left(\rho c^{\dagger}_i c_j\right),\ ] ] using wick s theorem ; here @xmath17 represent some fermion degrees of freedom , e.g. , lattice site , momentum , spin etc . since the reduced density matrix @xmath18 must also reproduce wick s theorem when the indices are restricted such that @xmath19 then it can be shown@xcite that the reduced density matrix must take the exponential form @xmath20 where @xmath21 is a normalization factor that ensures @xmath22 and the indices @xmath23 label states within @xmath1 . @xmath24 is often referred to as the entanglement hamiltonian , and the matrix @xmath25 is determined by the correlation matrix through the expression @xmath26_{a a'}. \label{corr}\ ] ] from this construction we can define several sets of useful eigenvalue spectra . since the correlation matrix @xmath27 is hermitian , then @xmath25 is also hermitian . in practice one typically diagonalizes @xmath27 to obtain the eigenvalues @xmath28 ( which lie between @xmath29 and @xmath30 ) commonly referred to as the single - particle entanglement spectra . from the @xmath28 one can construct the eigenvalues of @xmath25 to obtain the set @xmath31 ( which lie between @xmath32 and @xmath33 ) which are commonly referred to as the single - particle entanglement energies . from the @xmath31 one can derive the ( many - body ) entanglement spectrum @xmath34 of @xmath18 through the relation @xmath35 , where @xmath36 indicates a many - body state of of @xmath24 with occupation numbers @xmath37 for each single - particle entanglement energy state @xmath38 with energy @xmath39 the entanglement entropy can generically be written as @xmath40 using the expressions for @xmath41 in terms of @xmath42 ( which in turn can be written in terms of of the eigenvalues @xmath28 of @xmath27 ) one finally obtains the simplified expression@xcite @xmath43 the entanglement entropy , and additionally , all entanglement quantities of a free - fermion ground state @xmath0 can thus be understood by analyzing the @xmath44 note that , from eq . [ corr ] , the eigenvalues @xmath28 always lie between @xmath29 and @xmath30 . because of this , the entanglement of a system is determined by how close the @xmath45 get to @xmath46 we will make use of the single - particle entanglement spectrum to analyze the entanglement of the free - fermion models we study here in addition to the entanglement entropy itself . for interacting models this simplification no longer holds and one must calculate the @xmath47 directly from the many - body reduced density matrix . in this paper , we refer to the partition of the hilbert space as an _ entanglement cut _ that is performed on the system . going back to the expression eq . [ schmidt ] , we note that the degree to which @xmath0 is close to being factorizable in the partitioned basis is sensitive to the type of partition that is chosen . for example , if we were to choose a partition a and b for which all but one ( say n=1 ) of the @xmath13 s in the schmidt - decomposition are zero , then @xmath48 . in such a case , the state @xmath0 would not be entangled with respect to the partition that was chosen . but if we were to change the partition and choose different subspaces @xmath49 and @xmath50 we could find a partition such that all the @xmath13 s are equal in magnitude , and then we would find @xmath51 , which is a highly entangled state . because of this sensitivity of the entanglement to the choice of partition , we can obtain a more complete picture of the quantum properties of @xmath0 by making appropriate variations of the entanglement cut . typically , physical motivation will guide this choice . some types of partitions , other than the conventional spatial cut , that have been studied in previous work include orbital @xcite , particle @xcite , spin @xcite and momentum @xcite . in the case of one - dimensional disordered fermions , there are a number of possible cuts that can be considered which are physically relevant . spatial entanglement is the most conventional possibility . it is a basic measure of localization because it probes the degree to which extended states are able to correlate the two parts of the system . at the same time , localization is related with the interference between forward and backward traveling states that ultimately is at the heart of disorder - induced localization . hence , momentum entanglement will also serve as an insightful and complementary way to study the disordered ground state . finally , interesting anderson transitions occur in systems with internal degrees of freedom that are correlated with the electron motion , such as in topological insulators . performing entanglement cuts with respect to such internal degrees of freedom will also exhibit interesting behavior as the system transits into the localized phase . we consider all of these cases below . before we begin our main discussion , let us briefly build up some simple intuition about how entanglement encodes information relevant to disordered fermions . to begin , we consider how localization is exemplified in the spatial and momentum entanglement spectrum . the simplest situation we can consider is a quantum state of a single particle that is spatially delocalized . in other words , the particle moves freely throughout the system . the simplest example of this type of state is , of course , a plane - wave given by @xmath52 , where @xmath53 is the momentum wave vector and @xmath54 the corresponding spatial correlation matrix reads @xmath55 we want to cut the system in half , which is effected by restricting the indices to the range @xmath56.$ ] the @xmath57 entanglement modes that result from the eigenvalues of this correlation matrix , for any choice of @xmath53 , are @xmath58 and @xmath29 , with the latter being @xmath59-fold degenerate . it follows that a single particle in a plane - wave state has @xmath60 spatial entanglement entropy when the system is partitioned in half . this illustrates the intuition that generic delocalized states typically contribute order @xmath61 entropy to the spatial entanglement entropy , which indicates that the two halves of the system become correlated . on the other hand , the corresponding momentum entanglement correlation matrix for the same state reads @xmath62 as an example , we can choose to trace out momentum states in the range @xmath63,$ ] one reason being that , in some of the models that we will consider , this range corresponds to left - moving states and the remaining modes are all right - moving . hence , we are essentially probing correlations between right and left movers . using this type of partition , the momentum entanglement modes are @xmath30 and @xmath29 , with again the latter being @xmath59-degenerate . we thus obtain a momentum entanglement entropy @xmath64 . this is , of course , a trivial result because the state @xmath65 is trivially a product state in momentum space . now , upon the introduction of disorder , such plane wave states will scatter and eventually localize at sufficiently strong disorder . we should then contrast the case of a delocalized state with that of a completely localized state at a position @xmath66 given by @xmath67 . in this case , the spatial and momentum correlation matrices read @xmath68 in this completely localized case , it is the spatial entanglement entropy which vanishes . the momentum entanglement , on the other hand , has a @xmath58 entanglement mode that leads to a momentum entanglement entropy @xmath60 when using the same cut as above . hence , the spatial and momentum entanglement behave in a dual fashion with respect to localized and delocalized states . in the general situation of an occupied fermi sea there will be more than one single - particle state filled . in this case , one might conjecture that a ground state with more than one delocalized particle will increase the spatial entanglement in an additive manner , for example , each additional occupied plane wave state could add an extra @xmath61 contribution when cut in half . it turns out , however , that the behavior for more than one particle can be affected significantly due to interference effects . the spatial correlation matrix for the case of @xmath69 particles occupying plane - wave states with wave vectors @xmath70 is given by @xmath71 each matrix element represents a sum of phases that depend on the particular choice of the set @xmath72 . because of this , the matrix elements of the correlation matrix can be either enhanced or suppressed depending on which states are occupied . an analogous case occurs for momentum entanglement when more than one spatially localized state is occupied . so , while the connection between localization / delocalization and entanglement immediately seems plausible , the details can be complicated . in this work we will study some interesting effects that this interference in the correlation matrix can have on the entanglement of the system . finally , we mention how the entanglement can carry signatures of the configuration of the internal degrees of freedom in the presence of disorder . in practice , this will be particularly relevant when the momentum vector and the internal degree of freedom are correlated , such as what occurs in systems with spin - orbit coupling . one simple family of states we can consider is a spin @xmath58 particle in a state parameterized by momentum @xmath53 , given by @xmath73 . that is , this state is such that the spin rotates throughout the bz and is explicitly entangled in the spin degree of freedom in a momentum dependent way . as we will discuss later , this type of rotation property in momentum space can have an important impact on the physics of the system . if we introduce disorder , one might suspect that scattering processes can potentially destroy the correlation between the spin and the momentum degree of freedom . for example , as localization is induced , there will be a corresponding broadening of the momentum space features , and the specific configuration of the internal degree of freedom in momentum space might become blurred . if this happens , we then expect that such an effect will be apparent in the structure of both the orbital and momentum entanglement . to what extent this happens is something that we wish to understand in this work and will discuss in section [ sec : entanglement2 ] . now that we have introduced the notation , the concept of different entanglement cuts , and some simple examples , we will move on to discuss the entanglement properties of several types of 1d free - fermion models . in sec . [ sec : entanglement1 ] we will discuss the random @xmath69-mer models and the aubry - andr model , both of which exhibit metal - insulator transitions and have only a single on - site degree of freedom . we also discuss some effects of interactions on the aubry - andr model . in sec . [ sec : entanglement2 ] we will discuss a chiral symmetric topological insulator model in the aiii class in the presence of disorder potentials that preserve the chiral symmetry . this type of model , in addition to having interesting momentum and spatial entanglement structure , will also yield orbital entanglement since there are two degrees of freedom per site . in this section we focus on one - dimensional disordered models with delocalized states that can exhibit metal - insulator transitions . one dimensional disordered systems generically exhibit localization for any disorder strength as long as the system does not obey any particular symmetry @xcite . this would seem to imply that it is not possible to study delocalization transitions at a finite disorder strength in 1d . it has been shown , however , that there are special models with correlated disorder that preempt the occurrence of anderson localization , and thus possess metallic ground states . an example is the so - called random dimer model ( rdm ) @xcite and its generalizations , and the quasi - periodic system referred to as the aubry - andr model ( aam ) @xcite . although the aam is not actually a disordered system , it does have a localization transition for reasons similar to that of the rdm . both models exhibit metallic behavior at finite disorder potential strengths , and eventually become anderson insulators . we will thus explore the localization properties of these models using entanglement techniques . for such systems , there are two natural ways in which the hilbert space can be partitioned , namely the spatial and momentum entanglement cuts we mentioned previously . because of the nature of the delocalized state , it will turn out that for these classes of models the momentum entanglement cut will prove particularly insightful . the random n - mer models describe some of the simplest one - dimensional systems with disorder that allow for a metallic phase in one dimension@xcite . the hamiltonian of the general model is @xmath74 where @xmath75 labels the lattice sites , @xmath76 is the lattice size , @xmath77 is the tunneling matrix element between nearest neighbors , and @xmath78 takes on two values @xmath79 and @xmath80 . the important property of @xmath78 is that the value @xmath79 always has to be placed on @xmath69 consecutive sites so that `` n - mers '' are distributed throughout the lattice . the key to explain why this type of disorder leads to delocalized states in the energy spectrum is the fact that there is a complete suppression of scattering between certain momenta . to see this , note that the fourier elements of the disorder potential @xmath78 are given by @xmath81,\ ] ] where @xmath82 , the @xmath83 denote the random positions of the n - mers , @xmath84 depends on the particular disorder realization , and @xmath85 is independent of any randomness . the key observation is that the structure factor @xmath86 has zeroes at @xmath87 @xmath88 . scattering is consequently suppressed for all @xmath89 . if we calculate the energies of the points in the translationally invariant bandstructure that are nested by the @xmath90 of these suppressed scattering events we obtain @xmath91 in other words , near each of these special energies @xmath92 there will be a suppression of left and right - mover scattering . this suggests that localization is inhibited for states near the energies @xmath92 , which is indeed the case . furthermore , it can be shown that in the neighborhood of these so - called resonant energies , a finite fraction of the eigenstates have a localization length of the order of the lattice size . thus , the @xmath93 zeroes of @xmath94 lead to metallic phases in one - dimensional disordered fermions when @xmath95 is tuned to the neighborhood of @xmath96 if the disorder becomes sufficiently strong , all single - particle states will localize even at the resonant energies . for the n - mer models , the delocalized states exist up to critical values of the disorder strength @xmath97 given by @xcite @xmath98 where @xmath99 . in what follows , we will set @xmath100 and @xmath101 , so that the disorder strength will be measured by @xmath80 . we will focus on the n=1 ( uncorrelated disorder ) , n=2 ( random dimer model ) , and n=3 ( random trimer model ) cases . the rdm has an extended state at @xmath102 which survives up to the critical disorder @xmath103 . for the random trimer model ( rtm ) there are two extended - state regions , but we will focus on the extended state that exists at @xmath104 which survives for @xmath105 $ ] . having discussed the basic aspects of the random @xmath69-mer models , we now analyze how the metal - insulator transition manifests itself in the entanglement properties . the momentum entanglement and some spatial entanglement for these models was studied earlier in ref . [ ] , and additionally the spatial entanglement was considered in ref . [ ] . to study these models , there are two parameters at our disposal that allow us to tune the localization transitions , namely the fermi energy @xmath95 and the disorder strength @xmath106 by varying both of these parameters , we can drive the system between the metallic and insulating states . we will start with the fermi energy @xmath95 . in fig . [ fig_set17 ] , we show four columns : ( i ) the fourier components of the scattering potential showing the zeroes of scattering , ( ii , iii ) the spatial and momentum entanglement spectrum as the fermi level is varied , and ( iv ) the localization length as a function of energy . in appendix [ appa ] we explain how the localization length is calculated . the rows correspond to the system with uncorrelated disorder ( first row ) , the rdm ( second row ) , and the rtm ( third row ) . for all of these plots , the disorder strength is @xmath107 , which means that both the rdm and the rtm have resonant states in their energy spectrum . the localization length is shown here to confirm the energies for which delocalization occurs . from the spatial entanglement spectra in fig . [ fig_set17 ] , one can see by eye that the entanglement modes exhibit level repulsion when the fermi - level is in the delocalized region , i.e. , when the ground state is metallic . this is consistent with the results of ref . [ ] which made similar observations when the fermi - level was tuned to a delocalized state in a chern insulator band . the level repulsion causes the entire region between @xmath108 $ ] to be filled with eigenvalues since the levels are more rigidly spaced . in regions near the edges of the bandwidth , where localized states should dominate , the entanglement eigenvalues do not appear to have any particular structure . the filling of the region between @xmath108 $ ] is much more sparse as the levels can cluster together since they are not repelling . for the case of uncorrelated disorder ( @xmath109 ) there are no obvious features in the entanglement nor in the localization length . although the signatures in the spatial entanglement spectrum do not appear to be particularly clear , the disorder - averaged entanglement entropy does reveal the delocalized nature of the extended state . in particular , it is well known that the entanglement entropy in a critical state of a subsystem of size @xmath110 ( with total system size @xmath76 ) varies as @xmath111+s_0 , \label{scal}\ ] ] where @xmath112 is a non - universal constant , and @xmath113 is a universal coefficient that depends on the critical properties of the ground state ( and is the central charge when considering conformal field theories ) . we show the disorder - averaged spatial entanglement entropy for the rdm and rtm in fig . [ fig_criticality]a , b . the scaling function is shown as the red curve , and the blue dots are the numerical calculation when @xmath95 is at aligned to the delocalized resonant state . the green dots correspond to when the disorder strength is strong enough to localize everything . the match between the scaling function and the numerical calculation is clear . for the clean system , i.e. , the 1d tight - binding chain , @xmath114 for all fermi - energies away from the band - edges . in certain disordered systems , such as the disordered xx spin-@xmath58 chain , this coefficient acquires an additional @xmath115 factor@xcite . for the rdm and rtm , however , we obtain @xmath114 even with disorder . this result is also obtained in ref . . the simple reason that we find this result is due to the fact that the rdm and rtm potentials do not couple the states at the resonance energy and thus these fermions remain free . the low - energy theory will remain a @xmath114 dirac fermion . let us now proceed to discuss the momentum entanglement in these models . the momentum entanglement spectrum shows much sharper behavior than the spatial entanglement spectrum : in the metallic state the entanglement modes are largely clustered near @xmath29 and @xmath30 , whereas they fill the region in between @xmath29 and @xmath30 as the fermi level is moved away from resonance and into localized states . as we mentioned before , the scattering profiles of the rdm and rtm potentials exhibit exact zeroes in the fourier components which are responsible for stabilizing the delocalized states of such models . in particular , there is a clear connection between the existence of the metallic phase at finite disorder , and the behavior of its corresponding momentum entanglement . since momentum entanglement yields information about the build up of correlations between left and right movers , the fact that momentum entanglement is suppressed in the metallic state can be interpreted as a decrease of correlations between left and right movers , which thus preempts the anderson insulator at these energies . this connection is natural , since the return probability is an important ingredient in the localization transition @xcite . note that , in all of these examples , the momentum entanglement turns out to be sharply suppressed even though the occupied fermi sea of the ground state contains single - particle states below the resonant energies that are actually localized . as we mentioned in a previous section , a single localized state by itself leads to an entanglement of order @xmath115 for this type of cut . one would naturally expect that all of the entanglement from the entire set of occupied localized states would add up . however , the fact that there is not a large accumulation of entanglement when @xmath95 is at resonance suggests that there is significant destructive interference in the correlation matrix similar to what we discussed for eq . [ interference ] . to understand this observation , one can think of the zeroes of the scattering potential as a mechanism that prevents mixing between states in the energy spectrum that are separated by the resonant energies . for example , for the case of the rdm , the suppressed scattering in the neighborhood of @xmath102 is very effective in keeping states below and above this resonant energy from mixing with the opposite group at weak disorder . as a consequence , the set of states below zero energy ( call this set @xmath116 ) hybridize primarily among themselves . effectively , the absolute suppression at @xmath102 divides the spectrum into two pieces @xmath117 and @xmath118 that are not mixed for weak - disorder . this means that , upon constructing a slater determinant with @xmath119 we can approximate the single - particle states by @xmath120 . as a result , there is a significant simplification in the form of the slater determinant ground state @xmath121where the last equality is true up to a global phase ( which does not enter the calculation of the density matrix ) . in other words , the fermi sea will approximately reproduce the ground state of delocalized states of the clean system . in fact , while the energies of the states below @xmath95 will be affected by disorder , the same orbitals that were occupied in the clean - limit will be occupied in the weak - disorder limit . in terms of the correlation matrix , when @xmath95 is set at resonance , this is equivalent to saying that contributions from localized wave functions in the momentum correlation matrix interfere destructively , because for every @xmath53 , @xmath122 is included , and this leads to a matrix that is approximately diagonal in momentum space . parallel to this , we note that there are single momentum entanglement modes at @xmath58 in all of the examples we present here ( see e.g. , fig . [ fig_set18]b ) . although it is hard to see by eye in the figures we present here , closer inspection reveals that these @xmath58 modes depend on whether the number of occupied states is even or odd . to understand this behavior , consider the weak disorder limit . in this case , the degenerate plane waves at @xmath53 and @xmath122 will hybridize easily and form the states @xmath123 which are split in energy by an amount that is linear in the disorder strength ( though the effective disorder strength vanishes if @xmath124 ) . now , if the ground state only has one of these excitations occupied , say the state @xmath125 , then the momentum entanglement spectrum will automatically have a @xmath58 mode . however , if the ground state has both states filled , then we obtain the factorized state @xmath126 , which has zero momentum entanglement . by successively filling the hybridized @xmath127 states , we will thus obtain a sequence of @xmath58 modes for an odd number of occupied states regardless of the scattering properties of the disorder potential . since the hybridization is immediate for the degenerate modes , even for extremely small disorder , they will appear as soon as the disorder is increased from zero . eventually , when the disorder is strong enough to make scattering between @xmath53-states at different energies play a more important role , such entanglement modes are obscured because of new entanglement modes entering from the edges near @xmath29 and @xmath30 to fill the region near @xmath58 . after our discussion of the effects of tuning @xmath95 we now turn to the description of the metal - insulator transition as the disorder strength @xmath80 is varied to tune the transition . we take the rdm as an example and set the fermi energy at its resonant value @xmath128 . [ fig_set18]a , f shows the momentum entanglement spectrum and entropy , respectively as a function of @xmath106 fig . [ fig_set18]b , c , d , e shows the entanglement spectrum as a function of fermi energy for a few selected disorder strengths referenced by dashed lines in figures fig . [ fig_set18]a . we show a similar array of figures for the momentum entanglement entropy calculated from these same entanglement spectra in fig . [ fig_set18]f as a function of disorder strength and fig . [ fig_set18]g , h , i , j as a function of fermi energy for different disorder strengths . from fig . [ fig_set18]a we see that , as the disorder strength @xmath80 is varied , the momentum entanglement modes can change quite drastically . near the clean limit @xmath129 there is minimal entanglement and the eigenvalues are accumulated near @xmath30 and @xmath29 , as expected . as @xmath80 increases , the entanglement modes begin to spread out smoothly from @xmath29 and @xmath30 toward @xmath58 , which indicates an increase in momentum entanglement . correspondingly , the entanglement entropy grows smoothly until the point where the delocalized states disappear @xmath130 at this critical disorder strength , a number of entanglement modes reach @xmath58 and there is a discontinuous change in the slope of the entanglement entropy as a function of @xmath80 . this seems to indicate that the anderson insulator transition occurs when there is a saturation of momentum entanglement modes near @xmath58 followed by a slowdown in the rate of increase of momentum entanglement . this qualitative change suggests itself as a way to characterize the anderson localization transition . we note that all of this structure is apparent without having to perform disorder averages , as all of these figures are for a single disorder realization . the qualitative change in the entanglement entropy could be useful , and efficient , to determine the phase diagram of systems that undergo anderson transitions . we will explore this possibility further in later sections . we will now move on to discuss another special 1d free - fermion model , namely the aubry - andr model ( aam ) . the hamiltonian for this system consists of one - dimensional fermions in a quasicrystal potential and has the form @xmath131 where the aubry - andr potential is given by @xmath132 here , @xmath133 is conventionally chosen to be an irrational number , usually the golden ratio . the main property of such a quasicrystal is that it leads to a localization transition when the potential strength is tuned past a known value . this can be shown by comparing the spatial and momentum space versions of the hamiltonian and noting that both models are dual to each other : low values of @xmath134 lead to localization in momentum space whereas high enough values of @xmath134 lead to localization in position space . the self - dual point is consequently @xmath135 , which is the critical point ( we will usually set @xmath100 ) . with the potential turned off the system is simply a single - band tight - binding hamiltonian with the conventional dispersion @xmath136 for this dispersion the states for @xmath137 are left - movers and those from @xmath138 are right - movers . thus , when we perform a momentum space partition we will pick region a to be one of these subspaces , and region b to be the other so that we separate the left - movers from the right - movers . additionally , there is one important subtlety to note : since in a finite lattice with periodic boundary conditions we must have @xmath139 , we have to approximate @xmath133 using a fraction of relative prime numbers . for this work , we will use a lattice size @xmath140 for the non - interacting calculations , which means that the best approximation to the golden ratio using periodic boundary conditions is @xmath141 . for the interacting case we use @xmath142 at half - filling , and choose @xmath143 . to compare with our previous analysis for the random @xmath69-mer models , the potential profile and the corresponding fourier components of the aubry - andr potential are shown in figs . [ aapotential ] a , b . furthermore , in figs . [ aapotential]c , d we show the momentum entanglement spectrum as a function of fermi energy as well as the localization length as a function of energy , both for the case @xmath144 . the energy gaps that we see in fig . [ aapotential]c correspond to actual gaps in the single - particle energy spectrum of the system . these gaps are created by the nonzero scattering elements of the disorder potential shown in fig . [ aapotential]b . within each of the resulting energy bands the scattering is suppressed , which leads to delocalized states in each band . this is evidenced by the divergence of the localization length in each band . in this sense , the aam shares some similarities with the n - mer models . there is , however , an important difference : it is a well - known property of the aam that _ all _ single - particle states remain delocalized when @xmath145 , which means that there is no mobility gap and thus no mobility edge in the spectrum . it is only when @xmath134 crosses the critical point that _ all _ single - particle states in the energy spectrum become localized . an additional goal of this section is to use the aam to explore the effect of interactions on the momentum entanglement at the localization transition . knowing the location of the critical point ( @xmath146 ) in the noninteracting system will simplify the identification of the signatures in the entanglement which are relevant to the localization transition ; and it is known that the aam exhibits a localization transition even in the interacting case @xcite . interestingly , we will find that both spatial and momentum entanglement are still viable tools for identifying this transition , even though it is not at all obvious that the useful momentum - entanglement characteristics will survive when interactions are turned on . however , before we move on to the interacting case , let us narrow down some qualitative features of the entanglement of the non - interacting aam that will be useful for an interacting generalization . we now consider signatures of the disorder - induced phase transition in the entanglement of the system . in fig . [ fig_set12 ] , we show both the spatial and momentum entanglement spectra ( fig . [ fig_set12]a , b ) and entanglement entropies ( fig . [ fig_set12]c , d ) as a function of disorder ( @xmath134 ) for @xmath128 . for this particular model , there is a clear signature of the localization transition in the entanglement spectrum for both types of entanglement cuts . the spatial entanglement shows a pattern of modes very similar to the case of a translationally invariant system for weak potential strength . that is , in the translationally invariant limit , or for weak disorder , most of the modes are clustered near 0 and 1 except a few that have peeled off from the unentangled band - edges " and additionally o(1 ) modes that lie in the mid - gap region . toward the critical potential value , the entanglement modes start to deviate sharply toward one and zero , indicating that the eigenstates localize in the system in a sharp manner at the transition point . the spatial entanglement entropy appears to be discontinuous at the transition , and the first derivative of the entanglement entropy clearly shows a sharp peak precisely at @xmath146 as evidenced in fig . [ fig_set12]e . the momentum entanglement spectrum , for weak potential strength , is largely clustered toward one and zero in the delocalized state with no mid - gap modes , and has very little entropy . as the potential strength is increased toward the transition point , the entanglement modes gradually start spreading toward a value of @xmath58 . at the transition point , there is a sharp increase of the density of mid - gap entanglement modes near 1/2 , very reminiscent of what we observed in the random n - mer models . the first derivative of the momentum entanglement entropy correspondingly shows a maximum at the transition point as evidenced in fig . [ fig_set12]f . since we know the exact location of the critical point , we can try to use the non - interacting signatures near @xmath147 ( e.g. , the trends of the spatial and momentum entanglement entropy ) to help identify a localization transition when interactions are included . we turn to this case now . after clearly illustrating the behavior of the non - interacting aam we now move on to the case with repulsive interactions . the simplest type of interaction that we can consider for spinless electrons is a nearest - neighbor hubbard interaction of the form @xmath148where @xmath149 is the particle density on site @xmath150 as we mentioned previously , we numerically diagonalize a system of size @xmath142 at half - filling , and choose @xmath143 . there have been several studies on the aam with interactions including ref . [ ] where a two - particle aam with interactions was studied , and ref . [ ] where the interacting aam was studied at infinite temperature in the context of many - body localization . in order to correlate the behavior of the spatial and momentum entanglement with whether there is a tendency for localization in the system , we compute two additional independent quantities that are not directly related to entanglement . one if these is the so - called configuration - basis inverse participation ratio computed in ref . . if we express the ground state in terms of slater determinants in the spatial basis , we can schematically write this as @xmath151 , where @xmath113 corresponds to a specific spatial configuration of the fermions in the lattice . the normalized configuration - basis participation ratio is then given by @xmath152 , where @xmath153 and @xmath154 is the number of basis states . this quantity measures the weight distribution of the ground state in the basis of spatial slater determinants . in the delocalized state , @xmath155 is expected to be of order @xmath30 , whereas in the localized state @xmath155 should become vanishingly small . the second quantity we compute to probe localization is the change in ground state energy @xmath156 after twisting of the boundary conditions from @xmath29 to @xmath157 . in this case , @xmath158 should be nonzero in the delocalized state and it should become vanishingly small in the localized state . in fig . [ fig_set28fdec]a , b we show the results of calculating @xmath159 and @xmath160 as a function of disorder strength for four values of the hubbard interaction @xmath161 . for the non - interacting case @xmath162 , we obtain the behavior we expect as the system crosses the localization transition : there is a significant decrease in both quantities at the phase transition . as the interaction strength is increased , the curves maintain their overall shape , except that they appear to shift towards higher values of the disorder . the two lower plots figs . [ fig_set28fdec]c , d , show the derivatives of @xmath159 and @xmath160 with respect to the disorder strength . these plots exhibit a dip , which approximately corresponds to the point at which localization transition occurs . the dip moves toward higher values of the disorder strength with increasing @xmath163 , which indicates that as the interaction is increased , the parameter space over which delocalization is obtained is enlarged by interactions . we should emphasize that this conclusion might not extrapolate to the thermodynamic limit as there has been some evidence that the enhancement of delocalization in two - particle interacting aam models vanishes in this limit @xcite . notwithstanding this , it will be useful to consider these finite - size results , and to try to correlate the behavior of @xmath159 and @xmath160 with both the spatial and momentum entanglement . now we are will calculate the entanglement properties of the interacting aam . as we mentioned previously for interacting systems , the simple relation between the eigenvalues of the correlation matrix and the entanglement entropy does not hold because the ground state will not , in general , be a single slater determinant . instead , it is necessary to compute the eigenvalues of the reduced density matrix of the many - body system in order to obtain the entanglement entropy . we will briefly present here the method for calculating the many - body momentum entanglement . to calculate the many - body momentum entanglement spectrum we start by writing the ground state as @xmath164 where @xmath165\left[\prod^{n_r}_{k'\in k_r}c^{\dagger}_{k'}\right]{| 0 \rangle}$ ] , and @xmath166 ( @xmath167 ) is the number of left ( right ) movers in a given basis state @xmath168 . the matrix elements of the density matrix @xmath5 are then given by @xmath169^{k'_r , k'_l}_{k_r , k_l}= \psi^*(k_r , k_l)\psi(k'_r , k'_l)\ ] ] and so the reduced density matrix for the left movers is @xmath170^{k'_l}_{k_l}= \sum_{k_r}\psi^*(k_r , k_l)\psi(k_r , k'_l)\ ] ] which describes the left - moving part of the interacting ground state @xmath0 . the set of eigenvalues of @xmath171 which we previously called @xmath172 from the schmidt - decomposition , are then used to compute the entanglement entropy via @xmath173 . an analogous calculation can be performed for the spatial entanglement entropy . let us now compute the entanglement for the same values of the hubbard interaction we have used in fig . [ fig_set28fdec ] . we start by analyzing the spatial and momentum many - body entanglement spectra @xmath34 as a function of disorder potential strength . in fig . [ fig_manybody ] , we show the set @xmath174 as is customary in the literature . the closer the @xmath175 for each mode gets to @xmath30 , the higher the entanglement corresponding to that mode . in the spatial entanglement modes , where we have cut the system in half , the overall structure appears qualitatively the same for the four values of the interaction strength : in the region @xmath176 , the modes are spread out , with some modes near @xmath177 , whereas for @xmath178 the modes appear to move to higher ( less entangled ) values . the main effect of the interactions is to lower the entanglement modes toward @xmath177 for all potential strengths . it is not , however , clear from these figures whether there is a definite behavior as the interaction strength is increased . the entanglement entropy computed using these modes will provide a clearer picture as we see below in fig . [ intentropy ] . in the momentum entanglement spectrum , on the other hand , there appears to be a gap around @xmath177 when the system is in the delocalized state . in the localized state there is a large shift of many - body entanglement modes toward @xmath177 , which signals a significant increase in momentum entanglement since the system is becoming localized . the main effect of the hubbard interactions appears to be to systematically increase the momentum entanglement for all potential strengths . in particular , note that the momentum entanglement gap around @xmath177 gets smaller as the potential strength is increased . in parallel to this , however , one can see that the gap actually increases slightly in the region @xmath178 as the interaction becomes larger . since we know from the non - entanglement quantities @xmath159 and @xmath160 that interactions shift the point at which localization occurs toward higher values of @xmath134 , the behavior of the momentum entanglement spectrum suggests that this entanglement gap is tracking the delocalized phase . this would constitute a generalization to interacting systems of the connection between suppressed momentum entanglement and the metallic state . having made these observations about the many - body entanglement spectra , let us now take a look at the corresponding entanglement entropies . this is shown in fig . [ intentropy]a , b . it is clear that there is a drop ( rise ) in spatial ( momentum ) entanglement for all values of the interaction strength in the neighborhood of the non - interacting @xmath146 , which we expect should be connected with the localization transition . by taking the derivative of these curves , which we show in fig . [ intentropy]c , d , we can track the disorder strength for which both types of entanglement change their trend . in the derivative of the spatial entanglement we observe a dip and in that of the momentum entanglement , we observe a peak . both the dip and the peak shift toward larger values of the disorder strength as a function of the hubbard interaction strength . the behavior of the entanglement entropies is consistent with that of @xmath159 and @xmath160 . this instills confidence that both the spatial and the momentum entanglement are able to track the tendency to localize in the interacting system . in particular , the many - body momentum entanglement spectrum and the corresponding entanglement entropy appears to yield clear signatures of localization in such an interacting system even though the interactions will generate scattering between all momentum states . it is probable that there are other systems for which this behavior will also hold , especially in other interacting one - dimensional models . we are now going to switch gears to study a slightly different problem . whereas so far we have been considering the dual relationships between the spatial and momentum entanglement in simple models , we will now study a model that has internal degrees of freedom which are correlated with the single - particle momentum . in these systems we will explore how disorder affects the internal configurations of one - dimensional fermions . to study this , we will focus on disorder - induced phase transitions between two distinct insulating phases . a particular example of this type of transition is that which occurs between a topological insulator and trivial anderson insulator . topological states of non - interacting fermions in translationally invariant systems typically arise because of a coupling between the momentum and the internal spin / orbital space . the mapping between the brillouin zone momentum and the orbitals of the particular occupied states via the bloch functions can endow the electronic structure with robust topological properties . a natural question is then , upon disordering the system , how robust is the entanglement between the momenta and the internal degrees of freedom ? we will see that entanglement , and in particular a hybrid momentum / orbital partition , even without translation invariance , will be useful . let us now move on to discuss a specific example . if we restrict ourselves to non - spatial discrete symmetries ( namely particle - hole , time - reversal or chiral symmetry ) , there exist five classes that realize nontrivial ground states in one dimension and they are either classified by a @xmath179 or a @xmath180 topological invariant . the two simplest classes are the ones that only realize one of these symmetries : class d which satisfies exclusively particle - hole symmetry and class aiii which satisfies exclusively chiral symmetry ( hamiltonians with only time - reversal symmetry are all trivial in 1d ) . to simplify our discussion , we will focus on the latter . systems belonging to class aiii are those described by hamiltonians that anti - commute with an operator @xmath181 . the simplest such model in this class can be represented by a two - band model . systems with more bands can be adiabatically deformed so that the topological properties are determined by copies of such a two - band model . we thus consider the following hamiltonian @xmath182 c_k=\sum_{k } c^{\dagger}_k\left[\sum_a d_a(k)\sigma^a \right ] c_k\label{aiii}\end{aligned}\ ] ] where @xmath183 ( @xmath184 ) are the identity and the pauli matrices , and @xmath185 is a momentum - dependent , real , four - component vector . this hamiltonian has two energy bands given by @xmath186 . since the pauli matrices anti - commute , chiral symmetry can be realized by setting @xmath187 and any one of the remaining @xmath188 components to zero . although this choice is arbitrary , we choose to set @xmath189 for convenience . for this choice , the chiral operator is @xmath190 . a simple model in class aiii that has a non - trivial topological insulator phase is given by the bloch hamiltonian @xmath191 depending on the value of the parameter @xmath75 , the ground state will either be a trivial band insulator ( @xmath192 and @xmath193 ) or a topological insulator ( @xmath194 ) protected by chiral symmetry . this symmetry class has an integer topological invariant @xmath195 and in the two - band model we will discuss , the invariant is given by the expression @xmath196 which counts the number of times the @xmath197 vector winds around the unit circle in @xmath197-space as @xmath53 traverses the brillouin zone . this is the topological invariant that characterizes the nontrivial nature of the ground state . in the topological phase @xmath198 our model has @xmath199 one physical manifestation of a nonzero @xmath200 is the formation of protected boundary states when the system has open boundaries . these boundary modes indicate a non - zero charge polarization equal to @xmath201 as shown in ref . since we will focus on the case @xmath202 case , the system will have @xmath203 . in appendix [ appb ] , we explain how we can compute @xmath204 when the system is disordered , which we will do to determine the topological phase boundary for strong disorder . more generally one can use the real - space winding number invariant introduced in ref . [ ] . when @xmath205 the winding indicates that in the clean system the internal degree of freedom is entangled with the momentum , and we will use this idea to explore the entanglement properties of the disordered system . we will add chiral - symmetry preserving disorder via the on - site potential given by @xmath206 c_{n \sigma'}.\ ] ] for simplicity , we will primarily use the aubry - andr quasicrystal potential for @xmath207 because there is no need to disorder - average . we will also show some results for the case in which @xmath207 represents uncorrelated disorder towards the end of the discussion as a comparison . we proceed by showing the results of spatial , momentum , and orbital entanglement cuts and use the results to generate a hybrid entanglement cut that combines a momentum and orbital cut . in this section we perform three types of entanglement cuts , namely the same spatial and momentum entanglement cuts we made in previous sections and , in addition , we calculate the orbital entanglement by tracing out one of the components of the internal degree of freedom . as mentioned , the results in this section will further motivate a fourth type of entanglement cut that mixes the momentum and orbital cuts . before proceeding , we should emphasize that the spatial entanglement spectrum has been widely used to characterize topological insulator states @xcite . when the system is topological , the spatial entanglement spectrum exhibits single - particle mid - gap entanglement modes that are topologically protected . the fundamental reason for these modes is that the correlation matrix is equivalent to a spectrally flattened version of the hamiltonian , so that upon an entanglement cut , boundary modes naturally arise@xcite . we will focus more on the momentum and orbital cuts since those have not been emphasized as much in the literature . in figs . [ fig_set15]a , b , c we show density plots of the the spatial , momentum , and orbital entanglement entropy , respectively , over the @xmath208 parameter space . the phase boundary is given by the thick black line in each plot , which we obtained by calculating the polarization of the system numerically . the topological state is realized when the parameters are set within this black line where the polarization is exactly @xmath203 , and the rest of parameter space corresponds to a trivial state where @xmath209 . from these plots , it seems clear that the phase diagram is most clearly identified in the spatial entanglement and orbital entanglement plots . when the system is topological there is a nearly constant , nonzero spatial entanglement entropy , as well as a significant degree of orbital entanglement . when the parameters cross the phase boundary , these two types of entanglement decrease significantly and become vanishingly small . the match is not exact however , since there are some regions that appear to have spatial entanglement which lie outside of the topological phase boundary . it is unclear if these regions of nonzero spatial entanglement outside the topological regime are a peculiarity of the model , or are a signature of something more interesting . we leave this for future work . in contrast to these two types of entanglement , the momentum entanglement entropy does not appear to correlate clearly with the presence of the topological state . what appears to happen is that the momentum entanglement starts off from zero , as it must in the clean limit , and increases as the disorder strength increases . there is a region in the parameter space at which the momentum entanglement exhibits a marked increase , which suggests that the anderson insulator is realized beyond those points . however , this surge in momentum entanglement occurs regardless of whether the system begins in the topological or trivial phase . for example , if one fixes @xmath75 to lie deep in the topological or trivial regimes the momentum entanglement entropy eventually surges as @xmath134 is increased for both cases . to better understand our observations about the entanglement entropy maps , let us take a look at the entanglement spectra for a few cases . we show examples of the spectra plotted vs. @xmath134 for three fixed values of the mass parameter in figs . [ fig_set15]d - l in the phase diagram . the particular choices of the mass are indicated with horizontal lines in the corresponding figs . [ fig_set15]a , b , c . let us discuss the notable features observed in these figures . the first line we consider is @xmath210 which is deep in the topological phase for weak disorder . the entanglement spectra for this line is given by the blue plots ( second row ) : fig.[fig_set15]d , e , f . along this line , the topological phase transition occurs at around @xmath211 . up to this value of disorder , we observe entanglement properties which are reminiscent of the clean system : the spatial entanglement receives contributions almost exclusively from two @xmath58 modes ; the momentum entanglement remains largely suppressed , albeit increasing gradually with disorder ; and the orbital entanglement modes evenly occupy the whole region between @xmath29 and @xmath30 ( compare the disordered orbital entanglement with the clean system shown in fig . [ fig_cleanconf ] ) . as the disorder gets close to @xmath146 , there are also clear signatures of when the topological phase transition occurs : at the critical point , the spatial and orbital entanglement modes begin to diverge toward @xmath29 and @xmath30 , while the momentum entanglement modes collapse toward @xmath58 . this is the type of behavior that is expected for the system as it transits toward the trivial anderson insulator phase . the second mass line we consider is @xmath212 at which something interesting occurs . for this value of the mass parameter , the system is in the trivial phase at zero disorder . what is special about this value of the mass parameter is that , although it starts off from a trivial phase , the disorder is able to induce a topological state that occurs approximately between @xmath213 and @xmath214 . the entanglement spectra for this line are shown in red plots ( third row ) in fig . [ fig_set15]g , h , i . in this region of intermediate disorder , all three types of entanglement spectra exhibit the type of behavior we observed for the @xmath210 case in the topological regime . beyond @xmath215 , the trivial anderson insulator is obtained , with the corresponding entanglement properties matching the expected behavior for such a phase . the final line we consider has @xmath216 . for this value of the mass parameter , the system is in the trivial phase at zero disorder , and remains in that phase for all values of the disorder strength . the entanglement spectra for this line is given by the green plots ( fourth row ) in fig . [ fig_set15]j , k , l . the behavior for all three types of entanglement is the same as what we would expect for a localized insulator . in the momentum entanglement spectrum in fig . [ fig_set15]k we see the feature on which we commented earlier , namely that even when starting in the trivial insulator phase there is a surge of momentum entanglement , presumably due to the transition from a band insulator to a localized anderson insulator . from these results , we would like to bring attention to the behavior of both the momentum and orbital entanglement , as this will help motivate another type of entanglement cut . first , note that , even though the momentum entanglement does not reveal clearly the difference between the topological and trivial states clearly even at zero disorder , it does appear to reveal the point at which the trivial anderson insulator is obtained when the disorder becomes strong enough . this is a nontrivial result , because even while the system is topological , the single - particle states are localized ( they can be represented as exponentially localized wannier functions ) , which might naively have led us to assume that the momentum entanglement spectrum should saturate at @xmath58 even for weak disorder . however , from our results it is clear that the saturation near @xmath58 occurs only at a finite disorder strength . the connection of momentum entanglement saturation with the occurrence of the trivial anderson insulator is consistent with what we observed for the metal - insulator insulator transitions of the previous sections . this thus suggests that this connection is generic , at least for one - dimensional systems . we will exploit this observation in what follows . regarding the orbital entanglement , we now argue that this type of entanglement is intimately related with the winding property of the ground state . to understand this , let us consider the orbital entanglement in the clean limit . since the wave functions of the clean system can be computed analytically , the orbital entanglement can be obtained explicitly as @xmath217 . \label{orbtent}\end{aligned}\ ] ] we show the orbital entanglement as a function of the mass parameter in fig . [ fig_cleanconf ] . the curves followed by the orbital entanglement modes are very reminiscent of the behavior observed in figs . [ fig_set15 ] f , i , l , with the important difference that in the earlier set of figures it is the disorder which is being varied , whereas in the clean limit we only vary the mass parameter . we thus see that the entanglement modes @xmath218 exhibit spectral flow from @xmath29 up until @xmath30 as @xmath53 goes from @xmath29 to @xmath219 only when @xmath220 i.e. , in the topological phase . since the momentum part of the correlation matrix is diagonal in the clean limit , these entanglement modes originate exclusively from the entanglement of the internal degree of freedom at each value of @xmath53 . furthermore , the flow of entanglement modes when the system is topological occurs essentially because of the winding of the @xmath197 vector as the brillouin zone is traversed . this suggests that the flow of entanglement modes in figs.[fig_set15 ] f , i , l are describing the unwinding of the internal degree of freedom as the system transits from a topological to a trivial state . observations of this nature were also pointed out for two - dimensional topological models in ref . [ ] , albeit exclusively for clean systems . this , together with the behavior of the momentum entanglement , suggests a useful quantity would be a mixed type entanglement which we discuss in the next section . to define the hybrid momentum - orbital entanglement let us start by considering the clean system . let us allow for the variation of the momentum wave vector @xmath221 at which we perform the entanglement cut in the brillouin zone , so that we keep @xmath222.$ ] in our partition we keep all @xmath53-states between @xmath223 $ ] and we keep all the corresponding orbital degrees of freedom except for @xmath224 for which we also remove one of the orbitals . thus , in one subspace we have @xmath225 states where @xmath226 is the number of allowed momenta in the set @xmath223 $ ] , and in the other subspace we have @xmath227 remaining states where @xmath228 is the total number of lattice sites . in the clean case , the correlation matrix is block - diagonal in the momentum @xmath53 so the the entanglement spectrum is composed of the set entanglement modes @xmath229 at each @xmath53 where @xmath230 is @xmath231-fold degenerate and the non - trivial eigenvalue @xmath232 arises from the orbital cut at @xmath233 in fig . [ fig_cleanmomcut ] , we make a comparison between the spatial entanglement and the hybrid entanglement as a function of the mass parameter when the disorder is set to zero . in the hybrid entanglement there is a flow of entanglement modes between @xmath29 and @xmath30 when the system is topological as expected from eq . [ orbtent ] . in the trivial phase there is not a complete spectral flow as a function of @xmath233 this cut is not so revealing in the clean system as it just re - expresses eq . [ orbtent ] if we plot the entanglement spectra for the whole range of possible @xmath233 let us now add disorder to the system . we consider the aubry - andr potential first . in the sequence of plots shown in fig . [ fig_wdmom]a , we vary the disorder strength . for weak disorder , the entanglement spectrum largely retains the same structure we found in the clean system . as the disorder keeps increasing , and localization sets in , the momentum entanglement modes that were at @xmath30 and @xmath29 start to drift towards @xmath58 , signaling an increase in momentum entanglement . interestingly , the entanglement modes that traverse the entanglement gap are still present , and suggest that the system is still topological . this is corroborated with a calculation of the topological invariant , which is still quantized to @xmath202 at that disorder strength . finally , when the disorder strength crosses the critical point , the momentum entanglement modes fill the gap . for comparison , and to illustrate the generality of our observations , we also show the the case of uncorrelated disorder , for one disorder configuration , in fig . [ fig_wdmom]b . it is nice to confirm that this type of entanglement can also be illustrative without disorder averaging . the overall behavior of the anderson transition for uncorrelated disorder matches what is observed for the transition in the aubry - andr potential for the topological insulator . from these observations , we find that there is a real sense in which the topological winding of the ground state is still encoded in the disordered system . even though the disorder induces scattering and tends to randomize the internal configuration of the single - particle states , the correlation between momentum and internal degree of freedom persists throughout the topological state . this manifests itself consistently as a spectral flow of orbital entanglement modes as a function of the momentum entanglement cut . the topological phase transition toward the trivial state eventually occurs when the scattering is strong enough that entanglement modes saturate @xmath58 , which we have identified throughout this work as leading to the trivial anderson insulator . it would be interesting to see how these results can be applied to higher - dimensional topological systems . in higher dimensions the scattering kinematics are much less constrained and it is not clear how well the topological information would be encoded in the momentum entanglement . however , here we see that we can capture the correlations between the momentum and the internal degrees of freedom which is a fundamental feature of free - fermion topological phases , so perhaps this tool will still be useful . it may also be useful to consider this type of hybrid cut for numerically identifying 1d interacting topological phases of fermions or even in spin - systems . in this work we have discussed the signatures in the entanglement of disordered fermions when they undergo anderson localization transitions . disordered metallic states in 1d that have suppressed scattering exhibit marked signatures in the momentum entanglement spectrum . delocalized states occur when momentum entanglement is suppressed and the anderson insulator is obtained when there is a saturation of momentum entanglement modes near @xmath58 . the momentum entanglement entropy is able to yield the disorder strength for which localization occurs , even in the presence of hubbard - type interactions . regarding disordered 1d topological states , we have found that the nontrivial nature of the ground state can be captured by performing an appropriate simultaneous entanglement cut in momentum and orbital space . this type of entanglement cut suggests that the ground state still has some correlation between the momentum and the orbital degree of freedom which manifests in the nontrivial topological boundary states of the system . it would be interesting to further extend this study to higher - dimensional systems and with other symmetries . in 2d and in 3d there are interesting models that exhibit metal - insulator transitions as well as diverse nontrivial topological properties . consequently , one might encounter further interesting and useful features in the momentum entanglement that probe the quantum properties of the ground state . however , because of the much wider range of scattering kinematics , we expect that the extension of the pure momentum entanglement will be challenging . some work in this direction was carried out recently in ref . [ ] where indeed they find the higher - dimensional extension to be challenging . we would like to thank e. prodan and j. song for useful discussions , and the uiuc icmt for support . this work was supported by onr award n0014 - 12 - 1 - 0935 . in order to compute the localization length , we follow the calculation in terms of transfer matrices detailed in @xcite . this can be achieved by first computing numerically the matrix @xmath116 given by @xmath234^{1/2l}\ ] ] where the transfer matrix @xmath235 at the position @xmath69 and energy @xmath236 is given by @xmath237 the eigenvalue of the matrix @xmath116 which is closest to unity can be written as @xmath238 . finally , the exponent @xmath239 is the so - called lyapunov exponent which is related to the localization length by @xmath240 . we can obtain the polarization of a system of 1d fermions by computing the eigenvalues of the projected position operator @xmath241 where @xmath242 is the projection operator of the occupied states and @xmath243 is the position operator . in a finite system with periodic boundary conditions , the position operator is written in the exponential form @xmath244 where @xmath245 are single - particle position basis states . using the set of eigenvalues @xmath246 of @xmath247 , the polarization is then given by @xmath248 which just measures the relative shift of the wannier centers with respect to the lattice sites .

we present empirical relations between the global dust reddening and other physical galaxy properties including the h luminosity , h surface brightness , metallicity and axial ratio for star - forming disc galaxies . the study is based on a large sample of 22 000 well - defined star - forming galaxies selected from the sloan digital sky survey ( sdss ) . the reddening parameterized by color excess is derived from the balmer decrement . besides the dependency of reddening on hluminosity / surface brightness and gas phase metallicity , it is also correlated with the galaxy inclination , in the sense that edge - on galaxies are more attenuated than face - on galaxies at a give intrinsic luminosity . in light of these correlations , we present the empirical formulae of as a function of these galaxy properties , with a scatter of only 0.07 mag . the empirical relation can be reproduced if most dust attenuation to the hii region is due to diffuse interstellar dust distributing in a disc thicker than that of hii regions . the empirical formulae can be incorporated into semi - analytical models of galaxy formation and evolution to estimate the dust reddening and enable comparison with observations more practically . [ firstpage ] galaxies : ism galaxies : abundance hii regions dust , extinction .

1112.0063

dust is a crucial component of galaxies in modifying the observed properties of galaxies by absorbing and scattering starlight , and also re - emits the absorbed energy in the mid- and far - infrared bands . the extinction cross section of dust generally decreases with increasing wavelength , i.e. , the extinction is more severe at shorter wavelength , particularly at ultra - violet ( uv ) band . as a result , the spectral energy distributions ( sed ) of galaxies appear redder , which is so - called `` reddening '' . the effects of dust should be properly accounted for when interpreting the observations of the galaxies , for example the luminosity or star formation rate ( sfr ) of a galaxy . for a galaxy with spectroscopic data covering suitable wavelengths , it is usually possible to estimate the amount of dust extinction from the spectrum . for example , the average dust reddening of hii regions can be derived from balmer decrement , and be used to correct for the dust attenuation to the emission - line luminosity . in other cases , one has to rely on empirical relations to get an estimate of dust extinction in a statistical way . additionally , it is necessary to incorporate dust into the galaxy formation and evolution model . currently there are two ways to account for dust effects : full radiative transfer calculations assuming dust properties and geometric distribution , or adopting simple recipes obtained empirically . since the dust formation in the galactic environment is a complicated chemical process , it has not yet been implemented in the current galaxy evolution models . the dust properties and distribution are input in the models and should be tested with observations . alternatively , many authors have adopted the empirical relations to compute the perpendicular optical depth of a galactic disc , and then assign a random inclination angle for each galaxy to get the final dust corrections ( e.g. guiderdoni & rocca - volmerange 1987 ; kauffmann et al . 1999 ; somerville & primack 1999 ; de lucia , kauffmann & white 2004 ; de lucia & blaizot 2007 ; kang et al . 2005 ; kitzbichler & white 2007 ) . therefore , a well - defined empirical recipe of dust reddening will help the comparison of model predictions with observations . previous studies have suggested that dust reddening in star - forming galaxies is correlated with the sfr , which can be well estimated with the intrinsic h@xmath0 luminosity ( e.g. kennicutt 1998 ; calzetti et al . galaxies with large sfr ( or high luminosity ) show strong global extinction in the emission lines ( e.g. , wang & heckman 1996 ; charlot & fall 2000 ; calzetti 2001 ; stasiska & sodr 2001 ; afonso et al . 2003 ; kewley et al . 2004 ; zoran , barkana , & thompson 2006 ; calzetti et al . 2007 ; garn et al . 2010 ) , and also large infrared to ultraviolet flux ratios ( e.g. , iglesias - paramo et al . 2006 and reference therein ) . in addition , it has been recognized that dust reddening might also be a function of gas - phase metallicity , i.e. the reddening and extinction increase with metallicity . for instance , the average reddening to hii regions in an individual galaxy is correlated with the metallicity in the disc ( e.g. , quillen & yukita 2001 ; boisser et al . for the case of whole galaxies , the relationship between reddening and metallicity also exists ( e.g. heckman et al . 1998 ; buat et al . 2002 ; asari et al . it is also supported by the fact that the low - metallicity galaxies such as blue compact galaxies are usually less reddened ( e.g. kong 2004 ) . recently , garn & best ( 2010 ) also found a significant correlation between dust extinction and metallicity , however , they claimed that the dependency of reddening on stellar mass is more fundamental . they built a sample of about 90 000 star - forming galaxies using sloan digital sky survey ( sdss ) data , and compare the relationship between the dust extinction of h@xmath0luminosity and sfr , metallicity as well as stellar mass , respectively . they concluded that the dust extinction @xmath3 can be best predicted from the stellar mass , with a scatter of @xmath10.3 mag . besides , the dependency of dust attenuation on inclination has been also investigated . while the correlation between dust attenuation of optical stellar continuum and disc inclination has been well established for disc galaxies ( e.g. driver et al . 2007 ; shao et al . 2007 ; unterborn & ryden 2008 ) , that of emission lines is still uncertain ( e.g. yip et al . 2010 ) . these works focused on how the dust reddening depends on one property of galaxies , luminosity , metallicity or stellar mass , which are generally correlated with each other ( e.g. garnett & shields 1987 ; skillman et al . 1989 ; zaritsky et al . 1994 ; zahid et al . 2011 ; tremonti er al . disentangling the dependencies of dust reddening on these galaxy properties can not only help us to derive a more accurate empirical formulae for dust reddening , but also provide more insight to the dust formation within galaxy environment . in this work , we study a large sample of star - forming disc galaxies in the local universe with a median redshift of 0.07 , selected from the spectroscopy database of sdss . we obtain empirical formulae for dust reddening as a function of intrinsic h@xmath0 luminosity / surface brightness , gas - phase metallicity and the axial ratio of the disc . in section [ sec : sample ] we describe our sample selection , and in section [ sec : methods ] describe the spectral analysis and methods used to estimate the parameters such as the dust reddening , intrinsic h@xmath0 luminosity , surface brightness , metallicity and axial ratio etc . we present the resultant expression for the empirical formulae of dust reddening , and compare our results with previous work in section [ sec : relation ] . in section [ sec : toymodel ] we reproduce the observed trend with a toy model of parallel - slab disc , which gives some insights into the geometry of dust distribution and dust - to - gas ratio . conclusions are given in section [ sec : conclusion ] . throughout the paper , we will assume a cosmology of @xmath4 km s@xmath5mpc@xmath5 , @xmath6 and @xmath7 . we start from the spectroscopic sample of galaxies in the sdss data release five ( dr5 , adelman - mccarthy et al . sdss produces imaging and spectroscopic survey with a wide - field 2.5 m telescope at apache point observatory , new mexico ( york et al . 2000 ) . the survey provides imaging in five broad bands _ u , g , r , i , z _ , with magnitude limits of 22.2 in _ r _ band , and spectroscopic targets are selected using a variety of algorithms , including the `` main '' sample of galaxies with @xmath8-band magnitude brighter than 17.77 ( strauss et al . 2002 ) with fibers of 3 diameter . the spectra range from 3200 to 9200 at a resolution @xmath9 . dr5 spectroscopic area covers 5740 deg@xmath10 , and there are @xmath1675,000 spectra classified as galaxies . to select clean galaxies , we use the sdss photometric flags to eliminate the targets that are only one part of a large galaxy or a part of merging galaxies . we list the sample - selection cuts and corresponding number of remaining objects within each subsample in table [ tab : sample ] . then the galaxy spectra are corrected for the galactic extinction using the dust extinction map ( schlegel et al . 1998 ) with an extinction curve of fitzpatrick ( 1999 ) with @xmath11 . we select emission - line galaxies where the h@xmath0 has been detected at high significance , i.e. , s / n@xmath1220 . to include galaxies of high reddening , we use a looser criterion on h@xmath13 : s / n@xmath1210 . the emission lines are measured with the procedures described in details in section [ sec : specfit ] . in order to use the ratios of [ oiii]/h@xmath13 and [ nii]/h@xmath0 on line - ratio diagnostic diagrams ( e.g. bpt diagrams ; baldwin , phillips & terlevich 1981 ) for spectral classification , we also impose the criteria [ oiii ] s / n@xmath1210 and [ nii ] s / n@xmath125 . it should be noted that the requirement of lower limit in the s / n ratio for [ nii ] may drop some low metallicity galaxies , and that for [ oiii ] may miss high metallicity galaxies . we will justify the s / n criteria in section [ sec : erranaly ] . galaxies with significant broad h@xmath0 component are rejected . the broad - line active galactic nuclei ( agns ) are defined as objects for which adding an additional broad component of h@xmath0 to the emission - line model can significantly improve the fit to the h@xmath0+[nii ] blend ( refer to dong et al . 2005 , 2007 ; also zhou et al . 2006 ) . in practice , the galaxies with broad h@xmath0 component detected at the @xmath14 significance level are regarded as candidates of broad - line agns , and removed from the sample . we also remove the narrow - line active galaxies by using the bpt diagram ( kauffmann et al . 2003b , hereafter ka03 ; kewley et al . 2006 ) based on [ nii]/h@xmath0 . the galaxies below the ka03 pure star - formation line on [ nii]/h@xmath0 diagram are referred as our sample of star - forming galaxies . most of these galaxies lie below the extreme - starburst line on the [ sii]/h@xmath0 and [ oi]/h@xmath0diagrams , with a fraction of 99 percent and 94 percent , respectively . using a more strict criterion on selecting hii galaxies given by stasiska et al . ( 2006 ) will result in less metal - rich galaxies ( see [ sec : metal ] ) . as shown by kewley et al . ( 2005 ) , if the nuclear spectrum contains less than twenty percent of the total galaxy light , we will likely over - estimate the global metallicity and reddening , and under - estimate the global sfr by a significant fraction . therefore , we remove the galaxies for which the fiber magnitude at @xmath15-band is greater than the total magnitude by 1.7 mag . in the following analysis , we will pick out the late - type ( presumably disc - dominated ) galaxies from the star - forming galaxies based on the likelihoods provided by sdss pipeline . the photometry pipeline provides the likelihoods ( dev_l , exp_l , and star_l ) associates with the de vaucouleurs , exponential , and psf fits , respectively . the fractional likelihoods for the exponential fit is calculated as @xmath16 and similarly for @xmath17(exp_l ) and @xmath17(star_l ) . it is suggested that the fractional likelihood greater than 0.5 for any of the three model fits is generally good as a threshold for object classification ( stoughton et al . 2002 ) . for a galaxy , @xmath17(star_l ) is generally zero . we define a galaxy as a disc galaxy if the logarithmic likelihood for an exponential fit ( @xmath8-band ) is larger than that of a de vaucouleurs fit by 0.2 dex ( corresponding to @xmath17(exp_l)@xmath18 ) , while as an elliptical galaxy if the logarithmic likelihood for a de vaucouleurs fit is larger than that of an exponential fit by 0.2 dex ( @xmath17(dev_l)@xmath18 ) . this parameter is also correlated with the compactness index such as @xmath19 , which is commonly used in quantitative classification ( e.g. shimasaku et al . 2001 ; strateva et al . with this criterion , 23919 star - forming galaxies are classified as disc galaxies , and 7650 as elliptical galaxies , and 595 as unclassified type . [ cols="<,>,>",options="header " , ] in order to investigate the origin of the scatter in the empirical relations , we examine the correlation between @xmath20 and other galaxy parameters , such as stellar mass , the 4000 break index , ew of balmer absorption lines , the electron density in hii region or concentration index . the @xmath20 for either ( @xmath21 , @xmath22 , _ b / a _ ) model or ( @xmath23 , @xmath22 , _ b / a _ ) model is not correlated with any of the parameters above , implying that the dust reddening in a disc galaxy is primarily determined by only three physical properties : luminosity ( sfr ) / surface brightness ( sfr surface density ) , metallicity and disc inclination . we also attempt another multi - parameter model ( @xmath24 , @xmath22 , _ b / a _ ) with similar procedure described above , and find that the scatter of @xmath20 is 0.072 mag , which is slightly larger than the two models above ( equation [ eq : powfit ] and [ eq : powfit_br ] ) . however , @xmath20 for this model is still weakly correlated with @xmath21 ( @xmath25 , @xmath26 ) . this suggests that ( @xmath21 or sfr , @xmath22 ) combination works better than ( @xmath24 , @xmath22 ) combination . physically , at a given metallicity , the amount of dust is proportional to the amount of cold gas , which is closely correlated with sfr . we compare our models with previous empirical relations . garn & best ( 2010 ) investigate the dependency of dust extinction at h@xmath0 on sfr , metallicity and stellar mass respectively , in star - forming galaxies based on sdss database . they conclude that the stellar mass is the most fundamental parameter in determining the dust extinction in the local universe , and their relation between stellar mass and dust extinction seems to hold up to redshift of 1.5 ( sobral et al . they utilize different models to fit the extinction @xmath27 , with each model taking into consideration one of the three parameters : sfr , @xmath24 and metallicity . they find that the @xmath24 model gives the best fit , with the minimum of scatter ( 0.28 mag ) in the extinction residual @xmath28 , less than the scatter for sfr model and metallicity model , both of which are 0.33 mag . when calculating the @xmath27 , they adopt the calzetti et al . ( 2000 ) dust attenuation - curve . we repeat their analysis for our selected sample of disc galaxies and compare the scatter of @xmath29 among different models . in each model a low - order ( third order ) polynomial is adopted to fit the extinction , and the distribution of @xmath29 is well fitted by a gaussian , with widths @xmath300.27 , 0.28 and 0.28 mag for the @xmath24 , @xmath21 ( sfr ) and metallicity model respectively . higher order polynomial fits will not decrease the scatter . we check the scatter of @xmath29 for our ( @xmath21 , @xmath22 , _ b / a _ ) or ( @xmath23 , @xmath22 , _ b / a _ ) model assuming calzetti s attenuation law , and the gaussian width is only @xmath31 mag ( corresponding to @xmath32 mag for the @xmath20 distribution ) . we also check for the ( @xmath24 , @xmath22 , _ b / a _ ) model , and find that the scatter of @xmath29 is 0.24 mag , which is better than the single - parameter models , but worse than the other two multi - parameter models . our empirical relations predict dust reddening better than the single - parameter models , probably because we include disc inclination in the analysis . in order to investigate what dominates the relationships of the variables , we apply principle component analysis on the physical variables . the derived principle components ( pcs ) are linear combinations of the variables , and represent for the directions of maximum variance in the data . for h@xmath0 luminosity and surface brightness , @xmath2 , @xmath24 and metallicity , the quantities are normalized to zero mean and standard deviation of unity , while for the axial ratio original value is used . we test with different subsets of variables adopted in each multi - parameter model as well as all of the variables , and find similar results . take the case of ( @xmath21 , @xmath22 , _ b / a _ ) model for example , the first two pcs contribute to about 80 percent of the total variance . in pc1 , the variables @xmath21 , @xmath22 and @xmath2 weight almost equivalently . this is consistent with what garn & best ( 2010 ) found . but in pc2 the axial ratio weights the most ( -0.98 ) and @xmath2 weights the second ( 0.30 ) . the result is consistent with what we previously found . although the axial ratio only weakly correlated with @xmath2 ( @xmath33 , @xmath26 ) , after removing the @xmath2 dependence on h@xmath0 luminosity and metallicity , the negative correlation between @xmath34 and @xmath2 emerges . we have shown that the global reddening of the emission lines from a disc galaxy can be reasonably determined by its h@xmath0 luminosity or surface brightness , gas metallicity and disc inclination . the @xmath35 scatter of @xmath20 is about 0.07 mag , which is contributed by several sources , including the measurement uncertainty in the balmer decrement , the error in the metallicity , the uncertainties in other variables and the intrinsic scatter in the relation . we will discuss the uncertainties in the following . let us take a review of the uncertainties in the physical quantities . the formal error for @xmath21 measurement is 0.04 dex , and for @xmath23 is 0.05 dex . as discussed in section [ sec : lhacor ] , the uncertainty induced by aperture - correction and attenuation - correction is less than 0.12 dex . the typical error of @xmath34 is 0.02 . the uncertainty in metallicity calibration dominates the uncertainty of metallicity , estimated to be @xmath36 dex . we use monte carlo simulation to estimate the error of @xmath2 for equation [ eq : powfit ] and [ eq : powfit_br ] taking the uncertainties for the best - fit parameters and the measurement errors for the physical variables . the errors of @xmath2 predicted by equation [ eq : powfit ] and [ eq : powfit_br ] are both about 0.056 mag . if taking the uncertainty in aperture - correction and attenuation - correction ( 0.12 dex at most ) into account in @xmath21 and @xmath23 error , the simulated @xmath2 errors slightly changes to 0.058 mag . remember that the measurement error of @xmath2is 0.03 mag , the scatter of @xmath20 ( 0.07 mag ) can be almost explained by the uncertainties of the physical quantities . considering all these uncertainties , @xmath2 predicted from the empirical relations are quite consistent with the observed values . from figure [ fig : debv - dist ] we can see that the @xmath20 distribution can be well described by a gaussian profile , indicative of a random distribution . however , the intrinsic scatter of the empirical relation is still unclear , because of the large uncertainty in the metallicity calibration . therefore , to estimate the intrinsic scatter of the relation , it will be of help to decrease the uncertainty in the metallicity calibration and obtain the metallicity more precisely . one may find there is slight excess over the positive tail of the gaussian distribution . we examine the objects ( about one percent of the whole sample ) with @xmath37 , i.e. the observed @xmath2 is under - estimated with our empirical models by 0.2 mag , and find that these objects are prone to large balmer decrement , low h@xmath13 s / n and low h@xmath13 ew . as previously mentioned , we require s / n@xmath38 for h@xmath0 , and s / n@xmath39 for h@xmath13 and [ oiii ] when selecting our sample in section [ sec : sample ] . the s / n criteria for h@xmath0 will exclude some galaxies with low sfr and low h@xmath0 ew . thus the number of galaxies with low h@xmath0 luminosity and low sfr will be reduced . in fact , when the criteria s / n@xmath39 is applied for h@xmath13 , most of the remaining sample ( @xmath199.9 percent ) will have h@xmath0s / n@xmath38 . in order to investigate whether the h@xmath13 and [ oiii ] s / n criteria bias our results , we attempt a looser s / n(@xmath40 ) criteria for both lines , and the sample is expanded by 50 percent to have 34815 galaxies . then we fit the data with both the ( @xmath21 , @xmath22 , _ b / a _ ) model and ( @xmath23 , @xmath22 , _ b / a _ ) model . the results show that the fit parameters are quite similar as those we have obtained in section [ sec : relation ] , but the scatter in @xmath20 increases to about 0.08 mag . note that 10 percent uncertainty in h@xmath0/h@xmath13 will result in an uncertainty of 0.086 mag in @xmath2 , which explains the increased scatter . thus including the low s / n objects merely introduces more scatter in the relation . two different sources contribute to the extinction for hii regions : dust associated with the individual hii region and diffuse foreground dust in the galaxy . let us first consider the dust in the precursor molecular clouds ( mc ) . the intrinsic extinction in individual hiiregion depends on the relative distribution of young massive stars , gas and dust in the star - formation regions , which are also closely related to the age and size of the precursor mc , as well as the stochastic formation of high mass stars for a given initial mass function . an hii region is embedded in the cold gas / dust envelope before the cold gas / dust is blown away . the time scale for such a process is expected to depend on the size and density of the precursor cloud : longer for more massive and dense hii region . panuzzo et al . ( 2003 ) discussed the attenuation to the h@xmath0 emission line for different evaporating time scale , in comparison with the lifetime of massive stars that produce the ionizing continuum . they simply adopt a central ionizing star in a spherical cloud , which corresponds to the maximum attenuation case . if the time scale is short enough , most hiiregions are produced outside of the mc , and the main attenuation is due to foreground dust . in the case of long evaporation time , an hii region is buried in the mc during most of their life time and the extinction associated with the hii region will be important . in a galaxy , all these cases are likely present . furthermore , the young ionizing stars are not necessarily forming in the cloud centre . these complications make any prediction of the reddening in an individual cloud rather difficult . however , as far as a large number of hii regions are concerned , the statistical behavior may be quite well defined . the h@xmath0 luminosity function of hii regions in spiral galaxies can be described by double power - law functions , with a flatter power - law below the break luminosity of 38.6 ( in units of ergs s@xmath5 , logarithmic ) , and a steeper one above it ( e.g. kennicutt et al . 1989 ; bradley et al . 2006 ) . in our sample , most galaxies have h@xmath0 luminosities above 40 ( in units of ergs s@xmath5 , logarithmic ) with a median of 41.5 , and the maximum is 43.3 . therefore , a galaxy consists of numerous ( order of hundreds to millions of ) hii regions . the hiiregions follow certain distributions of ages and initial masses , determined by the star - formation processes , because the star - formation time in a galaxy is likely much longer than the lifetime of an individual hii region , i.e. the life time of ob stars . if these distributions are the same for each galaxy and there are a large number of hii regions , one may expect that the average extinction within hii regions should be similar in each galaxy . in this respect , it is not expected that the dust reddening depends on the h@xmath0 luminosity or the inclination of the disc . in contrast , for example , if the mc mass function and stellar imf depend on the global properties of the galaxies , such as the gas surface density , then one may expect that the average extinction is correlated with the global properties . however , this is poorly known so far . next , we consider the extinction due to diffuse interstellar dust in a galaxy . it is expected that this extinction depends on the gas surface density , inclination , metallicity of the interstellar medium and size of the galaxy . to simplify the model , we will adopt smooth distributions for the dust , gas and star - formation regions with the following assumptions : ( 1 ) : : dust - to - gas density ratio is proportional to the gas metallicity ( issa et al . 1990 ) , i.e. , gas and dust are coupled : @xmath41 where @xmath42 and @xmath43 are dust density and gas density , respectively ; ( 2 ) : : both gas and dust distributions in the vertical ( @xmath44 ) and radial ( @xmath45 ) direction can be described by an exponential law ( e.g. de jong 1996 ; thomas et al . 2004 ; misiriotis et al . 2006 ) , i.e. , @xmath46 where @xmath47 is the gas density at @xmath48 and @xmath49 , @xmath50 and @xmath51 are the characteristic lengths in the vertical and radial directions for gas distribution . but the dust distribution may have different length scales due to radial metallicity gradient : @xmath52 where @xmath53 is the dust density at @xmath48 and @xmath49 , @xmath54 and @xmath55 are the characteristic lengths in the vertical and radial directions for dust distribution ; ( 3 ) : : the hii distribution is also exponential in both radial and vertical direction , but with different characteristic lengths ( e.g. de jong 1996 ) : @xmath56 by integrating @xmath57 over @xmath44 direction , we yield a surface density of h@xmath0 , @xmath58 . specifically , the h@xmath0surface density at @xmath49 is @xmath59 the gas disc is rather thin in general . as a result , dust attenuation can be modeled as plane - parallel except at extremely high inclinations . young stars are formed in dense molecular cores , which might be distributed in a plane thinner than the gas disc . for a plane - parallel model , the optical depth is expressed as @xmath60 where @xmath61 is the opacity coefficient , and @xmath62 is the inclination angle , and @xmath63 represents the line of sight perpendicular to the galaxy disc , while @xmath64 represents the line of sight parallel to the galaxy disc . the optical depth to h@xmath0 emission from the middle plane of the disc to the upper disc surface at @xmath49 is defined as @xmath65 it is symmetric between the middle plane to the upper and lower disc surface , thus the overall optical depth through the whole disk vertically at @xmath49 should be @xmath66 . then we consider the optical depth from the point at @xmath44(@xmath67 ) on the vertical axis ( @xmath49 ) to the upper surface of the disc is @xmath68 and the optical depth from that point to the lower disc surface can be expressed as @xmath69 light from the hii regions at the position ( @xmath44 , @xmath45 ) needs to travel through the disc , from both upper and lower sides . with these descriptions , we can calculate the h@xmath0 luminosity , h@xmath0 surface brightness and average attenuation at h@xmath0and h@xmath13 by integrating the radiation transfer function over both vertical and radial directions within the disc . then the h@xmath0 luminosity is @xmath70 + exp[-\frac{\tau_{2}(z)}{\cos i } \exp(-r / r_\mathrm{d } ) ] \}. \label{eq : lhal0}\end{aligned}\ ] ] substitue equation [ eq : rhoha ] , [ eq : tau_up ] , [ eq : tau_low ] , @xmath71 and @xmath72 into equation [ eq : lhal0 ] , then we get @xmath73 + \nonumber \\ & \hspace*{2em } \exp[-\frac{\tau_\mathrm{{h\ensuremath{\alpha}},0}}{\cos i}\exp(-a/\eta)(2-y^{1/\kappa } ) ] \ } , \label{eq : lhal}\end{aligned}\ ] ] where @xmath74 is set by the threshold surface density of star formation @xmath75 ; and the threshold gas surface density is adopted as @xmath76 pc@xmath10(martin & kennicutt 2001 ) ; @xmath77 is the surface density of gas at @xmath49 ; @xmath78 and @xmath79 . in order to reduce the number of free parameters ( @xmath80 , @xmath81 , @xmath82 , @xmath83 , @xmath84 and @xmath85 ) , we use the empirical relation between the sfr surface density @xmath86 and gas surface density @xmath87 ( known as schmidt - kennicutt law ; kennicutt et al . 1998 ) : @xmath88 reminding that the gas and hii distributions are both exponential , we will get @xmath89 . with above relations and the conversion between @xmath21 and sfr ( equation [ eq : sfr - lha ] ) , we can write : @xmath90 assuming the dust - to - gas ratio to be the same as in the galactic disc , i.e. , @xmath91 ( bohlin 1978 ) and a galactic extinction curve ( fitzpatrick 1999 ) , we obtain the effective optical depth to h@xmath0 through the disc perpendicularly as @xmath92 once we have @xmath93 , it is straightforward to calculate the extinction at h@xmath0 using equation [ eq : lhal ] as following @xmath94 + \nonumber \\ & & \hspace*{2em } \exp[-\frac{\tau_\mathrm{{h\ensuremath{\alpha}},0}}{\cos i}\exp(-a/\eta)(2-y^{1/\kappa } ) ] \ } / \nonumber \\ & & \hspace*{3em } 2[1-\exp(-\xi)-\xi \exp(-\xi ) ] , \label{eq : ahal}\end{aligned}\ ] ] similarly , replacing @xmath81 with @xmath95 , one obtains the extinction at h@xmath13 . thus h@xmath0/h@xmath13 ratio can be calculated , and then @xmath21 can be corrected with the extinction estimated from the balmer decrement , as the extinction correction we make to the observed @xmath21 . in order to examine if such a model can reproduce the observed correlations , we compute the average surface brightness @xmath23 and color excess @xmath2estimated by h@xmath0/h@xmath13 from the model based on the following parameters : @xmath80 , @xmath22 , @xmath82 , @xmath84 and @xmath85 . to facilitate comparison with observation , we obtain @xmath23 following the same method used for observation ( equation [ eq : br ] ) by dividing the attenuation - corrected @xmath21 with the surface within the half - light radius , defined as the radius within which the h@xmath0 luminosity is half of the total @xmath21 . the @xmath2 is calculated from @xmath27 and @xmath96 assuming fitzpatrick s extinction curve . we first fix @xmath84 to be 1 . since increasing metallicity and inclination both cause an increase in extinction , these two factors are coupled in the model . therefore , we incorporate them into one parameter @xmath97 . we compute the models for a grid of parameters with @xmath98 , @xmath99 , @xmath100 . the models are over - plotted on the observation sample on the @xmath23 versus @xmath2 plane ( figure [ fig : br - ebv ] ) . for clarity , we only show the observation sample in five bins of @xmath101 with a bin - width of 0.2 , and on each panel the subsample is binned in @xmath23 and shown in squares with error bars . we can see the observation trend on each panel , that @xmath2increases as @xmath23 increases . the model curves are calculated with the corresponding value of median @xmath101 for each subsample . on each panel , @xmath2 gets larger when @xmath85 increases , and each curve with constant @xmath85 first increases and then becomes flat as @xmath23 increases . at small @xmath85 , i.e. the dust layer is much thinner than the stellar disc , @xmath2 saturates at a small value because the balmer - line emission from the disc of larger height tends to dominate , resulting in a small @xmath2 , while at large @xmath85 more balmer - line emission comes from the inner region close to the midplane of the disc , thus tend to have a large @xmath2 . the saturation is due to the fact that the observed light from outside weights more as the optical depth increases . generally the @xmath102 model reproduces the observed correlations at low @xmath101 . however , as @xmath101 increases , the constant @xmath85 model is lower than the observation trend at the high - end of @xmath23 . at medium @xmath101 ( 1.5@xmath12.5 ) , the observation can be better presented by the model with larger @xmath85 ( e.g. @xmath103 ) . this could be explained by the vertical expansion of dust distribution in the disc while higher surface brightness indicates intenser star - formation activity . the dust may be driven away from the midplane of the disc by radiation pressure or the stellar winds , thus the dust layer become thicker than that for low @xmath23 . but @xmath2 saturates at about 0.3 mag when @xmath103 , which means that the effective optical depth saturates when the height of dust layer gets to twice height of hii regions . another possible explanation is that the kennicutt - schimidt law we adopt in the model is an averaged star formation law that linked between the surface density of gas and star formation as the form of @xmath104 . leroy et al . ( 2008 ) suggested that the star formation law may vary as a function of local conditions at different galactocentric radius , including metallicity and the dominated gas content ( hi or h@xmath105 ) . the sf law has smaller @xmath106 value at higher @xmath86 in the inner region of a galaxy ( e.g. bigiel et al . 2008 , their figure 11 ) , thus large dust reddening in the inner region dominates the observed @xmath2 . as a result , the dust reddening is higher than that the present model predicts . in the model above , we assume @xmath107 , i.e. the dust and hii region has the same characteristic scale in the radial distribution , but the gas / dust may distribute at larger radii than the star - forming regions ( e.g. bigiel et al . 2008 ) . thus we also consider @xmath108 , and find that the observation with large @xmath101 can be well explained by model with @xmath102 at low @xmath23 and @xmath103 at high @xmath23 . it is possible because in a spiral galaxy centre , where the metallicity is higher than outside , we are observing the star - formation dominated centre , and the gas ( or dust ) distribution is more extended . in this simple toy model , the optical depth @xmath109 is assumed to be proportional to the metallicity @xmath22 . it comes from the assumption that @xmath110 . however , there is evidence that this assumption may be not the case in practice . boissier et al . ( 2004 ) studied six nearby late - type galaxies in fir and uv images , and find that the dust - to - gas ratio is proportional to @xmath111 , which was flatter than the linear correlation we adopted above . this suggests that the dust - to - gas ratio may not evolve linearly with @xmath22 . in addition , we assume in the model that the disc height is very thin , i.e. the disc height is quite small compared to the disc radius , thus the radial gradient in gas density can be neglected when calculating @xmath109 . under that assumption , the effect of inclination on optical depth can be expressed as @xmath112 . however , in practice , the disc may not have a uniform height , for instance , it may be thicker in the centre and thinner at large radii , or the presence of bulge in the disc centre . also , the distribution of dust and star - formation regions may be not smooth , for example when the spiral arms are present in the disc galaxy . in those cases , the inclination factor might be flatter than @xmath112 . accounting for these realistic effects , we adjust @xmath113 to @xmath114 in equation [ eq : tau0 ] and @xmath112 to @xmath115 in equation [ eq : lhal ] and [ eq : ahal ] , then test several values for the power - law index is adopted to be the same as that of @xmath22 , so as to be incorporated into @xmath101 . ] . it turns out that with @xmath116 the simple model reproduces the observation trend well ( figure [ fig : br - ebv - z07 ] ) . when @xmath101 is low , the model of constant @xmath85 ( 0.5 or 1 ) reproduces the observation trend well . while @xmath101 is higher , the observation can be reproduced with the model of @xmath103 . , except that the model is calculated with equation [ eq : tau0 ] and [ eq : ahal ] , but adjusting @xmath113 to @xmath117 , and @xmath82 to @xmath118 . , scaledwidth=48.0% ] to summarize , the simple toy models reproduce the observed trends between dust reddening and h@xmath0 surface brightness , that reddening increases with @xmath23 . the comparisons between observation and the toy models indicate that , the relative distribution of hii regions to gas / dust both in vertical and radial direction of the galaxy disc , may vary in different environment of star - formation activity levels or metallicities . the dust might be distributed further beyond the hii regions when the star formation activity is intenser , or the dust is distributed in a thicker disc ( @xmath119 ) than the hii regions when the metallicity gets higher . in addition , our results imply that dust - to - gas density ratio may be not proportional to the metallicity @xmath22 , but to a power - law of @xmath22 , @xmath120 with @xmath121 . we present the empirical formulae of intrinsic reddening as a function of intrinsic h@xmath0 luminosity or surface brightness , metallicity and disc inclination ( equation [ eq : powfit ] , [ eq : powfit_br ] and table [ tab : lha ] and [ tab : br ] ) for a large sample of @xmath122 000 well - defined star - forming disc galaxies selected from the sdss . with the empirical formulae , the reddening parameterized by @xmath2 could be predicted within 1@xmath122 uncertainty of 0.07 mag . the @xmath20 , defined as the observed reddening estimated from the balmer decrement minus predicted value , does not correlate with the three parameters used in the formulae . we also find that @xmath20 is independent of the stellar mass , 4000 break strength or electron density within the galaxy . the observed trend between the dust reddening and h@xmath0 surface brightness could be reproduced by a plane - parallel slab toy model , in which the dust is scaled with gas , and all the distributions of the dust , gas and star - formation regions are smooth and follow exponential laws in the disc . we find from the comparisons of the models to the observation , the relative vertical scale of dust distribution to hii regions distribution may vary with h@xmath0 surface brightness or metallicity . the higher intensity of star formation could drive the dust to distribute in a thicker layer , or the dust disc is thicker than the disc of the hiiregions when the metallicity is higher . however , the real galaxies may be not fully described by this simple toy model . for example , when the spiral arms are present , or the sf regions are clumpy , the assumptions of smooth distributions will be improper . in addition , our result implies a different scaling law of dust - to - gas ratio as a function of metallicity @xmath22 from the linearly relation ( equation [ eq : rhod1 ] ) . our empirical relations , which suggest dust reddening is partially dependent on metallicity ( @xmath123 ) , can be an observational constraints on the output of radiative - transfer model calculations , and hence provides constraints on the assumed dust - to - gas ratio in the model . since metallicity - dependence has been introduced , we believe that our empirical formulae should be applicable to the early evolutionary stage of a galaxy , when the gas metallicity might be low . in the starburst galaxies at high redshift , dust surrounding star - forming region has been proven to be prone to sub - milimeter and far - infrared observations . but most of those galaxies are relatively dim in optical , implying severe extinction and reddening to these galaxies . the observed galaxies at high redshifts are affected by reddening , in the way that the galaxies with heavy attenuation and reddening may escape from the detection . the importance of the reddening correction depends on the wavelength band that is used . for example , uv continuum luminosity is usually used to infer sfr in high - redshift galaxies . it is very sensitive to the attenuation / reddening . calzetti et al . ( 2000 ) find that the extinction to emission lines and to stellar continuum in optical band are different , and the continuum reddening is only about 40 percent of that for emission lines . however , the extinction to uv continuum is likely similar to the hii region because uv continuum is also emitted by young hot stars , which are likely surrounded by hiiregions . as a result , our results may be used to estimate the reddening for these high - redshift galaxies with caution , that the metallicity should be well - calibrated and be within the range covered by our sample . following equation [ eq : powfit ] and the discussions in section [ sec : relation ] , sfr in luminous star formation or high metallicity galaxies might have been under - estimated ( e.g. heckman et al . 1998 ; hopkins et al . 2001 ; panuzzo et al . 2007 ) , or those galaxies are even completely lost in optical surveys . the quantitative relationship of reddening with the metallicity , luminosity or surface brightness , and the inclination for disc galaxies can also be incorporated into the current semi - analytic models of galaxy formation and evolution ( e.g. croton et al . 2006 ; de lucia et al . 2004 ; kang et al . 2005 ) . in those models , a simple chemical enrichment scheme has been included , and the metal abundance has been actually predicted . with the known metallicity and the star formation rate ( intrinsic luminosity ) provided by the models , the reddening can be estimated accurately with our formulae , and then used to shape the spectral energy distribution and predict the emerging luminosity more realistically . we would like to thank the anonymous referee for useful suggestions that improved the paper . we thank gleniese mckenzie , peng jiang for helpful comments on the manuscripts . this work is supported by nsfc ( 10973013 , 11073017 , 11033007 ) , 973 program ( 2009cb824800 ) and the fundamental research funds for the central universities . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , 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 . adelman - mccarthy , j. k. , et al . 2007 , apjs , 172 , 634 afonso , j. , hopkins , a. , mobasher , b. , & almeida , c. 2003 , apj , 597 , 269 allende prieto , c. , lambert , d. l. , & asplund , m. 2001 , apj , 556 , l63 alonso - herrero , a. , rieke , g. h. , rieke , m. j. , colina , l. , prez - gonzlez , p. g. , & ryder , s. d. 2006 , apj , 650 , 835 asari , n. v. , cid fernandes , r. , stasiska , g. , torres - papaqui , j. p. , mateus , a. , sodr , l. , schoenell , w. , & gomes , j. m. 2007 , mnras , 381 , 263 asplund , m. , grevesse , n. , sauval , a. j. , allende prieto , c. , & kiselman , d. 2004 , a&a , 417 , 751 baldwin , j. a. , phillips , m. m. , & terlevich , r. 1981 , pasp , 93 , 5 balogh , m. l. , morris , s. , yee , h. , carlberg , r. , & ellingson , e. 1999 , apj , 527 , 54 bianchi , l. , & efremova , b. v. 2006 , aj , 132 , 378 bigiel , f. , leroy , a. , walter , f. , brinks , e. , de blok , w. j. g. , madore , b. , & thornley , m. d. 2008 , aj , 136 , 2846 blanton , m. r. , et al.2001 , aj , 121 , 2358 blanton , m. r. , & roweis , s. 2007 , aj , 133 , 734 boissier , s. , boselli , a. , buat , v. , donas , j. , & milliard , b. 2004 , a&a , 424 , 465 bradley , t. r. , knapen , j. h. , beckman , j. e. , & folkes , s. l. 2006 , a&a , 459 , l13 brinchmann , j. , charlot , s. , white , s. d. m. , tremonti , c. , kauffmann , g. , heckman , t. , & brinkmann , j. 2004 , mnras , 351 , 1151 bruzual , a. g. 1983 , apj , 273 , 105 bruzual , a. g. , & charlot , s. 2003 , mnras , 344 , 1000 buat , v. , boselli , a. , gavazzi , g. , & bonfanti , c. 2002 , a&a , 383 , 801 calzetti , d. 2001 , pasp , 113 , 1449 calzetti , d. , armus , l. , bohlin , r. c. , kinney , a. l. , koornneef , j. , & storchi - bergmann , t. 2000 , apj , 533 , 682 calzetti , d. , kinney , a. l. , & storchi - bergmann , t. 1994 , apj , 429 , 582 calzetti , d. , et al.2007 , apj , 666 , 870 calzetti , d. , et al.2010 , apj , 714 , 1256 chabrier , g. 2003 , pasp , 115 , 763 charlot , s. , & fall , s. m. 2000 , apj , 539 , 718 croton , d. j. , et al.2006 , mnras , 365 , 11 dale , d. a. , & helou , g. 2002 , apj , 576 , 159 de jong , r. s. 1996 , a&a , 313 , 377 de lucia , g. , & blaizot , j. 2007 , mnras , 375 , 2 de lucia , g. , kauffmann , g. , & white , s. d. m. 2004 , mnras , 349 , 1101 denicol , g. , terlevich , r. , & terlevich , e. 2002 , mnras , 330 , 69 dong , x .- b . , zhou , h .- y . , wang , t .- g . , wang , j .- x . , li , c. , & zhou , y .- y . 2005 , apj , 620 , 629 dong , x .- b . , et al . 2007 , apj , 657 , 705 driver , s. p. , popescu , c. c. , tuffs , r. j. , et al . 2007 , mnras , 379 , 1022 fischera , j. , dopita , m. a. , & sutherland , r. s. 2003 , apj , 599 , l21 fitzpatrick , e. l. 1999 , pasp , 111 , 63 flores , h. , hammer , f. , elbaz , d. , cesarsky , c. j. , liang , y. c. , fadda , d. , & gruel , n. 2004 , a&a , 415 , 885 gallazzi , a. , charlot , s. , brinchmann , j. , white , s. d. m. , & tremonti , c. a. 2005 , mnras , 362 , 41 garn , t. , & best , p. n. 2010 , mnras , 409 , 421 garn , t. , et al . 2010 , mnras , 402 , 2017 garnett , d. r. , & shields , g. a. 1987 , apj , 317 , 82 groves , b. a. , heckman , t. m. , & kauffmann , g. 2006 , mnras , 371 , 1559 guiderdoni , b. , & rocca - volmerange , b. 1987 , a&a , 186 , 1 heckman , t. m. , robert , c. , leitherer , c. , garnett , d. r. , & van der rydt , f. 1998 , apj , 503 , 646 ho , l. c. 2005 , apj , 629 , 680 hopkins , a. m. , connolly , a. j. , haarsma , d. b. , & cram , l. e. 2001 , aj , 122 , 288 iglesias - pramo , j. , et al . 2006 , apjs , 164 , 38 issa , m. r. , maclaren , i. , & wolfendale , a. w. 1990 , a&a , 236 , 237 kang , x. , jing , y. p. , mo , h. j. , borner , g. 2005 , apj , 631 , 21 kauffmann , g. , colberg , j. m. , diaferio , a. , & white , s. d. m. 1999 , mnras , 303 , 188 kauffmann , g. , et al . 2003 , mnras , 341 , 33 kauffmann , g. , et al . 2003 , mnras , 346 , 1055 kennicutt , r. c. , jr . 1998 , apj , 498 , 541 kennicutt , r. c. , jr . , edgar , b. k. , & hodge , p. w. 1989 , apj , 337 , 761 kewley , l. j. , & dopita , m. a. 2002 , apjs , 142 , 35 kewley , l. j. , dopita , m. a. , sutherland , r. s. , heisler , c. a. , & trevena , j. 2001 , apj , 556 , 121 kewley , l. j. , & ellison , s. l. 2008 , apj , 681 , 1183 kewley , l. j. , geller , m. j. , & jansen , r. a. 2004 , aj , 127 , 2002 kewley , l. j. , geller , m. j. , jansen , r. a. , & dopita , m. a. 2002 , aj , 124 , 3135 kewley , l. j. , groves , b. , kauffmann , g. , & heckman , t. 2006 , mnras , 372 , 961 kewley , l. j. , jansen , r. a. , & geller , m. j. 2005 , pasp , 117 , 227 kitzbichler , m. g. , & white , s. d. m. 2007 , mnras , 376 , 2 kraemer , s. b. , ho , l. c. , crenshaw , d. m. , shields , j. c. , & filippenko , a. v. 1999 , apj , 520 , 564 kroupa , p. 2001 , mnras , 322 , 231 kong , x. 2004 , a&a , 425 , 417 lacey , c. , & silk , j. 1991 , apj , 381 , 14 lee , j. c. , salzer , j. j. , & melbourne , j. 2004 , apj , 616 , 752 leroy , a. k. , walter , f. , brinks , e. , et al . 2008 , aj , 136 , 2782 liang , y. c. , yin , s. y. , hammer , f. , deng , l. c. , flores , h. , & zhang , b. 2006 , apj , 652 , 257 lonsdale , c. j. , et al . 2003 , pasp , 115 , 897 lu , h. , zhou , h. , wang , j. , wang , t. , dong , x. , zhuang , z. , & li , c. 2006 , aj , 131 , 790 malhotra , s. , & rhoads , j. e. 2004 , apj , 617 , l5 markwardt , c. b. 2009 , astronomical data analysis software and systems xviii , 411 , 251 martin , c. l. , & kennicutt , r. c. , jr . 2001 , apj , 555 , 301 mcgaugh , s. s. 1991 , apj , 380 , 140 misiriotis , a. , xilouris , e. m. , papamastorakis , j. , boumis , p. , & goudis , c. d. 2006 , a&a , 459 , 113 moustakas , j. , & kennicutt , r. c. , jr . 2006 , apj , 651 , 155 moustakas , j. , kennicutt , r. c. , jr . , & tremonti , c. a. 2006 , apj , 642 , 775 nagao , t. , maiolino , r. , & marconi , a. 2006 , a&a , 459 , 85 odonnell , j. e. 1994 , apj , 422 , 158 osterbrock , d. e. , & ferland , g. j. 2006 , astrophysics of gaseous nebulae and active galactic nuclei , 2nd . ed . by d.e . osterbrock and g.j . ferland . sausalito , ca : university science books , 2006 panuzzo , p. , bressan , a. , granato , g. l. , silva , l. , & danese , l. 2003 , a&a , 409 , 99 panuzzo , p. , granato , g. l. , buat , v. , et al . 2007 , mnras , 375 , 640 pettini , m. , & pagel , b. e. j. 2004 , mnras , 348 , l59 pilyugin , l. s. 2003 , a&a , 399 , 1003 pilyugin , l. s. , thuan , t. x. 2005 , apj , 631 , 231 quillen , a. c. , & yukita , m. 2001 , aj , 121 , 2095 rieke , g. h. , alonso - herrero , a. , weiner , b. j. , prez - gonzlez , p. g. , blaylock , m. , donley , j. l. , & marcillac , d. 2009 , apj , 692 , 556 salim , s. , rich , r. m. , charlot , s. , et al . 2007 , apjs , 173 , 267 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , apj , 500 , 525 shao , z. , xiao , q. , shen , s. , et al . 2007 , apj , 659 , 1159 shi , f. , kong , x. , & cheng , f. z. 2006 , a&a , 453 , 487 shimasaku , k. , et al . 2001 , aj , 122 , 1238 skillman , e. d. , kennicutt , r. c. , & hodge , p. w. 1989 , apj , 347 , 875 smail , i. , chapman , s. c. , blain , a. w. , & ivison , r. j. 2004 , apj , 616 , 71 sobral , d. , best , p. , matsuda , y. , et al . 2011 , mnras accepted , arxiv:1109.1830 somerville , r. s. , & primack , j. r. 1999 , mnras , 310 , 1087 stasiska , g. , cid fernandes , r. , mateus , a. , sodr , l. , & asari , n. v. 2006 , mnras , 371 , 972 stasiska , g. , mateus , a. , jr . , sodr , l. , jr . , & szczerba , r. 2004 , a&a , 420 , 475 storchi - bergmann , t. , calzetti , d. , & kinney , a. l. 1994 , apj , 429 , 572 stoughton , c. , lupton , r. h. , bernardi , m. , et al . 2002 , aj , 123 , 485 strateva , i. , et al . 2001 , aj , 122 , 1861 strauss , m. a. , weinberg , d. h. , lupton , r. h. , et al . 2002 , aj , 124 , 1810 swinbank , a. m. , smail , i. , chapman , s. c. , blain , a. w. , ivison , r. j. , & keel , w. c.2004 , apj , 617 , 64 thomas , h. c. , alexander , p. , clemens , m. s. , green , d. a. , dunne , l. , & eales , s. 2004 , mnras , 351 , 362 tremonti , c. a. , et al . 2004 , apj , 613 , 898 unterborn , c. t. , & ryden , b. s. 2008 , apj , 687 , 976 wang , b. , & heckman , t. m. 1996 , apj , 457 , 645 white , s. d. m. , & frenk , c. s. 1991 , apj , 379 , 52 white , s. d. m. , & rees , m. j. 1978 , mnras , 183 , 341 worthey , g. , & ottaviani , d. l. 1997 , apjs , 111 , 377 wu , h. , cao , c. , hao , c .- n . , liu , f .- s . , wang , j .- l . , xia , x .- y . , deng , z .- g . , & young , c. k .- s . 2005 , apj , 632 , l79 yasuda , n. , et al.2001 , aj , 122 , 1104 yip , c .- w . , szalay , a. s. , wyse , r. f. g. , et al . 2010 , apj , 709 , 780 york , d. g. , adelman , j. , anderson , j. e. , jr . , et al . 2000 , aj , 120 , 1579 zahid , h. j. , kewley , l. j. , & bresolin , f. 2011 , apj , 730 , 137 zaritsky , d. , kennicutt , r. c. , jr . , & huchra , j. p. 1994 , apj , 420 , 87 zhou , h. , wang , t. , yuan , w. , et al . 2006 , apjs , 166 , 128 zhu , y .- n . , wu , h. , cao , c. , & li , h .- 2008 , apj , 686 , 155 zoran , o. , barkana , r. , & thompson , r. i. 2006 , mnras , 368 , l47

a total dominating set in a graph is a set of vertices such that every vertex of the graph has a neighbor in the set . we introduce and study graphs that admit non - negative real weights associated to their vertices such that a set of vertices is a total dominating set if and only if the sum of the corresponding weights exceeds a certain threshold . we show that these graphs , which we call total domishold graphs , form a non - hereditary class of graphs properly containing the classes of threshold graphs and the complements of domishold graphs , and are closely related to threshold boolean functions and threshold hypergraphs . we present a polynomial time recognition algorithm of total domishold graphs , and characterize graphs in which the above property holds in a hereditary sense . our characterization is obtained by studying a new family of hypergraphs , defined similarly as the sperner hypergraphs , which may be of independent interest . * keywords : * total domination ; total domishold graph ; split graph ; dually sperner hypergraph ; threshold hypergraph ; threshold boolean function ; forbidden induced subgraph characterization . ( 2010 ) : 05c69

1303.0944

a possible approach for dealing with the intractability of a given decision or optimization problem is to identify restrictions on input instances under which the problem can still be solved efficiently . one generic framework for describing a kind of such restrictions in case of graph problems is the following : given a graph @xmath0 , does @xmath0 admit non - negative integer weights on its vertices ( or edges , depending on the problem ) and a set @xmath1 of integers such that a subset @xmath2 of its vertices ( or edges ) has property @xmath3 if and only if the sum of the weights of elements of @xmath2 belongs to @xmath1 ? property @xmath3 can denote any of the desired substructures we are looking for , such as matchings , cliques , stable sets , dominating sets , etc . if weights as above exist and are given with the graph , and the set @xmath1 is given by a membership oracle , then a dynamic programming algorithm can be employed to find a subset with property @xmath3 of either maximum or minimum cost ( according to a given cost function on the vertices / edges ) in @xmath4 time and with @xmath5 calls of the membership oracle , where @xmath6 is the number of vertices ( or edges ) of @xmath0 and @xmath5 is a given upper bound for @xmath1 @xcite . the advantages of the above framework depend both on the choice of property @xmath3 and on the constraints ( if any ) imposed on the structure of the set @xmath1 . for example , if @xmath3 denotes the property of being a stable ( independent ) set , and set @xmath1 is restricted to be an interval unbounded from below equivalently , @xmath1 is of the form of the form @xmath7 $ ] for some non - negative integer @xmath8 , we obtain the class of _ threshold graphs _ @xcite , which is very well understood and admits several characterizations and linear time algorithms for recognition and for several optimization problems @xcite . if @xmath3 denotes the property of being a dominating set and @xmath1 is an interval unbounded from above , that is , it is of the form @xmath9 for some non - negative integer @xmath8 , we obtain the class of _ domishold graphs _ @xcite , which enjoy similar properties as the threshold graphs . on the other hand , if @xmath3 is the property of being a _ maximal _ stable set and @xmath1 is restricted to consist of a single number , we obtain the class of _ equistable graphs _ @xcite , for which the recognition complexity is open ( see , e.g.,@xcite ) , no structural characterization is known , and several @xmath10-hard optimization problems remain intractable on this class . for instance , it is shown in @xcite that the maximum independent set and the minimum independent dominating set problems are apx - hard in the class of equistable graphs . as the above examples show , the resulting class of graphs can be either _ hereditary _ ( that is , closed under vertex deletion)as in the case of threshold or domishold graphs , or not as in the case of equistable graphs . when the resulting graph class is not hereditary , it is natural to consider the hereditary version of the property , in which the requirement ( the existence of weights and the set @xmath1 ) is extended to all induced subgraphs of the given graph . in this paper , we introduce and study the case when @xmath3 is the property of being a total dominating set and @xmath1 is an interval unbounded from above . given a graph @xmath11 , a _ total dominating set _ ( a td set , for short ) is a subset @xmath12 of the vertices of @xmath0 such that every vertex of @xmath0 has a neighbor in @xmath12 . this notion has been extensively studied in the literature , see , e.g. the monographs @xcite and the survey paper @xcite . a graph @xmath13 is said to be _ total domishold _ ( td for short ) if there exists a pair @xmath14 where @xmath15 is a weight function and @xmath16 is a threshold such that for every subset @xmath17 , @xmath18 if and only if @xmath19 is a total dominating set in @xmath0 . a pair @xmath14 as above will be referred to as a _ total domishold structure _ of @xmath0 . notice that the above definition allows @xmath0 to have isolated vertices . every graph with an isolated vertex is total domishold , even though it does not have any td sets . the complete graph of order @xmath6 is total domishold . indeed , a subset @xmath20 is a total dominating set of @xmath21 if and only if @xmath19 is of size at least two , and consequently the pair @xmath22 where @xmath23 for all @xmath24 is a total domishold structure of @xmath21 . on the other hand , the @xmath25-cycle @xmath26 is not a total domishold graph ( cf . theorem [ prop : forbidden - induced - subgraphs ] in section [ sec : htd - partial - characterizations ] ) . it is easy to see that adding a new vertex to the @xmath25-cycle and connecting it to exactly one vertex of the cycle results in a total domishold graph . therefore , contrary to the classes of threshold and domishold graphs , the class of total domishold graphs is not hereditary . this motivates the following definition : a graph @xmath0 is said to be _ hereditary total domishold _ ( htd for short ) if every induced subgraph of it is total domishold . we initiate the study of td and htd graphs . we identify several operations preserving the class of td graphs , which , together with results from the literature @xcite , imply that the class of htd graphs properly contain the classes of threshold graphs and the complements of domishold graphs ( section [ sec : properties ] ) . we exhibit close relationships between total domishold graphs , threshold boolean functions and threshold hypergraphs ( in section [ sec : td - graphs - bfs - hypergraphs ] ) , and use them to develop a polynomial time recognition algorithm for total domishold graphs , a polynomial time algorithm for the total dominating set problem in the class of total domishold graphs , and a @xmath27-approximation algorithm for the dominating set problem in the class of total domishold graphs ( in section [ sec : algorithmic - aspects ] ) . as our main result , we completely characterize the htd graphs ( in section [ sec : htd - partial - characterizations ] ) , both in terms of forbidden induced subgraphs , and in terms of properties of certain boolean functions derived from induced subgraphs of the graph . we conclude the paper with some open problems in section [ sec : conclusion ] . in this section we recall some useful definitions and results . for a positive integer @xmath6 , we will use the notation @xmath28 $ ] for the set @xmath29 . * graphs and graph classes . * a _ clique _ in a graph is a set of pairwise adjacent vertices , and an _ independent _ ( or _ stable _ ) set is a set of pairwise non - adjacent vertices . a graph @xmath0 is _ chordal _ if it does not contain any induced cycle of order at least @xmath25 , _ split _ if its vertex set can be partitioned into a clique and an independent set , and _ @xmath30-polar _ if it admits a partition of its vertex set into two ( possibly empty ) parts @xmath31 and @xmath32 such that @xmath31 is a clique and @xmath32 induces a subgraph of maximum degree at most @xmath33 . for a set of graphs @xmath34 , a graph @xmath0 is said to be _ ( or just _ @xmath35-free _ if @xmath36 ) , if it does not contain any induced subgraph isomorphic to a member of @xmath34 . every member of @xmath34 is said to be a _ forbidden induced subgraph _ for the ( hereditary ) set of @xmath34-free graphs . the neighborhood of a vertex @xmath37 in a graph will be denoted by @xmath38 , and its closed neighborhood by @xmath39 : = n_g(v)\cup \{v\}$ ] , omitting the subscript @xmath0 if the graph is clear from the context . for a subset @xmath40 of vertices , we denote by @xmath41 ( or just by @xmath42 ) the set @xmath43 . a vertex in a graph @xmath0 is _ universal _ if it is adjacent to every other vertex in @xmath0 and _ isolated _ if it is of degree @xmath44 . by @xmath45 we will denote the disjoint union of graphs @xmath0 and @xmath46 . the _ join _ of graphs @xmath0 and @xmath46 is the graph obtained from the disjoint union @xmath45 by adding all edges of the form @xmath47 . for a graph @xmath0 , we denote by @xmath48 the disjoint union of two copies of @xmath0 . we denote by @xmath21 , @xmath49 and @xmath50 the complete graph , the path and the cycle on @xmath6 vertices . the following characterization of threshold graphs due to chvtal and hammer will be used in some of our proofs . [ thm : chvatalhammer ] for every graph @xmath0 the following conditions are equivalent : a. [ itemcha ] @xmath0 is threshold . b. [ itemchb ] @xmath0 has no induced subgraph isomorphic to @xmath51 or @xmath52 . c. [ itemchd ] there is a partition of @xmath53 into two disjoint sets @xmath54 and an ordering @xmath55 of vertices in @xmath56 such that no two vertices in @xmath56 are adjacent , every two vertices in @xmath31 are adjacent , and @xmath57 . without loss of generality , we will assume that @xmath56 is a maximal independent set of @xmath0 , that is , that @xmath58 . * boolean functions . * let @xmath6 be positive integer . given two vectors @xmath59 , we write @xmath60 if @xmath61 for all @xmath62 = \{1,\ldots , n\}$ ] . a boolean function @xmath63 is _ positive _ ( or : _ monotone _ ) if @xmath64 holds for every two vectors @xmath59 such that @xmath60 . a positive boolean function @xmath65 is said to be _ threshold _ if there exist non - negative real weights @xmath66 and a non - negative real number @xmath8 such that for every @xmath67 , @xmath68 if and only if @xmath69 ( see , e.g. , @xcite ) . such a pair @xmath14 is called a _ separating structure _ of @xmath65 . every threshold positive boolean function admits an integral separating structure @xcite . threshold boolean functions have been characterized by chow @xcite and elgot @xcite , as follows . for @xmath70 , a positive boolean function @xmath71 is said to be _ @xmath72-summable _ if , for some @xmath73 , there exist @xmath74 ( not necessarily distinct ) false points of @xmath65 , say , @xmath75 , and @xmath74 ( not necessarily distinct ) true points of @xmath65 , say @xmath76 , such that @xmath77 . ( a _ false point _ of @xmath65 is an input vector @xmath67 such that @xmath68 ; a _ true point _ is defined analogously . ) function @xmath65 is said to be _ @xmath72-asummable _ if it is not @xmath72-summable , and it is _ asummable _ if it is @xmath72-asummable for all @xmath78 . [ thm : bf : characterization ] a positive boolean function @xmath65 is threshold if and only if it is asummable . the problem of determining whether a positive boolean function given by its complete dnf is threshold is solvable in polynomial time , using dualization and linear programming . this result is summarized in the following theorem . [ thm : threshold - bfs ] + there exists a polynomial time algorithm that determines , given the complete dnf of a positive boolean function @xmath79 , whether @xmath65 is threshold . if this is the case , the algorithm also computes an integral separating structure of @xmath65 . the existence of a _ purely combinatorial _ polynomial time algorithm for recognizing threshold positive boolean functions given by their complete dnf is an open problem @xcite . ( by a `` purely combinatorial algorithm '' we mean an algorithm that does not rely on solving a linear program . ) notice that a combinatorial algorithm with exponential running time follows from theorem [ thm : bf : characterization ] . for further definitions related to boolean functions , we refer to the monograph by crama and hammer @xcite . * hypergraphs . * a _ hypergraph _ is a pair @xmath80 where @xmath81 is a finite set of _ vertices _ and @xmath82 is a finite set of subsets of @xmath81 , called _ ( hyper)edges _ @xcite . for our purposes , we will allow multiple edges ( unless specified otherwise ) , that is , @xmath82 can be a multiset . a hypergraph @xmath80 is _ threshold _ if there exist a weight function @xmath83 and a threshold @xmath16 such that for all subsets @xmath84 , it holds that @xmath85 if and only if @xmath2 contains no edge of @xmath86 @xcite . such a pair @xmath14 is called a _ separating structure _ of @xmath86 . to every hypergraph @xmath80 , we can naturally associate a positive boolean function @xmath87 , defined by the positive dnf expression @xmath88 for all @xmath89 . conversely , to every positive boolean function @xmath90 given by a positive dnf @xmath91 , we can associate a hypergraph @xmath92 as follows : @xmath93 $ ] and @xmath94 . it follows directly from the definitions that the thresholdness of hypergraphs and of boolean functions are related as follows . [ prop : relation ] a hypergraph @xmath80 is threshold if and only if the positive boolean function @xmath95 is threshold . a positive boolean function given by a positive dnf @xmath91 is threshold if and only if the hypergraph @xmath96 is threshold . moreover , reformulating the characterization of threshold positive boolean functions given by theorem [ thm : bf : characterization ] in the language of hypergraphs , we obtain the following characterization of threshold hypergraphs . [ thm : characterization ] a hypergraph @xmath80 is _ not _ threshold if and only if there exists an integer @xmath6 with @xmath97 and @xmath6 ( not necessarily distinct ) subsets @xmath98 of @xmath53 , each containing an edge of @xmath86 , and @xmath6 ( not necessarily distinct ) subsets @xmath99 of @xmath53 , each containing no edge of @xmath86 , such that for every vertex @xmath100 , @xmath101 basic properties ---------------- in this section , we establish some basic properties of td graphs . [ prop : isolated - vertex ] every graph with an isolated vertex is td . if @xmath0 has an isolated vertex , then @xmath0 does not have any td sets , and hence the pair @xmath102 , where @xmath23 for all @xmath103 is a total domishold structure of @xmath0 . as shown by the @xmath25-cycle , td graphs are not closed under join . however , they are closed under join with @xmath104 , that is , under adding a universal vertex . this is stated formally in proposition [ prop : dominating - vertex ] below and proved using the following auxiliary lemma . [ lem : positive - weights ] every td graph admits a total domishold structure in which all weights are positive . let @xmath14 be a total domishold structure of a td graph @xmath13 . the value of @xmath105 where @xmath106 denotes the power set of @xmath53 and @xmath107 denotes the set of all total dominating sets of @xmath0 , is well defined and positive . let @xmath108 and @xmath109 be defined as : @xmath110 for all @xmath111 , and @xmath112 . we claim that @xmath113 is a total domishold structure of @xmath0 . on the one hand , if @xmath114 , then @xmath115 . on the other hand , if @xmath116 , then @xmath117 and consequently [ prop : dominating - vertex ] let @xmath0 be a graph , and let @xmath118 be the graph obtained from @xmath0 by adding to it a vertex adjacent to all vertices of @xmath0 . then , @xmath0 is td if and only if @xmath118 is td . the proof will follow from the observation that the sets @xmath107 and @xmath119 of total dominating sets of @xmath0 and @xmath118 , respectively , are related as follows : @xmath120 where @xmath37 is the added vertex . suppose first that @xmath0 is td . by lemma [ lem : positive - weights ] , @xmath0 admits a total domishold structure @xmath14 with @xmath121 for all @xmath103 . let @xmath122 be defined as follows : * for all @xmath103 , let @xmath123 ; * let @xmath124 . we claim that @xmath125 is a total domishold structure of @xmath118 . indeed , if @xmath126 then we consider two cases . if @xmath127 , then @xmath114 and @xmath128 . if @xmath129 , then @xmath130 for some @xmath103 , and hence @xmath131 . similarly , if @xmath132 , we consider two cases . if @xmath127 , then @xmath133 and therefore @xmath134 if @xmath129 , then @xmath135 ( since otherwise we would have @xmath136 by the positivity of @xmath137 ) , and thus @xmath126 . the other direction is straightforward . since @xmath138 , any pair @xmath14 such that @xmath125 is a total domishold structure of @xmath118 and @xmath137 is the restriction of @xmath139 to @xmath140 , is a total domishold structure of @xmath0 . [ cor : threshold ] every threshold graph is htd . the characterization of threshold graphs due given by theorem [ thm : chvatalhammer](c ) implies that every threshold graph contains either an isolated vertex or a universal vertex . therefore , an induction on the number of vertices together with propositions [ prop : isolated - vertex ] and [ prop : dominating - vertex ] shows that every threshold graph is td . since the class of threshold graphs is hereditary @xcite , every threshold graph is htd . in general , td graphs are not closed under disjoint union : the path @xmath141 is td , but the graph @xmath142 is not ( cf . theorem [ prop : forbidden - induced - subgraphs ] in section [ sec : htd - partial - characterizations ] ) . however , they are closed under adding a ( td ) graph with a _ unique _ ( inclusion - wise ) minimal td set . [ prop : unique - td_disjoint - union ] let @xmath0 and @xmath46 be graphs such that @xmath46 has a unique minimal td set . then , @xmath45 is td if and only if @xmath0 is td . let @xmath1 be the unique minimal td set in @xmath46 . then , the sets @xmath107 and @xmath119 of total dominating sets of @xmath0 and @xmath143 are related as follows : @xmath144 suppose that @xmath0 is td , with a total domishold structure @xmath14 . let @xmath145 and define @xmath146 as @xmath147 it is not hard to verify that the pair @xmath148 a total domishold structure of @xmath118 . conversely , if @xmath113 is a total domishold structure of @xmath118 , then @xmath149 , where @xmath137 is the restriction of @xmath139 to @xmath140 , is a total domishold structure of @xmath0 . [ cor : isolated - edge ] let @xmath0 be a graph , and let @xmath150 . then , @xmath0 is td if and only if @xmath118 is td . a graph @xmath0 is said to be _ co - domishold _ if its complement is domishold . since threshold graphs are exactly the domishold co - domishold graphs @xcite , the following result generalizes corollary [ cor : threshold ] . [ cor : co - domishold ] every co - domishold graph is htd . this can be proved similarly as corollary [ cor : threshold ] , using corollary [ cor : isolated - edge ] in addition to propositions [ prop : isolated - vertex ] and [ prop : dominating - vertex ] , and the facts that : ( 1 ) the class of co - domishold graphs is hereditary ( this is because the class of domishold graph is hereditary @xcite ) ; ( 2 ) every co - domishold graph contains either an isolated vertex , a universal vertex , or a connected component isomorphic to @xmath151 @xcite . note that not every htd graph is co - domishold . for example , the @xmath25-vertex path @xmath152 is easily verified to be htd but it is not domishold @xcite , and hence ( since it is self - complementary ) also not co - domishold . as observed in the introduction , the set of td graphs is not hereditary . we conclude this section by strengthening this observation , showing that the set of td graphs is not contained in any nontrivial hereditary class of graphs ( even if we disallow graphs with isolated vertices ) . [ prop : non - hereditary ] for every graph @xmath0 there exists a td graph @xmath118 without isolated vertices such that @xmath0 is an induced subgraph of @xmath118 . let @xmath0 be a graph . first , add to @xmath13 a new vertex , say @xmath37 , and connect @xmath37 only to isolated vertices of @xmath0 . second , add a new private neighbor to every vertex of the resulting graph . denoting by @xmath118 the obtained graph , it is clear that @xmath0 is an induced subgraph of @xmath118 . by construction , the set @xmath153 is the unique minimal total dominating set in @xmath118 . therefore , the pair @xmath14 , where @xmath154 is given by @xmath23 if @xmath155 and @xmath156 , otherwise , and @xmath157 , is a total domishold structure of @xmath118 . in this section , we exhibit close relationships between total domishold graphs , threshold boolean functions and threshold hypergraphs . we start with a characterization of td graphs in terms of the thresholdness of a derived boolean function . given a set @xmath53 and a binary vector @xmath158 , the _ support set _ of a vector @xmath158 is the set @xmath159 . also , by @xmath160 we denote the vector @xmath161 given by @xmath162 for all @xmath163 . given a graph @xmath11 , its _ neighborhood function _ is the positive boolean function @xmath164 that takes value @xmath33 precisely on vectors @xmath158 whose support set @xmath165 contains the neighborhood of some vertex of @xmath0 . formally , @xmath166 for every vector @xmath167 . ( if @xmath168 then @xmath169 . ) [ lem : reduction ] a graph @xmath11 with @xmath170 is total domishold if and only if its neighborhood function @xmath171 is threshold . moreover , if @xmath172 is an integral separating structure of @xmath171 , then @xmath173 with @xmath174 for all @xmath62 $ ] is a total domishold structure of @xmath0 . first , recall that a positive boolean function @xmath79 is threshold if and only if its dual function @xmath175 is threshold , and that if @xmath172 is an integral separating structure of @xmath65 , then @xmath176 is a separating structure of @xmath177 @xcite . therefore , it suffices to argue that @xmath0 is total domishold if and only if @xmath178 is threshold . let @xmath167 and let @xmath165 be the support set of @xmath179 . by definition , @xmath180 if and only if @xmath181 , which is the case if and only if @xmath182 contains the neighborhood of some vertex . in other words , @xmath180 if and only if @xmath19 is not a total dominating set . hence , if the dual function @xmath178 is threshold with an integral separating structure @xmath172 , then @xmath183 with @xmath174 for all @xmath62 $ ] is a total domishold structure of @xmath0 , and conversely , if @xmath14 is an integral total domishold structure of @xmath0 , then @xmath184 with @xmath185 for all @xmath62 $ ] is a separating structure of @xmath178 . finally , if @xmath172 is an integral separating structure of @xmath171 , then @xmath176 is a separating structure of @xmath178 and hence @xmath173 with @xmath174 for all @xmath62 $ ] is a total domishold structure of @xmath0 . in view of proposition [ prop : relation ] , td graphs can also be characterized in terms of the thresholdness of a derived hypergraph . we now show that the converse is also true , that is , we can test whether a given hypergraph is threshold by testing whether a derived graph is total domishhold . in fact , the derived graph can be assumed to be a split graph . recall that a graph is _ split _ if its vertex set can be partitioned into a clique and an independent set . first , we derive a simple but useful property of threshold hypergraphs . a _ universal vertex _ in a hypergraph @xmath186 is a vertex of @xmath186 contained in all hyperedges of @xmath186 . [ prop : univesal - vertex ] let @xmath187 be a hypergraph , and let @xmath188 be the hypergraph obtained from @xmath186 by adding to it a universal vertex , that is , @xmath189 and @xmath190 . then , @xmath86 is threshold if and only if @xmath191 is . suppose first that @xmath86 is threshold , with an integral separating structure @xmath14 . then , the pair @xmath113 , where @xmath192 is defined as @xmath193 and @xmath194 , is a separating structure of @xmath191 . conversely , if @xmath191 is threshold , with an integral separating structure @xmath113 , then the pair @xmath14 , where @xmath83 is the restriction of @xmath139 to @xmath195 and @xmath196 , is a separating structure of @xmath191 . given a hypergraph @xmath187 , the _ split - incidence graph _ of @xmath86 ( see , e.g. , @xcite ) is the split graph @xmath197 with @xmath198 , where @xmath199 , @xmath81 is a clique , @xmath200 is an independent set , and @xmath201 is adjacent to @xmath202 if and only if @xmath203 . [ prop : hypergraph - to - graph ] for every hypergraph @xmath86 , it holds that @xmath86 is threshold if and only if its split - incidence graph @xmath204 is total domishold . let @xmath205 be a hypergraph . we prove the statement by induction on @xmath206 . the case @xmath207 is trivial since @xmath86 is threshold , and @xmath204 is isomorphic to either @xmath104 ( if @xmath208 ) or to @xmath151 ( if @xmath209 ) , hence total domishold . suppose now that @xmath210 . if @xmath186 has a universal vertex @xmath211 , then let @xmath212 be the hypergraph obtained by deleting @xmath37 from @xmath186 , that is , @xmath213 and @xmath214 . by the inductive hypothesis , @xmath191 is threshold if and only if its split - incidence graph @xmath215 is total domishold . notice that the split - incidence graph of @xmath191 is isomorphic to the graph obtained from the split - incidence graph of @xmath186 by deleting from it a universal vertex ( namely @xmath37 ) . hence , proposition [ prop : dominating - vertex ] and lemma [ prop : univesal - vertex ] imply that @xmath86 is threshold if and only if its split - incidence graph is total domishold . we may thus assume that @xmath86 does not have any universal vertices . let @xmath197 be its split - incidence graph , with @xmath200 as above . let @xmath216 denote the neighborhood function of @xmath0 , @xmath217 let @xmath218 be the function given by @xmath219 and let @xmath220 denote the restriction of @xmath221 to @xmath195 , that is , @xmath222 is given by @xmath223 we will show that the following conditions are equivalent , which will imply the proposition : @xmath224 hypergraph @xmath86 is threshold . @xmath225 function @xmath95 is threshold . @xmath226 function @xmath221 is threshold . @xmath227 function @xmath171 is threshold . @xmath228 graph @xmath0 is total domishold . the equivalence between @xmath224 and @xmath225 follows by proposition [ prop : relation ] . the equivalence between @xmath225 and @xmath226 is easy to establish directly from the definition , using the fact that @xmath221 does not depend on any variable of the form @xmath229 where @xmath202 ( see @xcite ) . the equivalence between @xmath226 and @xmath227 can be justified as follows . since @xmath86 has no universal vertices , for every vertex @xmath201 in the clique there exists a vertex @xmath202 in the independent set such that @xmath230 . hence , @xmath231 . in particular , this implies that the functions @xmath171 and @xmath221 are logically equivalent . finally , the equivalence between @xmath227 and @xmath228 follows from proposition [ lem : reduction ] . proposition [ prop : hypergraph - to - graph ] admits a natural converse . a _ split partition _ of a split graph @xmath0 is a pair @xmath232 such that @xmath31 is a clique , @xmath56 is an independent set , @xmath233 and @xmath234 . given a split graph @xmath0 with a split partition @xmath232 , its _ @xmath56-neighborhood hypergraph _ is the hypergraph @xmath235 where @xmath236 . recall that @xmath237 is a multiset , that is , if two vertices in @xmath56 have the same neighborhoods in @xmath0 , then we keep in @xmath237 both copies of the neighborhood . [ cor : graph - to - hypergraph ] for every split graph @xmath0 with a split partition @xmath232 , it holds that @xmath0 is total domishold if and only if its @xmath56-neighborhood hypergraph @xmath238 is threshold . the proposition follows immediately from proposition [ prop : hypergraph - to - graph ] and the observation that @xmath0 is isomorphic to the split - incidence graph of its @xmath56-neighborhood hypergraph @xmath238 . we conclude this section with a theorem summarizing several statements equivalent to a fact that a given graph @xmath0 is total domishold . to state the theorem , we need two more definitions . to every graph @xmath239 , we associate a split graph @xmath240 defined as follows : @xmath241 , where @xmath242 @xmath53 is a clique , @xmath243 is an independent set , and @xmath100 is adjacent to @xmath244 if and only if @xmath245 . ( in the above definition of @xmath243 , relation @xmath246 denotes strict set inclusion . ) the following derived hypergraph will also be useful . given a graph @xmath0 , the _ reduced neighborhood hypergraph _ of @xmath0 is the hypergraph @xmath247 where @xmath248 and @xmath237 is a set ( that is , it does not have any duplicated elements ) defined by @xmath249 [ prop : equivalences ] for every graph @xmath0 , the following statements are equivalent : 1 . graph @xmath0 is total domishold . the neighborhood function @xmath171 of @xmath0 is threshold . the boolean function @xmath250 given by the complete dnf @xmath251 of @xmath171 is threshold . 4 . the hypergraph @xmath96 ( where @xmath251 is the complete dnf of @xmath171 ) is threshold . 5 . the reduced neighborhood hypergraph @xmath252 of @xmath0 is threshold . the split - incidence graph @xmath253 of @xmath252 is total domishold . the graph @xmath240 is total domishold . the equivalence between @xmath224 and @xmath225 was established in proposition [ lem : reduction ] . the equivalence between @xmath225 and @xmath226 is trivial , since @xmath171 is equivalent to the function given by its complete dnf @xmath251 . the equivalence between @xmath254 and @xmath227 follows from proposition [ prop : relation ] . the equivalence between @xmath227 and @xmath228 follows from the observation that reduced neighborhood hypergraphs @xmath96 and @xmath252 are isomorphic to each other . the equivalence between @xmath228 and @xmath255 follows from proposition [ prop : hypergraph - to - graph ] . finally , the equivalence between @xmath255 and @xmath256 follows from the observation that the split - incidence graph @xmath253 is isomorphic to the split graph @xmath240 . in this section , we prove our main result : two characterizations of hereditary total domishold graphs . our proofs will be based on a new family of hypergraphs defined similarly as the sperner hypergraphs . recall that a hypergraph @xmath80 is said to be _ sperner _ ( or : a _ clutter _ ) if no edge of @xmath86 contains another edge , or , equivalently , if for every two distinct edges @xmath257 and @xmath65 of @xmath86 , it holds that @xmath258 this motivates the following definition . [ def : td ] a hypergraph @xmath80 is said to be _ dually sperner _ if for every two distinct edges @xmath257 and @xmath65 of @xmath86 , it holds that @xmath259 [ lem : dually - sperner - threshold ] every dually sperner hypergraph is threshold . suppose for a contradiction that there exists a dually sperner hypergraph @xmath260 that is not threshold . by theorem [ thm : characterization ] , there exists an integer @xmath97 and @xmath6 ( not necessarily distinct ) subsets @xmath98 of @xmath81 , each containing an edge of @xmath86 , and @xmath6 ( not necessarily distinct ) subsets @xmath99 of @xmath81 , each containing no edge of @xmath86 , such that for every vertex @xmath201 , condition @xmath261 holds . for every @xmath62 $ ] , let @xmath262 be an edge of @xmath86 contained in @xmath263 . let @xmath264 $ ] be such that @xmath265 for all @xmath62 $ ] . in particular , this implies that for every @xmath62 $ ] , it holds that @xmath266 which , since @xmath86 is dually sperner , implies @xmath267 contains the edge @xmath268 , we have , for all @xmath62 $ ] , the inequality @xmath269 adding up the inequalities @xmath270 for all @xmath62 $ ] , we obtain @xmath271}|e_{i^*}\setminus b_i|\,.$ ] this implies the following contradicting chain of equations and inequalities @xmath272}|e_{i^*}\setminus b_i| = \sum_{i\in [ n]}\sum_{v\in e_{i^*}\setminus b_i}1 = \sum_{v\in e_{i^*}}\sum_{i\,:\,v\not\in b_i}1 = \sum_{v\in e_{i^*}}\left(n-|\{i\,:\,v\in b_i\}|\right)\ ] ] @xmath273}\sum_{v\in e_{i^*}\setminus a_i}1 = \sum_{i\in [ n]}|e_{i^*}\setminus a_i|\]]@xmath274}|e_{i^*}\setminus e_i| = \sum_{{\substack{i\in [ n]\\i\neq i^*}}}|e_{i^*}\setminus e_i| \le \sum_{{\substack{i\in [ n]\\i\neq i^*}}}1 = n-1\,.\ ] ] the first equality in the second line follows from condition @xmath261 , while the first inequality in the third line follows from the fact that @xmath275 , which implies @xmath276 . the last inequality follows from ( [ eq1 ] ) . this contradiction completes the proof . let us say that a graph @xmath0 is _ @xmath27-asummable _ if its neighborhood function @xmath171 is @xmath27-asummable , and _ hereditary @xmath27-assumable _ if every induced subgraph of it is @xmath27-asummable . it follows from theorem [ thm : bf : characterization ] and theorem [ prop : equivalences ] that every total domishold graph is @xmath27-asummable . notice that the converse is not true : there exists a @xmath27-asummable ( split ) graph @xmath0 that is not total domishold . this can be easily derived using the results of section [ sec : td - graphs - bfs - hypergraphs ] and the fact that not every @xmath27-asummable positive boolean function is threshold ( theorem 9.15 in @xcite ) . in theorem [ prop : forbidden - induced - subgraphs ] below , we prove several characterizations of htd graphs , which , in particular , imply that every hereditary @xmath27-asummable graph is htd . [ prop : forbidden - induced - subgraphs ] for every graph @xmath0 , the following are equivalent : 1 . [ item1 ] @xmath0 is hereditary total domishold . [ item2 ] @xmath0 is hereditary @xmath27-assumable . [ item3 ] @xmath0 is @xmath277-free , where @xmath278 are the graphs depicted in fig . [ fig : forbidden - induced - subgraphs ] . . ] the implication @xmath279 follows from proposition [ lem : reduction ] and theorem [ thm : bf : characterization ] . for the implication @xmath280 , we only need to verify that none of the graphs @xmath278 is @xmath27-asummable . arguing by contradiction , assume that there is a graph @xmath281 that is @xmath27-asummable . take two vertices of degree @xmath27 in @xmath35 , say @xmath282 and @xmath37 , such that @xmath282 and @xmath37 have disjoint neighborhoods ( e.g. , the two black vertices in the depiction of @xmath35 in fig . [ fig : forbidden - induced - subgraphs ] ) . denote their respective neighbors by @xmath283 and @xmath284 . ( if @xmath285 , then let @xmath286 and @xmath287 ) . it is easy to see that the union of the two neighborhoods @xmath288 can be partitioned into two disjoint sets , namely @xmath289 and @xmath290 , none of which contains the neighborhood of another vertex in the graph . for a set @xmath291 , let @xmath292 denote the characteristic vector of @xmath19 , that is , @xmath293 clearly , @xmath294 and @xmath295 are true points of @xmath296 , and @xmath297 and @xmath298 are false points of @xmath296 . since @xmath299 , the neighborhood function @xmath296 is @xmath27-summable , a contradiction . it remains to show the implication @xmath300 . since the class of @xmath277-free graphs is hereditary , it is enough to show that every @xmath277-free graph is total domishold . let @xmath0 be an @xmath277-free graph . by theorem [ prop : equivalences ] , @xmath0 is total domishold if and only if its reduced neighborhood hypergraph @xmath252 is threshold . we will prove that @xmath252 is dually sperner , and the conclusion will follow from lemma [ lem : dually - sperner - threshold ] . suppose for a contradiction that the reduced neighborhood hypergraph @xmath252 is not dually sperner . then , there exist two hyperedges @xmath257 and @xmath65 of @xmath252 such that @xmath301 . by the definition of @xmath252 , there exist two vertices @xmath282 and @xmath37 of @xmath0 such that @xmath302 and @xmath303 . moreover , the following condition holds : @xmath304 we analyze two cases , depending on whether @xmath282 and @xmath37 are adjacent or not . _ case 1 : @xmath305 . _ let @xmath306 and @xmath307 be two distinct elements in @xmath308 , and let @xmath309 and @xmath310 be two distinct elements in @xmath311 . clearly , vertices @xmath312 are all pairwise distinct . moreover , @xmath313 while @xmath314 . let @xmath72 be the number of edges in @xmath0 connecting a vertex in @xmath315 with a vertex of @xmath316 . notice that @xmath317 . suppose that @xmath318 . then , depending on the number of edges in the set @xmath319 , the subgraph of @xmath0 induced by @xmath320 is isomorphic to either @xmath321 , @xmath322 , or @xmath323 , a contradiction . suppose that @xmath324 . then , depending on the number of edges in the set @xmath319 , the subgraph of @xmath0 induced by @xmath320 is isomorphic to either @xmath325 , @xmath326 , or @xmath327 , a contradiction . suppose that @xmath328 . we may assume without loss of generality that @xmath329 and @xmath330 . suppose first that @xmath331 . then @xmath332 , and depending on the number of edges in the set @xmath319 , the subgraph of @xmath0 induced by @xmath320 contains either an induced @xmath333 , @xmath334 , or @xmath335 , a contradiction . suppose now that @xmath336 . then @xmath337 . since @xmath0 is @xmath26-free , @xmath338 , and depending on whether @xmath339 or not , the subgraph of @xmath0 induced by @xmath320 is isomorphic to either @xmath340 or @xmath341 , a contradiction . suppose that @xmath342 . we may assume without loss of generality that @xmath343 and @xmath330 . since @xmath0 is @xmath26-free , @xmath338 and @xmath339 . but now , the subgraph of @xmath0 induced by @xmath320 is isomorphic to @xmath344 , a contradiction . finally , suppose that @xmath345 . since @xmath0 is @xmath26-free , @xmath338 and @xmath339 . but now , the subgraph of @xmath0 induced by @xmath320 is isomorphic to @xmath346 , a contradiction . let @xmath347 and @xmath307 be two distinct elements in @xmath308 , and let @xmath348 and @xmath310 be two distinct elements in @xmath311 . since @xmath0 is @xmath26-free , @xmath336 , and the vertices @xmath349 induce a @xmath152 in @xmath0 . applying condition to @xmath350 and @xmath351 implies that @xmath352 . hence , there exists a vertex , say @xmath353 , such that @xmath354 but @xmath355 . clearly , @xmath356 . we claim that @xmath353 is not adjacent to @xmath307 . indeed , if @xmath357 , then @xmath0 contains either an induced @xmath26 ( if @xmath358 ) or an induced @xmath359 ( otherwise ) , a contradiction . a symmetric argument shows that @xmath0 contains a vertex @xmath360 that is adjacent to @xmath307 and non - adjacent to both @xmath37 and @xmath310 . clearly , @xmath361 . if @xmath362 , then depending on the number of edges in the set @xmath363 , the subgraph of @xmath0 induced by @xmath320 contains either an induced @xmath333 , @xmath334 , or @xmath335 , a contradiction . if @xmath364 , then depending on the number of edges in the set @xmath363 , the subgraph of @xmath0 induced by @xmath320 is isomorphic to either @xmath325 , @xmath326 , or @xmath327 , a contradiction . theorem [ prop : forbidden - induced - subgraphs ] implies a nice structural feature of htd graphs . recall that a graph is said to be _ @xmath30-polar _ if it admits a partition of its vertex set into two ( possibly empty ) parts @xmath31 and @xmath32 such that @xmath31 is a clique and @xmath32 induces a subgraph of maximum degree at most @xmath33 . the following result is an immediate consequence of theorem [ prop : forbidden - induced - subgraphs ] and the forbidden induced subgraph characterization of @xmath30-polar graphs due to gagarin and metelski @xcite . [ cor : polar - chordal ] every htd graph is a @xmath30-polar chordal graph . in @xcite , gagarin and metelski characterized the set of @xmath30-polar graphs with a set of @xmath365 forbidden induced subgraphs ( see also @xcite ) . only three graphs in this list are chordal : @xmath142 , @xmath366 , and @xmath367 ( that is , the graphs @xmath368 , and @xmath369 from fig . [ fig : forbidden - induced - subgraphs ] ) . this implies that the class of @xmath30-polar chordal graphs is exactly the class of @xmath370-free graphs . notice that @xmath371 , @xmath372 , @xmath373 , and @xmath325 is an induced subgraph of every cycle of order at least @xmath374 . therefore , the class of @xmath375-free graphs is a subclass of the class of @xmath30-free polar chordal graphs . in particular , by theorem [ prop : forbidden - induced - subgraphs ] , the same is true for htd graphs . notice that the converse of corollary [ cor : polar - chordal ] does not hold : graphs @xmath376 are @xmath30-polar chordal graphs that are not td . as shown in theorem [ prop : forbidden - induced - subgraphs ] , the class of hereditary total domishold graph is characterized by finitely many forbidden induced subgraphs . therefore , htd graphs can be recognized in polynomial time . the next theorem and its proof establish the existence of a polynomial time recognition algorithm also for total domishold graphs , reducing the problem to the problem of recognizing threshold positive boolean functions given by a their complete dnf . [ thm : poly - time - recognition ] there exists a polynomial time algorithm for recognizing total domishold graphs . if the input graph @xmath0 is total domishold , the algorithm also computes an integral total domishold structure of @xmath0 . theorem [ thm : threshold - bfs ] and proposition [ lem : reduction ] imply that the following polynomial time algorithm correctly determines whether @xmath0 is total domishold , and if this is the case , computes a total domishold structure of it . first , compute the complete dnf @xmath251 of the neighborhood function @xmath171 of @xmath0 . more specifically , let @xmath377 where @xmath378 is the set of neighborhoods of vertices of @xmath0 that do not properly contain any other neighborhood . second , apply the algorithm given by theorem [ thm : threshold - bfs ] to the input @xmath251 . if the algorithm detects that @xmath171 is not threshold , then @xmath0 is not total domishold . otherwise , the algorithm will compute an integral separating structure @xmath172 of @xmath171 , in which case proposition [ lem : reduction ] implies that @xmath173 with @xmath174 for all @xmath62 $ ] is a total domishold structure of @xmath0 . in view of proposition [ prop : relation ] , theorem [ thm : threshold - bfs ] also implies the existence of a polynomial time algorithm for recognizing threshold hypergraphs . recall that the known algorithm is based on dualization and linear programming ; the existence of a purely combinatorial polynomial time algorithm for recognizing threshold boolean functions / hypergraphs is an open problem @xcite . in view of the reduction from the proof of theorem [ thm : poly - time - recognition ] , it is a natural question whether there exists a purely combinatorial polynomial time algorithm for recognizing total domishold graphs . proposition [ prop : hypergraph - to - graph ] and corollary [ cor : graph - to - hypergraph ] imply the following : there exists a purely combinatorial polynomial time algorithm for recognizing total domishold split graphs if and only if there exists a purely combinatorial polynomial time algorithm for recognizing threshold hypergraphs ( equivalently : for recognizing threshold positive boolean functions given by their complete dnf ) . let us now examine some consequences of theorem [ thm : poly - time - recognition ] . the _ total domination number _ @xmath379 of a graph @xmath0 without isolated vertices is the minimum size of a total dominating set in @xmath0 , and the _ total dominating set problem _ is the problem of computing the total domination number of a given graph without isolated vertices . the problem is @xmath10-hard in general , and remains hard even for restricted graph classes such as bipartite graphs or split graphs @xcite . on the positive side , polynomial time algorithms have been designed for several graph classes @xcite . with the exception of dually chordal graphs @xcite and ddp - graphs @xcite , all known polynomial time algorithms for the total dominating set problem we are aware of deal with hereditary graph classes . the following result provides another example of a non - hereditary graph class for which the problem is polynomial . [ prop : mtd - problem - polynomial ] the total dominating set problem is solvable in polynomial time for total domishold graphs . applying theorem [ thm : poly - time - recognition ] , we may assume that the input graph @xmath0 is given together with an integral total domishold structure @xmath14 . the following greedy approach can now be used to find a minimum - sized total dominating set @xmath19 : 1 . sort the vertices according to their weights so that @xmath380 . 2 . start with the empty set , @xmath381 , and , as long as @xmath382 , keep adding to @xmath19 vertices in the above order @xmath383 . 3 . return @xmath19 . the correctness and the polynomial running time of this algorithm are immediate . while the total dominating set problem is @xmath10-hard for chordal graphs ( in fact , even for split graphs @xcite ) , proposition [ prop : mtd - problem - polynomial ] shows that the problem is polynomial in the class of htd graphs , which , by corollary [ cor : polar - chordal ] , is a subclass of chordal graphs . we now examine some consequences of proposition [ prop : mtd - problem - polynomial ] for the dominating set problem . the _ domination number _ @xmath384 of a given graph @xmath0 is the minimum size of a dominating set in @xmath0 . the _ dominating set problem _ is the problem of computing the domination number of a given graph . for general graphs , it is @xmath385-hard to compute the domination number @xcite . moreover , for every @xmath386 , there is no polynomial time algorithm approximating the domination number for graphs without isolated vertices within a factor of , unless @xmath387 @xcite , and it is @xmath385-hard to approximate the domination number within a factor of ( see @xcite ) . [ prop:2-approx ] there exists a @xmath27-approximation algorithm for the dominating set problem in the class of total domishold graphs . given a td graph @xmath0 , let @xmath19 be the set of isolated vertices in @xmath0 , and let @xmath388 . we may assume that @xmath0 has at least one edge ( otherwise we have @xmath389 ) , and hence @xmath118 is non - empty graph with a total dominating set . by proposition [ prop : mtd - problem - polynomial ] , we can compute a minimum total dominating set @xmath12 in @xmath118 in polynomial time . let @xmath390 . clearly , @xmath391 is a dominating set in @xmath0 . on the other hand , denoting by @xmath392 a minimum dominating set in @xmath0 , we see that @xmath393 is a minimum dominating set in the graph @xmath118 . it is easy to see that for every graph @xmath46 without isolated vertices , its domination and total domination numbers are related as follows : @xmath394 . in particular , @xmath395 , which implies @xmath396 and consequently @xmath397 therefore , the algorithm that computes and outputs @xmath391 is a @xmath27-approximation algorithm for the domination problem on total domishold graphs . we leave it as an open problem to determine the exact complexity status of the dominating set problem in the classes of td and htd graphs . simple reductions from general graphs show that the problems of computing the independence number , the clique number and the chromatic number of a given td graph are @xmath10-hard , and also @xmath385-hard to approximate to within a factor of @xmath398 . ( see appendix [ hardness ] . ) on the other hand , since htd graphs are chordal , all these problems are polynomial for htd graphs . in conclusion , we mention some open problems related to total domishold graphs . is there a purely combinatorial polynomial time algorithm for recognizing total domishold graphs ? for recognizing split total domishold graphs ? an answer to either of these questions would resolve the open problem regarding the existence of a purely combinatorial polynomial time algorithm for recognizing threshold positive boolean functions given by a complete dnf @xcite . as showed in this paper , the class of htd graphs forms a proper subclass of @xmath30-polar chordal graphs and a proper superclass of the complements of domishold graphs . a more detailed structural analysis of htd graphs seems to deserve further attention . for instance , theorem [ prop : forbidden - induced - subgraphs ] implies that a split graph @xmath0 is htd if and only if it is @xmath346-free . moreover , in a companion paper @xcite we prove that by replacing in the list of forbidden induced subgraphs of htd graphs the graphs @xmath340 and @xmath344 from fig . [ fig : forbidden - induced - subgraphs ] with two of their respective induced subgraphs on @xmath399 vertices , we obtain a class of graphs every connected member of which arises from a threshold graph by attaching an arbitrary ( non - negative ) number of leaves to each of its vertices . in particular , this structure implies a linear time recognition algorithm for graphs in this class . further insight into the structure of general htd graphs might imply the existence of a more efficient recognition algorithm of htd graphs than the one with running time @xmath400 implied by theorem [ prop : forbidden - induced - subgraphs ] . is there a linear time algorithm for recognizing hereditary total domishold graphs ? we conclude with a question motivated by the existence of a @xmath27-approximation algorithm for the dominating set problem in the class of total domishold graphs ( proposition [ prop:2-approx ] ) . determine the computational complexity of the dominating set problem on total domishold / hereditary total domishold graphs . we would like to thank endre boros for stimulating discussions and helpful comments . 10 c. benzaken and p.l . hammer , linear separation of dominating sets in graphs , in bollobs , b. ( ed . ) , advances in graph theory , annals of discrete mathematics , 3 , north - holland , 1978 , pp . 110 . c. berge , _ hypergraphs _ , north - holland , 1989 . a. brandstdt , v.d . chepoi and f.f . dragan , the algorithmic use of hypertree structure and maximum neighbourhood orderings , in proc . wg 1994 , lecture notes in computer science 903 ( 1995 ) 6580 . n. chiarelli and m. milani , linear separation of total dominating sets in graphs , to appear in proc . wg @xmath401 , lecture notes in computer science , springer , @xmath401 . n. chiarelli and m. milani , _ on a class of graphs between threshold and total domishold graphs _ , in preparation . m. chlebk and j. chlebkova , approximation hardness of dominating set problems in bounded degree graphs , _ information and computation _ 206 ( 2008 ) 12641275 . chow , boolean functions realizable with single threshold devices , in : proceedings of the ire 49 ( 1961 ) 370371 . v. chvtal and p.l . hammer , aggregation of inequalities in integer programming , in hammer , p. l. ; johnson , e. l. ; korte , b. h. et al . , studies in integer programming ( proc . bonn 1975 ) , annals of discrete mathematics , 1 , amsterdam : north - holland , pp . 145162 . f. cicalese , m. milani and u. vaccaro , on the approximability and exact algorithms for vector domination and related problems in graphs , _ discrete appl . math . _ 161 ( 2013 ) 750767 . corneil and l. stewart , dominating sets in perfect graphs , _ discrete math . _ 86 ( 1990 ) 145164 . y. crama and p.l . boolean functions : theory , algorithms , and applications _ , cambridge university press , 2011 . elgot , truth functions realizable by single threshold organs , in : ieee symposium on switching circuit theory and logical design , 1961 , pp . 225245 . u. feige . , a threshold of @xmath402 for approximating set cover , _ journal of acm _ 45 ( 1998 ) 634652 . gagarin and yu . m. metelski , characterization of ( 1,2)-polar graphs ( in russian ) , _ vestsi nats . navuk belarusi ser . fiz .- mat . navuk _ 3 ( 1999 ) 107112 . garey and d.s . johnson , _ computers and intractability : a guide to the theory of np - completeness _ , w. h. freeman , 1979 . golumbic , _ algorithmic graph theory and perfect graphs _ , _ ann . of discrete math . elsevier , 2004 . haynes , s. hedetniemi and p. slater , _ fundamentals of domination in graphs _ , marcel dekker , 1998 . t.w . haynes , s. hedetniemi and p. slater , _ domination in graphs : advanced topics _ , marcel dekker , 1998 . m.a . henning , recent results on total domination in graphs : a survey , _ discrete math . _ 309 ( 2009 ) 3263 . m.a . henning and a. yeo , _ total domination in graphs _ , springer , 2013 . kant , v. limouzy , a. mary and l. nourine , on the enumeration of minimal dominating sets and related notions , submitted . d. kratsch , domination and total domination on asteroidal triple - free graphs , _ discrete applied math . _ 99 ( 2000 ) 111123 . levit , m. milani and d. tankus , on the recognition of @xmath72-equistable graphs , in proc . wg 2012 , lecture notes in computer science 7551 ( 2012 ) 286296 . mahadev and u.n . peled , _ threshold graphs and related topics _ , ann . of discrete math . , 56 . north - holland publishing co. , amsterdam , 1995 . m. milani , j. orlin and g. rudolf , complexity results for equistable graphs and related classes , _ ann . _ 188 ( 2011 ) 359370 . c. payan , a class of threshold and domishold graphs : equistable and equidominating graphs , _ discrete math . _ 29 ( 1980 ) 4752 . u.n . peled and b. simeone , polynomial - time algorithms for regular set - covering and threshold synthesis , _ discrete applied math . _ 12 ( 1985 ) 5769 . d. zuckerman , linear degree extractors and the inapproximability of max clique and chromatic number , _ theory of computing _ 3 ( 2007 ) 103128 . i.e. zverovich and o.i . zverovich , independent domination in hereditary classes , _ theoret . _ 352 ( 2006 ) 215225 . here we provide the hardness proofs establishing the computational complexity of computing the independence , the clique and the chromatic numbers of a given total domishold graph . recall that the independence number of a graph @xmath0 is denoted by @xmath403 and equals the maximum size of an independent set in @xmath0 , the clique number of a graph @xmath0 is denoted by @xmath404 and equals the maximum size of a clique in @xmath0 , and the chromatic number of a graph @xmath0 is denoted by @xmath405 and equals the minimum integer @xmath72 such that @xmath140 can be partitioned into @xmath72 independent sets . for general graphs , all these three parameters are @xmath385-hard to compute , and also @xmath385-hard to approximate to within a factor of @xmath398 , for every @xmath386 @xcite . we reduce from the independent set problem in general graphs . given a graph @xmath0 , we construct a graph @xmath118 from @xmath0 by adding to it first an isolated , and then a universal vertex . by propositions [ prop : dominating - vertex ] and [ prop : isolated - vertex ] , the obtained graph is td . it is straightforward to verify that the independence numbers of @xmath0 and @xmath118 are related by the equation @xmath406 . the result follows . we reduce from the clique problem in general graphs . given a graph @xmath0 that is the input to the clique problem , we may assume without loss of generality that @xmath0 has at least one edge , and that it has no isolated vertices . we construct a graph @xmath118 from @xmath0 by adding a private neighbor to each vertex of @xmath0 . the obtained graph @xmath118 has a unique minimal td set , namely @xmath140 , and hence setting @xmath407 and defining @xmath408 as @xmath409 produces a total domishold structure of @xmath118 . hence @xmath118 is td . furthermore , we clearly have @xmath410 , and the result follows . given a graph @xmath0 that is the input to the chromatic number problem , we may assume without loss of generality that @xmath0 has no isolated vertices . we construct a graph @xmath118 from @xmath0 by adding a private neighbor to each vertex of @xmath0 . as above , the obtained graph @xmath118 is td . furthermore , we have @xmath411 , and the result follows .

we consider weakly interacting bosonic gases with local and non - local multi - body interactions . by using the bogoliubov approximation , we first investigate contact interactions , studying the case in which the interparticle potential can be written as a sum of-body-interactions , and then considering general contact potentials . results for the quasi - particle spectrum and the stability are presented . we then examine non - local interactions , focusing on two different cases of-body non - local interactions . our results are used for systems with- and-body-interactions and applied in the homogeneous limit with the values given in . finally , the effect of conservative-body terms in dipolar systems and soft - core potentials ( that can be simulated with rydberg dressed atoms ) is also studied .

1701.08367

the bogoliubov theory of weakly interacting bose gases @xcite provides an essential tool to investigate the effect of interactions in bosonic systems @xcite and it plays a key role in the study of properties of bose - einstein condensates @xcite . its results for the ground - state energy and condensate depletion are in agreement in the weakly interacting limit with the findings obtained by other methods subsequently developed , including rigorous treatments @xcite . an important point is that also when it does not ( quantitatively ) work , it is useful to have results which are of guide for a qualitative understanding , as in the case of helium@xmath4 @xcite , or to have estimates of the ground - state energy , as in the case of the @xmath5 lieb - liniger model for small couplings @xcite . moreover , being a self - consistent approach , where the number of condensate particles has to be self - consistently determined , it gives information on the issue whether there is condensation or not , as in low - dimensional systems @xcite . finally , the bogoliubov transformation used to diagonalize the quadratic hamiltonian obtained by the bogoliubov approximation is used in a variety of other systems , including spin - wave theory of antiferromagnets @xcite and superconductors @xcite . in this paper we study the generalization of the bogoliubov theory to local and non - local / finite - range @xmath2-body and general multi - body interactions . our reasons for such an investigation are the following : * on one side , we are motivated by recent experiments in which matter wave breathers were observed using an attractively interacting bose - einstein condensate @xcite . from a comparison between experimental findings and gross - pitaevskii results , it was concluded that @xmath2-body interactions may play an important role and an estimate of ( the real part of ) @xmath6 was provided @xcite . * the effect of @xmath2-body interaction terms in dipolar systems is very interesting as well , as it was recently investigated in @xcite for a harmonically trapped dilute dipolar condensate with a short - range @xmath2-body interaction . it was shown that this system exhibits a condensate state and a droplet state , discussing how the droplet crystal may be an excited state arising from heating as the system crosses the phase transition . * a tool to study ultracold strongly interacting systems such as @xmath5 bose gas @xcite or unitary fermi gases @xcite is to introduce effective local interactions . recent examples are provided by the recent study of the monopole excitations for the @xmath5 bose gas @xcite using an effective gross - pitaevskii equation of the form @xmath7 where @xmath8 is the external potential , @xmath9 is the density and the non - linear term @xmath10 is extracted from the solution of the bethe ansatz integral equations for the ( homogeneous ) @xmath5 bose gas @xcite . another example is the study of small - amplitude josephson oscillations of a @xmath11 unitary fermi gas in a double well potential @xcite , where experimental data were compared with an equation of the form with @xmath12 the pair density , @xmath8 the double well one - body potential and @xmath10 extracted from monte carlo numerical results @xcite ( an example of a possible parametrization of @xmath10 across the bec - bcs crossover is in @xcite ) . it is clear that in the weakly interacting limit it is @xmath13 : this corresponds for the @xmath5 bose gas to the limit @xmath14 , where @xmath15 is the lieb - liniger coupling constant @xcite , and for fermions in the bec - bcs crossover to the bec limit @xmath16 , @xmath17 being the scattering length . when deviations from the weakly interacting limit are incorporated through an @xmath18 which is no longer proportional to @xmath12 , if the function @xmath18 admits a series expansion of the form @xmath19 , then there are effective multi - body local interactions ( corresponding to integer values @xmath20 ) . therefore , to treat such multi - body ( albeit effective ) interaction terms one needs to study in the bogoliubov theory such higher - body terms , as we do systematically below . * finally , when ( local or non - local / finite - range ) 2-body terms are present jointly with higher - body contributions ( as @xmath2-body ones ) , they may compete to make the system stable or unstable and it is of interest to determine stability conditions and the spectrum of the quasi - particles . a typical example is given by an attractively interacting bose gas , having @xmath21 , which can be made stable by a repulsive @xmath2-body term - another example is the dipolar gas in presence of a @xmath2-body term . in the following we derive within the bogoliubov theory the stability conditions in presence of general multi - body local interactions ( considering the case of non - local @xmath2-body interactions as well ) , then we discuss in detail some specific examples . the paper is organized as follows . in section ( [ sec : local ] ) we consider general multi - body contact interactions for a homogeneous weakly interacting gas treated in the bogoliubov approximation . we consider first the case of an @xmath0-body @xmath1-interaction and successively we consider a local interaction which can be expanded in series of general @xmath0-body terms . for this class of interactions we compute the spectrum , the stability condition and the ground - state energy ( also for the @xmath5 case ) . we finally present results for contact interactions that can not be expanded in series . in section ( [ sec : finite - range ] ) we discuss the case of non - local interactions : after briefly reviewing the well - known case of a @xmath3-body non - local interaction , we study two different cases of @xmath2-body non - local interactions , and a comparison between these two cases is performed with a gaussian pair - wise interaction . the results of sections ( [ sec : local ] ) and ( [ sec : finite - range ] ) are used in section ( [ sec:2_3 ] ) to the case of a model with @xmath3- and @xmath2-body contact interactions and the obtained findings are applied to the experimental setup described in @xcite . in section ( [ sec : other ] ) we discuss some further realistic interaction potentials of interest for current experimental setups with ultracold atoms . we present results for a @xmath3-body non - local potential plus @xmath3- and @xmath2-body @xmath1-interactions , with applications on dipolar systems , e.g. magnetic atoms and polar molecules , and soft - core potentials , that can be simulated with rydberg dressed atoms , to discuss the effect of a @xmath2-body interaction term . finally we present our conclusions in section ( [ sec : concl ] ) , while more technical material is presented in the appendices . in this section we consider general local @xmath1-interparticle potentials with multi - body interactions . we start by considering a model for a gas of @xmath22 bosonic particles interacting only via local repulsive @xmath0-body @xmath1-interactions . the general hamiltonian for @xmath0-body interactions reads @xmath23 where the bosonic field operators satisfy the canonical commutation relations @xmath24 = 0 = \left [ \hat{\psi}^{\dagger}({\mathbf{r } } ) , \hat{\psi}^{\dagger}({\mathbf{r ' } } ) \right]$ ] and @xmath25 = \delta({\mathbf{r } } - { \mathbf { r'}})$ ] . we assume @xmath0-body local contact interactions having the form : @xmath26 where @xmath27 is a coefficient of dimension @xmath28=[e]\cdot[l]^{3n-3}$ ] . in the usual case of 2-body @xmath1-interaction one has @xmath29 , where @xmath30 is the mass of the bosons @xcite . with potential the hamiltonian reads @xmath31 using for the field operator the expansion @xmath32 where @xmath33 is the volume of the system ( chosen to be a cube of side @xmath34 with periodic boundary conditions ) and @xmath35 , with @xmath36 , the hamiltonian assumes the form @xmath37 with @xmath38 given as usual by @xmath39 where @xmath40 , and the interaction part reading as @xmath41 with the definition @xmath42 now , proceeding as usual , we make the bogoliubov prescription @xcite @xmath43 as usual this implies that @xmath44 , where @xmath45 is the condensate number , or the largest eigenvalue of the one - body density matrix @xcite . we then proceed by neglecting in @xmath46 products of 3 or more @xmath47 with @xmath48 . to start with , we consider @xmath2-body interactions , that is , @xmath49 . from it follows the total momentum conservation . therefore , one has to arrange @xmath3 nonzero momenta between @xmath2 initial and @xmath2 final possible momenta . to enumerate all the possibilities , we may start by considering all the momenta of the creation operators equal to zero : l*4cr @xmath50 & @xmath51 & @xmath52 & @xmath53 & @xmath54 & @xmath55 + 0 & 0 & 0 & 0 & @xmath56 & @xmath57 + & & & @xmath56 & 0 & @xmath57 + & & & @xmath56 & @xmath57 & 0 + i.e. the contribution to @xmath46 is @xmath58 multiplied by @xmath2 , that is the combinatorial multiplicity , @xmath59 , as illustrated in the table above . for a generic @xmath0 , it is therefore straightforward to conclude that the coefficient @xmath60 has to be replaced by @xmath61 . going ahead , the next possible dispositions are l*4cr @xmath50 & @xmath51 & @xmath52 & @xmath53 & @xmath54 & @xmath55 + @xmath56 & 0 & 0 & @xmath56 & 0 & 0 + & & & 0 & @xmath56 & 0 + & & & 0 & 0 & @xmath56 + 0 & @xmath56 & 0 & @xmath56 & 0 & 0 + & & & 0 & @xmath56 & 0 + & & & 0 & 0 & @xmath56 + 0 & 0 & @xmath56 & @xmath56 & 0 & 0 + & & & 0 & @xmath56 & 0 + & & & 0 & 0 & @xmath56 + from which one can infer that the corresponding contribution to @xmath46 is @xmath62 with multiplicity @xmath63 . for a general @xmath0 the multiplicity is @xmath64 . finally , the remaining possibilities are when all the momenta of the annihilation operators are vanishing : l*4cr @xmath50 & @xmath51 & @xmath52 & @xmath53 & @xmath54 & @xmath55 + @xmath56 & @xmath57 & 0 & 0 & 0 & 0 + @xmath56 & 0 & @xmath57 & & & + 0 & @xmath56 & @xmath57 & & & + whose contribution to the operatorial part is @xmath65 with multiplicity 3 , as in the first case considered above ; hence , for a general @xmath0 , the multiplicity is @xmath61 . thus , the hamiltonian in the bogoliubov approximation for a general @xmath0 reads @xmath66 defining the density @xmath67 and the condensate fraction @xmath68 , one gets @xmath69 \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{{\mathbf{p } } } + u_n n_0^{n-1 } \frac{n(n-1)}{2 n ! } \sum_{{\mathbf{p } } \neq 0 } \biggl [ \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{-{\mathbf{p}}}^{\dagger } + \hat{a}_{{\mathbf{p } } } \hat{a}_{-{\mathbf{p } } } \biggr ] \mbox{. } \label{eqn:2_bogol_4}\ ] ] introducing the total particle number operator @xmath70 and enforcing the total number conservation ( or , in other terms , subtracting the chemical potential @xcite ) , one finally obtains @xmath71 we can rewrite @xmath72 as @xmath73 where the sum on @xmath74 indicates that it has to be taken over one half of momentum space , and @xmath75 the next step is to perform the bogoliubov transformation @xmath76 where @xmath77 ( general formulas for the coefficients @xmath78 and @xmath79 are given in the next subsection ) . we obtain @xmath80 where the following quantity has been introduced : @xmath81 so that the quasi - particle spectrum is given by @xmath82 of course , for interactions involving only @xmath0-body @xmath1-interactions the stability depends just on the sign of @xmath83 : if @xmath83 is positive ( negative ) , the argument in the square root of is positive for all @xmath56 ( negative for small @xmath56 ) , and the system is stable ( unstable ) . when more interactions are present , then one has to impose for stability a suitable combination of the parameters @xmath27 to be positive , as discussed in the next subsection . we consider in this subsection a model where there is a sum of @xmath3-body , @xmath2-body, ... ,@xmath0-body @xmath1-interactions ( where @xmath0 is arbitrary ) . in order to generalize the formulas presented in section ( [ sec : n - body_only ] ) , we consider @xmath22 bosons interacting via contact repulsive interactions described by the hamiltonian @xmath84 where the @xmath85-th parameter @xmath86 has physical dimension @xmath87 \ ! \cdot \ ! [ l]^{3(\ell-1)}$ ] , whose strength refers to @xmath85-body interaction . again using eq . , the hamiltonian reads @xmath88 with @xmath38 given by eq . and @xmath89 proceeding as in sec . ( [ sec : n - body_only ] ) , the hamiltonian takes the form @xmath90 it follows @xmath91 where @xmath92 the quasi - particles are introduced according to eqs . , and it has to be @xmath93 in order to guarantee the commutation relations @xmath94 = \delta_{{\mathbf{p } } { \mathbf{p'}}}$ ] . with the parametrization @xmath95 , @xmath96 , one gets @xmath97 from which follows @xmath98 where @xmath99 and @xmath100 the diagonalization yields @xmath101 the excitation spectrum for small @xmath102 gives @xmath103 , with the sound velocity @xmath104 given by @xmath105 the stability condition can be deduced from the sign of @xmath106 , the stability requiring @xmath107 . the density of particles in the excited states is @xmath109 so that the depletion fraction can be written as @xmath110 the previous expression is the usual one from the @xmath3-body contact interaction with the substitution @xmath111 : notice however that if one wants to use it to determine self - consistently @xmath112 via the relation @xmath113 , one has to take into account the dependence of the coefficient @xmath106 ( entering @xmath104 ) on the condensate density @xmath114 according to eq . . to compute the ground - state energy @xmath115 in @xmath116 one has to regularize the large-@xmath102 divergence @xcite . the correct way to write the ground - state energy is @xmath117 where we used @xmath118 and the sum over @xmath56 is up to the cut - off scale @xcite . finally we get : @xmath119 the computation presented in the previous subsections applies as well to the one - dimensional hamiltonian @xmath120 ( of course no finite condensate fraction is obtained in @xmath5 ) . denoting in the @xmath5 case the length of the system by @xmath34 and the particle density by @xmath121 , one gets the ground - state energy @xmath122 where @xmath123 , @xmath124 and @xmath125 in this way the ground - state energy becomes @xmath126\mbox{. } \label{eqn : gs_1d}\ ] ] defining @xmath127 after calculating the integral in eq . , which converges to a finite nonzero value , the ground - state energy per particle in the bogoliubov approximation is found to be @xmath128 this result is the generalization up to @xmath0-body contact interactions , of the bogoliubov result obtained in @xcite for only @xmath3-body repulsive @xmath1-interactions in @xmath5 , which for small values of @xmath129 is in agreement with the exact result @xcite . in this subsection we briefly discuss two further generalizations of the bogoliubov theory to two local interaction potentials . in section ( [ sec : n - body_only ] ) we considered a hamiltonian of the form @xmath130 with @xmath131 integer and larger or equal than @xmath3 . if @xmath132 is a real number ( with @xmath133 ) and using in eq . @xmath134 instead of @xmath135 ( @xmath136 being the gamma function ) , then one can show that in the bogoliubov approximation the following form for the hamiltonian still holds : @xmath137 where @xmath138 the results of section ( [ sec : section1-subsection2 ] ) still hold with @xmath139 instead of @xmath106 , i.e. the quasi - particle spectrum is given by @xmath140 . finally , we may consider a general contact hamiltonian of the form @xmath141 with @xmath142 and @xmath143 a function of the density operator , with the normal ordering to be taken in the second term of the right - hand side of . details are given in appendix [ sec : spectrum_general_f ] . one gets @xmath144 where @xmath145 and the parameter @xmath106 entering the quasi - particle spectrum @xmath146 given by @xmath147 in the previous section we considered general multi - body contact interactions , showing that the well - known results for the @xmath3-body @xmath1-interactions in the homogeneous case are generalized in the bogoliubov approximation by the substitution @xmath111 , where @xmath106 is given by eq . [ or , according to the considered case , by or ] . in principle , one should determine self - consistently @xmath112 from eq . , but since the bogoliubov approximation works when @xmath118 , then one can make the substitution @xmath148 in @xmath106 , resulting for a sum of @xmath0-body contact interactions in the substitution @xmath149 . the case of higher - body non - local interactions is instead different and the final result ( e.g. , the quasi - particle spectrum ) depends on the specific form of the interactions . in the following we explicitly show this for two different cases of @xmath2-body non - local interactions , cases that we treat after briefly recalling the corresponding well - known results for @xmath3-body non - local interactions . we start considering a hamiltonian for a gas of @xmath22 bosons in a region of volume @xmath150 and interacting via a @xmath3-body local repulsive , non - local potential @xmath151:@xmath152 with @xmath153 . in momentum space , the hamiltonian reads @xmath88 with @xmath38 given by eq . and @xmath154 with the change of variables @xmath155 , @xmath156 and using the bogoliubov approximation , the hamiltonian becomes @xmath157 where @xmath158 having introduced the fourier transform @xmath159 : @xmath160 the hamiltonian is readily diagonalized obtaining @xmath161 where the excitation spectrum is now @xmath162 according to eq . , for @xmath2-body interactions the hamiltonian reads in general @xmath163 , where @xmath164 and @xmath165 in the following we consider two different kinds of non - local potentials and derive their excitation spectrum . we consider a potential of the form @xmath166 where the dimension of the @xmath167 entering eq . is @xmath168=[e]^{1/2}$ ] [ with @xmath153 ] . the hamiltonian in the bogoliubov approximation reads @xmath170 \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{{\mathbf{p } } } + \frac{n_0 ^ 2}{6 } \biggl [ 2v_0 v_{{\mathbf{p } } } + v_{{\mathbf{p}}}^2 \biggr ] \ ! \bigl ( \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{-{\mathbf{p}}}^{\dagger } + \hat{a}_{{\mathbf{p } } } \hat{a}_{-{\mathbf{p } } } \bigr ) \ ! \!\biggr\ } \mbox{,}\ ] ] where we used the convention for the fourier transform ( and we denote @xmath171 by @xmath172 ) . after some further manipulations , the final result is @xmath173 where @xmath174 with @xmath175 after diagonalizing , the quasi - particle energy spectrum @xmath176 is seen to be @xmath177 the potential is a sum of three terms , each of them given by the possible pairs which can be formed between the particles . one can also consider a potential which is the product of the three pair interactions : @xmath178 where the single factor has now dimension @xmath168=[e]^{1/3}$ ] , again @xmath179 . the fourier transform @xmath159 , defined in eq . , has dimension @xmath180=[e]^{1/3 } \cdot [ l]^3 $ ] . the interaction hamiltonian is written as @xmath181 with @xmath182 using the relation and performing the bogoliubov approximation , it is found that @xmath183 \ ! \biggl\ { & \bigl [ 3 v_{{\mathbf{p}}_{12}}^3 + 2 v_{{\mathbf{p}}_{12}}^2 \ , \bigl ( v_{{\mathbf{p}}_{12 } + { \mathbf{p } } } + v_{{\mathbf{p}}_{12 } - { \mathbf{p } } } \bigr ) + v_{{\mathbf{p}}_{12 } } \ , \bigl ( v_{{\mathbf{p}}_{12 } + { \mathbf{p}}}^2 + v_{{\mathbf{p}}_{12 } - { \mathbf{p}}}^2 \bigr ) \bigr ] \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{{\mathbf{p } } } + \\ & + v_{{\mathbf{p}}_{12}}^2 \ , \bigl ( v_{{\mathbf{p}}_{12 } + { \mathbf{p } } } + v_{{\mathbf{p}}_{12 } - { \mathbf{p } } } \bigr ) \ ! \bigl ( \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{-{\mathbf{p}}}^{\dagger } + \hat{a}_{{\mathbf{p } } } \hat{a}_{-{\mathbf{p } } } \bigr ) + \\ & + v_{{\mathbf{p}}_{12 } } \ , \bigl ( v_{{\mathbf{p}}_{12 } - { \mathbf{p}}}^2 \ , \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{-{\mathbf{p}}}^{\dagger } + v_{{\mathbf{p}}_{12 } + { \mathbf{p}}}^2 \ , \hat{a}_{{\mathbf{p } } } \hat{a}_{-{\mathbf{p}}}\bigr ) \biggr\ } \mbox { .}\label{eqn : hamiltonian f.t . } \end{aligned } \end{aligned}\ ] ] to analyse the previous expression for @xmath46 , we denote by @xmath184 a general function of the @xmath185 s operators entering in . in general the terms @xmath186 and @xmath187 are different . nonetheless it is possible to show that @xmath188 indeed , by using the fact that @xmath189 and by doing the change of variables @xmath190 in the left hand side of , we can rewrite the latter as @xmath191 so that after relabelling the index @xmath192 , eq . is proved . then the interaction hamiltonian can be finally written as @xmath193 where @xmath194 the complete hamiltonian is therefore @xmath195 where @xmath196 the quasi - particle spectrum is therefore @xmath197 we derive eq . in an alternative form in appendix [ sec : alternative_derivation_3factors ] , where we extend to the @xmath2-body interaction the procedure followed in sec . ( [ sec:2-body_finite ] ) . in appendix [ sec : alternative_derivation_3factors ] we also show the equivalence of these two approaches . passing from sums to integrals in eq . , we obtain @xmath198 to see how the finite - range in @xmath3- and @xmath2-body potentials modifies the quasi - particle spectrum , we choose a specific form for it , namely a gaussian form @xmath199 applying it to the three cases of @xmath3-body finite - range [ section ] , @xmath2-body finite - range sum of three terms [ section ] and @xmath2-body finite - range product of three terms [ section ] . for a @xmath3-body finite - range potential , we put @xmath200 with @xmath168 = [ e]$ ] ( and clearly @xmath201 = [ l]^{-1}$ ] ) . the fourier transform is given by @xmath202 to make comparison between the @xmath2-body finite - range potentials we pass to dimensionless units , denoted by tildes : we set @xmath203 , @xmath204 and @xmath205 , with @xmath206 . in this way the quasi - particle spectrum can be written as @xmath207 for the @xmath2-body potential given in eq . we choose @xmath208 where for the case in consideration @xmath209=[e]^{1/2}$ ] . one then finds @xmath210 with @xmath211 ( and @xmath212 and @xmath213 defined as above ) . for the @xmath2-body potential given in eq . we choose @xmath214 where @xmath215=[e]^{1/3}$ ] . . assumes then the following simple dimensionless form : @xmath216 with @xmath217 and the same definition for @xmath212 and @xmath213 . a comparison between eqs . and shows that the functional form of the quasi - particle spectrum is different between @xmath3- and @xmath2-body finite - range interactions , and the two considered @xmath2-body finite - range interactions give well different results . to show the differences in the spectra with the same value of the dimensionless coupling strength , i.e. setting @xmath218 , a plot is presented for the sake of comparison in fig . [ confronto ] . -body finite - range potential ( solid line ) , by for the @xmath2-body finite - range potential ( dashed line ) and by eq . for the @xmath2-body finite - range potential ( dotted line ) . dimensionless units are used as explained in section ( [ example ] ) : in all the three cases the dimensionless parameter , respectively @xmath219 , @xmath220 and @xmath221 , is chosen to be @xmath222 . ] as a first application of the results presented in sections and we consider a model with @xmath3- and @xmath2-body contact interactions and we apply the obtained findings to the experimental setup described in @xcite . for a model with @xmath3- and @xmath2-body @xmath1-interactions , the hamiltonian reads @xmath223 in the bogoliubov approximation one finds @xmath224 with @xmath225 note that @xmath226 and @xmath227 have dimensions @xmath87 \ ! \cdot \ ! [ l]^3 $ ] and @xmath87 \ ! [ l]^6 $ ] , respectively . the ground - state energy is found to be @xmath228 with @xmath229 . the stability in function of @xmath230 and @xmath231 is simply given by @xmath232 , i.e. by @xmath233 finally , the depletion fraction is given by @xmath234 where in the right - hand side it has been used the fact that @xmath118 . as an application of the previous results we consider the setup described in @xcite : there one has a @xmath235 with a scattering length @xmath236 tuned in the range between @xmath237 and @xmath238 ( @xmath239 being the bohr radius ) . in @xcite the breathing frequency has been measured after a sudden variation of the scattering length . the experimental data were then fitted assuming the presence of a @xmath2-body interaction and an estimate of @xmath240 $ ] was presented for two sets of data : notice that with our convention @xmath241 $ ] ( we remind that in our study we are not considering @xmath2-body losses ) . theoretical estimates of @xmath6 were given as well in literature @xcite . regarding one - dimensional trapping potentials , we mention that the @xmath2-body recombination rate in @xmath5 bose gases was experimentally studied @xcite , and several theoretical studies addressed the problem of determining @xmath6 and its effects in the lieb - liniger model @xcite . in the experimental setup described in @xcite the external potential has the form @xmath242 $ ] : in the first set of data it is @xmath243 , @xmath244 , while in the second @xmath245 , @xmath246 , in both being @xmath247 . the fitted values of @xmath248 for the two sets were @xmath249 and @xmath250 @xcite . the stability plot is represented in fig . [ fig:1_stability ] , where we also report the points corresponding to the experimental values of @xmath251 studied in @xcite . a very simple estimate of the density @xmath114 can be obtained by setting @xmath252 and choosing as effective volume @xmath150 the product @xmath253 of the thomas - fermi quantities @xmath254 ( with @xmath255 ) defined in appendix [ sec : tf_is ] , where we study the cubic - quintic gross - pitaevskii equation in an isotropic potential parabolic trap . similarly to what is done for @xmath3-body interactions , to take into account in a simple way the effect of the anisotropy of the potential @xmath256 , we use the formula of the appendix [ sec : tf_is ] by substituting @xmath257 in place of @xmath258 and @xmath259 in place of @xmath260 [ where @xmath258 and @xmath17 refer to an isotropic potential @xmath261 , as studied in appendix [ sec : tf_is ] ] . for each value of @xmath251 we determine the thomas - fermi radius @xmath262 using eq . and then the quantities @xmath254 via eq . . we observe that when @xmath263 increases the ratio @xmath264 increases , but @xmath262 itself decreases ( due to the small but appreciable attractive @xmath3-body interactions ) , finally resulting in a decrease of the effective volume and an increase of @xmath265 . the results are reported in fig . [ fig:1_stability ] , where we introduced the dimensionless variable @xmath266 and @xmath267 , respectively proportional to @xmath226 and @xmath227 . it finds out from the figure that the points lie in the stability region : similar results ( even more deeper in the stability region ) would have been obtained if we had chosen the density @xmath268 at the trap center , which is another reasonable choice . considering for instance the first set of data of @xcite , we observe that with ( the very large value of ) @xmath269 , one gets @xmath270 ; furthermore , for a value @xmath240 $ ] which is , e.g. , @xmath271 times smaller than the one given in @xcite , one would have again @xmath270 , in both the cases well inside the stability region . - and @xmath2-body @xmath1-interactions according to eq . . above ( below ) the separation line the system is stable ( unstable ) . we are using the notations @xmath272 and @xmath273 , and the circles ( diamonds ) refer to the first and second sets of data presented in @xcite , with @xmath114 estimated as discussed in the text and @xmath257 , @xmath259 . ] in this section we discuss some further realistic interaction potentials which are of interest for current experimental setups with ultracold atoms . the discussed applications include dipolar systems , e.g. magnetic atoms and polar molecules , and soft - core potentials that can be simulated with rydberg dressed atoms . for each of these cases we analyse the energy of the elementary excitations in the homogeneous limit and derive for some interesting parameter regimes the stability diagram of the bogoliubov spectra . before discussing in detail these specific implementations let us consider a general model with a non - local @xmath3-body interaction potential accompanied by @xmath3- and @xmath2-body contact interactions . note that it is possible to extend this model straightforwardly adding @xmath1-interactions up to @xmath0-body to a @xmath3-body non - local interaction potential , but for the sake of simplicity we will not consider this general case . the hamiltonian of the system reads @xmath274 proceeding as in sections ( [ sec : local ] ) and ( [ sec : finite - range ] ) and expanding the quantum fields in terms of creation and annihilation operators , one arrives in the bogoliubov approximation at the following expression for the hamiltonian : @xmath275 where @xmath276 is the sum of the fourier components of each interaction potential times the condensate density @xmath114 to the proper power . we then arrive at the bogoliubov excitation spectrum : @xmath277 the condensate density can be obtained by solving the self - consistent equation : @xmath278 an example where the stability conditions can be carried out explicitly is the gaussian 2-body potential eq . plus a @xmath3- and @xmath2-body @xmath1-interaction . the bogoliubov spectrum then reads @xmath279 with @xmath280 , @xmath281 . in eq . dimensionless units are used as in section ( [ example ] ) , setting @xmath203 , @xmath204 and @xmath205 , with @xmath206 . setting for convenience @xmath282 , we easily derive the inequality ensuring the stability for each component of the fourier spectrum : @xmath283 we then arrive at the following conditions for the parameters @xmath219 and @xmath284 : @xmath285 which define the stability regions of the bogoliubov spectrum . the inequalities are represented in fig . [ fig : stability_plot_2b_gaussian_and_2 - 3b_delta ] . of course , for all repulsive non - local interactions and @xmath3- and @xmath2-body contact potentials the system is stable against linear perturbations around the uniform solution . when local and non - local potentials are competing ( opposite sign ) or when both @xmath286 and @xmath287 are negative , it is possible to find instabilities which may signal the onset of a structured ground - state configuration . in fig . [ fig : spectrum_2b_gaussian_and_2 - 3b_delta ] we show two spectra cases with negative short - range interaction @xmath284 where a roton - like minimum occurs . the blue dashed line spectrum is unstable to linear perturbations for finite momentum wave - vector , signalling the onset of a modulated ground - state . this spectrum lies on the separation line of fig . [ fig : stability_plot_2b_gaussian_and_2 - 3b_delta ] . and @xmath219 . above ( below ) the separation line the uniform state is stable ( unstable ) against linear perturbations . notice that competitive interactions or attractive potentials are needed to have instability . the red circle and the blue square correspond to the values of @xmath288 used in fig . [ fig : spectrum_2b_gaussian_and_2 - 3b_delta ] . ] and @xmath289 ( corresponding to the red circle in fig . [ fig : stability_plot_2b_gaussian_and_2 - 3b_delta ] ) . blue dashed line : spectrum for @xmath290 and @xmath291 , where the minimum softens and the uniform solution becomes unstable signalling the onset of a new modulated ground - state . this spectrum lies on the separation line of fig . [ fig : stability_plot_2b_gaussian_and_2 - 3b_delta ] and corresponds to the blue square in fig . [ fig : stability_plot_2b_gaussian_and_2 - 3b_delta ] . ] the results presented above can be applied in the case of model potentials like soft - core interactions that can be implemented in the laboratory with rydberg - dressed potentials to see the effect of @xmath2-body terms . these potentials were recently created in optical lattices with @xmath292 atoms excited to rydberg states @xcite . the interest for such interactions is general and involves the simulation of novel kinds of spin hamiltonians @xcite for the creation of exotic phases , like the supersolid @xcite , and for metrological applications @xcite . motivated by these experimental results and theoretical investigations , we focus on the study of the stability diagram of a @xmath3-body isotropic step - potential : @xmath293 physically realizable potentials generically display long - range tails , decaying generically as a power law ( @xmath294 or @xmath295 ) . however , the model potential of eq . ( [ soft - pot ] ) is a good approximation of such more complicated real potentials in the sense that the many - body properties found for such potential do not differ qualitatively from the realistic ones @xcite . it is important to recall that the gaussian model potential described above does not fall in the same class of soft - core potentials . the reason is that the fourier transform of the gaussian potential never changes sign , indicating that such potential alone can never display instabilities as the ones that are found for generic soft - core potentials . in the following we analyse in more detail the @xmath116 as well as the @xmath296 geometries in free space for the potential . in @xmath116 , the fourier transform of eq . is : @xmath297 where @xmath298 and @xmath299 is the spherical bessel function of the 1st kind @xmath300 . from eq . the excitation spectrum is @xmath301 in dimensionless form this excitation spectrum can be written as @xmath302 where we defined @xmath303 , @xmath304 , @xmath305 , @xmath306 , @xmath307 and @xmath308 . defining also @xmath309 , the stability condition is represented by @xmath310 [ eqn : inequality_bessel3d ] the fourier transform of eq . in @xmath296 is @xmath311 where @xmath312 is the bessel function of the 1st kind . the excitation spectrum is @xmath313 with @xmath314 . in dimensionless form this expression reads @xmath315 with @xmath316 , @xmath317 and @xmath318 with @xmath212 , @xmath319 and @xmath213 defined as above . setting @xmath309 , the stability condition is represented by @xmath320 in fig . [ fig : stability_plot_2d_rydberg ] we show the stability plots for the @xmath296 and @xmath116 geometry in terms of the regime parameters @xmath321 and @xmath284 , picking up two cases for the values @xmath322 , represented by a red circle and a blue square ; their respective spectra are reported in fig . [ fig : spectrum_2d_rydberg ] . notice that , contrarily to the gaussian potential , in @xmath296 competitive interactions , or attractive potentials are not necessary to have instability . and @xmath321 for a @xmath296 soft - core potential . above ( below ) the separation line the uniform state is stable ( unstable ) against linear perturbations . the red circle and the blue square correspond to the values @xmath322 of the plots of fig . [ fig : spectrum_2d_rydberg ] . the green dashed line sets the stability threshold for the @xmath116 soft - core potential given in eq . . above ( below ) the separation line the uniform state is stable ( unstable ) against linear perturbations . ] and @xmath323 ( corresponding to the red circle in fig . [ fig : stability_plot_2d_rydberg ] ) . blue dashed line corresponding to the blue square in fig . [ fig : stability_plot_2d_rydberg ] : spectrum for @xmath324 and @xmath325 , where the minimum softens and the uniform solution becomes unstable signalling the onset of a new modulated ground - state . this spectrum lies on the separation line of fig . [ fig : stability_plot_2d_rydberg ] . ] in this subsection we analyse the stability diagram of a homogeneous bosonic system interacting via a long - range @xmath3-body dipolar potential . such potentials have been investigated in the past years for the study of effects induced by non - local interactions in the physics of becs both in free space and in optical lattices @xcite . one of the major problems regarding dipolar interaction in free space is their anisotropic character which induces instabilities in @xmath116 homogeneous systems @xcite . on the other hand , the presence of an asymmetric harmonic trapping in combination with short - range repulsive interactions can eliminate such instabilities , opening the way to the study of interesting many - body physics with long - range interactions @xcite . recent experiments with dipolar becs showed that under certain conditions where instability is expected from a standard bogoliubov approach , dense clusters with many atoms can occur @xcite , which are expected to be superfluid @xcite . two interpretations have been proposed to explain the stabilization of this phase , namely the presence of weak @xmath2-body interactions @xcite and beyond mean - field effects ( lee - huang - yang type corrections ) @xcite . motivated by these recent developments , we analyse in further detail the stability of uniform superfluids in the presence of long - range dipolar interactions and @xmath3- and @xmath2-body contact potentials . the dipolar potential can be written as : @xmath326 where @xmath327 is the strength of the dipolar interaction and @xmath328 is the angle between the direction of polarization and the relative position of the particles . tuning the relative angle among the particles and the quantization axis the potential can be either attractive and repulsive . the fourier transform of eq . is @xmath329 where @xmath330 is the angle between @xmath56 and the polarization direction ( of the dipole - dipole interaction ) . the excitation spectrum is then found to be : @xmath331 \ , \epsilon_p^0 } \mbox { , } \label{eqn : anisotropic_spectrum1}\ ] ] with condensate density given by the equation @xmath332 \frac{p^2}{2 m } } } - 1 \biggr ) \mbox{.}\ ] ] the spectrum can be expressed in terms of the ratio @xmath333 of the dipolar length @xmath334 to the @xmath335-wave scattering length , which compares the relative strength of the dipolar and contact interactions : @xmath336 \ , \epsilon_p^0 } \mbox{. } \label{eqn : anisotropic_spectrum}\ ] ] in dimensionless units one gets @xmath337 \tilde{p}^2 } \text{,}\ ] ] where @xmath338 , @xmath305 , @xmath339 and @xmath308 , with @xmath340 . in fig . [ fig : stability_plot_3d_magnetic ] we plot the stability diagram of eq . as a function of the dimensionless parameters @xmath341 and @xmath342 for @xmath343 . we compute the specific values of these parameters for a condensate of @xmath344 also considering as the @xmath2-body contact interaction the value given in @xcite ( notice however that by using other values of @xmath227 , as the ones given in @xcite , the effect of such terms is anyway rather small ) . from fig . [ fig : stability_plot_3d_magnetic ] one sees that upon varying the scattering length the uniform phase goes from a stable ( blue diamond ) to an unstable ( red circle ) configuration . note that the @xmath2-body interaction enhances the stability region to values of @xmath345 larger than @xmath346 . and @xmath347 for @xmath343 for a @xmath116 dipolar potential and contact potentials . above ( below ) the separation line the uniform state is stable ( unstable ) against linear perturbations . data points correspond to three different values of the scattering length for @xmath348 atoms : @xmath349 ( red circle ) , @xmath350 ( green square ) , and @xmath351 ( blue diamond ) , where @xmath239 is the bohr radius . in the calculations we included a @xmath2-body contact interaction potential using the value @xmath352 considered in @xcite , which slightly enhances the stability region of the uniform phase to @xmath353 ( green square ) . more quantitatively , putting @xmath354 one would have the green square on the stability line for @xmath355 . here we set the average density @xmath356 and the dipolar length @xmath357 . ] in this work we have presented a systematic study of weakly interacting bosonic gases with local and non - local multi - body interactions in the bogoliubov approximation . we considered conservative multi - body interactions for which the number of particles is conserved : of course multi - body interactions are associated with the presence of particle losses @xcite , that we did not study , rather focusing on the determination of the stability conditions due to the competition between @xmath3- and higher - body interactions . a variety of interparticle potentials have been considered . we first considered contact interactions , studying the case in which the interparticle potential can be written as a general sum of @xmath0-body @xmath1-interactions , providing the quasi - particle spectrum , the ground - state energy and the stability conditions . results for general effective contact potentials are also presented . our findings show that the well - known results for the @xmath3-body @xmath1-interactions in the homogeneous case are generalized in the bogoliubov approximation by the substitution @xmath111 , where @xmath106 is a function of the condensate fraction @xmath114 given by eq . for potentials which are sums of @xmath0-body @xmath1-potentials and by eq . in the general case . since the bogoliubov approximation works well when @xmath118 , then one can make the substitution @xmath148 in @xmath106 , resulting for a sum of @xmath0-body contact interactions in the substitution @xmath358 . the case of higher - body non - local interactions is instead different from this respect and the final results depend on the specific form of the interactions . we explicitly considered two different cases of @xmath2-body non - local interactions . in the last part we discussed a few interaction potentials which are of interest for current experimental setups with ultracold atoms . implementations include systems with @xmath3- and @xmath2-body @xmath1-interactions , where we applied in the homogeneous limit our results with the values given in @xcite . we also considered the effect of ( conservative ) @xmath2-body terms in dipolar systems , e.g. magnetic atoms and polar molecules , and soft - core potentials that can be simulated with rydberg dressed atoms . for each of these cases we analysed the energy of the elementary excitations and derived for some interesting parameter regimes the stability diagram of the bogoliubov spectra . in the present paper we focused on higher - body interactions in the homogeneous limit , having in mind both @xmath2-body terms and general effective multi - body interactions of course ultracold experiments are done in confined traps , and we think that a systematic study of the bogoliubov equations in inhomogeneous potentials with general multi - body local and non - local interactions is an interesting direction of future research . discussions with l. barbiero , g. gori and l. salasnich are gratefully acknowledged . t.m . acknowledges cnpq for support through bolsa de produtividade em pesquisa n. 311079/2015 - 6 and the hospitality of the physics department of the university of padova . support form the european strep matterwave is acknowledged . in this appendix we consider the case of a general contact interaction described by a hamiltonian of the form @xmath359 one has @xmath360 where we wrote explicitly the operator @xmath361 in terms of the operators @xmath362 , @xmath363 and we used the bogoliubov approximation . we assume that the function @xmath364 can be expanded in series up to the second order . since @xmath365 and @xmath366 , it follows that @xmath367 , we can write : @xmath368 where @xmath369 neglecting products of 3 or more operators , the interaction part reads @xmath370 \biggl\ { & { \cal f } ( n_0 ) + \frac{\partial { \cal f } \ ! \left ( x \right)}{\partial x}\bigg\vert_{x = n_0 } \ ! \biggl ( \ ! \sqrt{\frac{n_0}{\omega } } \sum_{{\mathbf{p } } \neq 0 } \bigl ( e^{-\frac{i}{\hbar } { \mathbf{p } } \cdot { \mathbf{r } } } \ , \hat{a}_{{\mathbf{p}}}^{\dagger } + e^{\frac{i}{\hbar } { \mathbf{p } } \cdot { \mathbf{r } } } \ , \hat{a}_{{\mathbf{p } } } \bigr ) + \frac{1}{\omega } \sum_{{\mathbf{p}},{\mathbf{p ' } } \neq 0 } e^{-\frac{i}{\hbar } ( { \mathbf{p}}'\!-{\mathbf{p } } ) \cdot { \mathbf{r } } } \ , \hat{a}_{{\mathbf{p'}}}^{\dagger } \hat{a}_{{\mathbf{p } } } \biggr ) + \\ & + \frac{1}{2 ! } \frac{\partial^2 { \cal f } \ ! \left ( x \right)}{\partial x^2}\bigg\vert_{x = n_0 } \ ! \biggl ( \frac{n_0}{\omega } \ ! \sum_{{\mathbf{p}},{\mathbf{p ' } } \neq 0 } \bigl ( e^{-\frac{i}{\hbar } { \mathbf{p } } \cdot { \mathbf{r } } } \ , \hat{a}_{{\mathbf{p}}}^{\dagger } + e^{\frac{i}{\hbar } { \mathbf{p } } \cdot { \mathbf{r } } } \ , \hat{a}_{{\mathbf{p } } } \bigr ) \bigl ( e^{-\frac{i}{\hbar } { \mathbf{p } } ' \ ! \cdot { \mathbf{r } } } \ , \hat{a}_{{\mathbf{p}}'}^{\dagger } + e^{\frac{i}{\hbar } { \mathbf{p } } ' \ ! \cdot { \mathbf{r } } } \ , \hat{a}_{{\mathbf{p } } ' } \bigr ) \ ! \biggr ) \ ! \biggr\ } \colon \end{aligned}\\ & = { \cal f } ( n_0 ) \ , \omega \begin{aligned}[t ] & + \frac{\partial { \cal f } \ ! \left ( x \right)}{\partial x}\bigg\vert_{x = n_0 } \sum_{{\mathbf{p } } \neq 0 } \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{{\mathbf{p } } } + \\ & + \frac{1}{2 ! } n_0 \frac{\partial^2 { \cal f } \ ! \left ( x \right)}{\partial x^2}\bigg\vert_{x = n_0 } \sum_{{\mathbf{p } } \neq 0 } \bigl ( 2 \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{{\mathbf{p } } } + \hat{a}_{{\mathbf{p}}}^{\dagger}\hat{a}_{-{\mathbf{p}}}^{\dagger } + \hat{a}_{{\mathbf{p } } } \hat{a}_{-{\mathbf{p } } } \bigr ) \text { , } \end{aligned } \end{aligned}\ ] ] where we integrated out the space variable and used the conservation of momentum , having taken explicitly the normal ordering of the operators . using the conservation of the total number of particles , we can write @xmath371 so that @xmath372 & + \biggl ( \frac{\partial { \cal f } \ ! \left ( x \right)}{\partial x}\bigg\vert_{x = n_0 } \ ! \ ! \ ! \ ! \ ! \ ! \ ! - \ , \ , \ , \frac{\partial { \cal f } \ ! \left ( x \right)}{\partial x}\bigg\vert_{x = n } \biggr ) \ ! \sum_{{\mathbf{p } } \neq 0 } \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{{\mathbf{p } } } + \\ & + \frac{1}{2 ! } n_0 \frac{\partial^2 { \cal f } \ ! \left ( x \right)}{\partial x^2}\bigg\vert_{x = n_0 } \sum_{{\mathbf{p } } \neq 0 } \bigl ( 2 \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{{\mathbf{p } } } + \hat{a}_{{\mathbf{p}}}^{\dagger}\hat{a}_{-{\mathbf{p}}}^{\dagger } + \hat{a}_{{\mathbf{p } } } \hat{a}_{-{\mathbf{p } } } \bigr ) \text{. } \end{aligned } \end{aligned}\ ] ] denoting @xmath373 , we see that @xmath374 so that the difference in the second term of the previous equation is of higher order and it can be safely neglected . thus the complete hamiltonian can be written as @xmath375 from which eq . in this appendix we consider the @xmath2-body finite - range potential of eq . : @xmath376 where @xmath377 . in momentum space , the hamiltonian can be written as @xmath88 with the kinetic term @xmath38 given by eq . . to write the interaction term @xmath46 , we consider here the change of variables \{@xmath378 } @xmath379 \{@xmath380 } , where @xmath381 in the bogoliubov approximation , after some algebra we obtain @xmath382 \biggl\ { & \int \ ! \mathrm{d}{\mathbf{r}}_{12 } \ , \mathrm{d}{\mathbf{r}}_{31 } \ , \mathfrak{f}_{{\mathbf{p}}}({\mathbf{r}}_{12},{\mathbf{r}}_{31 } ) v(r_{12 } ) v(r_{31 } ) v(r_{12 } \ ! + \ ! r_{31 } ) \bigl ( \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{-{\mathbf{p}}}^{\dagger } \ ! + \ ! \hat{a}_{{\mathbf{p } } } \hat{a}_{-{\mathbf{p } } } \bigr ) \ ! + \\ & \ ! + \ ! \ ! \mathrm{d}{\mathbf{r}}_{12 } \ , \mathrm{d}{\mathbf{r}}_{31 } \bigl [ 3 + 2 \ , \mathfrak{f}_{{\mathbf{p}}}({\mathbf{r}}_{12},{\mathbf{r}}_{31 } ) \bigr ] v(r_{12 } ) v(r_{31 } ) v(r_{12 } \ ! + \ ! r_{31 } ) \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{{\mathbf{p } } } \biggr\ } \mbox { , } \end{aligned } \end{aligned}\ ] ] where we defined @xmath383 setting @xmath384 proceeding as in sec . ( [ sec:2-body_finite ] ) , one can write the hamiltonian in the form @xmath385 \biggl\ { & \biggl [ \epsilon_p^0 \ ! + \ ! \frac{n_0 ^ 2}{3 } \ ! \ ! \int \ ! \ ! \mathrm{d}{\mathbf{r}}_{12 } \ , \mathrm{d}{\mathbf{r}}_{31 } \ , \mathfrak{f}_{{\mathbf{p}}}({\mathbf{r}}_{12},{\mathbf{r}}_{31 } ) v(r_{12 } ) v(r_{31 } ) v(r_{12 } \ ! + \ ! r_{31 } ) \biggr ] \bigl ( \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{{\mathbf{p } } } \ ! + \ ! \hat{a}_{-{\mathbf{p}}}^{\dagger } \hat{a}_{-{\mathbf{p } } } \bigr ) + \\ & + \frac{n_0 ^ 2}{3 } \ ! \int \ ! \ ! \mathrm{d}{\mathbf{r}}_{12 } \ , \mathrm{d}{\mathbf{r}}_{31 } \ , \mathfrak{f}_{{\mathbf{p}}}({\mathbf{r}}_{12},{\mathbf{r}}_{31 } ) v(r_{12 } ) v(r_{31 } ) v(r_{12 } \ ! + \ ! r_{31 } ) \bigl ( \hat{a}_{{\mathbf{p}}}^{\dagger } \hat{a}_{-{\mathbf{p}}}^{\dagger } \ ! + \ ! \hat{a}_{{\mathbf{p } } } \hat{a}_{-{\mathbf{p } } } \bigr ) \biggr\ } \mbox { , } \end{aligned } \end{aligned}\ ] ] which is diagonalizable by the standard procedure . then the excitation spectrum is @xmath386 } \mbox{:}\ ] ] we obtain @xmath387 \mbox{. } \label{eqn : spectrum_cv } } \ ] ] to demonstrate the equivalence between eqs . and we have to show that their prefactors multiplying @xmath388 are equal . we start considering the factor of eq . , which we denote by @xmath389 : @xmath390 since @xmath391 inserting this expression in , it is readily seen that @xmath389 is equal to @xmath392 which is equal to @xmath393 using in the previous expression the properties @xmath394 and @xmath395 we retrieve the prefactor of the @xmath388 term in eq . . in this appendix we consider the time - independent cubic - quintic gross - pitaevskii equation @xmath396 where the parabolic trap @xmath397 is assumed isotropic : @xmath398 to make contact with the notation used in the main text , it is @xmath399 and @xmath400 . imposing the normalization condition and defining the thomas - fermi radius @xmath262 such that @xmath403 , one gets @xmath404 where @xmath405 is measured in units of the harmonic oscillator length @xmath406 ( with @xmath407 ) and the dimensionless quantities @xmath408 , @xmath409 and @xmath410 in eq . are given by @xmath411 , @xmath412 and @xmath413 . in the function @xmath414 is defined ( with @xmath415 ) as @xmath416 \mbox{. } \label{func}\ ] ] the expectation value of @xmath417 is given by @xmath418 where @xmath262 is defined by eq . and the function @xmath419 is given by @xmath420 \mbox{. } \label{func2}\ ] ] since @xmath421 , we get @xmath422 . we set @xmath423 and @xmath424 , with @xmath255 , so that the product @xmath253 , which we denote by @xmath425 , reads @xmath426 where @xmath427 is given by eq . . the previous formulas simplify for @xmath428 , when only the @xmath2-body interaction term [ i.e. , only the quintic term in the gross - pitaevskii equation ] is present . one gets @xmath429 where we introduced the dimensionless parameter @xmath430 ( remember that @xmath431=[e ] \ , \cdot \ , [ l]^6 $ ] ) . the expectation value of @xmath427 is simply given by @xmath432 eqs . and have to be compared with the usual results for the cubic gross - pitaevskii ( with @xmath433 ) , where one has respectively @xmath434 ( where @xmath435 ) and @xmath436

basic quantum information measures involved in the information analysis of quantum systems are considered . it is shown that the main quantum information measurement methods depend on whether the corresponding quantum events are compatible or incompatible . for purely quantum channels , the coherent and compatible information measures , which are qualitatively different , can be distinguished . a general information scheme is proposed for a quantum - physical experiment . in this scheme , informational optimization of an experimental setup is formulated as a mathematical problem .

quant-ph0210190

significant progress in quantum physics in the last decade has essentially influenced the evaluation of the role and qualitative content of quantum physics , though , certainly , its fundamentals have been pre - served . in the earlier studies of specific nonclassical features of quantum systems , an experimenter only had to ensure suitable macroscopic conditions ( via choosing an object , using macroscopic fields , providing for necessary temperature , etc . ) . but today , it has become possible to intentionally vary directly quantum states of elementary quantum systems . this possibility initiated a number of novel applied sciences and technologies , such as quantum cryptography , quantum communications , and quantum calculation physics [ 14 ] , which exploit nonclassical features of quantum system states . a comprehensive analysis of quantum specific features characterizing physical systems ( these features are involved in the above applications ) can be found in the latest reviews [ 510 ] and monographs [ 1114 ] . despite the variety of physical mechanisms used for generation , processing , and transmission of quantum information , all of them are principally based on the only essential difference between quantum and classical events . this is non - commutativity of quantum variables of systems under consideration , which is equivalent to the nonorthogonality of their quantum states . because of this circumstance , it is impossible to consider an arbitrary set of quantum events within the framework of classical logic , which is the result of the so - called incompatibility of nonorthogonal states . indeed , suppose two quantum states @xmath0 and @xmath1 are orthogonal . then , using them as an algebraic basis for an algebra with addition in the form of linear subspace union ( sp ) and product in the form of intersection , we find that the algebra of quantum events constructed on @xmath0 and @xmath1 contains only four subspaces @xmath2 , including the empty @xmath3 and 2d hilbert @xmath4 spaces . it is equivalent to the algebra of four elements @xmath5 , @xmath6 constructed as an aggregate of subsets contained in the set of two point elements @xmath7 and @xmath8 ( the indices of quantum states under consideration ) with addition of subsets in the form of their sum and product in the form of their intersection . this algebra , considered as an elementary example , corresponds to the classical ( bivalent , i.e. , aristotelian [ 15 ] ) logic underlying classical physics , where the rule of the excluded third holds : either @xmath9 or not-@xmath10 . in the above example , this rule is expressed as @xmath11 in terms of indices and as @xmath12 in terms of states . here , @xmath1 has the sense of negation of event @xmath0 and @xmath13 has the sense of a known certain event . note that both elementary events @xmath0 and @xmath1 belong to this class of events . if two states are nonorthogonal , in the relationship @xmath14 state @xmath1 is not negation of @xmath0 because it includes a nonzero projection on @xmath0 along with the infinite set of other states @xmath15 existing due to the superposition principle . in this case , @xmath0 and @xmath1 are , respectively , eigenvectors of two non - commuting operators @xmath16 and @xmath17 of certain physical quantities . these quantities can not be measured simultaneously since their quantum eigenstates are nonorthogonal . when the above physical quantities take their possible values , the corresponding sets of quantum states are incompatible because they can not be considered simultaneously within the framework of classical logic . below , we present an elementary analysis which reveals a close relationship between physical quantities and their corresponding quantum states . determination of the relationships directly between coupling states and the quantitative measurement of information carried by these states is the subject of the information theory . the purpose of this study is to analyze the choice of an adequate quantitative measure of quantum information and its possible role in physics . to this end , we investigate a number of basic physical models of quantum systems and schemes of physical experiments in the presence or absence of incompatibility of quantum states , which is their only fundamental specific feature causing various quantum effects . in section i , the main types of quantum information and corresponding quantum measures are classified using this criterion . in sections ii and iii , we successively consider coherent and compatible information . the latter type of quantum information is closely related to quantifying information efficiency of a given experimental scheme . this problem is discussed in section iv . the notion of quantum information formulated simultaneously with the basic laws of quantum physics is directly related to them and plays a key role in their interpretation . any quantum effect ( for example , the essentially microscopic process of the atom spontaneous radiation or a macroscopic transition to the super - conducting state ) can explicitly be related with the processes of quantum information transformation if this information is adequately associated with the corresponding ensembles of quantum states . one can also state that a prototype of quantum information had appeared before the classical shannon information theory was developed . it suffices to recall the born interpretation of the physical sense of the wave function or to analyze the information content of the quantum measurement postulate ( the wave function collapse ) [ 16 ] . the notion of quantum information was admitted to be important long ago . however , interest in its application has only recently been stimulated by the development of modern quantum optics experimental methods ensuring quantum system control . progress in this area of quantum physics enables one to employ quantum information not only as a useful abstract notion but , also , to manipulate it in actual experiments in a free manner . studies of quantum information processing initiated quantum information physics . this new field of physics is covered in the literature referred to in the bibliography index [ 17 ] . proceeding from the conventional description of quantum mechanics [ 16 , 18 - 20 ] , one can suppose that proper physics must deal only with quantum - physical quantities , whereas quantum states could rather be studied , irrespective of specific physical variables , within the framework of mathematics , one of whose fields is sometimes qualified as the classical information theory . however , a more comprehensive analysis shows that this is not true : as soon as quantum states are associated with eigenstates of physical variables characterizing an actual quantum model , they become carriers of physically meaningful information . for example , let us consider the mathematical structure of self - adjoint operators @xmath16 in hilbert space @xmath13 , which are used in quantum - mechanical representations of physical variables . then , the spectral decomposition of operator @xmath18 describes its splitting into mathematical objects of two types : the set of physically possible values @xmath19 and the set of corresponding quantum states @xmath20 . the latter objects contain the most general physical information which is independent of specific values @xmath19 and characterizes only certain physical events . each of these events lies in the fact that a physical quantity takes one of the values @xmath19 . information relationships between quantum states are determined by the dynamic properties of a physical system and , evidently , provide for the most fundamental description of its dynamic characteristics . these relationships may characterize the intrinsic dynamics of a quantum system and its interaction with other systems . initially , they are represented as equations for wave functions or quantum state operators . the theoretical information approach is based on introducing an adequate quantitative measure of information exchange . in the general case , being independent of specific physical variables , this measure is superior to them in the analysis of general dynamic properties of a quantum system . if a specific scheme of a physical experiment is not discussed , a quantum system is described using its information characteristics which yield quantitative relationships between the quantum states of a given system and the quantum states of other systems that may interact with it . for example , the main information content of a two - level atom radiation is reduced to the fact that the quantum information carried by this atom is transferred to the corresponding quantum of the photon field . in this situation , the quantum receives information on the phase of the initial atom state , i.e. , the information exchange is essentially quantum and retains coherence of the transformed wave functions ( see section ii ) . thus , the information description is more comprehensive than in the case when it is represented in terms of the atom - field energy exchange . the fundamental concept of classical information measure is formulated in the shannon information theory [ 21 , 22 ] . for classical systems of physical events , a unified quantitative measure of information can be introduced , which is independent of neither the specific physical content of this information nor its usage . this measure enables one to express the asymptotic level of error - free transmitted information content as the optimized shannon information content , which is necessary for combining numerous channels in the presence of noise . this elegant theory is based on a special property of classical ensembles that is excluded from the original principles of quantum physics . this property is reproducibility of classical events : statistically , it makes no difference whether one and the same physical system or its information - equivalent copy is at the input and output . however , the latter situation is impossible in the quantum world . obviously , this circumstance initiated the discussion of the question whether the shannon information measure could be applied to quantum systems [ 2325 ] . in this paper , it is shown that the traditional definition of the shannon entropy and the corresponding information measure can successfully be used for analyzing quantum systems also in the case when fundamental differences between ensembles of classical and quantum events are properly taken into account . the quantum theory involves the principle of quantum state superposition implying the existence of an arbitrary linear combination @xmath21 of states @xmath0 and @xmath1 , which results in the presence of a continuum set in any quantum system . this set is hilbert space @xmath22 of states , most of which do not coincide with any orthogonal basis states @xmath20 associated with a certain physical quantity described by the operator @xmath23 . when a quantum system is transformed , the basis vectors and the entire hilbert space are also transformed . this circumstance , which is exploited in algorithms of quantum calculations , significantly enhances their efficiency due to high concurrency of operations [ 3 , 4 , 13 ] . however , the above continuum set does not contain an unlimited information content in the conventional classical sense . the point is that an arbitrary quantum state @xmath24 can be distinguished from another state @xmath1 only when the states are orthogonal . the probability of random coincidence of two arbitrary states is determined by the squared absolute value of the scalar product , so that the 2d probability density of two equiprobable states has the form @xmath25 where @xmath26 and @xmath27 are small volume elements on the sphere of wave functions satisfying the normalization @xmath28 and @xmath29 is the dimension of space @xmath13 . within the framework of the shannon information theory , one can find effective number @xmath30 of states @xmath7 distinguishable by variable @xmath8 and vice versa [ 26 ] . probability distribution describes the information exchange between two information variables @xmath7 and @xmath8 , which , according to the shannon theory , is characterized by effective number @xmath30 of error - free transmitted messages that are formed of subsets @xmath31 of indices corresponding to quantum states @xmath7 . this effective number ( calculated per symbol ) is attained for infinitely long sequences of independently transmitted unit symbols . when each @xmath31 is associated with appropriate states @xmath7 , all of these states will be distinguished without error ( in the above sense ) , so that @xmath30 is the number of distinguishable states . it is determined by the corresponding shannon information content @xmath32 using the formula @xmath33 . for joint probability distribution ( 1 ) ( see [ 27 ] ) , @xmath34 bit for @xmath35 and @xmath36 bit ( @xmath37 is the euler constant ) for @xmath38 . thus , at any space dimension @xmath29 , @xmath39 , i.e. quantum uncertainty reduces the effective number of distinguishable states even below a value of two corresponding to one bit . this is caused by high quantum uncertainty in pure states @xmath40 , which , on the one hand , are analogs of deterministic classical states when only their orthogonal sets are considered and , on the other hand , contain internal quantum uncertainty . in the latter case , other nonorthogonal states may coexist and pure states determine the corresponding probability distributions @xmath41 for variables @xmath42 such that @xmath40 is not their eigenfunction . particularly , the corresponding entropy of the @xmath43-partite state @xmath44 asymptotically approaches @xmath45 at @xmath46 , i.e. , the pure character of the state virtually does not influence the uncertainty of all of the quantum states . the mixed state @xmath47 exhibits the maximum uncertainty equal to @xmath45 bit . thus , the first specific feature of quantum information due to incompatibility of quantum states is the impossibility of extracting a noticeable information content with spaces of large dimension d without selecting states . quantum information always must be selected in transmission channels transforming a considerable information content in a distinguishable form . let us consider a simple example illustrating the qualitative characteristics of quantum systems due to incompatibility of all quantum states . suppose two two - level atoms are in one and the same state ( fig . while the operational sense of the expression _ in one and the same state _ is quite clear , its qualitative sense essentially does not coincide with that of this expression in the classical case . when this example is considered from the classical standpoint , only two basis states ( @xmath48 ) of each atom are taken into account . then , when describing the statistic of these atom states , it makes no difference whether they are assumed to correspond to different atoms or to one and the same atom . the point is that , in a combined system , only one state has a nonzero probability and the knowledge of state @xmath49 of each atom corresponds to the exact knowledge of a possible state of the other atom . thus , when only populations are considered in the classical case , atoms can be equivalent copies of each other . in the quantum case , it is impossible to copy all quantum states in a similar manner . in addition to the above two states in an atom , there exist other states @xmath0 with nonzero probabilities @xmath50 such that @xmath51 is not the proper basis for averaged physical quantities . this circumstance is due to quantum uncertainty , which is always present in an ensemble of quantum states ( fig . it is well known that , for a harmonic oscillator , this uncertainty manifests itself in nonzero energy of vacuum fluctuations @xmath52 . for two - level atoms , it takes the form of nonzero values @xmath53 , where pauli matrices @xmath54 have the sense of quadrature cosine and sine components of atom oscillators . in spite of the fact that these atoms are in one and the same state , their corresponding eigenstates are different because the above state refers to different physical systems . each of these systems contains its own ensemble of mutually incompatible quantum states , which are described by nonorthogonal eigenvectors corresponding to different noncommuting operators of physical variables as shown in fig . 1b . indeed , mean - square residuals @xmath55 and @xmath56 are nonzero because their operators do not commute with the operator of difference of populations @xmath57 which corresponds to strictly nonzero difference @xmath58 . this means that the eigenstates of these quantum variables for the atoms under consideration do not coincide , i.e. , all of their mutually corresponding quantum states are by no means copies of each other . this simple example implies the important general conclusion that , in the presence of incompatibility , i.e. , nonorthogonality in ensembles of quantum states of two different atoms , these ensembles are always different . hence , information on the states of a quantum system at a certain instant which is obtained by means of any other system considered at the same instant is never complete . complete quantum information on all quantum states of a system at a specified instant can be provided only by this system itself . the complete information content can appear in another place ( or at another instant ) when it is automatically cancelled at the initial position , which occurs , for example , during teleportation [ 2 ] . quantum information can be teleported only to a single receiver , which provides for the possibility of designing absolutely intercept - secure communication systems based on quantum cryptography . the above qualitative concept of uniqueness of quantum information can be justified quantitatively . let us consider the squared difference of projectors on the mutually corresponding wave functions of two atoms considered at one and the same instant . integrating the operator of this squared difference over all possible wave functions , we obtain the expression yielding a strictly positive value : @xmath59 where integration is performed over the bloch sphere including states @xmath0 with index @xmath60 , @xmath61 is a small volume element , and @xmath62 is the total volume of integration . this expression is similar to the corresponding classical formula for the rms discrepancy @xmath63 between classical indicators @xmath64 and @xmath64 of events @xmath65 and @xmath66 related with random variables @xmath67 and @xmath68 . for any joint probability distribution @xmath69 that describes random variables coinciding everywhere , the mean of random function @xmath70 is zero , i.e. , for these probability distributions , all of the events @xmath65 and @xmath66 are realized simultaneously and the above quantities are copies of each other . being an exact quantum analog of the discrepancy between two ensembles of classical events , bipartite operator ( 2 ) has two proper subspaces which are formed of singlet and triplet bell states @xmath71 and correspond to the eigenvalues @xmath72 of the mean - square residual . the value corresponding to the singlet state is three times greater than the value corresponding to the triply degenerate triplet state . strict positivity of this operator , i.e. , the absence of the zero eigenvalue , means that , for any joint density matrix , its averaging yields a nonzero result , which determines the rms discrepancy of all quantum states . hence , it is impossible to mutually copy all quantum states of various systems irrespective of their states . = 0.3 = 0.3 the foregoing implies that the key difference between the classical and quantum types of information depends on whether the states associated with the information of interest are compatible or incompatible . the states of different systems considered at one and the same instant are always compatible . hence , in the presence of internally incompatible states , they can not copy each other , and the total ensembles of quantum states of one and the same system which are considered at two different instants are most often incompatible and , moreover , can copy each other in the absence of noise retaining the uniqueness of their quantum fluctuations at every instant . however , the states of two different systems observed at different instants may be compatible or incompatible depending on the transformation which couples these two instants . this factor is rather important for the basic definitions of quantitative quantum information measures ( see sections ii and iii ) . thus , according to the most fundamental classification of quantum information , which involves the compatibility / incompatibility property of state ensembles under consideration , the following four types of information can be recognized : classical information , semiclassical information , coherent information , and compatible information . in the case of classical information , all states are compatible and , within the original form of the shannon information theory , are considered as classical on default [ 21 , 22 ] . however , classical information can always be transmitted via a quantum channel and is of interest to quantum physicists as well . a classical channel is specified by conditional probability distribution @xmath73 of states of output @xmath74 at fixed states of input @xmath75 . for semiclassical information , all input information is specified by classical states @xmath76 and output states are characterized by internal quantum incompatibility as quantum states in hilbert space @xmath13 . nevertheless , output states are automatically compatible with input states . in the general case , a quantum channel is described by an ensemble of mixed quantum states @xmath77 depending on classical parameter @xmath76 [ 28 , 29 ] . variables @xmath76 are equivalent to input variables @xmath75 the set of all wave functions @xmath78 is equivalent to output states @xmath74 , and the density matrix of @xmath77 is equivalent to conditional probability distribution @xmath73 of a classical channel . in the case of coherent information , the spaces of input and output states are characterized by internal quantum incompatibility . being related by channel superoperator @xmath79 , which transforms the input density matrix to the output density @xmath80 [ 30 , 31 ] , these spaces are mutually incompatible . transformation @xmath81 determines the flow of quantum incompatible states transmitted from the channel input to output and is a completely quantum analog of classical conditional distribution @xmath73 , which , in a similar manner , linearly transforms classical input probability distribution @xmath82 to output distribution @xmath83 . in the case of compatible information , the input and output contain internally incompatible states but are mutually compatible . while the first three types of information and the coherent information measure ( which was introduced not long ago ) are well known [ 22 , 28 , 30 ] , compatible information has been introduced in an explicit form most recently as a special type of information measure [ 32 ] . it is determined for a combined bipartite system with the compatible input and output characterized by internal quantum incompatibility . coherent and compatible types of information com - prise all possible qualitatively different types of information in completely quantum channels . our study of possibilities provided by applying the information approach to actual experiments shows that only compatible information is an adequate tool for analyzing the information efficiency of an abstract scheme of a quantum - physical experiment . according to the classification presented in section i , one possible quantitative information measure for a completely quantum channel is coherent information introduced by schumacher and lloyd [ 30 , 31 ] . it serves as a quantitative measure of incompatible information content that is transmitted from one state space to another . one can consider both the case of one and the same state space and the case of physically different spaces . a trivial case of coherent information exchange is the dynamic evolution of a closed system described by unitary operator @xmath84 : @xmath85 . here , all pure states @xmath40 admissible by initial density matrix @xmath86 are transmitted unchanged and transmitted coherent information content @xmath87 coincides with the initial information content . by definition , the latter is measured with the von neumann entropy @xmath88= s[\hat\rho_a]$ ] , i.e. , the corresponding information is described by the expression @xmath89 however , this definition requires additional reasoning in terms of the operational sense of the density matrix . in the self - consistent theory , this matrix is only a result of averaging the pure state of a combined system over auxiliary variables . then , expression ( 3 ) must be considered as the entanglement of input system @xmath90 and reference system @xmath91 for properly chosen pure state @xmath92 ( such that @xmath93 ) of combined system @xmath94 . thus , coherent information is measured in terms of mutually compatible states of two different systems @xmath90 and @xmath91 . at the same time , information is transferred from input @xmath90 to output @xmath95 , the latter differing from @xmath90 by a unitary transformation only . information channel @xmath79 with corresponding noisy environment @xmath96 ( fig . 2a ) must be included in the information system , which completes the description of its general structure [ 33 ] . = 0.35 = 0.45 coherent information for a channel of the general type is defined as follows [ 33 ] : @xmath97-s\bigl[({\cal n}{\otimes}{\cal i}){\mathop{\left|{\psi_{ar}}\right.\rangle}\nolimits } { \mathop{\langle\left.{\psi_{ar}}\right|}\nolimits}\bigr],\ ] ] which is in agreement with the shannon definition of classical information [ 34 ] . here , @xmath98 is the unit superoperator which is applied to the variables of the reference system leaving them unchanged . the second term in this expression describes the exchange entropy , which is nonzero only due to the interaction between subsystems @xmath94 and @xmath96 when @xmath99 . superoperator @xmath79 transforms , according to the relationships @xmath100 the states of input @xmath90 into the states of output @xmath95 , whose quantum states are compatible with the states of reference system @xmath91 as before because these systems are not entangled due to the above transformation . this enables one to consider @xmath95 and @xmath91 as kinematically independent systems described by joint density matrix @xmath101 . taking into account the above circumstance and the zero , by supposition , entropy of combined system @xmath102 , relationship ( 4 ) can be considered as the measure of entanglement between output b and combined system @xmath103 decreased by the entropy of exchange between the channel @xmath104 and noisy environment @xmath96 . thus , it follows from ( 4 ) that , in terms of physical content , coherent information is a specific measure of preserved entanglement between compatible systems @xmath91 and @xmath90 ( remaining after information is transmitted through the channel @xmath104 ) rather than that it directly serves as a measure of quantum incompatible state flow transmitted from @xmath90 to @xmath95 . in the general case , output @xmath95 may be physically different from @xmath90 and even may be described by a hilbert space of quite different dimension ( @xmath105 ) [ 35 , 36 ] , which is illustrated by the physical example of the information system presented in fig . 2b . for this system , input @xmath90 and reference system @xmath91 correspond to the ground two - level states of two entangled atom @xmath106-systems . information channel @xmath79 is provided by laser excitation of input system @xmath90 at the radiation - active upper level . combined with the vacuum state , two emitted photons correspond to output @xmath95 , whereas all of the remaining freedom degrees of the field combined with the excited atom state form noisy environment @xmath96 . the elementary carrier of quantum information is a two - level system , an analog of the classical bit which is conventionally called a qubit .. ] therefore , it is logical to call the quantitative measure of specifically quantum coherent information a qubit too . obviously , this unit of coherent information corresponds to the use of a binary logarithm in definition ( 3 ) , which yields @xmath107 qubit for a two - level system with density matrix @xmath108 characterizing the state with the maximum possible quantum entropy . with this state , all possible quantum states are presented equally and most completely . now , let us find out how the quantitative measure of quantum information can be used in physics . the quantum theory usually is applied to calculations of certain means of the form @xmath109 , where @xmath19 and @xmath20 are , respectively , eigenvalues and eigenvectors of operator @xmath16 . this decomposition is the result of averaging physical quantities represented in terms of probabilities @xmath110 of quantum states @xmath20 . all possible variables constitute an infinite set that is much richer than the set of all quantum states . therefore , irrespective of physical values , relationships between physical states provide for more general information on physical couplings in a more compact form . the characteristic features of the coherent information exchange correspond to physical relationships expressed in the most general form because they are associated with the most general properties of interaction between two quantum systems chosen as an input and output and related by a one - to - one transformation of all possible mixed input states . actually , the dependence of coherent information on the parameters of an information system is of even more fundamental character than relationships between specific physical quantities . this dependence is calculated for a number of fundamental models widely applied in quantum physics [ 3436 ] . as an example , let us consider the well - known dicke problem [ 37 ] , where the information exchange between atoms demonstrates the dynamics of the same oscillation type [ 35 ] as the energy exchange via emitted photons , which determines radiation damping in a system of two two - level atoms . in this case , the oscillation dynamics is typical of not only energy but also the set of other variables . hence , in order to describe the general properties of interatomic interaction , it is expedient to consider coherent information as a measure of pre - served entanglement rather than the set of various physical variables . the former is a characteristic of the internal quantum incompatibility exchange between mutually compatible state sets of the reference and input systems @xmath111 and @xmath112 . in the dicke problem , information exhibits its coherent character in the time dependence at a large difference between decay rates of the short - and long - lived dicke states . the interatomic exchange coherence is realized by coherent oscillations between both of these components . therefore , the lifetime of coherent information is determined by the short - lived component in contrast to , for example , the lifetime of the atom population determined by the long - lived state . unlike the other types of quantum information , coherent information enables one to distinguish two qualitatively different information exchange classes corresponding to the cases when classical information or quantum state entanglement is used . only in the latter case , coherent information is nonzero . therefore , it is just coherent information which can be adequately used for finding out to what extent a quantum information transmission channel retains the capability of utilizing the output as an input equivalent for realizing situations where the quantum character of an input signal is of essential importance . these problems are widely covered in the modern literature ( see [ 13 ] and references cited therein ) . being a measure of the entanglement of a quantum system that is preserved by a physical transformation , coherent information is now of certain practical interest for quantum information transmission and processing as well as for the analysis of specific physical models of quantum channels illustrated below by an example . proceeding from formal mathematical analogues , we can start describing a two - sided quantum information channel with the formal quantum generalization of the shannon classical mutual information @xmath113 : @xmath114 + s[\hat\rho_b]- s[\hat \rho_{ab}],\ ] ] this generalization makes sense only when joint density matrix @xmath115 is specified . this matrix is considered as a direct analog of classical joint probability distribution @xmath116 [ 38 ] . obviously , formula ( 6 ) can be applied to quantum systems if we assume that the states of systems @xmath90 and @xmath95 are mutually compatible . it is _ a fortiori _ true for one - time states of the corresponding physical systems when they are not overlapped by the parts of one and the same quantum system containing both an input and output .. when systems move at relativistic velocities , the appropriate relativistic corrections are necessary . these situations are topical in modern experiments with entangled quantum states [ 39 ] where effects associated with motion of a measuring system are recorded . ] the foregoing justifies the term _ one - time _ as referred to information that is specified using the joint density matrix as an initial characteristic . though one can formally define quantum information @xmath117 with expression ( 6 ) , its physical sense remains to be seen [ 40 , 41 ] . this can be attributed to the main qualitative difference between the classical and quantum information channels . in general , relationship ( 5 ) implies that the quantum input and output are incompatible as , for example , in the case when the states of one and the same quantum system are considered at two different instants . thus , @xmath90 and @xmath95 in ( 6 ) can not be the input and output that are involved in the definition of coherent information . therefore , in the quantum case , in order to apply density matrix @xmath115 , one has to physically specify systems @xmath90 and @xmath95 . this can be done using the schumacher definition of coherent information , which inevitably necessitates modifying expression ( 6 ) . = 0.45 adequate physical interpretations of systems @xmath90 and @xmath95 are provided by identifying them , respectively , with the reference system and with the output of a certain quantum channel associated with specified joint density matrix @xmath115 as shown in fig . 3 . in this case , they are automatically compatible . therefore , the actual input of the above channel corresponds to a certain state of input @xmath118 at an initial instant rather than to system @xmath90 . at the initial instant , we have information on a which is not distorted by the channel and is transformed by @xmath79 into the final state of output @xmath95 . in this case , @xmath118 and channel @xmath79 manifest themselves in the form of density matrix @xmath115 coupling the output and reference system rather than being explicitly introduced . specified density matrix @xmath115 properly corresponds to its analog , i.e. , density matrix @xmath119 in ( 5 ) ( which is involved in the schumacher method ) , when the structure consisting of the reference system and input is characterized by a pure state described by function @xmath120 such that the relationship @xmath121 holds for a certain channel @xmath79 . then , the density matrix @xmath122 of the reference system can automatically be represented as the corresponding partite density matrix @xmath123 of the system @xmath124 if one takes into account that the trace with respect to @xmath118 in ( 7 ) is @xmath79 invariant because this transformation does not affect subsystem @xmath90 . the foregoing implies that the corresponding one - time coherent information can be defined as the schumacher coherent information @xmath125-s[\hat\rho_{ab}],\ ] ] which , unlike information measure ( 6 ) , does not contain the term @xmath126 $ ] . as follows from the above description , in the general case , the term _ one - time _ does not directly mean that it refers to the coupling between the system states at one and the same instant . actually , reference system @xmath90 and output @xmath95 may be considered at different instants , and the only important condition is their compatibility . thus , one - time information couples two compatible systems all of whose variables are mutually compatible , i.e. , described by commuting operators , in contrast to coherent information , which , in the general case , couples the incompatible input and output . representing coherent information in another form , one - time information ( 8) ( in contrast to information ( 6 ) ) is not symmetrical with respect to @xmath90 and @xmath95 . moreover , coherent information may be negative . the latter circumstance is evident for density matrices @xmath115 corresponding to a purely classical exchange between basis sets of orthogonal states : @xmath127 . then , all entropies are reduced to classical ones : @xmath128{=}s_{ab}{=}-\sum p_{ij}\log p_{ij}$ ] , @xmath88{=}s_b{= } -\sum p_j\log p_j$ ] , and @xmath129 . a negative value of coherent information means that the exchange entropy is so large that it not only makes the preserved information content vanish but also exceeds the corresponding critical value . in the latter case , one can assume that @xmath130 . the information system presented in fig . 2b plays a special role in modern applications based on nonclassical features of quantum information , such as quantum cryptography and quantum computations . systems can be used as basic blocks in these applications . they are promising carriers of elementary quantum information units ( qubits ) that enable one to efficiently store quantum information and freely manipulate it by means of laser radiation [ 10 , 13 ] . for the information system presented in fig . 2b , the use of the second @xmath106-system as a reference system is physically justified because the entanglement of two corresponding qubits has a clear physical meaning as initially stored quantum information . particularly , the latter can be applied for performing basic logical operations in quantum computations involving an output system . the radiation quantum information transmission channel is of interest because , having converted an initial qubit into the photon field , one can exploit various possibilities provided by further high - speed transformations . it is of interest to find out how rapidly information can be reconstructed after the qubit - photon field channel is used once . the reader can find detailed computations of coherent information for this channel in [ 36 ] . figure 4a shows the dependence of coherent information on time and the laser field action angle for a symmetrical @xmath106-system . these results are obtained for the input qubit in the form of the state with the maximum entropy @xmath131 . this qubit state corresponds to the coherent information content that is independent of individual intensities of two resonant laser fields affecting the @xmath106-system . = 0.3 = 0.3 one can immediately see from fig . 4a that there exists the optimum information exchange level @xmath132 ( where @xmath133 ) when the information channel is used periodically at duration @xmath134 of the exchange cycle , so that the initial state is instantly renewed after each cycle . figure 4b demonstrates exchange rate @xmath91 calculated for a symmetrical @xmath106-system with the rates of radiative decay @xmath135 . the maximum value reached by @xmath91 is @xmath136 . thus , the atom - photon field information exchange limits the rate of using coherent information stored in @xmath106-systems , i.e. , the capacity of the @xmath106-system photon field coherent information channel . the order of magnitude of this rate is determined by the rate of radiative decay of an excited state , whereas its specific value depends on decay rates @xmath137 of both radiative transitions of the @xmath106-system . in the limit of a two - level radiation system , when @xmath138 or @xmath139 , the optimum rate is equal to @xmath140 . for one - time mean values of quantum physical quantities , the internal quantum incompatibility manifests itself just as statistical uncertainty , which can be taken into account using the equivalent classical probability distribution . with the probabilistic measure @xmath141 on the set of all quantum states , the mean value of any operator @xmath18 can be represented as @xmath142 , where @xmath143 . here , @xmath144 is a small volume element in the space of physically different states of a @xmath29-dimensional hilbert space @xmath112 ( @xmath145 ) , which is represented by the bloch sphere in the case of qubit , i.e , when @xmath35 ( see section i ) . relationship @xmath146 is the mean of the operator measure ( 10 ) which is a special case of a nonorthogonal decomposition of unit [ 42 ] , or a positive operator - valued measure ( povm ) [ 2 , 14 ] . generalized quantum measurement procedures are described by povms . unlike a direct measurement in the original system represented by the orthogonal decomposition of the unit , i.e. , by the ortho - projective measure in @xmath112 , a generalized quantum measurement is performed in the compound space @xmath147 . ha with the appropriate complementary state space @xmath148 and joint density matrix @xmath149 , which contains no other information on @xmath90 in addition to that contained in density matrix @xmath86 . system @xmath90 is characterized by uncertainty having the form of quantum incompatibility of the set comprising all of its quantum states . generalized quantum measurement ( 10 ) transforms this uncertainty into classical statistical uncertainty of the quantitatively equivalent set of compatible events in the system @xmath150 . with this representation , the coherent relationships typical of the original quantum system are transformed into the corresponding classical correlations , which have no quantum specificity . therefore , this measurement yields a result that is not equivalent to the original system and can not provide for further quantum transformations . this circumstance is the inevitable pay for information represented in the classical form allowing its free use . nevertheless , initial quantum correlations are taken into account in the statistic of resultant classical states . let us assume that two hilbert spaces @xmath112 and @xmath151 of corresponding quantum systems @xmath90 and @xmath95 are given and joint density matrix @xmath115 is specified in @xmath152 . in particular , @xmath90 and @xmath95 may correspond to the subsystems of the compound system @xmath153 specified at one and the same instant @xmath154 and can be considered as the input and output of an abstract quantum channel in an actual physical system . the determining property of systems @xmath90 and @xmath95 is their compatibility . hence , the joint measurement represented by two povms as @xmath155 introduces no new correlations between the input and output and can be interpreted as an indicator of information input - output relationships . the corresponding joint probability distribution is @xmath156\hat\rho_{ab}.\ ] ] then , the shannon information @xmath157 + s[p(d\beta)]- s[p(d\alpha , d\beta)]$ ] determines the compatible information content [ 32 , 43 ] . the physical sense of compatible information depends on a specific measurement procedure and characterizes the output quantum information . this information can be obtained via two povms , which ( in the form of classical carriers @xmath7 and @xmath8 ) select information on the quantum state of the input , which is transmitted to the output . as in the case of one - time information , the reference system of the schumacher scheme can serve as input @xmath90 ( see figs . 2a and 3b ) . then , input - output joint density matrix @xmath115 can be expressed in terms of the input partial density matrix and channel superoperator by formula ( 7 ) . let us consider the special case when @xmath7 and @xmath8 index all quantum states in @xmath112 and @xmath151 according to specific povms in forms ( 10 ) . in this case , compatible information is distributed over all quantum states and associated with the total internal quantum uncertainty of input states , which is automatically taken into account in probability distribution ( 9 ) . in particular , information that is contained in quantum correlations observed in the presence of quantum entanglement between @xmath90 and @xmath95 is also involved in joint probability distribution ( 11 ) . in addition , here , compatible information is characterized by operational invariance [ 44 ] , i.e. , all noncom - muting physical variables are equally taken into account in this information measure . the above representation of quantum information in terms of classical probability distributions can be interpreted as a modified quantum - mechanical representation in terms of classical physical variables applied in laser physics and discussed by r.j . glauber in his lectures [ 45 , 46 ] . it is natural to define information that corresponds to a generalized measurement specified in form ( 10 ) and comprises all quantum states of a system as nonselected since all quantum states are equally presented in it and quantum variables are not selected . the opposite situation occurs in the case of extremely selected information when orthogonal povms are used , which is typical of elementary cryptographic information exchange schemes [ 2 ] . for example , let us analyze nonselected information depending on the type of joint density matrix and its main parameter , which specifies the entanglement degree , for pure and mixed states . = 0.3 = 0.3 \(i ) a pure state is specified by the wave function @xmath158 with entanglement parameter @xmath159 . for the limit values @xmath160 and 1 , it yields a tensor product and completely entangled state , respectively . \(ii ) a mixed state is specified by the density matrix @xmath161 where @xmath162 is determined in ( 12 ) . for the limit values @xmath160 and 1 , we obtain , respectively , a mixed state with purely classical correlations and a pure complete entangled state . computations of nonselected information are illustrated by fig . the maximum value @xmath163 is attained for a completely entangled state and coincides with the available information content [ 47 ] calculated in [ 27 ] . in this context , the term _ availability _ is considered to mean the possibility of associating with the set of all possible quantum states of information distinguishable against the quantum uncertainty background . selected information corresponds to generalized measurements with povms @xmath164 and @xmath165 where not all of the quantum states are equally included . the results presented below are calculated for selected information in a two - qubit system obtained using measurements @xmath164 and @xmath165 which combine two different types : nonselected measurements @xmath166 and @xmath167 and orthoprojective measurements @xmath168 and @xmath169 corresponding to direct measurements of orthogonal quantum states . then , @xmath170 here , @xmath171 ; @xmath172 ; and @xmath84 describes rotation of the second qubit wave function . this rotation is specified by the transformation @xmath173 depending on rotation angle @xmath174 in basis @xmath51 which is proper for povm of the first qubit . discrete results of the output measurements @xmath175 complete continual results @xmath7 and @xmath8 , which corresponds to new variables with an extended spectrum of values : @xmath176 and @xmath177 . in other words , we measure variables with a combined value spectrum containing a discrete component and a continuous component . these variables comprise the continuum of all wave functions and a singled - out orthogonal 2@xmath29 basis . the limit cases @xmath178 and 1 correspond to nonselected and complete orthoprojective measurements , respectively . the density matrix has form ( 13 ) . the joint probability distribution @xmath179\ ] ] is represented by the components @xmath180 here , the terms @xmath181 and @xmath182 correspond to the information exchange between discrete and continual measured data for the first and second qubits . the above relationships imply the following normalization condition : @xmath183 with ( 13 ) and ( 14 ) , we have two parameters : the degree of selectivity @xmath184 of a combined measurement under consideration , the relative orientation of orthoprojective measurements @xmath185 with value extremes corresponding to the parallel and crossed orientations of the orthogonal bases of the first and second qubits , and the entanglement parameter @xmath186 ( see figs . 6 and 7 ) . the plots shown in these figures provide for the following results . the most unfavorable orientation @xmath187 reduces the selected information content down to zero at @xmath188 if the selective measurement guarantees a nonzero contribution , i.e. , if @xmath189 . at @xmath189 , the information content slightly depends on entanglement parameter @xmath159 . the information maximum @xmath190 bit is reached only for the degree of selectivity @xmath191 , i.e. , in the case of a direct measurement . = 0.35 = 0.35 = 0.35 = 0.35 note a simple correspondence between nonselected and extremely selected information , which immediately follows from the physical content of the corresponding quantum measurement types and holds not only for the considered 2d example but in the general case as well . suppose the completely selected ( in our example , @xmath191 ) measurement is performed at random . this means that we have no _ a priori _ information about the density matrix structure depending on the input and output information encoding , i.e. , on singled - out orientation directions of the corresponding spins . then , the information thus obtained , evidently , must be averaged over all possible orientations . the result of this averaging is exactly equal to the nonselected information con - tent , which is due to the relationship representing the qualitative content of input - output joint probability distribution ( 11 ) for the above two measuring procedures . in the case under consideration , when a completely selective measurement is considered , variables @xmath7 and @xmath8 in ( 11 ) correspond to discrete indices of basis states @xmath49 and @xmath192 . with arbitrary basis wave functions @xmath51 and @xmath193 for input @xmath90 and output @xmath95 , the corresponding povms are @xmath194 and @xmath195 , where quantities @xmath196 describe the rotation from the initial to measurement basis . hence , the input - output joint probability distribution is @xmath197 where @xmath7 and @xmath8 are the parameters of distribution @xmath198 that determine its dependence on orientations of measuring procedures . taking into account povm ( 10 ) and representing the expression for a nonselective measurement as a sum over projections with indices @xmath49 and @xmath192 of the wave functions @xmath199 and @xmath200 , we again obtain dependences ( 15 ) containing @xmath7 and @xmath8 as information ( in this case ) variables . therefore , distribution ( 15 ) over indices @xmath49 and @xmath192 simultaneously specifies orientation - angle probability distribution involved in the nonselective measurement procedure . hence , the integrals describing the mean selected information content exchanged via variables @xmath49 and @xmath192 and the integrals describing the nonselected information exchanged via continuous variables @xmath7 and @xmath8 are identical . thus , a nonselective measurement is equivalent to the set of completely selective measurements performed simultaneously for all possible orientation angles of the measurement basis . the corresponding compatible information automatically takes into account the uncertainty of the basis orientation at a completely selective measurement . the above analysis based on generalized quantum measurements stimulates further generalizations that promise a realistic concept of information content available with a given scheme of a physical experiment , which can certainly be considered as one of the most important purposes of the quantum information theory . the main difficulty is to mathematically describe an information model of a specific experimental scheme in a sufficiently general form . to this end , first , it is necessary to mathematically define the notions of input and output , which is , actually , the most complicated task . figure 8 demonstrates a block diagram illustrating the proposed solution . being affected by control interactions , the state of an object and its noisy environment changes . these interactions generate input quantum information associated either with the dynamic parameters of an object or with the set of certain quantum states that are of interest . output information is measured at the output of the channel described by superoperator transformation @xmath201 . superoperator measures @xmath202 and @xmath203 denote transformations realized by control interactions , and @xmath165 describes the generalized quantum measurement procedure in the form complying with the povm . = 0.3 this block diagram corresponds to the typical mathematical structure of the density matrix characterizing a complex system involving two transformations ( @xmath204 and @xmath203 ) which describe the control and measurement interactions , respectively : @xmath205 here , @xmath206 and @xmath207 are the initial and final density matrices of the set of degrees of freedom essential within the framework of the mathematical model corresponding to a chosen experimental scheme . superoperators @xmath204 , @xmath79 and @xmath208 describe physical information extraction , transmission to the input , and measurement , respectively . this markovian structure of transformations is not the most general . for simplicity , we assume that noisy environments corresponding to each transformation are independent , so that the density matrices can be obtained from @xmath209 and taken into account in the structure of superoperator transformations . only due to this simplification do we have the above combination of three superoperators and the input density matrix and , as a consequence , a relatively simple mathematical representation of the information structure in terms of the corresponding decompositions of superoperators @xmath210 and @xmath208 . however , in certain cases , it may be necessary to generalize relationship ( 16 ) . extraction of information always involves using physical interactions described by the corresponding transformations , which are unitary only when they con - tain all of the degrees of freedom employed . in addition , the interaction with the noisy environment must also be included , which results in non - unitary transformations . here , we discuss these transformations for two cases of a possible choice of desired physical information on a quantum system : ( i ) dynamic parameters @xmath9 and ( ii ) quantum states @xmath211 . in the first case , physical information is extracted by means of dynamic excitation of a system which is mathematically described by unitary operator @xmath212 . this operator may depend on control parameters @xmath213 . probabilistic measure @xmath214 properly specified must take into account _ a priori _ information on @xmath9 . then , superoperator @xmath204 can be represented as @xmath215 with @xmath216 where the symbol must be replaced by the transformed density matrix and broken brackets denote averaging over the noisy environment . in the second case , physical information can eventually be extracted in a storable form allowing for copying as a result of a certain generalized measurement corresponding to the set of positive superoperators @xmath217 in this case , the sum @xmath218 is the superoperator of a generalized measurement represented using the averaged standard decomposition @xmath219 , which preserves the trace of a completely positive superoperator [ 48 ] with adequately specified operators @xmath220 . since a may also describe continuous variables , one should use the generalized representation @xmath221 in the form of an integral with measure @xmath214 guaranteeing , with allowance for idempotency ( @xmath222 ) of orthoprojectors @xmath223 , that this decomposition corresponds to a certain povm : @xmath224 . in the most general form , the sets of superoperators ( 17 ) and ( 18 ) are represented using a certain positive superoperator measure ( psm ) : @xmath225 . this measure is a decomposition of a completely positive superoperator preserving its trace . the psm satisfies the complete positivity ( @xmath226 ) and normalization ( @xmath227)conditions . the latter can be represented in the equivalent form of the unit operator @xmath228 preserved under the effect of hermitean - conjugated psm @xmath229 . again , it is of interest to consider the special psm described by expression ( 10 ) with states specified in hilbert spaces @xmath112 and @xmath151 which correspond to transformations @xmath204 and @xmath230 . this psm associates the information content experimentally extracted directly from quantum states , which yields the most explicit description of fundamental constraints due to the quantum nature of information . in this situation , the output information is represented in a stable classical form , which potentially enables numerous users to employ it simultaneously . this characteristic feature of classical information may initially be involved on default in the meaning of the term _ information _ , at least , when concerned with experimental physical information unlike the physical con- tent of coherent information discussed in section ii . applying the above approach to the superoperator of a measuring system @xmath231 @xmath232 @xmath233 , where @xmath234 has form ( 18 ) , we can represent the input and output information as classical variables @xmath9 and @xmath235 describing the desired information in both cases ( i ) and ( ii ) . the joint probability distribution corresponding to these variables is @xmath236 evidently , this distribution is always positive and normalized to unity . it statistically associates the desired variables and output information that is extracted using an experimental setup . the information efficiency of the latter can be expressed in a quantitative form as the shannon classical information corresponding to the above probability distribution . this quantity can be used as an optimality criterion for optimizing the experiment by available control parameters . note that , for choice ( i ) , states @xmath211 and @xmath237 are not supposed to be mutually compatible and , in the general case , they may correspond to noncommuting variables . in the trivial limit case , they may coincide or differ by a unitary transformation , i.e. , all quantum information is sent with the error probability equal to zero . in this case , the internal quantum uncertainty of this system does not enable one to use distribution ( 19 ) , which establishes the one - to - one correspondence between @xmath9 and @xmath235 . if the states belong to physically different sub - systems , they , nevertheless , may contain quantum correlations due to the corresponding superoperator transformation of channel @xmath201 . the simplest example is the superoperator @xmath238 , which describes a unitary transformation entangling the input and output states . control parameters c may be fixed or chosen from a certain set @xmath239 of necessary values . in the latter case , information can be optimized according to the above criterion . the presence of unknown _ a priori _ distribution @xmath214 for dynamic parameters a in this information structure is not caused by quantum specific features of the problem , i.e. , the problem of a priori uncertainty has to be treated by the methods employed in the classical theory of optimal statistical decisions [ 49 ] . in the most general case , transformations @xmath234 ( see ( 18 ) ) of a measuring system can be described with an arbitrary psm . consequently , psms @xmath240 and @xmath241 cover a very wide range of possible types of object quantum state control when the above quantum measurement procedure is implemented in the experimental procedure under study . in this paper , the most general classification of quantum information is proposed . it is based on compatibility/ incompatibility of the input and output states of a quantum channel . according to this classification , all possible types of information are categorized as classical , semiclassical , coherent , and compatible types of information . the physical content of coherent information is the amount of information contained in internally incompatible states that is exchanged between two systems and quantified as the entanglement preserved between the output and the reference system . the entanglement with the latter is used to measure information at the input and output of a quantum channel specified by a superoperator transformation . here , we introduce the concept of one - time coherent information with the information channel represented by the corresponding joint density matrix . with this concept , two approaches to determining quantum information are adequately associated with each other . according to the first approach , the channel is specified by a superoperator transformation of the input density matrix , and according to the second , by the joint input output density matrix . the coherent information exchange rate calculated for the channel between a @xmath106-system and the field of free photons yields the upper bound equal to @xmath242 for a symmetric @xmath106-system and @xmath140 in the absence of this constraint . the necessity of introducing compatible information as an adequate characteristics of quantum information exchange between two compatible systems of quantum states is justified . compatible information is expressed in terms of the classical information theory despite the presence of internal quantum incompatibility of states in contrast to coherent information , which principally can not be reduced to classical representations . nevertheless , the determination of coherent information is genetically related to compatible information because it is based on distinguishing a pair of compatible systems similar to the input and output systems involved in the analysis of compatible information . one of them is the reference system , and the second is the input or output . thus , the presence of mutually compatible sets of quantum states is necessary for quantifying information of each of the above types . however , when interacting systems are considered , different types of information exhibit qualitatively different types of behavior . the reason for this is that , unlike coherent information , the presence of compatible information may be due to both purely quantum and classical input - output correlations . particularly , in the dicke problem , this circumstance results in possible existence of compatible information ( in contrast to coherent information ) after the short - lived collective dicke state decays . selection of quantum states is shown to be principal for obtaining a useful compatible information content . it is found that nonselected information is equivalent to completely selected information which averaged over all possible orientations the orthogonal bases of input and output complete quantum measurements , realizing complete selection of quantum states . it is shown that mutual and internal compatibility , i.e. , the property of input and output quantum information being classical , is a natural limitation of the physical content of the information flow in an experimental setup . this circumstance enables one to introduce a sufficiently general unified mathematical structure corresponding to a chosen scheme of an actual physical experiment and to quantify its information efficiency . in this situation , the information exchange between subsystems preparing quantum information and a measuring device is described by the probabilistic correspondence between classical variables determining the physical parameters of a quantum system under study and measured output variables . quantum information generation and readout are represented in the general mathematical form with two psms . this mathematical representation of quantum information exchange realized experimentally looks promising for applying the quantum information theory to physical experiments . the approaches proposed in this paper additionally justify the general statement on the physical concept of quantum information mentioned in the paper title``quantum information is physical too '' this work was supported in part by the russian foundation for basic research ( project no . 01 - 02 - 16311 ) , the _ fundamentalnaya metrologiya _ and _ nanotkhnologiya _ state science and technology programs of the russian federation , and intas ( grant no . 00 - 479 ) . 99 gisin , n. , ribordy , g. , tittel , w. , and zbinden , h. , lanl e - print quant - ph/0101098 . preskill , j. , lecture notes on physics 229 : quantum information and computation , http://www.the-ory . caltech.edu / people / preskill / ph229/. steane , a. , rep . phys . , 1998 , 2 , p. 117 . valiev , k.a . and kokin , a.a . , kvantovye kompyutery : nadezhdy i realnost ( quantum computers : hopes and reality ) , izhevsk : nits regulyarn . khaot . din . , 2001 . whitaker , m.a.b . quantum electron . , 2000 , vol . kadomtsev , b.b . , nauk , 1994 , vol . 164 , no . 5 , p. 449 . klyshko , d.n . , nauk , 1998 , vol . 168 , no . 9 , p. 975 . kilin , s.ya . , nauk , 1999 , vol . 169 , no . 5 , p. 507 . menskii , m.b . , nauk , 2000 , vol . 170 , no . 6 , p. 631 . bargatin , i.v . , grishanin , b.a . , and zadkov , v.n . , usp . nauk , 2001 , vol . 171 , no . 6 , p. 625 . belokurov , v.v . , timofeevskaya , o.d . , and khrustalev , o.a . , kvantovaya teleportatsiya obyknovennoe chudo ( quantum teleportation as a common miracle ) , izhevsk : nits regulyarn . khaot . din . kadomtsev , b.b . , dinamika i informatsiya ( dynamics and information ) , moscow : redak . nauk , 2000 . the physics of quantum information : quantum cryp - tography , quantum teleportation , quantum computa - tion , bouwmeester , d. , ekert , a. , and zeilinger , a. , eds . , new york : springer , 2000 . menskii , m.b . , kvantovye izmereniya i dekogerentsiya ( quantum measurements and decoherence ) , moscow : fizmatgiz , 2001 . korn , g. and korn , t. , mathematical handbook for sci - entists and engineers , new york : mcgraw - hill , 1968 , 2nd ed . translated under the title spravochnik po matematike dlya nauchnykh rabotnikov i inzhenerov , moscow : nauka , 1970 . sudbery , a. , quantum mechanics and the particles of nature . an outline for mathematicians , cambridge : cambridge univ . press , 1986 . translated under the title kvantovaya mekhanika i fizika elementarnykh chastits , moscow : mir , 1989 . cabello , a. , lanl e - print quant - ph/0012089 . landau , l.d . and lifshits , e.m . , kvantovaya mekha - nika : nerelyativistskaya teoriya , moscow : nauka , 1974 , 3rd ed . translated under the title course of theoretical physics , vol . 3 : quantum mechanics : non - relativistic theory , new york : pergamon , 1977 , 3rd ed . blokhintsev , d.i . , osnovy kvantovoi mekhaniki ( funda - mentals of quantum mechanics ) , moscow : nauka , 1983 . davydov , a.s . , kvantovaya mekhanika ( quantum mechanics ) , moscow : nauka , 1973 . shannon , c.e . , bell syst . j. , 1948 , vol . 27 , p. 379 ; bell syst . j. , 1948 , vol . 27 , p. 623 . gallager , r.g . , information theory and reliable com - munication , new york : wiley , 1968 . translated under the title teoriya informatsii i nadezhnaya svyaz , mos - cow : sovetskoe radio , 1974 . brukner , . and zeilinger , a. , lanl e - print quant - ph/ 0006087 . hall , m.j.w . , lanl e - print quant - ph/0007116 . brukner , . and zeilinger , a. , lanl e - print quant - ph/ 0008091 . jones , k.r.w . rev . a , 1994 , vol . 50 , no . 5 , p. 3682 . caves , c.m . and fuchs , c.m . , lanl e - print quant - ph/ 9601025 . kholevo , a.s . peredachi inf . , 1973 , vol . 2 , p. 31 . hall , m.j.w . rev . a , 1997 , vol . 1 , p. 100 . schumacher , b. and nielsen , m.a . a , 1996 , vol . 4 , p. 2629 . lloyd , s. , phys . rev . a , 1997 , vol . 3 , p. 1613 . grishanin , b.a . peredachi inf . , 2002 , vol . 1 , p. 31 . barnum , h. , schumacher , b.w . , and nielsen , m.a . rev . a , 1998 , vol . 57 , no . 6 , p. 4153 . grishanin , b.a . and zadkov , v.n . , zh . eksp . fiz . , 2000 , vol . 118 , no . 5 , p. 1048 . grishanin , b.a . and zadkov , v.n . , phys . rev . a , 2000 , vol . 3 , p. 032303 . grishanin , b.a . and zadkov , v.n . , laser phys . , 2000 , vol . 10 , no . 6 , p. 1280 . dicke , r.h . , 1954 , vol . stratonovich , r.l . , 1965 , vol . 8 , no . 1 , p. 116 . zbinden , h. , brendel , j. , tittel , w. , and gisin , n. , j. phys . a , 2001 , vol . 35 , p. 7103 . lindblad , g. , quantum aspects of optical communica - tion : springer - verlag lecture notes in physics , benja - ballah , c. , hirota , o. , and reynaud , s. , eds . , heidelberg : springer , 1991 , vol . 378 , p. 71 . holevo , a.s . , lanl e - print quant - ph/9809022 . grishanin , b.a . nauk sssr , tekh . kibern . , 1973 , vol . 11 , no . 5 , p. 127 . grishanin , b.a . and zadkov , v.n . , laser phys . , 2001 , vol . 11 , no . 12 , p. 1088 . brukner , . and zeilinger , a. , phys . lett . , 1999 , 17 , p. 3354 . glauber , r.j . , quantum optics and electronics , dewitt , c. , blandin , a. , and cohen - tannoudji , c. , eds . , new york : gordon and breach , 1965 , p. 53 . grishanin , b.a . , kvantovye sluchainye protsessy ( quantum stochastic processes ) , http://comsim1.phys.msu . su / index.html . schumacher , b. , complexity , entropy and the physics of information , zurek , w.h . redwood city : addison - wesley , 1990 , p. 29 . kraus , k. , states , effects and operations , berlin : springer , 1983 . wald , a. , in pozitsionnye igry ( position games ) , moscow : sovetskoe radio , 1967 , p. 300 . rudolph , t. , lanl e - print quant - ph/9904037 . schumacher , b.w . rev . a , 1995 , vol . 4 , p. 2738 .

we present a catalog of 105 rich and massive ( ) optically - selected clusters of galaxies extracted from 70 square - degrees of public archival imaging from the blanco 4-m telescope acquired over 45 nights between 2005 and 2007 . we use the clusters optically - derived properties to estimate photometric redshifts , optical luminosities , richness , and masses . we complement the optical measurements with archival xmm - newton and rosat x - ray data which provide additional luminosity and mass constraints on a modest fraction of the cluster sample . two of our clusters show clear evidence for central lensing arcs ; one of these has a spectacular large - diameter , nearly - complete einstein ring surrounding the brightest cluster galaxy . a strong motivation for this study is to identify the massive clusters that are expected to display prominent signals from the sunyaev - zeldovich effect ( sze ) and therefore be detected in the wide - area mm - band surveys being conducted by both the atacama cosmology telescope and the south pole telescope . the optical sample presented here will be useful for verifying new sze cluster candidates from these surveys , for testing the cluster selection function , and for stacking analyzes of the sze data .

1002.2226

a new era of galaxy cluster surveys , based on measuring distortions in the cosmic microwave background ( cmb ) , has begun . these distortions , known as the sunyaev - zeldovich effect @xcite , have been detected for the first time in untargeted surveys over large areas of the sky by two new mm - band experiments : the atacama cosmology telescope ( act ) and the south pole telescope ( spt ) . first results from act @xcite and spt @xcite offer a taste of the future potential these experiments hold for obtaining large samples of essentially mass - selected clusters to arbitrary redshifts and have also provided the first measurement of the microwave background at arc - minute angular scales @xcite . both act and spt aim to provide unique samples of massive clusters of galaxies , selected by mass nearly independently of redshift , over a large area of the southern sky . while the number density of sze - selected clusters can be used as a potentially strong probe of dark energy as well as for studies of cluster physics , it is crucial to understand the systematics of sze surveys by comparing with cluster identification using independent methods before the new cluster samples can be effectively used . for example the low amplitude of the sz component in the high-@xmath2 cmb power spectrum @xcite or stacked clusters in wmap @xcite are recent issues that provide additional motivation for an independent search for clusters over the region being surveyed in the sz . furthermore , although the new mm - band telescopes can be used to identify clusters , coordinated optical data are necessary for confirmation and to determine redshifts and other fundamental properties of the new clusters . the last decade has seen significant effort to produce large and well - selected optical catalogs of cluster of galaxies that can be used in cosmological , large - scale structure and galaxy evolution studies . the first systematic attempts to generate large samples of clusters , and to define their richness , came from the abell catalogs @xcite which searched for projected galaxy overdensities through visual inspection of photographic plates successfully identifying thousands of clusters . although optical catalogs can be relatively inexpensive and efficient at detecting low mass systems , early attempts were known to suffer from significant projection effects along the line of sight . the advent of ccd cameras and the digitization of large photographic plates has enabled the development of new search algorithms for galaxy clusters using a combination of space , brightness and color information ( i.e. photometric redshifts ) , minimizing projection issues ( see * ? ? ? * for a historical review of search methods ) . among these algorithms are the pioneering implementation of a spatial matched filter technique @xcite and its variants @xcite , the adaptive kernel technique @xcite , voronoi tessellation @xcite and methods exploiting the tight ridgeline in color - magnitude space of galaxies in clusters @xcite such as the red cluster sequence ( rcs ) @xcite and the maxbcg @xcite techniques . this new wave of studies have produced large sets of well - defined optical cluster catalogs covering thousands of square - degrees ( i.e. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) providing reliable richness - mass correlations @xcite and establishing independent cosmological constraints @xcite using optical catalogs . in this article we present new results from the southern cosmology survey ( scs ) , our on - going multi - wavelength survey coordinated with act . our first paper in this series , @xcite , described our scs imaging pipeline and presented a sample of new galaxy clusters from an @xmath3 deg@xmath4 optical imaging survey of the southern sky acquired at the blanco 4-m telescope . here , we complete our cluster analysis using all the 70 deg@xmath4 contiguous imaging available , which represents a comoving volume of 0.076 gpc@xmath5 at @xmath6 . throughout this paper we assume a flat cosmology with @xmath7 km s@xmath8 mpc@xmath8 , @xmath9 and matter density @xmath10 . our cluster analysis is based on all of the public unprocessed imaging data available for download at the time of writing from the observations carried out for the blanco cosmology survey ( bcs ) proposal . this was a noao large survey project ( 05b-0043 , pi : joe mohr ) that was awarded 45 nights between 2005 and 2007 to carry out intermediate depth @xmath11band observations on the cerro tololo interamerican observatory ( ctio ) blanco 4-m telescope using the @xmath12 pixel ( 0.36 deg@xmath4 ) mosaic - ii camera . the survey originally aimed to target two square @xmath13 deg@xmath4 southern sky patches centered near declinations @xmath14 and @xmath15 at right ascensions of 23 hr and 5 hr respectively , that were to be contained in a larger sze common region that both act and spt would survey . after three years of observations the survey acquired 78 and 118 mosaic - ii contiguous pointings at 23hr and 5hr respectively that correspond to 29 deg@xmath4 and 41 deg@xmath4 over each patch reaching a total of 70 deg@xmath4 imaged in all @xmath1 filters . the pointing layout and the sky area covered for each patch is shown in figure [ fig : densemap ] . the survey was awarded 15 extra nights at the end of 2008 to complete the proposed 100 deg@xmath4 ; our analysis does not include this extra year s worth of observations as the data are not publicly available yet . and width @xmath16 over the 23hr ( top panel ) 5hr ( bottom panel ) field . dark regions in the image represent denser areas . in red we show the area covered by each of the 78 and 112 mosaic - ii tiles at 23hr and 5hr respectively that comprise the area studied over three years of observations.,width=336 ] and width @xmath16 over the 23hr ( top panel ) 5hr ( bottom panel ) field . dark regions in the image represent denser areas . in red we show the area covered by each of the 78 and 112 mosaic - ii tiles at 23hr and 5hr respectively that comprise the area studied over three years of observations.,width=345 ] the raw unprocessed images were downloaded from the national virtual observatory portal and were processed following the same procedures as described in our initial analysis @xcite where we provide a full description of the data analysis followed and associated data products . here we briefly outline the steps involved and recent improvements from our original analysis . our pipeline automatically handles the file associations and all of the initial standard ccd imaging tasks for each night s run as well as the secondary calibration steps ( i.e. , skyflats , fringe correction , cosmic - ray rejection , badpixel masks , world coordinate calibration , etc . ) using a custom modified version of iraf / mscred @xcite . when observed , photometric standards from the southern hemisphere standards stars catalog @xcite were processed by the pipeline and a photometric zero - point was estimated in ab magnitudes . hereafter all magnitude quoted are in the ab system . in table [ tab : obs ] we show the observing dates , photometric conditions and observed bands that comprise the dataset we analyzed and that were used for this paper . we note that the information that went into creating this table was extracted from the raw data and header information of the data files . in general each mosaic ii pointing consisted of exposures of @xmath17 , @xmath18 , @xmath19 and @xmath20 in the @xmath21 and @xmath22bands respectively with offsets of @xmath23 arcmin ( within each filter ) . we adjusted the photometric zero points of non - photometric nights using the overlapping regions between tiles and their matched sources from adjacent photometric tiles providing a homogeneous photometric calibration across each region with typical variations below @xmath24 mags . the 2007 season data at 23hr was particularly hard to match due to poor observing conditions and large variations within the individual exposures in each tile . we tackled this problem by first scaling all individual exposures comprising a tile to a common median scale ; this step ensured that opposed edges of a combined frame could be effectively used to photometrically tie down adjacent tiles . this is a new feature of the pipeline . all science images were mosaicked , aligned and median combined using swarp @xcite to a plate scale of @xmath25/pixel . source detection and photometry measurements for the science catalogs were performed using sextractor @xcite . large variations in the galactic dust absorption on the 5hr field led us to implement dust correction for every source in each observed band in the science catalogs utilizing the infrared maps and c - routines provided by @xcite . this is particularly important to obtain unbiased colors for photometric redshifts . in figure [ fig : dustmap ] we show our custom generated @xmath26 galactic dust absorption maps for the 5hr and 23hr regions , where darker regions represent areas of higher obscuration . from this figure we can see that in the 23hr region foreground galactic absorption is negligible , while at 5hr it can be as high as @xmath27 magnitudes . finally , we determine photometric redshifts from the four - band optical images and the galactic extinction corrected magnitudes using bpz @xcite following the same procedure as discussed in @xcite . cccccccccc 18 nov 2005 & yes & & & 1.7 & 1.3 & 1.0 & 1.0 & 3.0 & 3.0 + 19 nov 2005 & yes & & & 4.7 & 5.0 & 4.5 & 4.5 & 1.0 & 1.0 + 20 nov 2005 & no & & & 5.7 & 5.7 & 8.5 & 8.0 & & + 22 nov 2005 & no & & & & & & & & + 24 nov 2005 & yes & 9.0 & 9.0 & 1.7 & 1.7 & 5.5 & 5.5 & & + 26 nov 2005 & yes & 9.5 & 9.0 & 1.0 & 1.0 & 4.5 & 4.5 & 1.0 & 1.0 + 28 nov 2005 & yes & 4.0 & 3.0 & 4.0 & 4.0 & & & 2.7 & 2.7 + 30 nov 2005 & yes & 12.5 & 12.5 & & & 5.0 & 5.0 & 0.3 & 0.3 + 02 dec 2005 & yes & & & & & 3.5 & 3.5 & & + 04 dec 2005 & no & 9.0 & 9.0 & & & 2.0 & 2.0 & & + 05 dec 2005 & no & 2.5 & 2.0 & 2.7 & 2.7 & 2.0 & 2.0 & 1.3 & 1.7 + 06 dec 2005 & yes & & & 4.0 & 4.0 & & & 1.3 & 1.0 + 08 dec 2005 & yes & & & 4.0 & 4.0 & & & 2.3 & 2.7 + 10 dec 2005 & yes & & & 5.7 & 5.7 & & & 3.0 & 3.0 + 11 dec 2005 & yes & & & 4.3 & 4.3 & & & 3.0 & 3.3 + 23 oct 2006 & yes & 9.0 & 10.5 & & & & & 5.0 & 5.0 + 25 oct 2006 & no & & & & & & & & + 27 oct 2006 & yes & 10.5 & 7.0 & & & & & 5.0 & 5.3 + 28 oct 2006 & no & & & & & & & 5.3 & 5.0 + 29 oct 2006 & yes & 8.0 & 8.0 & 2.0 & 2.0 & & & 4.0 & 4.3 + 30 oct 2006 & yes & & & & & & & 5.0 & 5.3 + 31 oct 2006 & yes & & & 6.0 & 6.0 & & & 3.0 & 2.7 + 12 dec 2006 & yes & & & 3.0 & 1.7 & & & & + 13 dec 2006 & yes & & & 3.0 & 3.3 & 1.0 & 0.5 & & + 14 dec 2006 & no & & & 2.7 & 2.7 & & & & + 15 dec 2006 & yes & 4.0 & 4.0 & 5.7 & 5.7 & & & & + 16 dec 2006 & no & & & 4.0 & 3.7 & & & & + 17 dec 2006 & yes & 7.0 & 6.5 & 4.3 & 5.0 & & & & + 18 dec 2006 & yes & 8.0 & 8.5 & & & & & & + 19 dec 2006 & yes & 3.0 & 3.0 & 6.3 & 6.3 & & & & + 20 dec 2006 & yes & & & 4.0 & 4.0 & & & & + 21 dec 2006 & yes & & & 8.7 & 8.7 & & & & + 22 dec 2006 & yes & 3.5 & 3.5 & 2.3 & 2.0 & & & & + 23 dec 2006 & yes & 6.5 & 6.0 & 6.0 & 6.3 & & & & + 11 sep 2007 & yes & 2.5 & 3.0 & 2.3 & 2.7 & 4.5 & 5.0 & 3.3 & 4.0 + 12 sep 2007 & yes & & & & & 1.0 & 1.0 & 1.0 & 1.0 + 13 sep 2007 & yes & & & & & 1.5 & 1.0 & 1.3 & 1.3 + 14 sep 2007 & yes & 1.5 & 1.5 & 1.7 & 1.3 & 4.5 & 5.0 & 4.7 & 4.7 + 15 sep 2007 & no & 2.0 & 2.0 & 0.7 & 1.3 & 2.5 & 2.5 & 1.0 & 1.0 + 16 sep 2007 & yes & & & & & 2.5 & 2.5 & & + 17 sep 2007 & yes & 1.0 & & & & & & 2.0 & 1.7 + 18 sep 2007 & yes & & & 2.3 & 2.0 & 3.0 & 2.5 & 7.0 & 6.7 + 19 sep 2007 & yes & & & & & & & 4.7 & 5.0 + 11 oct 2007 & yes & & & & & & & & + 12 oct 2007 & no & & & & & & & & + 30 oct 2007 & no & & & 1.0 & & & 0.5 & & + 31 oct 2007 & no & & & & & & & & + 01 nov 2007 & yes & & & 1.3 & 2.3 & 2.5 & 2.0 & & + 02 nov 2007 & yes & & & & & 5.0 & 5.0 & & + 03 nov 2007 & yes & & & 3.0 & 3.0 & 9.5 & 9.5 & & + 04 nov 2007 & yes & 5.0 & 5.5 & & & 3.0 & 3.0 & & + 06 nov 2007 & yes & & & & & & & 4.7 & 4.3 + 07 nov 2007 & yes & 1.0 & 2.0 & 0.7 & 0.7 & 1.0 & 1.0 & 1.7 & 2.0 + 08 nov 2007 & yes & & & & & & & & + [ tab : obs ] in this section we describe our efforts to select clusters of galaxies from multi - wavelength optical imaging . to this objective we follow an identical procedure as the one we laid out in @xcite as part of our initial study of @xmath3 deg@xmath4 using the 2005 observations at 23hr . we refer the reader there for an expanded and formal description of the method as well as for simulations that address the contamination and completeness of the search method . here , we expanded our procedure to the full 70 deg@xmath4 of blanco @xmath1 contiguous imaging . our cluster search algorithm uses a matched filter approach based on the one described in @xcite to identify significant overdensities . we define membership and estimate richness for each peak in the overdensity maps using the maxbcg prescription @xcite . our method folds in the contributions from : a ) a cluster spatial profile filter function , b ) a luminosity weight and c ) the bpz redshift probability distribution from each source to generate likelihood density maps or a `` filtered '' galaxy catalog over the area covered by the survey as a function of redshift . we generate likelihood density maps with a constant pixel scale of @xmath28 arcmin at @xmath16 intervals between @xmath29 over the surveyed regions . in figure [ fig : densemap ] we show an example of a likelihood density map centered at @xmath30 on which we superpose outlines of the 78 and 112 tiles that define the 23hr and 5hr regions studied . because the mass of a cluster is not a direct observable , some observable proxy for mass needs to be used in order to obtain estimates for a given cluster sample . such proxies include the x - ray flux and temperature @xcite , weak lensing shear @xcite and optical galaxy richness of clusters @xcite . for this analysis we use the latest mass tracers for clusters of galaxies that are based on optically observed parameters @xcite extracted from a sample of around @xmath31 clusters from the sloan digital sky survey ( sdss ) maxbcg catalog @xcite . to use these mass tracers we define membership in similar fashion as @xcite . we then visually inspected each candidate - cluster peak in the density maps and selected the brightest elliptical galaxy in the cluster ( bcg ) , which was taken to be the center and initial redshift of the system . we then use galaxies photometrically classified as e or e / s0s according to their bpz spectral type and within a projected radius of @xmath32 mpc and redshift interval @xmath33 to obtain a local color - magnitude relation for each color combination as well as the cluster mean redshift , @xmath34 , for all cluster members , using a @xmath35 median sigma - clipping algorithm . we use these to determine @xmath36 , the number of galaxies within @xmath37mpc of the cluster center . for our richness measurements we estimated the galaxy background contamination and implemented an appropriate background subtraction method following the same procedure described in @xcite ( see section 3.1 ) . we use a statistical removal of unrelated field galaxies with similar colors and redshifts that were projected along the line of sight to each cluster . we estimate the surface number density of ellipticals in an annulus surrounding the cluster ( within @xmath38 ) with @xmath39 and the same colors as the cluster members . we measure this background contribution around the outskirts of each cluster and obtain a corrected value @xmath40 which is used to compute @xmath41 and then corresponding values of 200 and @xmath42 . the magnitude of the correction ranges between @xmath43% . we will refer to the corrected values hereafter . finally we show , as examples of the depth and data quality , composite @xmath44 color images of several clusters in the 5hr and 23hr regions in figures [ fig : clusters05hr ] and [ fig : clusters23hr ] , respectively , that cover a wide range of redshifts . color image for nine newly discovered massive scs clusters in the 23hr field . we indicate the cluster s redshift using the median value from galaxy members within @xmath45 kpc of the cluster s center.,title="fig:",width=288 ] + color image for nine newly discovered massive scs clusters in the 23hr field . we indicate the cluster s redshift using the median value from galaxy members within @xmath45 kpc of the cluster s center.,title="fig:",width=288 ] + color image for nine newly discovered massive scs clusters in the 23hr field . we indicate the cluster s redshift using the median value from galaxy members within @xmath45 kpc of the cluster s center.,title="fig:",width=288 ] + the observational quantities required as input to the cluster mass scaling relation are 200 , 200 , and . the cluster richness , 200 , is the number of e and e / s0 galaxies within a given radius , originally defined by @xcite as @xmath46 mpc , with colors and luminosities that satisfy specific conditions for membership . similarly , @xmath42 is the total rest - frame integrated @xmath47-band luminosity , @xmath48-corrected to @xmath49 , of all member galaxies included in 200 , and @xmath50 is the similarly defined rest - frame @xmath47-band luminosity of the bcg . @xcite provide power - law functions for both the luminosity - mass and richness - mass relations ( see section 5.2.1 in their paper ) . both @xcite and @xcite found that the luminosity - mass and richness - mass relations were well described by power - law functions and they measured the normalizations and slopes in these relations using @xmath51 minimization . we computed the two fitting functions based on @xmath42 and 200 , ( see section 5.2.1 from @xcite for full details ) , which are described as : @xmath52 and @xmath53 where @xmath54 is the mass observational equivalent of @xmath55 ( i.e. the halo mass enclosed within a radius of spherical volume within which the mean density is 200 times the average density ) in units of @xmath56 , @xmath42 is in units of @xmath57 and the dependence is normalized by its mean value . this is also described by a power - law function for a given value of 200 and 200 : @xmath58 and @xmath59 the published best - fitting parameters for @xmath60 , @xmath61 and @xmath62 in eqs . ( [ eq : m1 ] ) and ( [ eq : m2 ] ) as well as the new erratum - corrected values of @xmath63 and @xmath64 ( r. reyes , private communication ) for eqs . ( [ eq : l1 ] ) and ( [ eq : l2 ] ) are shown in table [ tab : pars ] . these recent changes in the values of the @xmath63 and @xmath64 parameters in @xcite have implications for our mass estimation . specifically , the changes translate into a decrease in mass when compared to our previous analysis of @xcite which used the initial parameters from their pre - print paper . as we discuss below , this change also affects the recovery of clusters in @xcite for the sky region that overlaps with this study . band observed magnitude for the bcgs in our sample as a function of the redshift of the cluster ( filled circles ) . we also show the @xmath65 bcg - redshift relationship from @xcite as the dashed curve.,width=384 ] we use this parametrization to obtain the optical mass estimates 200 and 200 , based on 200 and 200 , respectively , for all of the visually inspected clusters . based on these estimates we defined our catalog of massive clusters by selecting all systems with either @xmath66 or @xmath67 and @xmath68 . this defines the sample : 61 systems in the 5hr region and 44 in the 23hr region for a total of 105 clusters over @xmath69 deg@xmath4 of multi - band imaging . this mass threshold is aimed to include the upcoming @xmath70 significant sze detections from spt and act . in tables [ tab : clustes23hr ] and [ tab : clustes05hr ] we display the optical properties , photometric redshifts , positions and mass estimates for all clusters at 23hr and 5hr respectively . in these tables we provide for each cluster the photometric redshift of the bcg as well as the mean photometric redshift for the system which was estimated using galaxies within 250 kpc of the bcg . as we discussed in @xcite the expected luminosity range of bcgs in rich clusters has been already observationally established from the sdss @xcite and it can be used as an additional probe to confirm the presence of a massive cluster . to this end in figure [ fig : magz ] we compare the observed @xmath71band magnitudes of all bcgs in our sample as a function of redshift to a parametrization of the observed @xmath71band of sdss bcgs ( dashed curve ) . this corresponds to the @xmath72 prescription from @xcite where we have used @xmath73 from @xcite and allowed it to evolve passively with redshift . we see in the figure that all of the sources lie very close or well below ( i.e. , are intrinsically brighter ) than the model curve as we would expect for the bcg in a rich cluster . in our previous study @xcite we presented a similar cluster analysis and mass estimation over @xmath3 deg@xmath4 area in the 23hr region which is fully contained in the current 70 deg@xmath4 dataset . however , due to the changes in the mass parametrization prescription we did not recover all the massive clusters from @xcite even though the nominal mass threshold was the same ( @xmath0 ) as here . therefore , only 3 out of 8 clusters from table 5 of @xcite are massive enough to appear in the current paper s table [ tab : clustes23hr ] , while the remaining clusters , although detected , now have masses that fall below the @xmath74 mass limit . similarly , when comparing with the weak lensing mass estimates of @xcite we can only match the same 3 sources with clusters in their analysis ( see table 2 of their paper ) . on the other hand , there is now better agreement between the updated @xmath75 values and the weak lensing masses for the matched clusters . out of the 105 massive clusters in the sample we report two systems ( see the two lower panels of figure [ fig : clusters23hr ] ) with obvious evidence of arcs near the bcg . of these two lensing clusters , scso 235138@xmath76545253 at @xmath77 is a potentially unique system in that it shows a large diameter ( @xmath78 ) , nearly complete einstein ring embedded in a massive cluster . although our density of lensing clusters is not widely different from the @xcite rcs sample of 5 lensing clusters over @xmath79 deg@xmath4 , all of the gladders et al . clusters are at @xmath80 , while both of our systems are at @xmath81 making them completely exclusive in redshift . moreover , none of the @xcite sample shows anything close to a complete einstein ring . rrrrrrrrrrr[b ] @xmath82 & 1.54 & 0.61 & 0.41 & 0.67 & @xmath83 & @xmath84 & @xmath85 & @xmath86 & @xmath87 & @xmath88 + @xmath89 & 1.64 & 0.58 & 0.43 & 0.66 & @xmath90 & @xmath91 & @xmath92 & @xmath93 & @xmath94 & @xmath95 + [ tab : pars ] rrrccrrrr scso j231340@xmath76521919 & 0.21 & 0.21 & @xmath96 & @xmath97 & @xmath98 & @xmath99 & @xmath100 + scso j231403@xmath76564710 & 0.60 & 0.60 & @xmath101 & @xmath102 & @xmath103 & @xmath104 & @xmath105 + scso j231455@xmath76555308 & 0.21 & 0.23 & @xmath106 & @xmath107 & @xmath108 & @xmath109 & @xmath110 + scso j231511@xmath76523322 & 0.36 & 0.39 & @xmath111 & @xmath112 & @xmath113 & @xmath114 & @xmath115 + scso j231629@xmath76554535 & 0.51 & 0.53 & @xmath116 & @xmath117 & @xmath118 & @xmath119 & @xmath120 + scso j231651@xmath76545356 & 0.36 & 0.36 & @xmath121 & @xmath122 & @xmath123 & @xmath114 & @xmath124 + scso j231717@xmath76565723 & 0.74 & 0.73 & @xmath125 & @xmath126 & @xmath127 & @xmath128 & @xmath129 + scso j231848@xmath76561711 & 0.51 & 0.51 & @xmath130 & @xmath131 & @xmath132 & @xmath133 & @xmath119 + scso j231930@xmath76563858 & 0.36 & 0.37 & @xmath134 & @xmath135 & @xmath136 & @xmath105 & @xmath100 + scso j232001@xmath76565222 & 0.80 & 0.80 & @xmath137 & @xmath138 & @xmath139 & @xmath140 & @xmath109 + scso j232423@xmath76565705 & 0.75 & 0.75 & @xmath141 & @xmath142 & @xmath143 & @xmath144 & @xmath145 + scso j232437@xmath76530047 & 0.73 & 0.75 & @xmath146 & @xmath147 & @xmath148 & @xmath114 & @xmath110 + scso j232529@xmath76532420 & 0.74 & 0.71 & @xmath149 & @xmath150 & @xmath151 & @xmath152 & @xmath153 + scso j232612@xmath76531858 & 0.15 & 0.13 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath99 + scso j232627@xmath76531512 & 0.74 & 0.74 & @xmath158 & @xmath159 & @xmath160 & @xmath161 & @xmath162 + scso j232637@xmath76533911 & 0.76 & 0.77 & @xmath163 & @xmath164 & @xmath165 & @xmath153 & @xmath166 + scso j232653@xmath76524149 & 0.11 & 0.11 & @xmath167 & @xmath168 & @xmath169 & @xmath99 & @xmath104 + scso j232724@xmath76533553 & 0.74 & 0.74 & @xmath170 & @xmath171 & @xmath172 & @xmath173 & @xmath174 + scso j232800@xmath76535152 & 0.74 & 0.75 & @xmath175 & @xmath147 & @xmath176 & @xmath177 & @xmath178 + scso j232811@xmath76533847 & 0.74 & 0.74 & @xmath179 & @xmath180 & @xmath181 & @xmath133 & @xmath119 + scso j232825@xmath76522814 & 0.73 & 0.70 & @xmath182 & @xmath183 & @xmath184 & @xmath185 & @xmath186 + scso j232827@xmath76531414 & 0.36 & 0.35 & @xmath187 & @xmath188 & @xmath189 & @xmath190 & @xmath177 + scso j232856@xmath76552428 & 0.57 & 0.57 & @xmath191 & @xmath192 & @xmath103 & @xmath193 & @xmath105 + scso j232916@xmath76522910 & 0.73 & 0.74 & @xmath194 & @xmath195 & @xmath196 & @xmath153 & @xmath173 + scso j233006@xmath76524035 & 0.73 & 0.71 & @xmath197 & @xmath198 & @xmath199 & @xmath161 & @xmath200 + scso j233227@xmath76535827 & 0.35 & 0.32 & @xmath201 & @xmath202 & @xmath203 & @xmath177 & @xmath204 + scso j233232@xmath76522016 & 0.36 & 0.37 & @xmath205 & @xmath206 & @xmath207 & @xmath208 & @xmath204 + scso j233329@xmath76521513 & 0.51 & 0.50 & @xmath175 & @xmath209 & @xmath210 & @xmath211 & @xmath109 + scso j233420@xmath76542732 & 0.56 & 0.55 & @xmath212 & @xmath213 & @xmath214 & @xmath124 & @xmath211 + scso j233544@xmath76535115 & 0.51 & 0.51 & @xmath215 & @xmath216 & @xmath217 & @xmath218 & @xmath219 + scso j233556@xmath76560602 & 0.64 & 0.63 & @xmath220 & @xmath221 & @xmath222 & @xmath157 & @xmath110 + scso j233607@xmath76535235 & 0.54 & 0.53 & @xmath223 & @xmath224 & @xmath225 & @xmath226 & @xmath227 + scso j233726@xmath76565655 & 0.50 & 0.52 & @xmath228 & @xmath229 & @xmath230 & @xmath124 & @xmath190 + scso j233807@xmath76531223 & 0.47 & 0.49 & @xmath231 & @xmath232 & @xmath233 & @xmath234 & @xmath235 + scso j233931@xmath76544525 & 0.73 & 0.71 & @xmath236 & @xmath112 & @xmath237 & @xmath124 & @xmath238 + scso j234012@xmath76541907 & 0.59 & 0.62 & @xmath239 & @xmath240 & @xmath214 & @xmath152 & @xmath241 + scso j234138@xmath76545210 & 0.55 & 0.56 & @xmath242 & @xmath243 & @xmath244 & @xmath178 & @xmath105 + scso j234156@xmath76530848 & 0.49 & 0.49 & @xmath245 & @xmath246 & @xmath247 & @xmath248 & @xmath249 + scso j234703@xmath76535051 & 0.56 & 0.55 & @xmath250 & @xmath251 & @xmath252 & @xmath193 & @xmath190 + scso j234917@xmath76545521 & 0.73 & 0.72 & @xmath253 & @xmath254 & @xmath255 & @xmath140 & @xmath161 + scso j235055@xmath76530124 & 0.46 & 0.48 & @xmath256 & @xmath257 & @xmath258 & @xmath234 & @xmath211 + scso j235138@xmath76545253 & 0.33 & 0.31 & @xmath259 & @xmath260 & @xmath160 & @xmath120 & @xmath261 + scso j235233@xmath76564348 & 0.74 & 0.72 & @xmath262 & @xmath263 & @xmath264 & @xmath128 & @xmath99 + scso j235454@xmath76563311 & 0.51 & 0.50 & @xmath265 & @xmath266 & @xmath267 & @xmath268 & @xmath200 + [ tab : clustes23hr ] rrrccrrrr[b ] scso j050854@xmath76513048 & 0.70 & 0.70 & @xmath141 & @xmath269 & @xmath270 & @xmath204 & @xmath268 + scso j050857@xmath76535837 & 0.76 & 0.78 & @xmath271 & @xmath272 & @xmath273 & @xmath200 & @xmath274 + scso j050902@xmath76520704 & 0.58 & 0.58 & @xmath275 & @xmath271 & @xmath276 & @xmath178 & @xmath234 + scso j050926@xmath76522227 & 0.67 & 0.70 & @xmath277 & @xmath278 & @xmath279 & @xmath157 & @xmath208 + scso j051023@xmath76544455 & 0.39 & 0.37 & @xmath280 & @xmath281 & @xmath282 & @xmath105 & @xmath144 + scso j051112@xmath76523112 & 0.73 & 0.73 & @xmath212 & @xmath163 & @xmath283 & @xmath129 & @xmath162 + scso j051136@xmath76561045 & 0.70 & 0.71 & @xmath277 & @xmath182 & @xmath284 & @xmath161 & @xmath285 + scso j051144@xmath76511416 & 0.48 & 0.48 & @xmath286 & @xmath287 & @xmath279 & @xmath100 & @xmath177 + scso j051145@xmath76515430 & 0.70 & 0.70 & @xmath288 & @xmath289 & @xmath290 & @xmath104 & @xmath105 + scso j051207@xmath76514204 & 0.48 & 0.46 & @xmath291 & @xmath292 & @xmath293 & @xmath105 & @xmath114 + scso j051225@xmath76505913 & 0.70 & 0.70 & @xmath294 & @xmath295 & @xmath296 & @xmath297 & @xmath173 + scso j051240@xmath76513941 & 0.66 & 0.66 & @xmath298 & @xmath299 & @xmath300 & @xmath104 & @xmath105 + scso j051245@xmath76502028 & 0.62 & 0.62 & @xmath301 & @xmath302 & @xmath303 & @xmath178 & @xmath304 + scso j051258@xmath76542153 & 0.67 & 0.68 & @xmath305 & @xmath306 & @xmath307 & @xmath193 & @xmath208 + scso j051400@xmath76512635 & 0.72 & 0.73 & @xmath308 & @xmath309 & @xmath310 & @xmath311 & @xmath312 + scso j051412@xmath76514004 & 0.67 & 0.66 & @xmath313 & @xmath314 & @xmath315 & @xmath316 & @xmath317 + scso j051457@xmath76514345 & 0.69 & 0.69 & @xmath197 & @xmath205 & @xmath318 & @xmath238 & @xmath319 + scso j051542@xmath76514017 & 0.73 & 0.73 & @xmath320 & @xmath321 & @xmath322 & @xmath208 & @xmath323 + scso j051558@xmath76543906 & 0.66 & 0.64 & @xmath324 & @xmath135 & @xmath325 & @xmath178 & @xmath105 + scso j051613@xmath76542620 & 0.38 & 0.36 & @xmath326 & @xmath327 & @xmath328 & @xmath329 & @xmath319 + scso j051637@xmath76543001 & 0.23 & 0.25 & @xmath330 & @xmath331 & @xmath332 & @xmath312 & @xmath333 + scso j051755@xmath76555727 & 0.66 & 0.66 & @xmath334 & @xmath335 & @xmath336 & @xmath128 & @xmath133 + scso j051933@xmath76554243 & 0.69 & 0.70 & @xmath141 & @xmath337 & @xmath338 & @xmath100 & @xmath339 + scso j051935@xmath76554916 & 0.75 & 0.75 & @xmath340 & @xmath242 & @xmath341 & @xmath104 & @xmath185 + scso j052051@xmath76561804 & 0.74 & 0.73 & @xmath342 & @xmath343 & @xmath344 & @xmath345 & @xmath219 + scso j052113@xmath76510418 & 0.64 & 0.61 & @xmath346 & @xmath347 & @xmath348 & @xmath349 & @xmath350 + scso j052200@xmath76502700 & 0.50 & 0.47 & @xmath351 & @xmath352 & @xmath353 & @xmath304 & @xmath120 + scso j052533@xmath76551818 & 0.72 & 0.72 & @xmath354 & @xmath355 & @xmath356 & @xmath357 & @xmath234 + scso j052608@xmath76561114 & 0.14 & 0.15 & @xmath358 & @xmath359 & @xmath360 & @xmath105 & @xmath161 + scso j052803@xmath76525945 & 0.68 & 0.69 & @xmath361 & @xmath362 & @xmath363 & @xmath329 & @xmath274 + scso j052810@xmath76514839 & 0.65 & 0.64 & @xmath364 & @xmath365 & @xmath366 & @xmath144 & @xmath200 + scso j052858@xmath76535744 & 0.70 & 0.70 & @xmath205 & @xmath367 & @xmath368 & @xmath316 & @xmath218 + scso j052951@xmath76551611 & 0.66 & 0.65 & @xmath228 & @xmath369 & @xmath370 & @xmath129 & @xmath218 + scso j053052@xmath76552056 & 0.73 & 0.71 & @xmath149 & @xmath371 & @xmath372 & @xmath373 & @xmath219 + scso j053154@xmath76552031 & 0.23 & 0.21 & @xmath374 & @xmath375 & @xmath376 & @xmath349 & @xmath162 + scso j053327@xmath76542016 & 0.23 & 0.25 & @xmath377 & @xmath378 & @xmath379 & @xmath129 & @xmath133 + scso j053437@xmath76552312 & 0.76 & 0.80 & @xmath380 & @xmath381 & @xmath382 & @xmath383 & @xmath384 + scso j053448@xmath76543534 & 0.65 & 0.65 & @xmath385 & @xmath386 & @xmath387 & @xmath204 & @xmath114 + scso j053500@xmath76532018 & 0.59 & 0.57 & @xmath388 & @xmath389 & @xmath390 & @xmath100 & @xmath274 + scso j053632@xmath76553123 & 0.72 & 0.72 & @xmath391 & @xmath392 & @xmath393 & @xmath235 & @xmath120 + scso j053638@xmath76553854 & 0.74 & 0.71 & @xmath394 & @xmath395 & @xmath396 & @xmath114 & @xmath234 + scso j053645@xmath76553302 & 0.74 & 0.71 & @xmath397 & @xmath398 & @xmath399 & @xmath241 & @xmath297 + scso j053655@xmath76553809 & 0.76 & 0.72 & @xmath400 & @xmath401 & @xmath402 & @xmath109 & @xmath403 + scso j053715@xmath76541530 & 0.49 & 0.51 & @xmath404 & @xmath405 & @xmath406 & @xmath193 & @xmath99 + scso j053732@xmath76542521 & 0.62 & 0.61 & @xmath407 & @xmath408 & @xmath409 & @xmath157 & @xmath166 + scso j053952@xmath76561423 & 0.36 & 0.36 & @xmath410 & @xmath411 & @xmath412 & @xmath238 & @xmath100 + scso j054012@xmath76561700 & 0.38 & 0.38 & @xmath413 & @xmath414 & @xmath325 & @xmath200 & @xmath114 + scso j054022@xmath76541622 & 0.51 & 0.48 & @xmath415 & @xmath231 & @xmath416 & @xmath129 & @xmath185 + scso j054052@xmath76551943 & 0.76 & 0.78 & @xmath417 & @xmath418 & @xmath296 & @xmath153 & @xmath218 + scso j054228@xmath76525002 & 0.65 & 0.66 & @xmath126 & @xmath419 & @xmath366 & @xmath140 & @xmath153 + scso j054332@xmath76505651 & 0.35 & 0.36 & @xmath420 & @xmath421 & @xmath422 & @xmath238 & @xmath178 + scso j054358@xmath76531349 & 0.24 & 0.25 & @xmath423 & @xmath424 & @xmath425 & @xmath99 & @xmath144 + scso j054401@xmath76511254 & 0.28 & 0.28 & @xmath131 & @xmath426 & @xmath427 & @xmath133 & @xmath190 + scso j054407@xmath76530924 & 0.25 & 0.26 & @xmath428 & @xmath429 & @xmath430 & @xmath129 & @xmath104 + scso j054436@xmath76550319 & 0.35 & 0.36 & @xmath431 & @xmath432 & @xmath433 & @xmath211 & @xmath434 + scso j054721@xmath76554906 & 0.59 & 0.59 & @xmath117 & @xmath435 & @xmath436 & @xmath316 & @xmath329 + scso j054742@xmath76554836 & 0.50 & 0.50 & @xmath437 & @xmath340 & @xmath307 & @xmath157 & @xmath114 + scso j054811@xmath76555601 & 0.64 & 0.64 & @xmath438 & @xmath439 & @xmath440 & @xmath178 & @xmath162 + scso j054931@xmath76522655 & 0.38 & 0.39 & @xmath441 & @xmath251 & @xmath442 & @xmath124 & @xmath443 + scso j054949@xmath76513503 & 0.28 & 0.28 & @xmath135 & @xmath413 & @xmath444 & @xmath316 & @xmath129 + scso j055017@xmath76534601 & 0.49 & 0.47 & @xmath299 & @xmath445 & @xmath446 & @xmath157 & @xmath211 + [ tab : clustes05hr ] we queried the nasa / ipca extragalactic database ( ned ) for catalogued clusters from _ rosat _ , @xcite , act @xcite and spt @xcite within a 3@xmath447 radius of the location of each scs cluster ( see table [ tab : clustersid ] ) . in some cases there was a catalogued galaxy from the 2 micron all sky survey ( * ? ? ? * ned i d : 2masx ) , which we report if it is within @xmath448 of the bcg . we also take note of radio sources within @xmath449 of our cluster positions since these could potentially bias the cluster sz signal . we only found radio sources from the sydney university molonglo sky survey ( sumss ) at 843 mhz @xcite . finally we find one unidentified bright x - ray source from the _ rosat _ all sky survey ( rass ) @xcite that is coincident with one of our clusters . it is interesting to note that although we recover two of the sze - selected clusters from the first spt results @xcite , there are two others from that study which we do not recover . one of these ( spt - cl 0509@xmath765342 ) falls below the mass threshold used here with mass estimates of @xmath450 and @xmath451 for 200 and 200 respectively . these are lower than the optical mass quoted in @xcite due to the change in the @xcite mass parametrization but still are comparable to the weak lensing mass range , @xmath452 , obtained by @xcite using the same optical dataset . the other cluster ( spt - cl0547@xmath765345 ) has a photometric redshift of 0.88 @xcite that puts it beyond the redshift threshold ( @xmath453 ) we use here . we searched for x - ray counterparts to the scs optical clusters using the _ rosat _ all sky survey ( rass ) data following the same procedure as in @xcite . the raw x - ray photon event lists and exposure maps were downloaded from the mpe _ rosat _ archive and queried with our own custom software . at the position of each scs cluster , rass count rates in the 0.52 kev band ( corresponding to pi channels 52201 ) were extracted from within a 3@xmath447 radius for the source emission and from within a surrounding annulus ( 5@xmath447 to 25@xmath447 inner and outer radii ) for the background emission . the background - subtracted count rates were converted to x - ray luminosity ( in the 0.52.0 kev band ) assuming a thermal spectrum ( @xmath454 kev ) and the galactic column density of neutral hydrogen appropriate to the source position , using data from the leiden / argentine/ hi bonn survey @xcite . x - ray masses within an overdensity of 500 times the critical density were then determined using the @xmath455 vs. @xmath456 scaling relation from @xcite . these were then converted to an overdensity of 200 with respect to the average density of the universe for comparison to the optically - derived masses using a simple multiplicative scaling factor of 1.77 . this factor is good to 10% over the redshift and mass range of our sample . in table [ tab : clustersrass ] we show redshifts , column densities , rates , luminosities , and masses only for those clusters with x - ray emission significant at 2 @xmath457 or higher , while figure [ fig : masses ] shows all the scs clusters with either actual values for the x - ray mass or ( for most cases ) upper limits . we also show points corresponding to a simple stacking of the x - ray and the optical masses for clusters in three redshift and three optical mass bins as the green and red dots , respectively . each point represents an average of @xmath45820 clusters each with some positive x - ray count rate , excluding the several higher significance individual cases plotted in the figure . it is encouraging that the points scatter by about a factor of two around the line of equal optical and x - ray masses , suggesting that the x - ray emission is in fact associated with hot gas in the potential well of a massive cluster . in at least two cases , we have recovered known clusters ( scso j051637@xmath76543001 and scso j232653@xmath76524149 ) . and two of the rass - correlated clusters ( scso j053154@xmath76552031 and scso j233227@xmath76535827 ) are among those shown in the color images ( i.e. , see the bottom left panel of fig.[fig : clusters05hr ] and the top panel of fig . [ fig : clusters23hr ] ) . -@xmath455 scaling relation . clusters with x - ray emission are shown as the box symbols , while the others are shown as upper limits . the solid line denotes equality ; the dashed lines indicate a factor of two range in mass . the green points show the average optical and x - ray masses for clusters stacked in three redshift bins ( 0.0 - 0.4 , 0.4 - 0.6 , 0.6 - 0.8 ) ordered from bottom to top , while the red points were stacked in three optical mass bins ( @xmath4593.2@xmath460 , 3.2@xmath461 - 4.5@xmath460 , and @xmath4624.5@xmath461 ) ordered from left to right . , width=336 ] there are a number of _ xmm - newton _ pointings that overlap with the scs especially in the 23-hr region . table [ tab : clustersxmm ] lists all the optical clusters from this study ( tables [ tab : clustes23hr ] and [ tab : clustes05hr ] ) that are located within the field of view of an archival _ xmm - newton _ pointing ( specifically the following obsids : 0205330301 , 0505380601 , 0505381801 , 0505382201 , and 0505383601 ) . in several cases , there is an associated x - ray source detected by the pipeline processing system ( pps ) which is run on all _ xmm - newton _ pointings to produce standard results and are provided to the observer . if a pps source exists , we list its i d , the offset in arcseconds between the x - ray source and the optical cluster position and whether or not the pps has flagged the x - ray source are being extended . to determine fluxes and luminosities , we extract spectra from circular regions chosen to maximize the extracted count rate for the extended x - ray sources . for the other sources we use a 1@xmath447 radius circle . background spectra come from a surrounding annular region with sufficient area to obtain good photon statistics . the rate , summed over the single pn and both mos detectors , is quoted in column 7 of table [ tab : clustersxmm ] . we also used standard _ xmm - newton _ software tools to calculate the instrumental response functions ( i.e. , the arf and rmf files ) for each cluster . x - ray luminosity values were determined using the extracted spectra and response files assuming a simple source emission model consisting of an absorbed thermal plasma model . for the absorption component we fixed the column density of neutral hydrogen to the galactic value toward the cluster position ( obtained in the same way as for the rass discussed above ) . three clusters have enough signal that their mean temperatures can be measured ( see below ) ; for the others we assume @xmath454 kev for the luminosity calculation . the final column lists the cluster mass inferred from the x - ray luminosity as discussed above . the _ xmm - newton _ spectra constrain the cluster temperatures for and to values of @xmath463 kev and @xmath464 kev , respectively ( errors are at the 68% confidence interval ) , assuming a metal abundance of 0.3 times the solar value . scso j051637@xmath76543001 ( abell s0520 ) is a bright cluster that was the target of a specific _ xmm - newton _ program . our analysis finds a best fit temperature of @xmath465 kev and metal abundance of @xmath466 relative to solar , which are both consistent with previous work ( e.g. , * ? ? ? the x - ray masses for these three systems just mentioned agree quite well with the optically - derived masses and , in the case of scso j051637@xmath76543001 , with the rosat - derived mass as well . for the three other clusters in the 23hr region in table [ tab : clustersxmm ] the x - ray masses are all lower than the optical masses by factors of 35 . for the remaining system ( scso j051613@xmath76542620 ) there is a catastrophic disagreement between the inferred optical and x - ray masses with the x - ray mass being more than an order of magnitude below the optical one . this is likely due to significant contamination of the optical number counts by cluster members from the nearby , rich system abell s0520 which is also close in redshift ( 0.2952 vs. 0.36 ) . we have fully processed using an independent custom - built pipeline @xmath467 tb of archival _ griz _ imaging data from the ctio blanco 4-m telescope acquired under the noao large survey program ( 05b-0043 , pi : joe mohr ) , as part of our own southern cosmology survey . this data volume corresponds to 45 nights of observing over 3 years ( 2005 - 2007 ) and covers 70 deg@xmath4 of the southern sky that has also been fully observed by act and spt . here we have presented the first results from the full nominal data set , namely a sample of 105 massive , optically - selected galaxy clusters . future studies will present the properties of lower mass clusters and groups as well as multi - wavelength studies of cluster physics utilizing selected clusters from this sample . the current sample is limited to systems with optically - derived masses greater than @xmath468 and redshifts less than @xmath469 . we have chosen this mass limit to be at or below the anticipated mass threshold of act and spt in order to encompass the upcoming significant sze detections . however we also have a redshift limit which is due to the depth of the imaging and the wavelength coverage of the filter set . thus we are missing the most interesting massive clusters at high redshift ( @xmath470 ) . however , as demonstrated in @xcite , the optical data analyzed here can be used to confirm the presence of a cluster when conducting a targeted positional search for a high significance sze candidate . the recent success of the mm - band wide - survey - area experiments ( act and spt ) in finding new clusters through untargeted sze surveys has been a strong catalyst for our work . we present this cluster sample to aid in the verification of sze cluster candidates and the characterization of the sze selection function which currently is observationally unexplored . moreover , we anticipate stacking the act mm - band maps at the positions of optical clusters to detect , statistically , the average sze signal for systems that fall below the act detection threshold for individual sources . given the large volume of this data set we believe it might be helpful to address other problems in astrophysics ; therefore we plan to make the blanco data products ( i.e. , photometric source catalogs and images ) available to the community in december 2010 at the following url http://scs.rutgers.edu . we would like to thank the blanco cosmology survey team for the planning and execution of the ctio blanco observations that were used in this paper . we have made use of the rosat data archive of the max - planck - institut fr extraterrestrische physik ( mpe ) at garching , germany as well as results obtained from the high energy astrophysics science archive research center , provided by nasa s goddard space flight center . this research has made use of data obtained from or software provided by the us national virtual observatory , which is sponsored by the national science foundation . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . partial financial support was provided by the national science foundation under the pire program ( award number oise-0530095 ) . we also acknowledge support from nasa / xmm grants nnx08ax55 g and nnx08ax72 g . rllc[b ! ] scso j051412@xmath76514004 & sumss j051411@xmath76513953 & 13@xmath471 & radios + scso j051637@xmath76543001 & 2masx j05163736@xmath765430017 & 3@xmath471 & galaxy + & abell s0520 , rxc j0516.6@xmath765430 & 2.4@xmath472 & gclstr ( @xmath473 ) + & act - cl j0516@xmath765432,spt - cl 0517@xmath765430 & 2.8@xmath447,0.5@xmath447 & sz - gclstr + scso j052113@xmath76510418 & sumss j052114@xmath76510419 & 13@xmath471 & radios + scso j052533@xmath76551818 & abell s0529 & 1.5@xmath472 & gclstr + scso j052608@xmath76561114 & abell s0530 & 2.4@xmath472 & gclstr + scso j052803@xmath76525945 & sumss j052805@xmath76525953 & 24@xmath471 & radios + & spt - cl 0528@xmath765300 & 0.4@xmath447 & sz - gclstr + scso j053327@xmath76542016 & 2masx j05332723@xmath765420154 & 2@xmath471 & galaxy + scso j053437@xmath76552312 & sumss j053442@xmath76552248 & 55@xmath471 & radios + scso j053632@xmath76553123 & sumss j053629@xmath76553147 & 30@xmath471 & radios + scso j053715@xmath76541530 & sumss j053718@xmath76541608 & 48@xmath471 & radios + scso j054012@xmath76561700 & sumss j054014@xmath76561723 & 31@xmath471 & radios + scso j054407@xmath76530924 & sumss j054406@xmath76530922 & 8@xmath471 & radios + scso j054949@xmath76513503 & sumss j054948@xmath76513454 & 9@xmath471 & radios + scso j055017@xmath76534601 & sumss j055019@xmath76534601 & 25@xmath471 & radios + scso j231455@xmath76555308 & 2masx j23145553@xmath765553093 & 5@xmath471 & galaxy + scso j232001@xmath76565222 & sumss j232001@xmath76565219 & 4@xmath471 & radios + scso j232653@xmath76524149 & rxc j2326.7@xmath765242 & 1.2@xmath472 & gclstr ( @xmath474 ) + & sumss j232654@xmath76524153 & 14@xmath471 & radios + scso j233227@xmath76535827 & 1rxs j233224.3@xmath76535840 & 27@xmath471 & xrays + scso j233544@xmath76535115 & sumss j233544@xmath76535113 & 3@xmath471 & radios + scso j234156@xmath76530848 & sumss j234156@xmath76530849 & 5@xmath471 & radios + scso j234703@xmath76535051 & sumss j234703@xmath76535052 & 3@xmath471 & radios + scso j234917@xmath76545521 & sumss j234917@xmath76545518 & 8@xmath471 & radios + scso j235055@xmath76530124 & sumss j235054@xmath76530141 & 17@xmath471 & radios + scso j235138@xmath76545253 & 2masx j23513813@xmath765452538 & 2@xmath471 & galaxy + & sumss j235138@xmath76545255 & 3@xmath471 & radios + [ tab : clustersid ] rccccr[b ! ] scso j051136@xmath76561045 & 0.70 & 1.61 & @xmath475 & @xmath476 & @xmath477 + scso j051637@xmath76543001 & 0.2952@xmath478 & 2.05 & @xmath479 & @xmath480 & @xmath481 + scso j052533@xmath76551818 & 0.72 & 4.08 & @xmath482 & @xmath483 & @xmath484 + scso j053154@xmath76552031 & 0.23 & 5.20 & @xmath485 & @xmath486 & @xmath487 + scso j053952@xmath76561423 & 0.36 & 4.90 & @xmath488 & @xmath489 & @xmath490 + scso j054407@xmath76530924 & 0.25 & 5.01 & @xmath491 & @xmath492 & @xmath490 + scso j054721@xmath76554906 & 0.59 & 6.98 & @xmath493 & @xmath494 & @xmath495 + scso j054949@xmath76513503 & 0.28 & 4.58 & @xmath496 & @xmath486 & @xmath487 + scso j232653@xmath76524149 & 0.1074@xmath478 & 1.28 & @xmath497 & @xmath498 & @xmath499 + scso j233227@xmath76535827 & 0.35 & 1.28 & @xmath500 & @xmath501 & @xmath502 + [ tab : clustersrass ] ccccccrccccc scso j051558@xmath76543906 & @xmath503 & @xmath504 & xmm j051600@xmath76543900 & @xmath505 & yes & 910 ( 2.2@xmath447 ) & @xmath506 & @xmath507 & 3.0 + scso j051613@xmath76542620 & @xmath508 & @xmath509 & & & & 300 ( 1@xmath447 ) & @xmath510 & @xmath511 & 0.5 + scso j051637@xmath76543001 & @xmath512 & @xmath513 & xmm j051635@xmath76543022 & @xmath514 & yes & 1390 ( 5.3@xmath447 ) & @xmath515 & @xmath516 & 14.4 + scso j231651@xmath76545356 & @xmath508 & @xmath517 & xmm j231653@xmath76545410 & @xmath518 & yes & 650 ( 2.2@xmath447 ) & @xmath519 & @xmath520 & 4.1 + scso j232856@xmath76552428 & @xmath521 & @xmath517 & xmm j232856@xmath76552429 & @xmath522 & no & 390 ( 1@xmath447 ) & @xmath523 & @xmath524 & 0.8 + scso j233420@xmath76542732 & @xmath525 & @xmath526 & & & & 380 ( 1@xmath447 ) & @xmath527 & @xmath528 & 0.9 + scso j233556@xmath76560602 & @xmath529 & @xmath526 & & & & 410 ( 1@xmath447 ) & @xmath530 & @xmath531 & 1.0 + [ tab : clustersxmm ]

to solve the ancient problem of fluids , i.e. , of states in which there is no difference between gas and liquid ( the so - called supercritical states ) , it is necessary to abandon several `` rules of the game '' , which are customary to physicists and to refine them by using rigorous mathematical theorems . to the memory of v. l. ginzburg

0912.5011

physicists and mathematicians have different language and different logics . namely , a mathematician first formulates the conditions , then the theorem ( the formula ) , and only after this proves it . a physicist , conversely , first derives a formula , which he makes rather convincingly , but he often conceals the conditions assuming them to be obvious . as a rule , mathematicians prove theorems , which have already been `` known '' to physicists in the sense that the physicists have been using them explicitly or implicitly for a long time . there is a series of rather well - known theorems bearing the names of their authors , which were formulated in a different language and , in fact , had been already proved by the physicists . the results of mathematicians who are translators to the mathematical language were especially easily and rapidly understood by the mathematical community . sometimes , a mathematician , who also obtained new results in physics and , in addition , tried to explain them in the physical language , was not taken as a mathematician by the mathematical community , and he had to prove this separately . the author also happened to obtain several results concerning the area of interests of those who doubted , see @xcite , @xcite , and the work @xcite continued in @xcite ( how levsha had to `` show a flea '' ) , in order to make the mathematical community to acknowledge him . similarly , it is difficult to make the physicists to believe that the mathematicians , who follow rigorous logics , can obtain results that are paradoxical from the point of view of customary understanding . physicists are not convinced even by the fact that the mathematical results correspond to well - known experiments and predict new experiments , which are in a rather good agreement with the theoretical calculations . one of my friends , a famous physicist , used to tell me : `` you do not convince me by your lemmas '' . in the second part of the famous book of feynman and hibbs about continual integrals , the unrigorous logics of feynman continuous integrals ( which was useful for physicists ) led to the rigorous results obtained by norman wiener twenty years ago ( no reference was made ) . the author had to `` prove '' the semiclassical asymptotics of the feynman continuous integrals ( he proved it indirectly somewhat earlier ) according to the feynman logics ( v. guillemin and s. sternberg repeated the proof in @xcite , but fairly quoted the word `` proof '' ) . the `` rules of the game '' developed by physicists are very useful , because they allow them to `` jump over '' the mathematical difficulties and hence to proceed significantly faster . their physical intuition developed in experiments in different areas of physics helps them to feel the result empirically , especially if the numerical data are known . the latter allows them to avoid complicated asymptotics in several small parameters and a nonstandard analysis . in the first aspect ( the intuition ) , i was always surprised by b. l. ginzburg ; in the second aspect ( the knowledge of numerical data ) , by ya . b. zeldovich , who , even speaking over phone , could mentally calculate whether or not a given theorem corresponds to physics . if a mathematician does not have any physical intuition ( for example , just as the author ) , then he can be severely disappointed considering some well - known physical law as an axiom and obtaining an answer contradicting the experimental data . therefore , prior to presenting mathematically rigorous results , which correspond to the unsolved problems of chemistry ( and even of alchemy , because the term `` fluids '' was introduced by alchemists on the critical isotherm behaves as the 100% methanol in several extraction processes . in technology , the supercritical water vapor is used in turbines in all thermal and nuclear power stations . ] ) , we show that some well - known , even `` eternal '' physical and mathematical `` verities '' turn out to be significantly refined in the case of rigorous mathematical verification . to distinguish the classical theory in its modern understanding from the quantum theory , it is necessary somewhat to change the physicist s ideology that the classical theory is the theory that existed in the 19th century before the appearance of the quantum theory . but , in fact , the classical theory is the theory obtained from the quantum theory in the limit as @xmath0 . so feynman was right stating that spin is a phenomenon of classical mechanics . indeed , this is the case in rigorous passing to the limit from quantum to classical mechanics . similarly , the ray polarization does not disappear as the frequency increases and hence is a property of geometric rather than wave optics , as was used to think because the light polarization was discovered as a result of discovering the wave optics . we consider the `` lifshits well '' , i.e. , the one - dimensional schrdinger equation with potential symmetric with respect to the origin and having two wells . its eigenfunctions are symmetric or antisymmetric with respect to the origin . as @xmath0 , this symmetry remains unchanged , and since the squared modulus of the eigenfunction corresponds to the probability that a particle stays in the wells , it follows that , in the limit as @xmath1 , i.e. , `` in the classics '' , for energies less than the barrier height , the particle occurs between the wells , in the two wells at once , although the classical particle can not penetrate through the barrier . nevertheless , this simple example shows the variation in the ideology of the `` classical theory '' . to understand this paradox , it is necessary to take into account that the symmetry must be very precise and the state stationarity means that this state appears in the limit of `` infinitely large '' time . as we show below , the same concerns the bose - distribution in the classical theory of gases . now we show how the existence of an additional parameter changes the representation in the classical limit . we dwell upon the notion , which is called `` collective oscillations '' in classical physics and quaseparticles in quantum physics . in classical physics , this is the vlasov equation of self - consistent ( or mean ) field , and in quantum physics , this is the hartree ( or hartree fock ) equation . so we want to note the following point that seems to be paradoxical . the solutions of the equation in variations for the vlasov equation _ do not coincide _ with the classical limit for the equations in variations for the mean - field equation in the quantum theory . this is because of the fact that the variations are related to another small parameter , namely , the variation parameter . and this is already the field of nonstandard analysis . n. n. bogolyubov , for example , in @xcite , studied the problem without an external field , and the asymptotics thus obtained coincides with the semiclassical asymptotics in an external field @xcite , @xcite . moreover , this is in fact the classical limit , because the parameter @xmath2 in this bogolyubov s paper can be compensated by a large parameter , namely , the wave number @xmath3 . this becomes especially obvious if the interaction is assumed to be zero . then an ideal gas is obtained , which in this case can be considered as a classical gas . just therefore , the author proved that the classical fluids in nanotubes have superfluidity ( see @xcite , @xcite ) , which was confirmed by a series of experiments ( see the references in @xcite ) . confusion is due to the fact that the constant @xmath4 has dimension of action , and , rather often , in the classical limit , in order to keep the dimensions , it participates even in the maxwell distribution @xcite . for example , the thomas fermi equations ( including the temperature equations ) are classical equations from the above point of view , but the fact that they contain the constant @xmath4 leads to confusion . therefore , the van der waals law of corresponding states is very important , it allows us to consider the reduced temperatures @xmath5 and pressures @xmath6 as dimensionless quantities . now we consider the @xmath7-particle gibbs distribution @xmath8 where @xmath9 is the hamiltonian . it is assumed that the level surfaces @xmath10 are simply connected . the theorem about this distribution was in fact proved in @xcite . it must be considered in the sense of kolmogorov complexity . this distribution is the distribution over the number of different experiments whose number @xmath11 is independent of @xmath7 , over systems of @xmath7 particles at the same temperature ( the average distribution over the number of experiments ) , and it is the distribution over the energy surfaces @xmath12 , @xmath13 , @xmath14 : @xmath15 where @xmath16 and @xmath17 are less than some average energy @xmath18 , @xmath19 , and the phase volume is @xmath20 , @xmath21 . we divide the phase space @xmath22 into finitely many domains @xmath23 where @xmath24 , @xmath25 , @xmath26 , @xmath13 , @xmath14 , and , correspondingly , the phase space @xmath27 has coordinates @xmath28 . we perform ordered sampling with return @xmath29 from the partition of domains in the space @xmath27 into the `` box '' @xmath30 , under the condition that @xmath31 from the physical viewpoint , the ordered sampling means that @xmath11 distinguishable @xmath32-dimensional particles are considered . let @xmath33 be the number of @xmath32-dimensional `` particles '' in the energy interval @xmath34 divided by @xmath35 . assume that the above conditions on the function @xmath36 are satisfied . we determined @xmath11 from the condition is considered as average over the the number of particles , then @xmath37 . here @xmath38 is the average energy over the number of experiments . ] @xmath39 where @xmath3 is the boltzmann constant . we determine @xmath18 in as @xmath40 . then we have the following theorem . [ theor2 ] the following relation holds : @xmath41 where @xmath42 is arbitrarily integer . here the probability @xmath43 is the lebesgue measure of the phase volume in parentheses in with respect to the entire phase volume @xmath44 bounded by . thus , the gibbs distribution is not a distribution over momenta and coordinates , but is a distribution over the energy levels . if the domain is multiply connected , then the problem can be solved only by semiclassical transitions . if @xmath45 , then we formally obtain the maxwell boltzmann distribution . but the latter is considered , as a rule , as the average distribution for @xmath7 particles . we discuss this treatment in remark 5 . boltzmann obtained his distribution integrating this distribution over @xmath46 . but this can be done under the following conditions only . for a great many particles , their dependence on a common potential field must be assumed to be slowly varying as a rule . indeed , if the particles are in a volume @xmath47 ( or in an area @xmath48 ) , then the thermodynamical asymptotics requires that the volume @xmath47 ( the area @xmath48 ) tend to infinity . but this means that the potential field varies very slowly , and it must be assumed that the function of the coordinates is of the form @xmath49{v}}).\ ] ] otherwise , we have no rights to pass from the gibbs distribution from the maxwell boltzmann distribution @xmath50 ( integrating it over the momenta ) to the boltzmann distribution of the form @xmath51 since the maxwell boltzmann distribution is not a distribution of the density of the number of particles with respect to the momenta and coordinates . this distribution can give only the number of particles between energy levels of the form @xmath52 but if @xmath53 is of the form @xmath54{v}}),\ ] ] then this integration can be done in the thermodynamical asymptotics as @xmath55 . one should not forget here that the boltzmann distribution is also far from being a distribution with respect to the coordinates @xmath56 , and that it is a distribution with respect to level surfaces of the function @xmath53 only . there is a mathematical error in the very definition of thermodynamical limit as the limit as @xmath57 and @xmath55 such that @xmath58 , where @xmath59 ( the density ) is finite . let the pressure be @xmath60 . this pressure does not lead to a too rarified gas ( and is admitted by the knudsen criterion ) . obviously , the above definition is incorrect because the number of particles is bounded by avogadro s number . as @xmath57 , we have @xmath61 as well . but @xmath62 and @xmath63 is not large under the above pressure . it would be correct to pose the problem , as was said above , neglecting the inverse quantity @xmath64 . in this case , a mathematically rigorous asymptotics in statistical physics would be obtained . the transfer of the bose distribution for photons to gas in the form of the bose einstein distribution is incorrect . since @xmath65 , where @xmath66 stands for the number of particles at the energy level @xmath67 , can not vary up to infinity ( @xmath68 ) , it follows that the chemical potential @xmath69 can take positive values , which must not be neglected @xcite . this restriction means that parastatistics must be considered . the most essential is the note that the distribution given in remark 4 with the property that @xmath7 and @xmath70 are finite permits preserving the parastatistics of the bose einstein - type distribution as @xmath0 , and just this property will be used in what follows . the fact that it holds in the classical limit is especially unusual for the physicists ( cf . remark 1 . ) . therefore , in addition to references to exact theorems , i also present several other arguments . for example , the physicists completely accepted the financial considerations presented in @xcite . `` a general important property of money bills is that the change of one money bill by another of the same denomination does not play any role . this property , which can be called `` money do not smell '' , permits uniquely determining the formula for nonlinear addition . we assume that we want to deposit two copecks into two equivalent banks . how many possibilities do we have ? we can deposit both copecks into the first bank , or deposit both copecks into the second bank , or deposit per one copeck in each of the banks . thus , we have three possibilities . but if we want to deposit two diamonds , then we have four possibilities , because we can interchange the diamonds . at the same time , it does not make any sense to interchange the copecks . the identity property of money bills of the same denomination permits changing the number of versions . this statistics , just as the statistics of identical bose - particles is called the quantum statistics ( bose - statistics ) , and the statistics of diamonds ( provided that they are not absolutely identical ) is called the classical statistics . but as we see , the bose - statistics can be applied to money , more precisely , to money bills.''@xcite , p. 5 . and this is despite the fact that the copecks may have different years of their issue and the money bills have individual numbers . objectively , they are different , but for our problems , this difference is not essential , because only the quantity of money bills is important for us . further , the physicists did not protest , when schoenberg used the creation and annihilation operators in classical mechanics @xcite . and this is possible only when considering the fock space of identical particles . we consider one more aspect of this problem in more detail , because the molecular dynamics of classical particles is nowadays commonly used . in the well - known textbook @xcite by landau and lifshits , the authors explain the identity principle for particles as follows : `` in classical mechanics , identical particles ( such as electrons ) do not lose their ` identity ' despite the identity of their physical properties . namely , we can visualize the particles constituting a given physical system at some instant of time as ` numbered ' and then observe the motion of each of them along its trajectory ; hence , at any instant of time , the particles can be identified . ... in quantum mechanics , it is not possible , in principle , to observe each of the identical particles and thus distinguish them . we can say that , in quantum mechanics , identical particles completely lose their ` identity ' '' ( p. 252 ) . there are similar explanations of the identity principle for particles in other textbooks as well . but , as a matter of fact , if the initial data for the cauchy problem does not possess a symmetry property , then the situation in quantum mechanics does not differ from that in classical mechanics . indeed , suppose that the hamiltonian function @xmath71 is symmetric with respect to permutation of @xmath72 and @xmath73 , @xmath74 . let @xmath75 denote the self - adjoint operator in the space @xmath76 corresponding to the hamiltonian @xmath77 ( weyl of jordan quantized ) . consider the corresponding schrdinger equation @xmath78 satisfying the initial conditions @xmath79 where @xmath80 , and we can assume that @xmath81 , where @xmath82 , and @xmath83 is a bell - shaped function vanishing outside a neighborhood of the point @xmath84 , and the neighborhoods are small so that these functions do not intersect . returning in time to the initial point , we can number all the bell - shaped functions . but in the projection on the real space @xmath85 containing all @xmath86 particles , the experimenter can not distinguish them without taking into account the full deterministic process with respect to time from zero to the given @xmath87 . similarly , in classical mechanics , of two point particles have intersected and the experimenter has no knowledge of their velocities ( instant photo ) , then he also can not distinguish them . he must know their original velocities , i.e. , use slow - motion filming . this means that he has to look into their `` past '' . but if the experimenter must determine which of the original particles with a prescribed velocity arrived at the given point , then he must observe the whole process , up to the point @xmath88 . finally , let us consider the particle identity philosophically . in statistical calculations of the number of inhabitants in a town , the permutation between a child and an old man does not change the total number of inhabitants . hence , from the point of view of the statistics of the given calculation , they are indistinguishable . from the point of view of the experimenters who observes the molecules of a homogeneous gas using an atomic microscope , they are indistinguishable . he counts the number of molecules ( monomers ) and , for example , of dimers in a given volume . dimers constitute 7% in the total volume of gas ( according to m. h. kalos ) . this means that the experimenter does not distinguish individual monomers , as well as dimers , from one another and counts their separate numbers . his answer does not depend on the method of numbering the molecules . these obvious considerations are given for the benefit of those physicists who relate the fact that quantum particles are indistinguishable with the impossibility of knowing the world . i do not intend to argue with this philosophical fact , but wish to dwell only on mathematics and statistics and distributions related to the number of objects . let us turn to the boltzmann statistics , as is described in `` mathematical encyclopedia '' @xcite . in his article on the boltzmann statistics , d. zubarev , the well - known specialist in mathematical physics and the closest disciple of n. n. bogolyubov , writes that , in boltzmann statistics , `` particles ... are distinguishable '' . however , a few lines below , zubarev states that `` in the calculation of statistical weight , one takes into account the fact that the permutation of identical particles does not change the state , and hence the phase volume must be decreases by @xmath89 times '' . of course , it is impossible to simultaneously take into account both remarks . but they are both needed to solve the gibbs paradox . as i pointed out on numerous occasions , from the mathematical point of view , the gibbs paradox is a counterexample to the maxwell boltzmann distribution regarded as a statistical distribution for a gas in which the molecules can not be turned back , but not a counterexample to a dynamical distribution @xcite . as for the famous discussion between boltzmann and some mathematicians @xcite , of course , if the particles of the gas are distinguishable and can be numbered , then they can also be mentally turned back and returned to a state close to their initial state , as the poincare theorem states . as an example we consider the last source of errors , because the computer has a finite memory how is the number of particles related to the computer error and the computer s ability to recover their initial data ? by way of example , we consider a set of unnumbered billiard balls of unit mass and the same color . first , consider one billiard ball and launch it from some ( arbitrary ) point with velocity not exceeding a certain sufficiently large value @xmath90 , i.e. , with energy not exceeding @xmath91 . however , since the computer has certain accuracy , it follows that the energy of the ball will take a finite integer number , @xmath92 , of values in the interval of energies @xmath93 $ ] : @xmath94 , @xmath95 , where @xmath96 corresponds to this accuracy . thus , we obtain a spectrum of energy values which can be regarded as a self - adjoint diagonal matrix of order @xmath92,where @xmath97 . by assigning such a discrete set of energies to a ball , we obtain the wave particle correspondence in classical mechanics , because the resulting matrix is unitary equivalent to any operator @xmath98 with such a spectrum in a hilbert space @xmath99 . how many balls must be launched so that the computer will be unable to determine their initial data ? the spectrum corresponding to @xmath7 balls is obtained by considering the tensor product of @xmath7 hilbert spaces : @xmath100 . the eigenvalues of this operator have the form @xmath101 if we only consider the eigenvalues symmetric with respect to the permutation of the particles of this operator , which corresponds to the identity of the balls , then the eigenvalue @xmath102 is of multiplicity equal to the number of all possible variants of the solution of the problem @xmath103 since the initial set of energies is `` without preferences , '' i.e. , is in general position , then all the multiplicities corresponding to are equiprobable . the computer calculation time is related to the computer accuracy @xmath96 with respect to energy . the problem is how to determined @xmath7 for which , at a particular instant of time , the computer can not recover the initial data in view of the inaccuracy of the classical pattern or that of the quantum mechanical pattern ( which is more accurate , but more cumbersome ) ; the latter pattern , in turn , is not accurately described by an interaction of lennard - jones type . thus , we can draw the following conclusion . the initial data in the classical and quantum mechanical problems are forgotten because of external noise . as a result , the problem is reduced to the distribution @xmath104 . in this problem , we assume _ a priori _ that the initial data are forgotten and , therefore , so is the numbering of classical particles . although the textbook @xcite erroneously interprets the difference between the quantum mechanical pattern and the classical one , it , nevertheless , gives a valid interpretation of the numbering of identical balls . therefore , we can take symmetric eigenfunctions for @xmath105 . therefore , it only remains to obtain the distribution of the number of particles @xmath66 using relations . if @xmath106 , then can be expressed as @xmath107 these relations coincide with those in the classical number - theoretic problem under the condition that @xmath108 is integer , which , of course , is of no importance in the asymptotics as @xmath109 and @xmath57 . thus , since the noise component has prevented us to recover the initial data and the number of particles @xmath7 is preserved as well as the total energy @xmath110 , without given any preference to any versions from , we assume all the variants satisfying the relation @xmath111 to be equiprobable . in this approach , we can also take into account the collision of billiard balls , the initial energy of all balls can only decrease due to friction and the passage of kinetic energy into thermal energy during collisions . the numb er of balls will remain the same and the total energy will not exceed the initial energy @xmath110 . this approach can also be applied to turbulence if we consider such a significant invariant as the kolmogorov spectrum . note another highly important difference between the statistical and dynamical approaches . if the experimenter presents some data , then , as a rule , this means that he has carried out many experiments and presented the data consistent with all the experiments . in addition , these data can be verified by other experimenters and only after that they will be universally accepted . this means that all the variants of relations are , in a sense , equiprobable . if inequality is replaced by the corresponding equality , then we obtain the well - known microcanonical distribution . but as shown in @xcite , the final results for the inequality and the equality coincide . therefore , we have finally derived the well - known microcanonical distribution for the number of particles , whose permutation does not affect the counting result . thus the notion of different ( distinguishable ) particles is incorrect if one speaks about their density . from the point of view of density of the set of objects under consideration , these objects are indistinguishable , namely , any transportation of these objects does not influence the density . therefore , one must not apply the boltzmann distribution in thermodynamics . since @xmath7 is a large number and the value @xmath70 is not large , one must use parastatistics ( see @xcite , @xcite , @xcite and formula below ) . so we arrived at some inconsistencies in the main laws of thermodynamics and statistical physics . one of the most important distinctions between them is that there is a positive chemical potential in the distribution for the `` bose gas '' of finitely many particles @xmath7 . of course , this is related to the famous notion of bose condensate . as was shown in @xcite , there exists a degeneracy temperature such that for @xmath112 the number of particles is maximal , @xmath113 , and if @xmath114 , then @xmath115 of particles get into the bose condensate with energy zero ( or minimal ) . we present an example of such a condensate in number theory . for this , we consider koroviev fagot s trick , described by m. a. bulgakov in the novel `` master and margarita '' , which was already discussed in our preceding papers @xcite . in a variety show , he scattered @xmath86 money bills of 1 tchervonets denomination among @xmath3 spectators . we assume that all versions of scattering of @xmath86 money bills over @xmath3 spectators are equiprobable ( the chaotic condition ) . this is equivalent to decomposition of an integer @xmath86 into @xmath3 terms . the logarithm of the number of possible versions @xmath116 is the entropy . we consider two examples related to fagot trick . let @xmath86 money bills be distributed over @xmath117 spectators , where @xmath117 is such that @xmath118 ( clearly , for @xmath119 and @xmath120 , the number of versions is equal to @xmath121 , and hence such a supremum exists . ) then , with a larger probability , @xmath117 spectators obtain a quantity of money bills . if for @xmath122 , the number of versions and hence the entropy decrease , then the entropy is maximal when the remaining spectators do not get any bill , i.e. , they get into the bose condensate . now we assume that the number of spectators is @xmath123 . then , roughly speaking , @xmath117 ( i.e. , a half ) spectators obtain several money bills , and @xmath117 spectators do not obtain anything . since @xmath117 spectators with a larger probability ( see @xcite ) obtain zero of money bills , the number of versions of the distribution of money bills over spectators is also equal to @xmath124 . we couple the spectators under the following condition : if somebody in a pair does not obtain any money bill , then the sum obtained by the other is divided , for example , in half . therefore , we now have @xmath117 pairs , and the number of all versions is equal to the number of versions in the first example . therefore , the kolmogorov complexity ( entropy ) in both examples is the same and equal to @xmath125 , where @xmath117 is the number of pairs . thus either case is equiprobable , i.e. it there were @xmath123 spectators , then @xmath117 of them would not get anything and if the spectators coupled , then all the pairs would obtain some money bills , but the number of versions of the distribution still remains equal to @xmath124 , and none of the examples can be preferred . both of the realizations is equiprobable . in other words , it may be that the spectators coupled , and it may be that @xmath117 spectators remain without any money bills . if we speak about the social economic efficiency , then it would be better if the spectators coupled . thus , it is a priori unknown what is the results of the phase transition of an ideal bose gas into condensate : pairs of particles or , as was previously assumed in physics , particles at the ground level . if we consider the repulsing potential , as n. n. bogolyubov did in his famous work in 1947 , then this problem remains unsolved . but if we consider the lennard jones potential , then , as the author showed in @xcite , there are arising pairs . in this case , one significantly uses the pontryagin andronov vitt theorem @xcite , which implies that the time of stay of a particle inside the barrier , which exists in the case of the lennard jones potential , depends on the noise level or the gas temperature . it is clear that the transition into the condensate state must occur with a larger probability , which a mathematician must estimate . these estimates were first performed in @xcite . if some particles form pairs ( dimers ) , then the free particles ( monomers ) occur in the fractal situation . this fractal dimension can in general be counted by the hausdorff formula . as is known , as the temperature increases , the number of dimers decreases , and hence the fractal dimension increases . and this means that one more pair of quantities must be introduced in thermodynamics : the extensive ( dimension ) quantity and the intensive quantity corresponding to it , which we call the `` press '' . this can be done rather easily , because the thermodynamical potential @xmath126 is known for parastatistics of any dimension . its derivative with respect to the dimension is just the desired intensive quantity @xcite . as for the problem shown in fig . [ fig1 ] , it is clear that , for @xmath127 , there exist only monomers , and between @xmath128 and @xmath129 , i.e. , between the barrier bottom and height , there exist both monomers and dimers . hence the fractal dimension between @xmath128 and @xmath129 is less than @xmath130 . let us consider the following natural variation of the notion of bose condensate for the classical parastatistical gas . since the parastatistical condensate results in the appearance of pairs ( dimers ) , the number of free degrees of freedom increases , and hence one can speak not about the energy disappearance in the parastatistical condensate but about the energy conservation by means of its redistribution . but if dimers , i.e. , molecule associations , appear , then rotational and oscillatory components are added . this decreases the total energy of monomers , and hence @xmath113 also decreases , which started an avalanche - like process until ( with a larger probability ) the translational energy of monomers disappears . and hence there is a phase transition in the energy redistribution and , naturally , a phase transition in the fractal dimension , which thus decreases . we must not seek a phase transition of such a type in the domain , where the gas liquid transition occurs according to the van der waals maxwell scheme , the scheme accepted by both the specialists in thermodynamics and the specialists in molecular physics of the old school . but the specialists in modern mathematical physics prefer to use the theory of percolation and renorm - group , as well as the possibilities due to computer computations . both in the theory of percolation and in the theory of nucleation , as well as in the van der waals maxwell theory , an important role is played by the surface tension , treated as an additional energy . if this additional component is not considered , then the just described purely mathematical approach and the phase transition must be sought in the domain of critical temperature , i.e. , in the domain , where the surface tension disappears . this structure , in which gas and liquid are indistinguishable , is called fluids from the time of alchemists . just as in remark 2 , we can write the microcanonical distribution , but under the condition that permutations in @xmath66 are forbidden . as a result , we obtain a parastatistical distribution that is very close to the bose distribution . as a rule , in physics in this case , the correct conditions of the sommerfeld radiation are posed , which turn the problem into a not self - adjoint problem . if we go out into the complex domain through the cut , then it is reduced to the problem of finding regge - type poles for a self - adjoint problem . the mathematicians considered , as an example , a potential on a finite interval in the half - plane for the one - dimensional wave equation with a finite well and with the radiation conditions on the boundary of the finite domain ( the domain where the potential takes the values @xmath131 $ ] ) . let @xmath132 be the value of @xmath133 on the barrier maximum . but the `` eigenfunctions '' corresponding to the regge poles ( the poled of the resolvent continued through the cut into the complex domain ) increase exponentially as @xmath134 . but the regge `` eigenfunctions '' for which the real parts of the poles are close to the quasilevels of the well are sufficiently large in the domain of the well . we prove that the solution of the cauchy problem for the wave equation whose initial conditions are consistent with the radiation conditions ( i.e. , the derivative with respect to @xmath87 depends on the initial function for @xmath88 ) tends to zero at each point @xmath133 of the support of the finite potential . in the problem under study , there is addition energy equal to the average energy @xmath136 ( where @xmath3 is the boltzmann constant and @xmath137 is the temperature ) multiplied by the number of particles @xmath7 ( another large parameter ) . thus , we must compare the quantity @xmath138 with @xmath139 . in fact , there is one more dissipation in gas , namely , the viscosity , which also can not be neglected in mathematical computations and which ensures that the initial problem is not self - adjoint ( the parameter corresponding to friction ) . the problem of determining the limit situation becomes significantly simpler in the language of nonstandard analysis . the rigorous theorems even in the case of this simplification are very c cumbersome . we present the main ideas . in the semiclassical limit , one can assume that the well up to the barrier maximum and the domain from this point to the boundary condition are in no way related to each other . only at the point @xmath18 equal to the barrier maximum in infinite time , the classical particle can `` creep '' over the barrier ( see fig . [ fig1 ] ) . the mathematicians consider the one - dimensional wave equation @xmath140 in the half - space @xmath141 , @xmath142 for @xmath143 . the potential @xmath144 has a smooth barrier with maximum at the point @xmath145 . the stationary problem has the form @xmath146 the radiation conditions outside the interval @xmath147 $ ] have the form @xmath148 . the initial condition is @xmath149,\ ] ] and @xmath150 is chosen so that it correspond to the travelling wave outside the interval as @xmath151 . the residues at the regge poles form a basis on this interval @xmath147 $ ] . now we consider the contrary problem , i.e. , the `` irradiation '' problem . in other words , we pose conditions that are complex conjugate to the sommerfeld conditions . this problem does not always has solutions . the well is being filled until @xmath157 if @xmath158 then the solutions do not exist , because the part @xmath159\ ] ] reflects from the point @xmath160 , and the `` irradiation '' condition is not satisfied . this phenomenon is similar to the following one . assume that a negatively charged body is `` irradiated '' by positive ions . then after a certain number of ions sticks to the body ( fills the well ) and neutralizes it , the other freely flying ions reflect from the body and the `` irradiation '' process stops . the conditions dual to the sommerfeld condition are not satisfied any more , because the radiation process originates simultaneously . in our case , the role of attracting charges is played by regge poles . the sommerfeld conditions are a specific case of the blackbody condition . this problem was posed even by frank and von mises in their survey as one of fundamental problems . the feynman integral does not contain any measure , because it is impossible to pose the boundary conditions for the feynman tube : the feynman tube boundary is a black body . by definition , if the trajectory enters this boundary , it should not be considered any more , in the wiener integral of tunes , this is also a black body , but compared with the wave problem , the black body in the diffusion problem is understood as a situation in which the particles stick to the wall , and this means the zero boundary conditions for the heat conduction equation . after the transition to @xmath46-representations , as was shown in our work with a. m. chebotarev @xcite , the zero boundary conditions give the feynman tube in the @xmath46-representation , and it is possible to determine the measure for the finite potential . the temporary energy capture in the laser illumination and its subsequent transformation into a directed beam also characterize the absorption - type trap of `` irradiation '' that further turns into radiation . probably , the problem of black holes in astronomy is a problem of the same type . in the one - dimensional case , on a finite interval , the modern mathematical apparats can solve this problem completely . in this case , if @xmath161 , where @xmath7 is the number of particles , then with a large probability ) one can determine the number of particles inside the well and the number of particles outside it . by the way , this also readily follows from the concept of microcanonical distribution . indeed , the multiplicity ( or the `` cell '' as it is called in @xcite ) of the eigenvalue @xmath162 is different inside and outside the well : it depends on the density of eigenvalues as @xmath0 on a small interval @xmath163 . obviously , this quantity is proportional to the width of each well on the interval @xmath147 $ ] , i.e. , to the length of @xmath164 $ ] and @xmath165 $ ] . assume that @xmath166 is the number of particles in the well @xmath164 $ ] , @xmath167 is the number of particles in the well @xmath165 $ ] , @xmath168 is the length of the interval @xmath164 $ ] , and @xmath169 is the length of the interval @xmath165 $ ] . then @xmath170 of particles will reflect . in numerous papers , the author has rigorously proved that there exists a parastatistical condensate similar to the bose condensate in the one - dimensional and two - dimensional cases as well as in the case of fractal dimension greater than zero . this holds not only for identical objects ( particles ) , but for objects ( particles ) that are assumed indistinguishable , such as those in the calculation of the `` number '' of teenagers , the `` number '' of women , the `` number '' of men , the `` number '' of money bills of one denomination , the `` number '' of cites with population of over 1000 people , etc . this same applies when we speak about density . in the last case , the difference with thermodynamics is in only that there can be no finite thermodynamic limit ( as stated in @xcite ) , i.e. , the ratio @xmath174 , where @xmath7 is the number of objects , and @xmath47 is their volume , area , or length , may tend to infinity . in addition , the author proved that the parastatistical condensate , which is close to the bose einstein condensate , does not necessarily lead to the accumulation of particles at the lowest energy level ( roughly speaking , when they `` stop '' ) , but can lead to the formation of pairs of particles ( dimes ) and clusters and to the conservation of energy , i.e. , passage of translational energy into rotational oscillatory energy . in the present paper , we show that this point of view for the thermodynamics of imperfect gases in the interval of temperature variation from the critical temperature to the boyle temperature is in good agreement with the experiment if for the interaction between the particles we consider the lennard jones potential of the form @xmath175 where @xmath176 is the energy of the well depth and @xmath177 is the effective radius . this agreement can not be random . the relations related to the microcanonical distribution do not take pairwise interaction into account . however , in the new axiomatics of the new ideal thermodynamics , we take into account the form of this interaction , namely , the lennard jones potential , and the properties of this potential . in the scattering problem with the lennard jones potential , the pair energy is of the form @xmath178 where @xmath176 is the depth of the well , @xmath177 is the effective radius , and @xmath59 is the impact parameter . hence @xmath179 . by replacing @xmath180 we get rid of @xmath177 . for a given @xmath59 , the minimum of @xmath181 and the maximum of @xmath182 on the graph in fig . [ fig1 ] are determined by the condition @xmath183 at some point @xmath184 , they coincide and , therefore , @xmath185 at some point @xmath186 . on the graph of @xmath187 ( fig . [ fig1 ] ) for @xmath188 , the inflection points corresponds to @xmath189 . in general , we associate the point @xmath59 with the value @xmath129 , which is the maximal tangent to the well and with the minimal values @xmath128 . a pair of particles is `` incident from infinity '' , i.e. , from the point @xmath190 , is tangent to the value @xmath128 , and slips into the well of depth @xmath191 . thus the actual well of scattering is turned over with respect to the axis of the energy increase . this is related to the fact that the attracting part in the lennard jones potential is negative . the calculation yields @xmath192 . for @xmath193 the well ( the trap ) vanishes . this criterion refers only to two colliding particles . the usual argument in molecular physics involves the symmetry of the average motion of the molecules in all six directions . therefore , @xmath194 of all particles move toward one another . since there are three axes , it follows that @xmath195 of all molecules collide .. their mean energy is @xmath196 where @xmath3 is the boltzmann constant . from physical considerations , the corresponding temperature , apparently , corresponds to the so - called boyle temperature . in fig . [ fig1 ] , which describes the behavior of the mass center , the penetration beyond the barrier of particles moving from infinity and scattering at one another is only possible in the quantum case . however , in the presence of noise , such a penetration is also possible for classical particles by the pontryagin andronov vitt theorem ( see @xcite ) . in our case , by assumption , all the variants of the microcanonical distribution are equiprobable . this condition yields significantly larger amplitudes of deviation in white noise . therefore , we can assume that if the number of particles @xmath7 exceeds some value of @xmath197 , then particles penetrate the barrier under the condition that the temperature is below @xmath198 and transform into oscillating or rotating pairs . after replacing @xmath204 , the equation for the points of possible extremum is of the form @xmath205 hence @xmath206 and the dependence @xmath207 is of the form @xmath208 the maximum value of @xmath209 is attained at the point @xmath210{5}$ ] . the graph of the function @xmath214 , where @xmath215 , has two points on the graph in fig . [ fig2 ] for @xmath216{5}$ ] and for @xmath217{5}$ ] . these points correspond to the minimum and maximum of the trap in fig . [ fig1 ] . note that the passage of a classical particle through the point @xmath218 , when the energy of the particle corresponds to the point @xmath218 ( see fig . [ fig2 ] ) , is not a passage through an ordinary barrier . in passing from quantum mechanics to classical mechanics , we see that this point is no longer the usual turning point , and the classical particle , in the limit , penetrates into the well in infinitely large time . in @xcite , we already discussed koroviev s trick in a variety show when he scattered @xmath86 money bills among @xmath3 spectators and we shoed that if the number of spectators is twice as large as some critical number @xmath117 , then , with a large probability , one - half of the spectators does not get a single bill . however , if they combine into pairs with a view to share the spoils , then most of them will get some bills with a large probability . from the point of view of kolmogorov complexity , both scenarios are equiprobable : none of them is preferable . as proved , given such a canonical distribution , there is also a critical number @xmath197 , as in the case of the bose condensate . by what has been said above , the axioms concerning the bose distribution that @xmath219 particles pass to the zero level ( stop ) are replaced by the following axiom : @xmath219 particles form dimers , trimers , and other clusters that do not affect pressure and form brownian particles in the gas . the point @xmath220 for which the depth of the well ( trap ) is maximum is @xmath221 . in our theory , this point corresponds to the critical temperature @xmath222 , where @xmath3 is the boltzmann constant . let us present the table for @xmath223 for different gases for which the formula for the gas is in correspondence with the depth @xmath224 of the well of the lennard jones potential ( not to be confused with the depth of a trap in the scattering problem ) . the deviations of the data for propane @xmath225 , ammonia @xmath226 , and carbon dioxide @xmath227 from the theoretical data are the largest in the table . apparently , the ammonia and carbon dioxide to not reach the well bottom because of the polarity effect . it follows from abstract thermodynamics that if the dimension of dimers and monomers is the same , the total energy inside and outside the well is also the same . therefore , when we subtract the energy of dimers in the well it follows from the total energy of the monomers , as is seen from our general concept , that , for monomers , the energy remains equal to @xmath228 . the physical thermodynamic relations imply that such a levelling - off occurs in infinite time . it is especially easy to see in the case of a plasma when it is easy to reproduce a pattern with a well and a barrier . , @xmath229 , @xmath137 is the temperature in kelvin degrees , @xmath230 is the volume in @xmath231 , @xmath232 is the gas constant , and @xmath233 is the compressibility factor . the isochores @xmath234 are shown by dotted lines . ] since the point @xmath235 corresponds to the fractal dimension @xmath236 , and the point @xmath237 to the fractal dimension @xmath238 , using abstract thermodynamics @xcite , we can compare the corresponding values of @xmath239 for @xmath240 with experimental curves in fig . [ fig3 ] and in fig . [ fig5 ] . as fig . [ fig3 ] shows , experimental curves for different gases differ somewhat . in fig . [ fig6 ] , we marked the values of the averaged isotherm @xmath241 by black points . in fig . [ fig5 ] , we demonstrate the law of corresponding van der waals states and use the following notation : @xmath242 , @xmath243 . then the dimensionless quantity , i.e. , the compressibility factor @xmath244 becomes a dimensional quantity in the units @xmath231 . to avoid this , we assume that the volume @xmath47 is also dimensionless , taking the ratio @xmath245 divided by unit volume equal to @xmath246 in fig . this volume is shown in the figure by dotted line marked by @xmath247 . then all the quantities are dimensionless , including the chemical potential . we assume that @xmath248 and @xmath249 . in what follows , the subscript @xmath250 is omitted . to find the isotherm @xmath252 , we apply the well - known ( in thermodynamics ) general formula for @xmath253 , @xmath254 , @xmath47 is the volume , @xmath255 is the pressure , and @xmath69 it the dimensionless chemical potential . as is known , in the @xmath3th - order parastatistics of fractal dimension @xmath267 , the distribution over energy has the form @xmath268/\theta}-1}-\frac{k}{e^{[(p^2/2m)-\mu]k/\theta}-1}\bigg\}\ , dp } { \int_0^\infty p^{\gamma}\bigg\{\frac{1}{e^{[(p^2/2m)-\mu]/\theta}-1}-\frac{k}{e^{[(p^2/2m)-\mu]k/\theta}-1}\bigg\}\ , dp}.\ ] ] for the bose - statistics , @xmath269 , for the fermi - statistics , @xmath119 . the parastatistics is characterized by a finite number @xmath3 . in this case , the number of particles is assumed to be large . we associated this number @xmath3 with the number of particles and obtain a new distribution : @xmath270 for @xmath271 and @xmath272 for @xmath273 . in our case , @xmath270 , where @xmath7 is the number of particles . although @xmath57 , the chemical potential @xmath69 still can be positive and depend on @xmath7 . we consider the parastatic energy distribution for @xmath270 in the case where @xmath274 and the dimension is equal to @xmath275 for @xmath241 @xmath276 near the point @xmath277 we expand the expression in braces in a series up to @xmath278 . we have @xmath279 for @xmath277 , this expression tends to infinity as @xmath280 . at zero , the integrand exponentially tends to zero if @xmath283 is integrable as @xmath284 . this implies that as @xmath57 , the expression in braces tends to @xmath285 and the integral in tends to @xmath286 . this implies that @xmath287 and since @xmath288 for @xmath289 , up to @xmath290 ( i.e. , up to 4% ) , we obtain the diagonal line on the graph @xmath291 for the isotherm @xmath292 . this corresponds to the strict incompressibility of fluid as the pressure increases , which , as we see , coincides with experimental graph in fig . [ fig5 ] . for @xmath293 ( @xmath294 ) , the graph of the critical isotherm decreases and becomes the diagonal only for @xmath295 . we note that the fluid critical volume at the phase transition instant , a jump in the fractal dimension , is equal to @xmath300 . this quantity is clearly determined . it also weakly depends on the repulsion degree of the lennard jones potential . therefore just this quantity is the main characteristics of the critical volume . as was already said , it is rather difficult to observe this quantity experimentally . it is much easier to observe the lower end of the vertical segment . if the trajectory of rotation of a pair of molecules about the central point is nearly circular , then the energy redistribution from translational to rotational is @xmath301 . for noble gases , the trajectory ellipticity is rather small , and hence the energy redistribution is @xmath302@xmath303 . but for carbon dioxide , it is more `` elliptic '' , and hence the fall may attend the value @xmath304 , but the original critical volume remains @xmath235 . the energy redistribution law results in the phase transition , see the vertical segment in fig . further , the condensate is also formed as the fluid becomes incompressible in the end , the fractal dimension is then @xmath251 . although distribution is similar to the bose distribution , the final condensate is an incompressible fluid consisting of clusters of dimension less than three , i.e. , it does not contain domains or micelles . note that , for gas dimers , the fractal dimension is equal to @xmath236 . as is readily seen from arguments in @xcite , if for the three - dimensional case , the poisson adiabatic curve is @xmath305 , then , for the dimension @xmath236 , the adiabatic curve will be @xmath306 , and , for `` liquid '' fluids of dimension @xmath251 the adiabatic curve is of the form @xmath307 . the point is that the experiment even with argon is highly unstable bear the critical isotherm . roughly speaking , it is considered as follows . a cylinder with a freely sliding piston on its upper cover is first fixed so that the volume is equal to the value of @xmath308 for argon @xmath309 @xmath231 . here part of the argon is in liquid state and part of it in vapor state ( see @xcite ) . further , the cylinder is heated up from @xmath310 to @xmath311 until the surface tension film disappears and a fluid is formed . so we obtain the lower point of the vertical segment in fig . further , slowly releasing the piston and supplying heat so that the temperature remains equal to @xmath312 all the time , we must come to the point @xmath313 . if , as the result of the experiment , the temperature drops below @xmath312 , then we have jumped to dimension @xmath251 , as is seen from the graph in fig . [ fig6 ] . therefore , here it is hard to abide by equilibrium thermodynamics , while theoretical calculations yield the dimension @xmath236 at first , and then the volume is increased until the temperature drops to @xmath312 , then the dimension @xmath236 in the experiment will be preserved . until @xmath314 , and then becomes @xmath130 in the phase transition of the second kind ( see below ) . ] . the incompressible part of the critical fluid is obtained for any dimension . and this variation in the dimension is a very important new phenomenon . the dimension reconstruction is a separate question . apparently , this phenomenon explains the well - known `` jamming '' effect for glass ( cf . thus , although the volume does not vary with pressure , but , in this case , the dimers clusters are `` chewed '' with a gradual reconstruction of clusters in the direction of a more strict ( ideal ) architecture . for the case of zero dimension and the architecture `` self - organization '' , see @xcite . the `` chewing '' effect , i.e. , the dimension variation with @xmath255 varying on the isotherm , occurs according to the law @xmath315_{\mu = p}}{\frac{\partial\omega}{\partial\gamma } } , \qquad \omega=-\int_0^{\infty}\varepsilon^{\gamma}\ln\big(1-e^{(\mu-\varepsilon)/t}\big)\,d\varepsilon , \qquad \gamma|_{p=1.5p_{\text{cr}}}=0.2.\ ] ] the fractal dimension @xmath267 decreases ( is `` chewed '' ) by the law ( see fig . [ fig7 ] ) @xmath316 as we known , @xmath317 corresponds to the fractal dimension @xmath251 . thus , the fractal dimension has the form @xmath275 , @xmath318 . for @xmath319 , the parastatistical term , where @xmath270 , must be taken into account in @xmath126 . we described the critical isotherm @xmath241 . the process of passing to other isotherms is described in @xcite , @xcite by successive steps of the @xmath137-mapping . in this paper , we do not consider this procedure , which requires a lot of computer computations . we consider the isochore @xmath320 for the scattering problem . for a fixed @xmath59 , it means attraction , i.e. , @xmath321 , where @xmath181 and @xmath182 are coordinates of two particles participating in scattering and denoted by the letter @xmath322 . thus , this is the problem of `` irradiation '' by a flow of monomers . as was already mentioned , if @xmath323 , where @xmath132 is the well width , then the monomers get stuck in the well , and if @xmath324 , then a part of monomers return ( i.e. , are reflected ) . in our case , the quantity @xmath325 at this `` transition point '' is equal to @xmath326 , and @xmath327 , and hence @xmath328 . in fig . [ fig5 ] , one can see that the isochore @xmath329 lying above this point begins to stretch . this means that the isochore endures a phase transition of the second kind . this occurs due to variations in the dimension and , respectively , in the entropy @xcite . this problem , just as the relation between the press and the pressure @xmath255 , must be studied separately , because it is related to some generalization of the basic thermodynamical notions @xcite . in the scattering problem , the volume is assumed to be equal to infinity . according to nonstandard analysis , infinities can be graded . in our case , we can consider the given isochore ( fig . ) on different scales , simultaneously extending or shortening both coordinate axes . it is only necessary to calculate the angle which they make with the isotherm @xmath330 . let us carry out these calculations . the fractal dimension @xmath267 does not vary along the isochore up to a point at which the trap width in the scattering problem for the interaction potential coincides with the distance from the barrier maximum point to the impact parameter @xmath59 ( fig . [ fig1 ] ) . in our case , this coincidence occurs at the point @xmath338 . we note that , for other lennard jones potentials with a different degree of repulsion , the point of phase transition of the critical volume varies only in thousandth fractions . the value @xmath339 at the critical point @xmath312 , @xmath340 is always equal to @xmath341 . thus , up to the value @xmath314 , we can reconstruct the family of isochores , and hence of isotherms , because , at each point of the isochore , @xmath342 and @xmath255 , and hence @xmath137 , are determined . since in the three - dimensional case ( the fractal dimension is @xmath130 ) @xmath339 for @xmath343 is attained at the point @xmath344 , and above the point @xmath314 , the dimensions begin to increase up to the three - dimensional , i.e. , if there are no dimers at the boyle and higher temperatures , then , according to the above argument , the addition to the original number of particles monomers is equal to @xmath345 with a certain constant , which we determine by the condition : for @xmath343 , the dimension must be equal to @xmath130 , i.e. , @xmath346 . hence @xmath347 for the lennard - jones type potential . we find @xmath267 from the condition @xmath348 now , in our case , @xmath267 varies from @xmath349 to @xmath350 and depends on @xmath128 , i.e , on @xmath351 , where @xmath352 . but @xmath353 where @xmath354 is determined by the isochore @xmath355 continued above the point @xmath314 . now we have determined the dependence @xmath239 for the new isochore whose dimension varies from @xmath236 to @xmath130 and the volume remains unchanged . the quantity @xmath356 , as a function of temperature , increases almost exponentially approaching the boyle temperature . to retain this fast increasing number of monomers in the volume @xmath235 , it is necessary to increase the pressure at the same rate . therefore , the pressure increases much faster than @xmath342 , and this results in elongation of the isochore for @xmath357 . the value of @xmath342 increases by @xmath358 units and attains @xmath359 , while the pressure increases more than twice . the other isochores are similar to those constructed above , because their abscissas and ordinates increase and decreases by the same factors . the angle at which they intersect the critical isotherm @xmath241 was calculated . to each point in the family of isochores there corresponds a pair of points @xmath244 and @xmath255 . hence , for a given @xmath47 , the temperature @xmath137 is determined and the locus of points corresponding to the isotherm @xmath253 is constructed . in conclusion , i note that the rigorous proof of all the above statements is very cumbersome . it is based on the use of the correction to variations in the self - consistent field in the scattering problem and the complex germ method ( in topology , the term `` maslov gerbe '' is used , in mathematical physics , the term `` complex germ '' ) , as well as the ultrasecond quantization , where the operators of couple creation are used . the author wishes to express his deep gratitude to a. r. khokhlov , a. a. kulikovskii , a. i. osipov , i. v. melikhov , l. r. fokin , yu . a. ryzhov , a. e. fekhman , a. v. uvarov , p. n. nikolaev , m. v. karasev , and a. m. chebotarev for fruitful discussions and also thanks a. v. churkin , d. s. golokov , and d. s. minenkov for help in computer computations . v. p. maslov , on discreteness criterion for the spectrum of the sturm liouville equation with operator coefficient ( to the paper by b. m. levitan and g. a. suvorchenkova ) . // functional anal . appl . , v.2 , no.2 , 1968 , pp . 6367 . v. p. maslov . on the superfluidity of classical liquid in nanotubes . // rjmp , i : v.14 , no.3 , 2007 , pp . 304318 ; ii : v.14 , no.4 , 2007 , pp . 401412 ; iii : v.15 , no.1 , 2008 , pp . 6165 ; iv : v.15 , no.2 , 2008 , pp . 280290 . v. p. maslov . on a general theorem of the set theory leading to the gibbs , bose einstein , and pareto distributions as well as to the zipf mandelbrot law for the stock market . notes , v.78 , no.6 , 2005 , pp . 870877 . y. kaneda , t. ishihara , m. yokokawa , and others . energy dissipation rate and energy spectrum in high resolution direct numerical simulations of turbulence in a periodic box . // physics of fluids , v.15 , no.2 , 2009 , pp .

we present the deepest near - infrared ( ) photometry yet obtained of the sagittarius dwarf spheroidal ( sgr dsph ) , using vista to survey 11 square degrees centred on its core . we list locations and-band magnitudes for over 2.9 million sources in the field . we discuss the isolation of the sgr dsph from the foreground and galactic bulge populations , identify the sgr dsph s horizontal branch in the near - infrared for the first time , and map the density of the galaxy s stars . we present isochrones for the sgr dsph and bulge populations . these are consistent with the previously - reported properties of the sgr dsph core : namely that it is dominated by a population between [ fe / h ] 1 dex and solar , with a significant [/fe ] versus [ fe / h ] gradient . while strong contamination from the galactic bulge prevents accurate measurement of the ( galactic ) north side of the sgr dsph , the dwarf galaxy can be well - approximated by a roughly ovaloid projection of characteristic size , beyond which the projected stellar density is less than half that of the region surrounding the core . the galaxy s major axis is perpendicular to the galactic plane , as in previous studies . we find slight evidence to confirm a metallicity gradient in the sgr dsph and use isochrones to fit a distance of 24.3 2.3 kpc . we were unable to fully constrain the metallicity distribution of the sgr dsph due to the bulge contamination and strong correlation of [/fe ] with metallicity , however we find that metal - poor stars ( [ fe / h ] 1 ) make up per cent of the sgr dsph s upper - rgb population . the bulge population is best fit by a younger population with [ fe / h ] 0 and [/fe ] 0 or slightly higher . we find no evidence for a split , peanut- or x - shaped bulge population in this line of sight ( , ) . [ firstpage ] stars : late type stars : agb and post - agb stars : evolution stars : fundamental parameters galaxies : individual : sgr dsph galaxy : bulge

1308.4323

the histories of local group galaxies offer a reflection of our own galaxy s history and of its future in the context of universal cosmology @xcite . tidal dissipation of satellite galaxies into their hosts , as part of ongoing hierarchical galaxy formation , directly affects the star - forming , dynamical and chemical evolution of the larger system ( e.g. @xcite ) . studying our nearest galactic neighbours reveals the processes of competitive galactic accretion into the milky way , of tidal stripping of dwarf galaxies , and of the incorporation of their stars into the heterogeneous population we see in the solar neighbourhood , galactic disc and halo ( e.g. @xcite ) . the sagittarius dwarf spheroidal ( sgr dsph ; @xcite ) , is our nearest confirmed galactic neighbour at 25 kpc distance @xcite . we see it caught in the galaxy s gravitational pull as it is being tidally torn apart . its tidal streams comprise 70% of its total light @xcite and span the entire sky , tracing multiple orbits around the galaxy @xcite . these streams are a primary tool in modelling the mass distribution of the milky way ( e.g. @xcite ) . the original dwarf galaxy seems to have been somewhat smaller than the small magellanic cloud ( @xmath810@xmath9 m@xmath10 ; @xcite ) , but its impact into the galaxy has nevertheless been suggested as the underlying cause of the milky way s current spiral structure @xcite . though well - approximated by a spheroidal shape showing little or no rotation @xcite , the surviving part of the galaxy has multiple populations and a noteable metallicity gradient ( @xcite , hereafter sdm+07 ; @xcite ) and large - scale density enhancements which appear to deviate from the overall spheroid @xcite . historically , several galactic globular clusters appear to have been associated with the sgr dsph and , while the cluster m54 has been suggested as the core of the sgr dsph galaxy , it is unexpectedly metal - poor for such an object and may simply be projected onto a region of high density @xcite . the galaxy s star - formation history is complex , with several spectroscopically identified but poorly delineated populations . sdm+07 have summarised them into four ( or possibly five ) photometrically - defined groups , which we investigate in section [ isosgrsect ] . there are strong correlations between metallicity and both age and @xmath2-element enhancement between m54 ( [ fe / h ] = 1.7 dex , [ @xmath2/fe ] @xmath1 + 0.2 dex , @xmath11 gyr ) and the youngest population around solar metallicity ( [ @xmath2/fe ] @xmath1 0.2 dex , @xmath12 gyr ) . a potential super - solar metallicity population of @xmath13 gyr may also exist , though the sdm+07 acknowledge that the evidence for it is not strong . the proximity of the sgr dsph to the galactic centre ( @xmath14 = 5.6 , @xmath15 = 14.1 ) has hampered efforts to characterise its structure and history . a large part of the galaxy lies behind the galactic bulge , which remains the primary contaminant across its face . so strong is the contamination that the galaxy can not be visually separated from the bulge : we must rely on colour magnitude diagrams , such as those used to discover it @xcite , or time - consuming radial velocity or variability observations ( e.g. @xcite , or the aforementioned rr lyr detections ) to identify its members and map its structure . in this work , we present @xmath0-band observations of the inner regions of the sgr dsph with the european southern observatory s visible and infrared survey telescope for astronomy ( vista ) , which we use to create the deepest - yet infrared colour magnitude diagram of the sgr dsph galaxy , with which we probe its stellar populations and their physical extent in the galaxy . the vista infrared camera ( vircam ) surveyed the core of the sgr dsph over 16 nights between 2012 apr 06 and 2012 jun 07 . vircam consists of each 16 imaging chips which create a pawprint that must be moved around the sky to create a contiguous imaging tile of around 1.5 square degrees . the survey consists of seven vircam tiles , covering an irregularly - shaped pattern chosen to match the densest parts of the sgr dsph identified by @xcite . figure [ mapfig ] shows the survey area . each tile was observed once in the @xmath16- and @xmath17-bands , and 12 times in the @xmath18-band . in this paper , we will use a time - average of those @xmath18-band observations , which will be used to provide variability information in a subsequent paper . during the exposures , vircam was set in fpjme nesting mode . in this mode , each tile is repeated for each filter , and for each filter the array is repointed such that the pawprint fills in the gaps in the tile pattern . the tile6zz pattern was used , which follows a zig - zag pattern which ensures that the pawprints cover every region of the tile twice , building redundancy . for each pawprint , the telescope was randomly jittered by 15@xmath19 four times , and within each jitter point the array was micro - stepped by half a pixel each way in each axis ( using the ustep2x2 micro - stepping pattern , resulting in four microstep pointings per jitter ) , with two exposures per micro - step . exposure times were set to 4 s in each channel to ensure maximum saturation of the galaxy s agb stars without loss of linearity , giving an effective exposure time of 64 s to most of the observed area . following on - site quality checks and calibration on - site at paranal and subsequently by the european southern observatory at garching , data reduction to the level of aperture photometry was carried out automatically by the cambridge astronomy survey unit ( casu ) . image processing by casu comprises of the usual image resotration process ( corrections for the detector reset , dark , linearity and flat - field functions ) , plus more - advanced image processing ( sky background and nebulosity subtraction , de - striping , jitter and pawprint stacking ) . source extraction is then performed on the stacked , tiled images by casu , which is then astronomically and photometrically calibrated to 2mass to generate their final catalogues . data retrieved from casu therefore comprised a list of astrometrically - calibrated object detections for individual tiles and epochs in individual filters . further analysis was required to correct the object fluxes for point - spread function distortions , merge the catalogues taken in different filters and epochs , remove objects duplicated between tiles , and systematically correct magnitudes for photometric zero points . we used magnitudes calculated within the recommended aperture # 3 , which corresponds to the flux received by a circular aperture of 1@xmath19 radius , and followed the distortion correction routine outlined in the vista documentation . the simultaneous @xmath0-band images allowed for a relatively simple merging procedure . objects identified in different filters were simply grouped by finding the nearest @xmath16- and @xmath17-band neighbour to each @xmath18-band detection . objects within 0@xmath208 of each other were considered to be detections of the same object ( cf . typical full - width half - maximum point - spread function size of 1@xmath207 , with a typical @xmath210@xmath201 variation over the image plane ) . in principle , we may miss exceptionally red objects this way , but the depth of the catalogue ensures that the reddest agb stars ( known already from 2mass ; @xcite ) will still be identified , thus the only missing objects will be faint , red background galaxies or exceptionally cool foreground objects , neither of which are associated with the sgr dsph . the same principle was followed to merge the multi - epoch @xmath18-band images , resulting in a catalogue for each tile with 12 @xmath18-band magnitudes and one @xmath16-band and one @xmath17-band magnitude . the median difference between the @xmath18-band magnitudes at each epoch was found , and an offset applied to give each @xmath18-band epoch the same zero - point magnitude , though we note that this has not yet been absolutely calibrated . we discuss this in the appendix . tile merging to form a master catalogue followed a similar principle . detections in different tiles were flagged as the same object when they fell within 1@xmath202 arcseconds of each other . this was done iteratively , as some objects were observed in up to four tiles , thus we have up to four independent detections of the same object in some cases . the source tile number was retained during this process to allow a per - tile zero - point correction later . we can , to first order , check the effectiveness of the source matching by comparing the source densities in regions where the tiles overlap to regions covered by only one tile . if the matching is too liberal , then the overlap regions will be under - dense in sources ; if it is too conservative , they will remain over - dense . the top panel of figure [ mapfig ] shows that the tile overlap areas have the same source density as the centres of the tiles , thus the matching process is effective to first order . the same figure highlights considerable differences in source density between tiles . this originates in the original casu pipeline reduction , and is a result of differing signal - to - noise limits between tiles , which were taken on different epochs with different seeing conditions and different levels of lunar and other background light . when we limit the data to only bright sources ( bottom panel of figure [ mapfig ] ) , the discrepencies between tiles become smoother and the tiling pattern disappears , indicating that our completeness is similar across all tiles down to the magnitude of the sgr dsph horizontal branch ( @xmath22 ) . an identical algorithm to the tile merging processing was followed for matching literature catalogues , again with a 1@xmath202 arcsecond maximum offset . the merged catalogue was cross - correlated with the 2mass , ucac4 ( zacharias et al . , in prep . ) and _ wise _ @xcite catalogues to give a multi - mission catalogue spanning from the optical ( @xmath23-band ) to mid - infrared ( 22 @xmath24 m ) . the final merged catalogue was then calibrated to the 2mass vega - based photometric zero - point system , taking into account the difference in filter transmissions between the two and assuming a reddening of @xmath25 ( valid for m54 ; @xcite ) or @xmath26 . details on the source density limits , comparison to 2mass and filter corrections are given in the appendix . [ cols="^,^,^,^,^,^,^,^,^,^,^ " , ] the final catalogue ( table [ cattable ] ) contains 2921920 vista objects , of which 1047713 have detections in all three bands ( the sharp decrease is due to the much lower efficiency in detecting intrinsically fainter sources given higher sky background at longer wavelengths ) . of the vista objects , 433245 have identified 2mass counterparts , 48028 have ucac4 counterparts , and 222191 have identified _ wise _ counterparts . figure [ mapfig ] shows the projected density of vista sources . source density can be seen to increase to the north - west ( upper right ) where the galactic bulge contributes an increasing number of stars . gaps can be seen where coverage is decreased or absent due to removal of a poorly - functioning detector on the vircam ccd . the high - density region at @xmath27 , @xmath28 is m54 , which marks the galaxy s core . figure [ cmdfig ] presents the associated colour magnitude diagrams , with the distribution of component populations plotted in figure [ map2fig ] . considerable contamination is evident along the line of sight from both the galactic plane and galactic bulge . the foreground main sequence is mostly confined to @xmath29 mag . its stars are distributed more uniformly across the surveyed area than the other populations , but shows concentrations towards high galactic latitudes , which lie towards the top - right of the mapped panels in figure [ map2fig ] . the sgr dsph giant branches are well defined and easily separable from the other populations at the bright end , near the rgb tip . the asymptotic giant branch ( agb ) continues above the rgb and curves off to the red side of the diagrams , with the curve being quite sharp when plotted as @xmath17 versus @xmath30 . all these red ( ( @xmath31 ) @xmath32 1.3 ) agb stars are thought to be carbon - rich @xcite . the stars immediately above the rgb tip contain a mixture of oxygen- and carbon - rich agb stars within the sgr dsph . the base of the agb can be faintly seen at around @xmath33 and @xmath34 mag , also at @xmath35 and @xmath36 mag . the sgr dsph giant stars are concentrated around m54 ( figure [ map2fig ] , centre - left panel ) , but also reveal the greater extent of the galaxy further from the cluster . structure can be seen on various scales at roughly twice the level given by poisson noise . unfortunately , the differing sensitivity limits and irregular footprint of our observations means we are limited in the quantitive detail we can place on the outer regions of the sgr dsph galaxy . additionally , the low contrast of these features means determining the shape , extent and origin of that structure is not reliably possible . altering the lower - magnitude boundary of stars included in this map does not have any appreciable impact on the existence or otherwise of any structures . the bulge giant branch dominates the central and upper parts of the colour magnitude diagram . its stars clearly concentrate towards the galactic centre , with a density variation of a factor of around ten over the mapped area . in colour magnitude space , it disappears near @xmath37 = 16 mag , where the three populations converge . few bulge stars are seen at southerly galactic latitudes . here , stars identified as bulge giants are likely giants in the galactic plane : even though the plane is typically closer ( therefore brighter ) than then bulge , there is still sufficient overlap that the plane contaminates the bulge giant population . the full colour magnitude diagrams ( figure [ cmdfig ] ; see also later , figure [ biasfig ] ) extend down to the bulge main - sequence turn off , which is just visible at the faint limit of the @xmath17-band data , but is more obvious near @xmath16 = 18 mag in the shorter - wavelength data . two final density enhancements are present in figure [ cmdfig ] , labelled as the bulge red clump and the sgr dsph horizontal branch in figure [ map2fig ] . in reality , these represent the combination of two different phenomena . evolutionarily , the first is the rgb bump , caused by chemical discontinuities in the hydrogen - burning shell as a result of first dredge - up . the second is the horizontal branch ( hb ) , or start of the helium - burning period of evolution . this is termed the `` red clump '' in metal - rich and higher - mass populations where no significant horizontal extent to the branch is seen . in the bulge , we mainly see the rgb red bump ( see section [ isobulgesect ] ) , as the red clump stars are more vertically spread in the colour magnitude diagrams . in the sgr dsph , we primarily see the hb stars : the rgb bump is just visible below the hb in the @xmath16 vs. ( @xmath38 ) colour magnitude diagram . the extent of the sgr dsph hb is rather less in @xmath39 than in @xmath30 . this extension in @xmath30 is unlikely to be real and probably results from decreased sensitivity in the longer - wavelength bands . using colour - cuts of @xmath40 and @xmath41 mag , we have produced the map of sgr dsph hb stars in the centre - right panel of figure [ map2fig ] . although it has more sources , this map shows a strong concentration of sources around m54 . however , it also suffers slightly more from contamination , leading to an erronous increase in sources in the top right . in order to fit accurate isochrones to the observed colour magnitude diagrams , we have endeavored to separate the bulge and sgr dsph populations from each other , by bisecting our sample in galactic latitude around m54 . the population at southerly galactic latitudes should contain mostly sgr dsph stars , the population at northerly galactic latitudes ( closer to the galactic plane ) will be dominated increasingly by the galactic bulge . as we have no field of solely sgr dsph nor bulge stars to use as a template , we can only determine the bias of a particular combination of colour and magnitude being present in the southern or northern fields ( therefore being in the sgr dsph or bulge ) . we define the following statistic : @xmath42 where @xmath43 and @xmath44 are the density of sources per unit area of sky in the ( galactic co - ordinate ) northern and southern regions . this defines the amount of northerly bias present in the sample . the total source density is strongly biased towards regions near the galactic equator , with the northern region containing 63% of the stars , and the southern half 37% . the sgr dsph has its major axis in the direction of galactic latitude , and m54 is its nominal centre @xcite . if its population is symmetrically distributed across in both halves , its population should have values of @xmath23 of 0.50 . the bulge should be predominantly in the northern regions , thus it should have @xmath45 to 1 . the galactic disc should show some preference for the bulge populations due to a longer line of sight , but also contains a local population of stars which are more uniformly distributed . it should therefore have values of @xmath23 between those of the sgr dsph and the bulge . figure [ biasfig ] illustrates the north - south bias for our sample . the sgr dsph giant branch is clearly defined in this plot , visible down to the hb , which juts blueward of the giant branch near @xmath46 mag . the sgr dsph population becomes much harder to identify below the hb as it merges into the base of the bulge rgb . the upper bulge rgb is also well - defined , but merges into the galactic disc rgb on its blueward edge . the galactic ms , bulge rgb and sgr dsph hb all co - incide , making it difficult to identify the blue edge of the sgr dsph hb . a uniformly distributed extension of stars near @xmath47 , @xmath46 mag may be a metal - poor extension of the sgr dsph population , or it may truncate near @xmath48 mag . stars in m54 , with metallicities much below the sgr dsph average , lie on the redward side of the bulge rgb . any test for asymmetry in the sgr dsph population is hampered by the fact that the survey area is not symmetrical about the galactic north south axis . with this concession in mind , we can only perform a limited test for asymmetry between the two halves . taking stringent limits on the stars within the sgr dsph giant branch , namely @xmath49 mags with @xmath50 mags and @xmath51 mags , we find 2759 stars with an average @xmath52 , implying no measurable asymmetry in the sgr dsph galaxy population . in practice , since the centroid of the northern half of the observations is considerably further from m54 than the southern centroid , we should expect it to have a slightly lower density , thus @xmath23 should be slightly less than 0.5 . thus , it seems likely that there are more stars in the sgr dsph at higher galactic latitudes than there are at more - negative galactic latitudes , implying m54 is not the true centroid of the sgr dsph . although we have no control field to subtract , figure [ biasfig ] shows that sections of the bulge giant branch reach values of @xmath23 which approach unity . a variety of ranges can be placed onto the colour magnitude diagram where @xmath23 is maximised . this effectively sets a lower limit to the fraction of bulge stars in the northern section . the 1-@xmath53 lower limit is maximised near the region @xmath54 and @xmath55 , containing 214 stars , where @xmath23 is 0.921 @xmath4 0.068 . this sets a 1-@xmath53 limit whereby at least 85.2% of stars with those parameters ( and by implication all bulge stars ) are in the northern region . to first order , we can use this to extract both the sgr dsph and bulge populations from our sample . the confusing issue of the galactic plane means that some of this population will be present in both extracted populations . assuming 92.1% of the bulge and 50% of the sgr dsph stars are in the northern half , given a total number of stars @xmath56 in a particular colour magnitude bin , the number of stars in that bin belonging to each population should be given by : @xmath57 the lower panels of figure [ biasfig ] show the results of this statistical extraction . bulge stars do appear to have been cleanly extracted from the rest of the data , while the foreground stars have gravitated towards the sgr dsph population as a result of having similar values of @xmath23 . in each case , the faintest stars have fallen within the sgr dsph population , due to the fainter limiting magnitude in these less - crowded tiles . an attempt was made to separate the foreground stars from the sgr dsph stars using similar techniques . this test used the average projected radius from m54 as the diagnostic , instead of galactic latitude . however , it proved not to be very decisive , given the limited spatial coverage of our observations compared to the size of the sgr dsph . the five - population summary present by sdm+07 represents the cumulation of a number of previously published studies , along with their own work . these estimates can be broadly summarised as follows : * distance : * early estimates using red giants have put the sgr dsph consistently at distances from 25.2 @xmath4 2.8 kpc @xcite to 26.3 @xmath4 1.8 kpc @xcite . more recent work using rr lyrae has sugggested a slightly closer distance ( 24.8 @xmath4 0.8 kpc ; @xcite ) , while _ hubble space telescope _ data has been re - analysed by @xcite to give a significantly further distance ( 2830 kpc , inclined towards our line of sight ) . * metallicity distribution : * the metallicity distribution has been determined by fitting globular cluster loci or isochrones to colour magnitude diagrams by several authors ( @xcite ; sdm+07 ; @xcite ) . the earlier works among these sampled few stars in small regions around the galaxy , and did not accurately consider stars @xmath2-element enhancement . thus while they found a distribution of metallicities in the galaxy , they did not always agree on the bulk metallicity . spectroscopic determinations of the metallicity distribution have been clearer . the metallicity distribution is found to peak between [ fe / h ] = 0.4 and 0.7 dex @xcite . however , a high - metallicity tail extends toward solar metallicity @xcite and a low metallicity tail down to the metallicity of m54 @xcite . * metallicity gradient : * spectroscopic and photometric studies have also shown a significant metallicity gradient . while the remaining sgr dsph core is fairly well - mixed @xcite , there is an absence of the metal - rich populations in the tidal tails , which typically have metallicities around [ fe / h ] = 1.1 dex @xcite . this suggests selective stripping of metal - poor stars for the last @xmath83 gyr @xcite . * @xmath2-element distribution : * spectroscopic surveys have also shown a smooth trend of decreasing [ @xmath2/fe ] with increasing [ fe / h ] , with the variation at each metallicity being dependent mostly on the study concerned than within each dataset . reported values are between @xmath1 + 0.3 to + 0.2 dex in the metal - poor stars , @xmath1 + 0.1 dex to 0.1 dex in the metal - intermediate population , and @xmath10.2 dex to 0.3 dex by the time the stars reach solar metallicity @xcite . * age distribution : * @xcite and sdm+07 compiled age estimates of the populations as well as presenting their own . the metal - poor population is thought have an age of 912 gyr , with estimates favouring the later value . the metal - rich population is generally found to be 46 gyr old , though @xcite find the star - formation rate to have peaked 810 gyr ago . the age and nature of the younger populations is more speculative , with star formation generally thought to have continued at a lower rate until less than 1 gyr ago . photometric data containing only post - main - sequence stars suffers from considerable degeneracy between [ fe / h ] , [ @xmath2/fe ] and age . with the photometric data considered here , we can not solve for all three simultaneously . however , if we are prepared to adopt the [ @xmath2/fe ] versus [ fe / h ] and [ fe / h ] versus age relationships given by the above literature , we can estimate the metallicity distribution of stars in our field by comparison to stellar isochrones . further degeneracy exists between age and distance , and between age , metallicity and reddening . reddening over the field is well - constrained to be @xmath58 mag , which gives @xmath59 mag and @xmath60 mag , where these values represent the mean and typical range across the field estimated from cobe dust maps ( see appendix for more details ) . the sgr dsph is behind any likely reddening sources , which are expected to arise in the local galactic plane . we adopt 25 kpc as the distance of the sgr dsph . finally , uncertainties exist in the isochrones themselves . different isochrones , using different stellar models and evolutionary codes produce marginally different results . in older populations , one significant uncertainty is post - main - sequence mass loss . even canonically - identical stars , such as those in globular clusters , have been found to experience significantly different post - main - sequence mass - loss histories @xcite . isochrones also treat mass loss differently : for example , the padova models @xcite assume a formulism based on @xcite , but the actual rgb mass loss is only experienced by the models instantaneously at the rgb tip , though they allow for reddening of the rgb tip due to circumstellar dust cen which , as a supposed tidally - stripped dwarf galaxy nucleus , may best represent the core of the sgr dsph , we have found some evidence of rgb dust production @xcite . however , this is questionable as it suffers from the same problems arising in a recent exchange of papers with origlia et al . these papers concern the globular cluster 47 tuc , often used as a comparison to the sgr dsph as it has a similar metallicity . in the mcdonald et al . papers , confirmed by momany et al . , while we find dust production extends a bolometric magnitude below the rgb tip , dusty stars in this region are consistent with being on the agb . ] . we use the dartmouth synthetic isochrones and horizontal branch tracks @xcite for preference as they generally provide a good fit to stars in our age metallicity regime ( e.g. @xcite ) . figure [ sgrisofig ] shows the extracted colour magnitude diagram for the sgr dsph , overplotted with dartmouth synthetic isochrones and horizontal branch tracks @xcite . we have included here isochrones with ages , [ fe / h ] and [ @xmath2/fe ] to broadly represent the population catagorisations of sdm+07 , but incorprating [ @xmath2/fe ] ratios from the recent literature cited above : * in red , to represent the m54 population , an isochrone at 13 gyr , [ fe / h ] = 1.7 dex , [ @xmath2/fe ] = + 0.2 dex ; * in green , to represent the outer , metal - poor population , isochrones at 10 and 12 gyr , [ fe / h ] = 1.3 dex , [ @xmath2/fe ] = + 0.2 dex ; * in blue , to represent the metal - intermediate population , isochrones at 4 and 8 gyr , [ fe / h ] = 0.7 dex , [ @xmath2/fe ] = 0 dex ; * in magenta , to represent the metal - rich population , isochrones at 2 and 3 gyr , [ fe / h ] = 0 dex , [ @xmath2/fe ] = 0.2 dex ( note that the @xmath2-depletion and age differences render these isochrones almost as blue as the metal - intermediate population ) ; * in cyan , to represent the suspected very - metal - rich population , an isochrone at 1 gyr , [ fe / h ] = + 0.5 dex , [ @xmath2/fe ] = 0.2 dex ( note that this isochrone does not reach observed rgb tip ) . comparative theoretical hb tracks are shown at estimates zero - age hb masses , smoothed at either end by gaussian of width 0.05 m@xmath10 . in the same order and colours , these are : * @xmath61 [ fe / h ] = 1.7 dex , [ @xmath2/fe ] = + 0.2 dex , @xmath62 = 0.60 m@xmath10 ; * @xmath61 [ fe / h ] = 1.3 dex , [ @xmath2/fe ] = + 0.2 dex , @xmath62 = 0.65 and 0.70 m@xmath10 ; * @xmath61 [ fe / h ] = 0.7 dex , [ @xmath2/fe ] = 0 dex , @xmath62 = 1.10 m@xmath10 ; * @xmath61 [ fe / h ] = 0 dex , [ @xmath2/fe ] = 0.2 dex , @xmath62 = 1.15 m@xmath10 ; * @xmath61 [ fe / h ] = + 0.5 dex , [ @xmath2/fe ] = 0 dex , @xmath62 = 1.95 m@xmath10 . note that [ @xmath2/fe ] = 0.2 dex models are not available for this metallicity . figure [ sgrisofig ] broadly confirms that isochrones representing the populations described by sdm+07 model the sgr dsph galaxy well . the most - metal - poor population , corresponding to m54 , is sufficiently low density and sufficiently blue to be poorly - distinguishable from the reddest , most - highly - extincted bulge population . however , redward of this , there is a continuous distribution of stars . this distribution has a distinct peak towards redder colours where the higher metallicities , forming the bulk population , are expected . the sharpness of the red edge of the giant branch is amplified by the known [ @xmath2/fe ] versus [ fe / h ] anti - correlation . the distance to the sgr dsph can not be constrained by our observations to better than the literature values , as we are limited in the number of clear evolutionary stages at which we can fix the various properties of the isochrones . our best constraint is provided by the @xmath17-band magnitude of the rgb tip , which we fit to lie at @xmath630.04 mag ( figure [ sgrrgbhistfig ] ) . our applied gaussian spread in the rgb tip positions ( @xmath40.10 mag ) allows for a depth to the galaxy of @xmath64 per cent of its distance , which will we expect to reduce substantially once intrinsic variation , differential reddening and photometric variability are taken into account . this equates to @xmath651.2 kpc at the distance of the sgr dsph . the rgb tips of the dartmouth isochrones representing the metal - intermediate population , which the above spectroscopic studies find to represent the bulk of the population , lie at @xmath66 and 10.70 mag ( for 4 and 8 gyr , respectively ) if we assume a distance of 25 kpc . those representing the metal - rich population have a wider tolerance , lying at @xmath67 and @xmath68 mag ( for 2 and 3 gyr , respectively ) . a comparison using the independent parsec isochrones @xcite shows their corresponding isochrones at @xmath69 , 10.64 , 10.96 and 10.43 mag , respectively ( note that the parsec isochrones do not allow non - solar [ @xmath2/fe ] ratios ) , suggesting that the choice of isochrones can add at most a @xmath40.10 mag uncertainty to the distance . the uncertainty in the mean differential reddening ( taken to be @xmath70 mag ) is very small and can be neglected compared to these errors . combining these magnitudes , giving double weight to the more - common metal - intermediate ( [ fe / h ] = 0.7 ) population , we form an expected range of 10.74 @xmath4 0.19 mag for the position of the isochrone rgb tip at 25 kpc . scaling these to the observed value , we find a distance to the sgr dsph of 24.3 @xmath4 2.3 kpc to the sgr dsph . this is consistent with most previous estimates , but does not agree with the @xcite result of 2830 kpc . figure [ metalmapfig ] shows the distribution of the different populations highlighted in figure [ sgrisofig ] . note that we have not generated a region for the expected location of the most - metal - rich population as the isochrones at this metallicity are poorly calibrated . the population belonging to m54 is heavily contaminated by stars from the bulge , which cluster in the direction of the galactic centre ( the top - right of the maps ) . it is not possible to meaningfully extract the distribution of stars originated in m54 , which appears as a high - density ( black ) dot without extensive radial velocity surveys . however , the density of m54 sources to the left of the map is much less than any of the subsequent maps , showing a general absence of any population of m54 stars outside of the cluster itself . the population of m54 stars mixed in with the rest of the sgr dsph galaxy must be a very small fraction of the total number of stars . while @xcite place m54 in front of the sgr dsph galaxy , this disagrees with the consensus opinion that they are at the same distance and m54 s projected spatial co - incidence with the sgr dsph core implies that it is the galaxy s nucleus @xcite . this is despite its unusually - low metallicity for a galactic nucleus , and it is presumed to have tidally decayed into its current position @xcite . the lack of an obvious m54 population in the rest of the sgr dsph galaxy implies there has been little dispersion of the m54 population or mixing of it into the rest of the galaxy . one might expect such mixing if m54 arrived at the sgr dsph core via tidal accretion or a multiple - body encounter , as we see with sgr dsph stars being tidally dispersed into the milky way s halo . while m54 s low metallicity would appear to prevent it forming _ in situ _ , we would expect that the passage that brought it to its current position is a dynamically - simple , low - velocity merger that does not cause significant tidal stripping . we hypothesise that the current sgr dsph nucleus may have simply built up around the gravitational potential of m54 . the metal - poor population , approximately representing metallicities from [ fe / h ] = 1.6 to 0.9 dex , is also slightly contaminated by bulge stars . however , regions at higher galactic latitude ( left and bottom of the maps ) show more of these metal - poor stars than the m54 population , indicating that the metal - poor population is better spread throughout at least the high - galactic - latitude end of the sgr dsph . the metal - intermediate and metal - rich populations both show a roughly ellipsoidal shape , centred around m54 , with very little obvious contamination from the bulge . as with the other populations , the bulk population shows a central point centered on m54 itself . this central cusp was shown by @xcite to be a few arcminutes in diameter . it overlies a broader distribution of stars , which falls to half the central density at approximately two degrees from m54 in galactic latitude and one degree from m54 in galactic longitude . the relative ratios of the metal - rich to metal - intermediate components ( i.e. the blue and red sides of the visually - obvious sgr dsph giant branch ) show little variation with position ( figure [ metalmapfig ] , bottom - right panel ) . the differences in star count that do exist mostly trace the small differences in interstellar reddening in the _ iras _ dust maps @xcite . the ratio between metal - poor and bulk metallicity ( metal - intermediate and -rich ) populations shows more structure , although some of this is attributable to contamination from the galactic bulge in the metal - poor sample . even when only considering the left - hand side of the map , away from the bulge , there is still a general trend whereby pinker colours , representing a greater fraction of metal - rich stars , concentrate around m54 ( though not in the pixel containing m54 itself ) . this indicates a metallicity gradient in the galaxy . further study at larger radii would be helpful when the vista hemispheric survey completes its catalogue of this area . the four colour magnitude regions , in order of increasing metallicity , contain 2005 , 1565 , 1893 and 2031 stars . the most - metal - poor region , representing m54 , is dominated by bulge stars ( figure [ metalmapfig ] ) . we can crudely estimate from their spatial bias ( @xmath23 ) towards the galactic bulge that @xmath7164 per cent of stars in this colour magnitude region belong to the bulge , thus we discount it from our analysis . most of the 1565 objects in the metal - poor region are sgr dsph stars , though this population also suffers from some contamination . when compared to the two metal - rich populations , we find the metal - poor population contains at most 29 per cent of the sgr dsph s giant stars ( note that , for the mass range we are exploring , the rate of evolution is roughly mass - invariant on the giant branch ; e.g. @xcite ) . in reality , the number is likely to be somewhat less , due to the contribution from the bulge , and potential contamination in the metal - poor region by common - envelope binaries and agb and post - agb stars from the bulk population . figure [ colourhistfig ] shows the colour distribution of stars on the sgr dsph giant branch , along with the corresponding colours expected from dartmouth isochrones representing the three remaining populations . we here skew the colour magnitude diagram using a colour term , @xmath72 , which allows the observed sgr dsph giant branches to have a roughly constant value of @xmath73 mag : the ( @xmath31 ) colour of the giant branch at @xmath74 mag . the isochrones have been chosen to match the three observed populations , at [ @xmath2/fe ] = + 0.2 , 0 and 0.2 dex with ages of 11 , 6 and 2.25 gyr , but are shown at a variety of different metallicities from [ fe / h ] = -2 to -1 dex for the metal - poor stars , and [ fe / h ] = -1 to 0 dex for the metal - intermediate/-rich components . the degeneracy between [ fe / h ] and both [ @xmath2/fe ] and age can be seen by comparing the colours given by the three sets of isochrones at [ fe / h ] = -1 . for identical metallicities , the isochrones show a range in colour of @xmath75 mag . this degeneracy means we are unable to recover the metallicity distribution of the sgr dsph without much more precise constraints on the absolute relationship between age and metallicity , and between [ @xmath2/fe ] and [ fe / h ] . that is not to say that we can not make inferences from this diagram , but that they must remain as such and should not be taken as direct evidence in their own right . we note that the strength and coherence of the sgr dsph giant branch histogram peak is lost at fainter magnitudes , and that most stars occupy the space between @xmath76 mag where , canonically , the metal - poor population is expected to lie . this is not an artifact of decreasing signal - to - noise : stars in all these regions have roughly similar photometric errors , and the red side of the giant branch falls off at the same rate at faint magnitudes as at bright magnitudes . this can mostly be explained by the increasing contribution of bulge stars at bluer colours . bulge stars make a significant contribution up to @xmath77 mag in the brightest bin , where the histogram reaches a minimum . however , with decreasing brightness , this point moves progressively redwards in this colour space as bulge stars increase in number , with their reddest significiant contribution in the subsequent bins being ( from brightest to faintest ) @xmath78 , 0.93 , 0.96 and 1.00 mag . at magnitudes fainter than @xmath79 mag , the sgr dsph and bulge begin to merge in the colour magnitude diagram . underneath this , the distribution of sgr dsph stars remains largely unchanged ( figure [ overplothistfig ] ) . we can also infer that the sgr dsph population should be slightly @xmath2-element depleted . we have artificially removed the slope of the giant branch , but the isochrones have independent slopes . the younger isochrones at [ @xmath2/fe ] = 0.2 dex have a much more constant colour ( @xmath80 ) than the middle - aged isochrones at [ @xmath2/fe ] = 0 dex . this slope is more defined by [ @xmath2/fe ] than age , hence the sgr dsph population should have at least slightly sub - solar [ @xmath2/fe ] . however , this result depends strongly on the accuracy of the isochrones . finally , the narrowness of the sgr dsph giant branch is surprising , when given the range of metallicities where significant numbers of stars are found spectroscopically ( @xmath81 { { \lower-1.2pt\vbox{\hbox{\rlap{$<$}\lower5pt\vbox{\hbox{$\sim$}}}}}}0 $ ] dex ; @xcite , their figure 8) . based again on the accuracy on isochrones and ages involved , this implies an increase from [ @xmath2/fe ] = 0 dex at [ fe / h ] @xmath1 0.7 dex to [ @xmath2/fe ] = + 0.2 dex by [ fe / h ] @xmath1 0.2 dex . this is a little greater than the spectroscopically - derived values visible in figure 20 of @xcite , but probably within a factor of two of the slope of their abundance - averaged values . in summary , our observations are consistent with the current understanding of the sgr dsph population , though a spectroscopic survey to determine the metallicities and @xmath2-element enhancements of its stars would improve matters further . we anticipate the results of the apache point observatory galactic evolution experiment ( apogee ; @xcite ) will meet these criteria . the best - isolated group of galactic bulge stars is the rgb bump , which we show in figure [ bulgefig ] . the rgb bump appears roughly gaussian in our observations , and peaks near @xmath17 = 12.60 mags by @xmath30 = 0.71 and a gaussian width of @xmath82 = 0.25 mags and @xmath83 = 0.10 mags . we overplot in figure [ bulgefig ] the locations of the rgb bump and horizontal branch in various dartmouth isochrones . we begin here with an isochrone for solar - scaled abundances at [ fe / h ] = 0.5 , 5 gyr old , placed at 8200 pc behind @xmath84 mags of reddening . the extinction of the average bulge star is a little higher than that of m54 . the iras+cobe / dirbe dust maps @xcite show that the bulge - dominated regions of the survey area typically have @xmath85 between 0.13 and 0.23 mag . figure [ bulgefig ] shows that a variety of degeneracies prevent us from getting accurate data about the typical bulge star from this observation alone . however , we can constrain some parameters . strong @xmath2-element enhancements can be ruled out , as these would make the rgb bump too faint . metallicities of [ fe / h ] @xmath5 0.5 dex are ruled out too : partly by the rgb bump luminosity , but also by the horizontal branch tracks , which appear to lie on top of the giant branch . strong helium enhancements would also give rise to bluer horizontal branch models ( not plotted here but see , e.g. , @xcite ) , so this is also disfavoured . populations older than @xmath89 gyr are also disfavoured , as their rgb bumps are not bright enough . closer distances are favoured , which we discuss below . thus the dominant bulge population along this line of sight is best fit by a population of roughly solar abundances which is relatively young ( perhaps @xmath85 gyr or even less ) and closer to us than the galactic centre . it should be noted that we have not taken into account errors within the isochrones here , which are expected to broadly match the @xmath40.1 mag variations in @xmath17-band magnitude noted earlier ( and are expected to be relatively less in ( @xmath31 ) ) . we further note that we only examine the strongest signal from the bulge . populations with fewer stars or a wider spread of parameters will not leave such a strong signal in the colour magnitude diagram . our observations do not provide strong evidence for the fainter red clump seen previously @xcite . this has been attributed to a peanut- or x - shaped bulge with a major axis aligned with our line of sight . given our implication that the bulge population we observe is closer than the galactic centre , we suggest its absence is due to missing the more - distant population due to projection effects . our larger angle from the galactic centre and our positive offset in galactic longitude ( @xmath86 ; cf . @xcite , their figure 5 ) may pass over the nearer component of the bulge , but miss the far component . alternatively , it may indicate an absence of this reputed component at greater distances from the galactic plane . our conclusions on the galactic bulge population would be much better constrained by [ fe / h ] and [ @xmath2/fe ] measurements from spectra of bulge stars in this line of sight . differentiating the distances of the bulge rgb bump stars would also place strong constraints on the population : distance measurements by _ gaia _ are expected to be close to the useful statistical limit for individual stars , but an ensemble of stars in the rgb bump could provide strong constraints on the distance of this population . magnitude diagrams showing the main - sequence turn - off would also be helpful in constraining the age , removing some of the degeneracy among the other parameters . we advocate a spectroscopic survey of bulge stars in this field , combined with deep imaging to probe the main - sequence turn - off . the aforementioned apogee spectroscopic survey is anticipated to provide the necessary spectral coverage , while optical imaging has recently been acquired using the blanco-4 m dark energy camera , which will be presented in a future publication . images of the densest 11 square degrees of the sgr dsph at @xmath18 , @xmath16 and @xmath37 were compiled into a catalogue of 2921920 sources , and cross - matched to other major optical and infrared catalogues in the literature . this catalogue was used to confirm the properties of the sgr dsph and the foreground galactic bulge population . comparison to stellar isochrones confirms that the spheroidal sgr dsph core is dominated by a population with [ fe / h ] @xmath71 1 dex , with a significant anti - correlation of [ @xmath2/fe ] and [ fe / h ] . from the distribution of the galaxy s stars , we find it well represented by an oval shape measuring approximately 4@xmath87 in galactic latitude and 2@xmath87 in galactic longitude . we find some evidence for a metallicity gradient in the galaxy , and place it at 24.3 @xmath4 2.3 kpc using isochrone fitting . although we can not accurately constrain the metallicity distribution of the sgr dsph galaxy s stars , the metal - poor populations make up @xmath529 per cent of the observed bright giants in the galaxy . the rgb bump and helium - burning giant branch clump of the southern edge of the galactic bulge are matched with isochrones . a match is found only for isochrones representing a younger population , or one of particularly high metallicity , high [ @xmath2/fe ] and/or high he / h . we consider a younger population of near - solar composition most likely , possibly with some @xmath2-element enhancement . we do not see evidence for a split rgb bump in the bulge population , though this may be attributable to our line of sight . the ongoing vvv survey ( @xcite ) is mapping the bulge regions with vista in the nir , and will provide complementary information on the sgr populations in the innermost regions . we are thankful to mike irwin , mike read and the staff at the cambridge astronomical survey unit and vista science archive for their help and support in understanding the systematics associated with our data . based on observations made with eso telescopes at the la silla paranal observatory under programme i d 089.d-0113 and data products from the cambridge astronomical survey unit . dm is supported by basal cata center for astrophysics and associated technologies pfb-06 , and by project fondecyt no . we have performed here the simplest algorithm to match sources across images taken in different filters , that of nearest - neighbour matching . the only parameter used here is that of the distance between two detections at which one is prepared to declare that they are the same object . this radius is determined here empirically . we have taken two contiguoud . minniti@xmath88,s tiles observed at the same epoch in the same filter , and concatenated them into a single list of sources . the overlap region between the tiles contains some @xmath18700 duplicated detections . we then attempt to match sources over the entire concatenated list using a range of different maximum radii . we can identify the effectiveness of choosing a particular radius by comparing the number of false positive and false negative matches that we obtain . a proxy for false positives is relatively easy to determine : sources in non - overlapping regions of tiles should , by definition , only be observed once . if two objects are matched within a certain radius in these non - overlapping regions , we can define it as a false positive . this can then be scaled to the area of the overlap regions to estimate how many similar false positive matches we can expect in the overlapping areas . false negatives are a little more difficult to determine analytically , but can be approximated as the difference between the number of matches gained at that radius and the number gained at an arbitrarily large ( @xmath82@xmath19 ) radius after correcting for the number of false positives . the number of matches is little affected by choice of radius , changing by @xmath891% between a separation of 0@xmath205 and 1@xmath204 . we have adopted a conservative 0@xmath208 when matching sources . we estimate that @xmath890.07% of these matches will result in incorrect matching of detections between filters , though we may wrongly identify @xmath10.4% of detections as two different objects in this way . we have adopted a more - liberal 1@xmath202 when merging different tiles . we estimate that @xmath10.11% of sources merged in this way are actually two separate detections , while @xmath10.16% of objects should be merged but were not . while all percentages here are approximate , we note that these choices of radii were taken to make our catalogue more robust against poor photometry . we have also examined the effectiveness of target identification by comparing our catalogue to that of 2mass in a typical 2.25@xmath87 @xmath90 1.30@xmath87 region , away from the edges of the observed field , regions affected by a poorly - functioning vista chip , and m54 , but including regions of overlapping tiles . of the 102612 2mass sources with photometric qualities aaa through ccc , we recover 102051 ( 99.45% ) . of the missing 561 sources , all but 62 have a quality flag of c in the 2mass @xmath17 band , and only 43 have a 2mass quality rating of aaa . examination of these missing sources show that they are mostly detected , but outwith our 1@xmath202 matching radius . the difference in astrometry is likely to be due to blending by brighter objects in some cases , poor data quality in others , with a few nearby objects also affected by high proper motion . to photometrically place our observations on the 2mass zero - point system , we performed the following steps . first , we convolved the full grid of bt - settl model atmospheres @xcite with the 2mass and vista filter transmissions , computing the flux expected for each model in each filter . we use the [ fe / h ] = 0 dex , 10000 k , log @xmath91 = 4.0 dex model as a substitute for vega . models with @xmath92 = 0.3 to 0.4 mag were chosen for calibration for three reasons : ( 1 ) there is little variation in colour with metallicity or gravity for stars bluer than this colour ; ( 2 ) there is little slope in the colour - correction term over this colour range for giant stars ; and ( 3 ) there are a sufficiently - large number of observed stars in this colour range . we found the following linear relationships over this range : @xmath93 these relationships are valid for unextincted stars . the colour excess @xmath94 is roughly half of @xmath85 @xcite , though this can for different values of @xmath95 . the extinction towards the core of the sgr dsph , m54 , is given by @xcite as @xmath25 mag , thus we assume @xmath26 mag . we estimate that @xmath96 is roughly a third of @xmath85 . while the amount of extinction does vary across the observed region by typically @xmath97 mag @xcite , this change is relatively small and can be neglected , imparting an error to our zero - point magnitudes of @xmath98 times the gradients in eq . ( [ transeq ] ) . given @xmath99 mag , we therefore compare with stars observed by 2mass to have 0.38 @xmath100 0.48 mag . there are around 1000 such stars between @xmath101 mag per vista tile , with an average colour of @xmath102 mag in each case . most of these stars are giants in the galactic bulge . for each tile , we compute the magnitude difference between the 2mass and vista systems on a per - tile basis , and offset them appropriately using the transformation factors listed in eq . ( [ transeq ] ) . on this basis , we believe our zero points are accurate to approximately 23 , 1.5 and 1 millimagnitudes for the @xmath18 , @xmath16 and @xmath17 bands , respectively , not including small errors arising from uncertainty in filter transmission curves and instantaneous variations due to weather , etc . we finally correct our @xmath16- and @xmath17-band data to the 2mass system using the following approximations : @xmath103 this transformation should be accurate to within 10 millimagnitudes for 0.2 @xmath5 @xmath30 @xmath5 0.9 mag for almost all stars , with the possible exception of very cool oxygen- or carbon - rich stars . figure [ difffig ] shows the remaining differences between the vista and 2mass photometry , once all photometric calibrations have taken place . the data here are limited to those sources with an aaa data quality flag in 2mass and a mean difference of @xmath890.2 magnitudes in each band ( to avoid erroneously including variable and saturated stars ) . a clear variation of a few hundredths of magnitudes can be seen , causing vertical stripes in the offset maps . as both the 2mass and vista surveys have a near - north south alignment , choosing the survey which contributes most to these systematic errors is not trivial . the stripes do appear to correspond to the vista tile width of 1@xmath1045 ( cf . the 2mass tile width is 8@xmath1055 ) . examination of the vista data and data processing revealed no obvious errors indicating any issues in the vista data ( m. irwin , private communication ) . m. l. , van loon j. t. , mcdonald i. , gordon k. d. , babler b. , block m. , bracker s. , engelbracht c. , hora j. , indebetouw r. , meade m. , meixner m. , misselt k. , sewilo m. , shiao b. , whitney b. , 2010 , apj , 711 , l99 r. m. , skrutskie m. f. , van dyk s. , beichman c. a. , carpenter j. m. , chester t. , cambrsy l. , evans t. , fowler j. , gizis j. , howard e. , huchra j. , et al . , 2003 , 2mass all sky catalog of point sources . the irsa 2mass all - sky point source catalog , nasa / ipac infrared science archive . r. m. , wright e. l. , conrow t. , bauer j. , benford d. , brandenburg h. , dailey j. , eisenhardt p. r. m. , evans t. , fajardo - acosta s. , fowler j. , gelino c. , grillmair c. , harbut m. , hoffman d. , jarrett t. , kirkpatrick j. d. , leisawitz d. , liu w. , mainzer a. , marsh k. , masci f. , mccallon h. , padgett d. , ressler m. e. , royer d. , skrutskie m. f. , stanford s. a. , wyatt p. l. , tholen d. , tsai c. w. , wachter s. , wheelock s. l. , yan l. , alles r. , beck r. , grav t. , masiero j. , mccollum b. , mcgehee p. , papin m. , wittman m. , 2012 , explanatory supplement to the wise all - sky data release products . tech . rep . s. e. , belokurov v. , evans n. w. , gilmore g. , gieles m. , irwin m. j. , lewis g. f. , niederste - ostholt m. , pearrubia j. , smith m. c. , bizyaev d. , malanushenko e. , malanushenko v. , schneider d. p. , wyse r. f. g. , 2012 , apj , 750 , 80 d. , lucas p. w. , emerson j. p. , saito r. k. , hempel m. , pietrukowicz p. , ahumada a. v. , alonso m. v. , alonso - garcia j. , arias j. i. , bandyopadhyay r. m. , barb r. h. , barbuy b. , bedin l. r. , bica e. , borissova j. , bronfman l. , carraro g. , catelan m. , clari j. j. , cross n. , de grijs r. , dkny i. , drew j. e. , faria c. , feinstein c. , fernndez lajs e. , gamen r. c. , geisler d. , gieren w. , goldman b. , gonzalez o. a. , gunthardt g. , gurovich s. , hambly n. c. , irwin m. j. , ivanov v. d. , jordn a. , kerins e. , kinemuchi k. , kurtev r. , lpez - corredoira m. , maccarone t. , masetti n. , merlo d. , messineo m. , mirabel i. f. , monaco l. , morelli l. , padilla n. , palma t. , parisi m. c. , pignata g. , rejkuba m. , roman - lopes a. , sale s. e. , schreiber m. r. , schrder a. c. , smith m. , sodr jr . l. , soto m. , tamura m. , tappert c. , thompson m. a. , toledo i. , zoccali m. , pietrzynski g. , 2010 , new astronomy , 15 , 433 r. k. , hempel m. , minniti d. , lucas p. w. , rejkuba m. , toledo i. , gonzalez o. a. , alonso - garca j. , irwin m. j. , gonzalez - solares e. , hodgkin s. t. , lewis j. r. , cross n. , ivanov v. d. , kerins e. , emerson j. p. , soto m. , amres e. b. , gurovich s. , dkny i. , angeloni r. , beamin j. c. , catelan m. , padilla n. , zoccali m. , pietrukowicz p. , moni bidin c. , mauro f. , geisler d. , folkes s. l. , sale s. e. , borissova j. , kurtev r. , ahumada a. v. , alonso m. v. , adamson a. , arias j. i. , bandyopadhyay r. m. , barb r. h. , barbuy b. , baume g. l. , bedin l. r. , bellini a. , benjamin r. , bica e. , bonatto c. , bronfman l. , carraro g. , chen a. n. , clari j. j. , clarke j. r. a. , contreras c. , corvilln a. , de grijs r. , dias b. , drew j. e. , faria c. , feinstein c. , fernndez - lajs e. , gamen r. c. , gieren w. , goldman b. , gonzlez - fernndez c. , grand r. j. j. , gunthardt g. , hambly n. c. , hanson m. m. , heminiak k. g. , hoare m. g. , huckvale l. , jordn a. , kinemuchi k. , longmore a. , lpez - corredoira m. , maccarone t. , majaess d. , martn e. l. , masetti n. , mennickent r. e. , mirabel i. f. , monaco l. , morelli l. , motta v. , palma t. , parisi m. c. , parker q. , pealoza f. , pietrzyski g. , pignata g. , popescu b. , read m. a. , rojas a. , roman - lopes a. , ruiz m. t. , saviane i. , schreiber m. r. , schrder a. c. , sharma s. , smith m. d. , sodr l. , stead j. , stephens a. w. , tamura m. , tappert c. , thompson m. a. , valenti e. , vanzi l. , walton n. a. , weidmann w. , zijlstra a. , 2012 , a&a , 537 , a107 m. h. , majewski s. r. , law d. r. , sarajedini a. , dotter a. , marn - franch a. , chaboyer b. , anderson j. , aparicio a. , bedin l. r. , hempel m. , milone a. , paust n. , piotto g. , reid i. n. , rosenberg a. , 2011 , apj , 743 , 20 c. t. , bell e. f. , schlafly e. f. , juri m. , martin n. f. , rix h .- w . , bernard e. j. , burgett w. s. , chambers k. c. , finkbeiner d. p. , goldman b. , kaiser n. , magnier e. a. , morganson e. p. , price p. a. , tonry j. l. , 2013 , apj , 762 , 6 g. , johnson j. a. , frinchaboy p. m. , majewski s. r. , nidever d. l. , rocha pinto h. j. , girardi l. , andrews b. , chojnowski s. d. , cudworth k. m. , jackson k. , munn j. , skrutskie m. f. , beaton r. l. , blake c. h. , covey k. , deshpande r. , epstein c. , fabbian d. , fleming s. w. , garcia hernandez a. , herrero a. , mahadevan s. , meszaros s. , schultheis m. , sellgren k. , terrien r. , van saders j. , allende prieto c. , bizyaev d. , burton a. , cunha k. , da costa l. n. , hasselquist s. , hearty f. , holtzman j. , garcia perez a. e. , maia m. a. g. , oconnell r. w. , odonnell c. , pinsonneault m. , santiago b. x. , schiavon r. p. , shetrone m. , smith v. , wilson j. c. , 2013 , arxiv e - prints

we introduce a generalized rayleigh - quotient on the tensor product of grassmannians enabling a unified approach to well - known optimization tasks from different areas of numerical linear algebra , such as best low - rank approximations of tensors ( data compression ) , geometric measures of entanglement ( quantum computing ) and subspace clustering ( image processing ) . we briefly discuss the geometry of the constraint set , we compute the riemannian gradient of , we characterize its critical points and prove that they are generically non - degenerated . moreover , we derive an explicit necessary condition for the non - degeneracy of the hessian . finally , we present two intrinsic methods for optimizing a newton - like and a conjugated gradient and compare our algorithms tailored to the above - mentioned applications with established ones from the literature . riemannian optimization , grassmann manifold , multilinear rank , best approximation of tensors , subspace clustering , entanglement measure , newton method , conjugated gradient method . 14m15 , 15a69 , 65d19 , 65f99 , 65k10 , 81p68

1005.4854

the present paper addresses a constrained optimization problem , subsuming and extending optimization tasks which arise in various areas of applications such as ( i ) low - rank tensor approximation problems from signal processing and data compression , ( ii ) geometric measures of pure state entanglement from quantum computing , ( iii ) subspace reconstruction problems from image processing and ( iv ) combinatorial problems . the problem can be stated as follows : given a collection of integer pairs @xmath2 with @xmath3 for @xmath4 and a hermitian @xmath5 matrix @xmath6 with @xmath7 , find the global maximizer of the trace function @xmath8 . here , @xmath9 is restricted to the set of all hermitian projectors @xmath10 of rank @xmath11 , which can be represented as a tensor product @xmath12 of hermitian projectors @xmath13 of rank @xmath14 . thus , one is faced with the constrained optimization task @xmath15 where @xmath1 denotes the set of all hermitian projectors of the above tensor type and @xmath16 is a shortcut for @xmath17 . we will see that it makes sense to call the above objective function @xmath18 the _ generalized rayleigh - quotient _ of @xmath6 with respect to the partitioning @xmath16 . to the best of the authors knowledge , problem ( [ grq1 ] ) has not been discussed in the literature in this general setting . however , depending on the structure of @xmath6 as well as on the choice of @xmath16 , problem ( [ grq1 ] ) relates to well - known numerical linear algebra issues : \(i ) for hermitian matrices of rank-@xmath19 , i.e. @xmath20 , it reduces to a best low - rank approximation problem for the tensor @xmath21 which satisfies @xmath22 , cf . @xcite . classical application areas of such low - rank approximations can be found in statistics , signal processing and data compression @xcite . \(ii ) a recent application in quantum computing plays a central role in characterizing and quantifying pure state entanglement . here , the distance of a pure state ( tensor ) to the set of all product states ( rank-@xmath19 tensors ) provides a geometric measure for entanglement @xcite . \(iii ) moreover , the challenging task of recovering subspaces of possibly different dimensions from noisy data known as subspace detection or subspace clustering problem in computer vision and image processing @xcite can also be cast into the above setting . more precisely , for an appropriately chosen hermitian matrix @xmath6 the subspace clustering task can be characterized by problem ( [ grq1 ] ) in the sense that for unperturbed data the global minima of the generalized rayleigh - quotient are in unique correspondence with the sought subspaces . numerical experiments in section 4 support that even for noisy data the proposed optimization yields reliable approximations of the unperturbed subspaces . \(iv ) in @xcite a certain class of combinatorial problems are recast as optimization problems for trace functions on the special unitary group . for the case when @xmath6 is a diagonal matrix , optimization task is a generalization of the applications mentioned in @xcite . our solution to problem ( @xmath23 ) is based on the fact that the constraint set @xmath24 can be equipped with a riemannian submanifold structure . this admits the use of techniques from riemannian optimization a rather new approach towards constrained optimization exploiting the geometrical structure of the constraint set in order to develop numerical algorithms @xcite . in particular , we pursue two approaches : a newton and a conjugated gradient method . on a riemannian manifold , the intrinsic newton method is usually described by means of the levi - civita connection , performing iterations along geodesics , see @xcite . a more general approach via local coordinates was initiated by shub in @xcite and further discussed in @xcite . here , we follow the ideas in @xcite and use a pair of local parametrizations normal coordinates for the push - forward and qr - type coordinates for the pull - back satisfying an additional compatibility condition to preserve quadratic convergence . thus we obtain an intrinsically defined version of the classical newton algorithm with some computational flexibility . nevertheless , for high - dimensional problems its iterations are expensive , both in terms of computational complexity and memory requirements . therefore , we alternatively propose a conjugated gradient method , which has the advantage of algorithmic simplicity at a satisfactory convergence rate . in doing so , we suggest to replace the global line - search of the classical conjugated gradient method by a one - dimensional newton - step , which yields a better convergence behavior near stationary points than the commonly used armijo - rule . as mentioned earlier , depending on the structure of @xmath6 , the above - specified problems ( i ) , ( ii ) , ( iii ) and ( iv ) are particular cases of the optimization task ( @xmath23 ) . for the best low - rank approximation of a tensor the standard numerical approach is an alternating least - squares algorithm , known as higher - order orthogonal iteration ( hooi ) @xcite . recently , several new methods also exploiting the geometric structure of the problem have been published . newton algorithms have been proposed in @xcite , quasi - newton methods in @xcite , conjugated gradient and trust region methods in @xcite . for high - dimensional tensors , all riemannian newton algorithms manifest similar problems : too high computational complexity and memory requirements . our conjugated gradient method is however , a good candidate to solve large scale problems . it exhibits locally a good convergence behavior , comparable to that of the quasi - newton methods in @xcite at much lower computational costs , which considerably reduces the necessary cpu time . for the problem of estimating a mixture of linear subspaces from sampled data points , cf . ( iii ) , our numerical approach is an efficient alternative to the classical ones : _ ad - hoc _ type methods such as k - subspace algorithms @xcite , or probabilistic methods using a maximum likelihood framework for the estimation @xcite . the paper is organized as follows . in section 2 , we familiarize the reader with the basic ingredients of riemannian optimization . in particular , we address the following topics : the riemannian submanifold structure of the constraint set @xmath1 , its isometry to the @xmath25-fold cartesian product of grassmannians , geodesics and parallel transport and the computation of the intrinsic gradient and hessian for smooth objective functions . section 3 is dedicated to the problem of optimizing the generalized rayleigh - quotient @xmath0 , including also a detailed discussion on its relation to problems ( i ) , ( ii ) , ( iii ) and ( iv ) . moreover , an analogy to the classical rayleigh - quotient is also the subject of this section . we compute the gradient and the hessian of the generalized rayleigh - quotient and derive critical point conditions . we end the section with a result on the generic non - degeneracy of its critical points . in section 4 , a newton - like and a conjugated gradient algorithm as well as numerical simulations tailored to the previously mentioned applications are given . we start our study on the optimization task ( @xmath23 ) with a brief summary on the necessary notations and basic concepts . let @xmath26 be the set of all hermitian @xmath27 matrices @xmath6 , i.e. @xmath28 , where @xmath29 refers to the conjugate transpose of @xmath30 moreover , let @xmath31 be the lie group of all special unitary matrices and @xmath32 its lie - algebra , i.e. @xmath33 if and only if @xmath34 and , respectively , @xmath35 if and only if @xmath36 and @xmath37 the _ grassmannian _ , @xmath38 is the set of all rank @xmath39 hermitian projection operators of @xmath40 . it is a smooth and compact submanifold of @xmath26 with real dimension @xmath41 , whose tangent space at @xmath42 is given by @xmath43:=p\omega-\omega p\;|\;\omega\in\mathfrak{su}_{n}\},\ ] ] cf . hence , every element @xmath44 and every tangent vector @xmath45 can be written as @xmath46 where @xmath47 is the _ standard projector _ of rank @xmath39 acting on @xmath40 and @xmath48 denotes a tangent vector in the corresponding tangent space , i.e. @xmath49 , & \zeta_{m , n}=\left[\begin{array}{cc } 0 & z\\ z^{\dagger } & 0 \end{array}\right],\;\ ; z\in{{\mathbb c}}^{m\times ( n - m)}. \end{array}\ ] ] whenever the values of @xmath39 and @xmath50 are clear from the context , we will use the shortcuts @xmath51 and @xmath52 . with respect to the riemannian metric induced by the frobenius inner product of @xmath26 , the grassmannian @xmath53 is a riemannian submanifold and the unique orthogonal projector onto @xmath54 is given by @xmath55,\;\;\ ; x\in\mathfrak{her}_n.\ ] ] we define the _ @xmath56fold tensor product of grassmannians _ @xmath57 as the set @xmath58 of all rank-@xmath59 hermitian projectors of @xmath60 with @xmath11 and @xmath61 , which can be represented as a kronecker product @xmath62 here , @xmath16 stands for the multi index @xmath63 then , @xmath64 can be naturally equipped with a submanifold structure as the following result shows . the @xmath56fold tensor product of grassmannians @xmath64 is a smooth and compact submanifold of @xmath65 of real dimension @xmath66 . we consider the following smooth action @xmath67 of the compact lie group @xmath68 let @xmath69 be of the form @xmath70 where @xmath71 denotes the standard projector in @xmath72 then , the orbit @xmath73 of @xmath74 coincides with @xmath64 . by @xcite ( pp . 4446 ) we conclude that the @xmath56fold tensor product of grassmannians is a smooth and compact submanifold of @xmath65 . moreover , @xmath75 , where the stabilizer subgroup of @xmath74 is given by @xmath76 & = \{{\boldsymbol \theta}\in\mathrm{su}({\bf n})\;|\ ; \theta_j\pi_j\theta_j^{\dagger}=\pi_j,\ ; j=1,\dots , r\}. \end{array}\ ] ] it follows easily that the dimension of @xmath77 is @xmath78 $ ] and therefore , @xmath79 is the dimension of the @xmath56fold tensor product of grassmannians . @xmath80 let @xmath81 denote the tensor product of finite dimensional vector spaces @xmath82 and @xmath83 , cf . @xcite and let @xmath84 be the tensor product of @xmath85 and @xmath86 given by @xmath87 for all @xmath88 and @xmath89 moreover , let @xmath90 and @xmath91 be bases of @xmath82 and @xmath83 , respectively . then , the matrix representation of @xmath92 with respect to the product basis @xmath93 of @xmath81 is given by the kronecker product of the matrix representations of @xmath6 and @xmath94 with respect to @xmath90 and @xmath91 . this clarifies the relation between the `` abstract '' tensor product of linear maps and the kronecker product of matrices and justifies the term `` tensor product '' of grassmannians when we refer to @xmath1 . + @xmath95 it is a well - known fact that the grassmannian @xmath53 is diffeomorphic to the grassmann manifold @xmath96 of all @xmath97dimensional subspaces of @xmath40 , cf.@xcite . therefore , @xmath98 is diffeomorphic to @xmath99 where @xmath100 and @xmath101 + both items ( a ) and ( b ) readily generalize to an arbitrary number of grassmannians . we conclude this subsection by pointing out an isometry between the @xmath56fold tensor product of grassmannians @xmath1 and the _ direct @xmath56fold product of grassmannians _ @xmath102 the vector spaces @xmath65 and @xmath103 endowed with the inner products @xmath104 and @xmath105 induce a riemannian submanifold structure on @xmath106 and @xmath107 , respectively . [ diffeo ] the map @xmath108is a diffeomorphism between @xmath109 and @xmath110 moreover , @xmath111 is a global riemannian isometry when the right - hand side of ( [ innerprod3 ] ) is replaced by @xmath112 with @xmath113 for @xmath114 note that the isometry between @xmath115 and @xmath1 is very special , as in general the map @xmath116 fails even to be injective . for the proof of proposition @xmath117 we refer to the appendix . it is well - known that every riemannian manifold @xmath118 carries a unique _ riemannian _ or _ levi - civita connection _ @xmath119 , e.g. @xcite . by means of @xmath119 , one defines parallel transport and geodesics as follows . let @xmath120 be a vector field along a curve @xmath121 on @xmath118 , i.e. @xmath122 for all @xmath123 . then , @xmath124 is defined to be _ parallel along _ @xmath121 if @xmath125 for all @xmath126 . given @xmath127 , there exists a unique parallel vector field @xmath124 along @xmath121 such that @xmath128 and the vector @xmath129 is called the _ parallel transport _ of @xmath130 to @xmath131 along @xmath132 in particular , @xmath121 is called a _ geodesic _ on @xmath118 , if @xmath133 is parallel along @xmath121 . for the grassmann manifold @xmath53 , the curve @xmath134}pe^{t[\xi , p]}$ ] describes the geodesic through @xmath44 in direction @xmath135 , i.e. @xmath136 satisfies equation with initial conditions @xmath137 and @xmath138 . similarly , it can be verified that the parallel transport of @xmath139 to @xmath140 along the geodesic through @xmath42 in direction @xmath130 is given by @xmath141}\eta e^{t[\xi , p]}$ ] . these notions can be straight - forward generalized to the direct product of grassmannians @xmath115 . first , let us recall that the _ riemannian gradient _ at @xmath142 of a smooth objective function @xmath143 on a riemannian manifold @xmath118 is defined as the unique tangent vector @xmath144 satisfying @xmath145 @xmath146 , where @xmath147 denotes the differential ( tangent map ) of @xmath148 at @xmath42 . moreover , if @xmath119 is the _ levi - civita connection _ on @xmath118 , then the _ riemannian hessian _ of @xmath148 at @xmath42 is the linear map @xmath149 defined by @xmath150 for all @xmath151 now , if @xmath118 is a submanifold of a vector space @xmath82 , then and simplify as follows . let @xmath152 and @xmath153 be smooth extensions of @xmath148 and of the vector field @xmath154 , respectively . then , @xmath155 where @xmath156 is the orthogonal projection onto @xmath157 and @xmath158 denotes the standard gradient of @xmath152 on @xmath82 . for the generalized rayleigh - quotient @xmath0 on @xmath115 , explicit formulas of the gradient and hessian will be given in section 3.3 . let @xmath1 be the @xmath56fold tensor product of grassmannians with @xmath16 as in ( @xmath159 ) and let @xmath160 @xmath161 in the following , we analyze the constrained optimization problem @xmath162 which comprises problems from different areas , such as _ multilinear low - rank approximations of a tensor _ , _ geometric measures of entanglement _ , _ subspace clustering _ and _ combinatorial optimization_. these applications are naturally stated on a tensor product space . however , for the special case of the grassmann manifold they can be reformulated on a direct product space . to this purpose , we define the _ generalized rayleigh - quotient _ of the matrix @xmath6 as @xmath163 based on the isometry between @xmath1 and @xmath164 we can rewrite problem ( @xmath165 ) as an optimization task for @xmath0 @xmath166 in general this is not the case , as we have already pointed out in . the term generalized rayleigh - quotient " is justified , since for @xmath167 we obtain the classical rayleigh - quotient @xmath168 in the sequel we want to point out some similarities and differences between the generalized and the classical rayleigh - quotient . it is well known that under the assumption that there is a spectral gap between the eigenvalues of @xmath169 , there is a unique maximizer and a unique minimizer of the classical rayleigh - quotient of @xmath6 . unfortunately , this is no longer the case for the generalized rayleigh - quotient @xmath0 . global maximizers and global minimizers exist since the generalized rayleigh - quotient is defined on a compact manifold , but unlike the classical case , it admits also local extrema as the following example shows . for the case when @xmath6 is of rank one we refer to example 3 in @xcite . [ ex ] let @xmath170 be a diagonal matrix with @xmath171 and @xmath172 of the form @xmath173 & \textrm{and } \;\ ; p_2^*=\left[\begin{array}{cc } 0 & 0\\ 0 & 1 \end{array}\right ] . \end{array}\ ] ] the maximum of @xmath0 is obvious less or equal to @xmath174 . since @xmath175 , we have @xmath176 as the global maximizer of @xmath0 . from it follows that all @xmath177 with @xmath178 and @xmath179 diagonal , are critical points of @xmath0 . in particular @xmath180 is a critical point of @xmath0 with @xmath181 . moreover , one can check by computing the hessian of @xmath0 at @xmath180 , see , that @xmath180 is actually a local maximizer of @xmath0 . comparative to the classical rayleigh - quotient , this strange behavior results from the fact that not all @xmath182 permutation matrices are of the form @xmath183 , with @xmath184 . while for the classical rayleigh - quotient one knows that the maximizer and minimizer are orthogonal projectors onto the space spanned by the eigenvectors corresponding to the largest and , respectively , smallest eigenvalues of @xmath6 , it is difficult to provide an analog characterization for the global extrema of the generalized rayleigh - quotient for an arbitrary matrix @xmath6 . but , for particular @xmath6 and @xmath25 such a characterization is possible . + * ( a ) * if @xmath185 and @xmath6 is of rank one , i.e. @xmath186 , with @xmath187 , then the generalized rayleigh - quotient can be rewritten as @xmath188 = { \operatorname{tr}}(y^{\dagger}p_1yp_2).\ ] ] under the assumption that @xmath189 has full rank and distinct singular values there exist one maximizer and one minimizer . the maximizer @xmath190 of @xmath0 is given by the orthogonal projectors onto the space spanned by the @xmath191 left , respective right singular vectors corresponding to the largest @xmath192 singular values . similar for the minimizer , the singular vectors corresponding to the smallest @xmath192 singular values . + * ( b ) * if @xmath25 is arbitrary and @xmath6 diagonalizable via a transformation of @xmath193 , then we can assume without loss of generality that @xmath6 is diagonal . moreover , if @xmath6 can be written as @xmath194 diagonal , which is always the case when @xmath195 , @xmath196 , then the generalized rayleigh - quotient becomes a product of @xmath25 decoupled classical rayleigh - quotients . hence , there is one maximizer and one minimizer . however , there is a dramatic change if @xmath6 can not be written as a kronecker product of diagonal matrices . in this case @xmath0 has also local extrema , as example [ ex ] shows . from one can immediately formulate the following critical point characterization . [ prop_diag ] let @xmath197 be diagonal . then , @xmath198 is a critical point of @xmath0 if and only if @xmath199 are permutations of the standard projectors @xmath71 , for all @xmath200 . there is a wide range of applications for problem ( @xmath201 ) in areas such as signal processing , computer vision and quantum information . we briefly illustrate the broad potential of ( @xmath201 ) by four examples . the problem of best approximation of a tensor by a tensor of lower rank is important in areas such as statistics , signal processing and pattern recognition . unlike in the matrix case , there are several rank concepts for a higher order tensor , @xcite . for the scope of this paper , we focus on the _ multilinear rank _ case . a finite dimensional complex tensor @xmath203 of order @xmath25 is an element of a tensor product @xmath204 , where @xmath205 are complex vector spaces with @xmath206 such an element can have various representations , a common one is the description as an @xmath56way array , i.e. after a choice of bases for @xmath205 , the tensor @xmath203 is identified with @xmath207_{i_1=1 , \dots , i_r=1}^{n_1 , \dots , n_r}\in{{\mathbb c}}^{n_1\times n_2\times\dots\times n_r}$ ] , see e.g. @xcite . the @xmath208th way of the array is referred to as the @xmath208th _ mode _ of @xmath203 . a matrix @xmath209 acts on a tensor @xmath21 via _ mode@xmath210 multiplication _ @xmath211 , i.e. @xmath212 cf . @xcite . it is always possible to rearrange the elements of @xmath203 along one or , more general , several modes such that they form a matrix . let @xmath213 and @xmath214 be ordered subsets of @xmath215 such that @xmath216 . moreover , consider the products @xmath217 for @xmath218 and @xmath219 , respectively . then , the _ matrix unfolding _ of @xmath203 along @xmath220 is a matrix @xmath221 of size @xmath222 such that the element in position @xmath223 of @xmath203 moves to position @xmath224 in @xmath221 , where @xmath225 as an example , for a third order tensor @xmath226 we obtain the following matrix unfoldings as in @xcite @xmath227 , & a_{(2)}=\left[\begin{array}{cccc } a_{111 } & a_{112 } & a_{211 } & a_{212}\\ a_{121 } & a_{122 } & a_{221 } & a_{222 } \end{array } \right ] , \end{array}\\[3 mm ] a_{(3)}=\left[\begin{array}{cccc } a_{111 } & a_{121 } & a_{211 } & a_{221}\\ a_{112 } & a_{122 } & a_{212 } & a_{222 } \end{array } \right ] . \end{array}\ ] ] the _ multilinear rank _ of @xmath228 is the @xmath56tuple @xmath229 such that @xmath230 to refer to the multilinear rank of @xmath203 we will use the notation rank-@xmath202 or @xmath231 given a tensor @xmath232 we are interested in finding the best rank-@xmath233 approximation of @xmath203 , i.e. @xmath234 here , @xmath235 is the frobenius norm of a tensor , i.e. @xmath236 with @xmath237 here , @xmath238 refers to the matrix unfolding @xmath239 . in the matrix case , the solution of the optimization problem ( @xmath240 ) is given by a truncated svd , cf . eckart - young theorem @xcite . however , for the higher - order case , there is no equivalent of the eckart - young theorem . according to the tucker decomposition @xcite or the higher order singular value decomposition ( hosvd ) @xcite , any rank-@xmath202 tensor can be written as a product of a _ core _ tensor @xmath241 and @xmath25 stiefel matrices @xmath242 , i.e. @xmath243 thus , solving ( @xmath240 ) is equivalent to solving the maximization problem @xmath244 with @xmath245 , see e.g. @xcite . using @xmath246operation and kronecker product language , one has @xmath247 according to and the properties of the _ trace _ function , the best multilinear rank-@xmath202 approximation problem becomes @xmath248 with @xmath249 and @xmath250 the task of characterizing and quantifying entanglement is a central theme in quantum information theory . there exist various ways to measure the difference between entangled and product states . here , we discuss a geometric measure of entanglement , which is given by the euclidean distance of @xmath251 with @xmath252 to the set of all product states @xmath253 , i.e. @xmath254 since any minimizer of @xmath255 is also a maximizer of @xmath256 and vice versa , computing the entanglement measure ( [ eq12 ] ) is equivalent to solving @xmath257 with @xmath258 and @xmath259 note that actually constitutes a best rank@xmath260 tensor approximation problem @xcite . subspace segmentation is a fundamental problem in many applications in computer vision ( e.g. image segmentation ) and image processing ( e.g. image representation and compression ) . the problem of clustering data lying on multiple subspaces of different dimensions can be stated as follows : given a set of data points @xmath261 which lie approximately in @xmath262 distinct subspaces @xmath263 of dimension @xmath264 , identify the subspaces @xmath263 without knowing in advance which points belong to which subspace . every @xmath265 dimensional subspace @xmath266 can be defined as the kernel of a rank @xmath267 orthogonal projector @xmath268 of @xmath269 with @xmath270 as @xmath271 therefore , any point @xmath272 satisfies @xmath273 which is equivalent to @xmath274 thus , the problem of recovering the subspaces @xmath263 from the data points @xmath74 can be treated as the following optimization task : @xmath275 with @xmath276 and @xmath277 we mention that here we have used the same notation @xmath107 to refer to the direct @xmath56fold product of _ real _ grassmannians . for best multilinear rank tensor approximation and subspace clustering applications , numerical experiments are presented at the end of section 4 . let @xmath278 be a given array of positive real numbers and let @xmath279 be fixed . find @xmath280 columns and @xmath281 rows such that the sum of the corresponding entries @xmath282 is maximal , i.e. solve the combinatorial maximization problem @xmath283 we can permute @xmath280 columns and @xmath281 rows of @xmath284 by right and left multiplication with permutations of the standard projectors @xmath285 and @xmath286 respectively . hence , problem is solved by finding permutation matrices @xmath287 and @xmath288 which maximize : @xmath289 where @xmath290 is the sum over all entries and @xmath291 . the sum in can be written as @xmath292 where @xmath293 . the last equality in holds since @xmath294 is diagonal , too . according to proposition [ prop_diag ] , we have the following equivalence @xmath295 hence , we can embed the combinatorial maximization problem into our continuous optimization task . the generalization of to @xmath284 being an arbitrary multi - array is straight - forward . problems of this type arise in multi - decision processes such as the following . assume that a company has @xmath296 branches and each branch produces @xmath297 goods . if @xmath282 denotes the gain of the @xmath208th branch with the @xmath298th good , then one could be interested to reduce the number of producers and goods to @xmath280 and @xmath281 , respectively , which give maximum benefit . we continue our investigation of problem ( @xmath201 ) by computing the gradient and the hessian of @xmath0 . in the following lemma we establish multilinear maps @xmath299 , which will enable us to derive clear expressions for the gradient and the hessian of @xmath300 [ lemma1 ] let @xmath169 and @xmath301 then , for all @xmath302 there exists a unique map @xmath303 such that @xmath304 holds for all @xmath305 in particular , one has @xmath306 & = \dots = { \operatorname{tr}}\bigg(\psi_{a , r}(x_1,\dots , x_{r-1 } , i_{n_r } ) x_r\bigg ) . \end{array}\ ] ] moreover , for @xmath307 the maps @xmath299 exhibit the explicit form @xmath308 fix @xmath309 and consider the linear functional @xmath310 by the riesz representation theorem , there exists a unique @xmath311 such that @xmath312 for all @xmath313 . therefore , the map @xmath299 is given by + @xmath314 . it is straightforward to show that @xmath299 is multilinear in @xmath315 . now , choosing @xmath316 and @xmath317 in immediately yields . moreover , follows from the trace equality @xmath318 thus the proof of lemma [ lemma1 ] is complete . the linear maps @xmath299 constructed in the above proof are almost identical to the so - called partial trace operators a well - known concept from multilinear algebra and quantum mechanics @xcite . next , we show how to compute @xmath319 for given @xmath320 if @xmath6 is not a pure tensor product @xmath321 . [ t_d ] let @xmath169 and @xmath320 . then , the @xmath322position of @xmath323 is given by @xmath324 where @xmath325 denotes the standard basis of @xmath326 . let @xmath327 . then , the element in the @xmath224 position of the matrix @xmath328 is given by @xmath329 & = { \operatorname{tr}}\biggl(a(x_1\otimes\cdots\otimes x_j\mathrm{e}_{t}\mathrm{e}_{s}^{\top}\otimes\cdots\otimes x_r ) \biggr)\\[3 mm ] & = { \operatorname{tr}}\biggl(a ( x_1\otimes\cdots\otimes x_r)(i_{n_1}\otimes\cdots\otimes \mathrm{e}_{t}\mathrm{e}_{s}^{\top}\otimes\cdots\otimes i_{n_r } ) \biggr ) . \end{array}\ ] ] hence , follows from the identity @xmath330 . let @xmath169 and @xmath320 . a straightforward consequence of the identity @xmath331 for all @xmath332 , shows that @xmath333 is hermitian . for simplicity of writing , whenever @xmath334 is understood from the context , we use the following shortcut @xmath335 now , we can give an explicit formula for the riemannian gradient of @xmath0 and derive necessary and sufficient critical point conditions . [ critic ] let @xmath336 and let @xmath0 be the generalized rayleigh - quotient on @xmath337 then , one has the following : @xmath338 the gradient of @xmath0 at @xmath42 with respect to the riemannian metric @xmath339 is @xmath340 @xmath341 the critical points of @xmath0 on @xmath115 are characterized by @xmath342 = 0 \ ] ] i.e. @xmath199 , @xmath302 is the orthogonal projector onto an @xmath343dimensional invariant subspace of @xmath344 . @xmath345 fix @xmath346 and let @xmath347 denote the canonical smooth extension of @xmath0 to @xmath348 then , @xmath349 for all @xmath350 from ( [ innerprod3 ] ) , we obtain that the gradient of @xmath347 at @xmath42 is given by @xmath351 thus , according to and ( [ subgrad ] ) , @xmath352 @xmath353 @xmath346 is a critical point of @xmath0 iff @xmath354 . this is equivalent to @xmath355=\displaystyle [ p_j,\widehat{a}_j]p_j , \ ] ] for all @xmath114 by multiplying once from the left with @xmath199 and once from the right with @xmath199 , we obtain that @xmath356 and @xmath357 . hence , the conclusion @xmath358=0 $ ] holds for all @xmath200 . as a consequence of theorem @xmath359 , we immediately obtain the following necessary and sufficient critical point condition . [ critic1 ] let @xmath336 and let @xmath360 be such that @xmath361 where @xmath71 is the standard projector in @xmath72 we write @xmath362 \psi_j'''^{\dagger } & \psi_{j}''\\ \end{array } \right],\ ] ] with @xmath363 and @xmath364 . then @xmath42 is a critical point of @xmath0 if and only if @xmath365 for all @xmath114 for the rest of this section we are concerned with the computation of the riemannian hessian of @xmath0 and also with its non - degeneracy at critical points . [ thhess ] let @xmath169 and @xmath366 . then , the riemannian hessian of @xmath0 at @xmath42 is the unique self - adjoint operator @xmath367 \xi:=(\xi_1,\dots,\xi_r)\mapsto { \bf h}_{\rho_a}(p)(\xi ) : = \big({\bf h}_{1}(\xi),\dots , { \bf h}_{r}(\xi)\big ) , \end{array}\ ] ] defined by @xmath368 where @xmath344 is the shortcut for @xmath369 . let @xmath370 denote a smooth extension of @xmath371 to @xmath372 . according to ( [ grad ] ) , we can choose @xmath373 then , @xmath374 & + \sum\limits_{k=1 , k\neq j}^r \mathrm{ad}^2_{p_j } \psi_{a , j } ( p_1,\dots , i_{n_j},\dots , x_k , \dots , p_r ) , \end{array}\ ] ] for all @xmath375 and @xmath376 in @xmath348 notice that , the derivative of the linear map @xmath377 in direction @xmath378 ( @xmath379 ) is @xmath380 . applying ( [ ad ] ) and ( [ subgrad ] ) , the riemannian hessian of @xmath0 at @xmath381 is given by and . here , we have used the following two facts : \(i ) clearly , @xmath382 is skew - hermitian and hence @xmath383 is in the tangent space @xmath384 for all @xmath385 \(ii ) a straightforward computation shows that @xmath386 is in the orthogonal complement of @xmath384 and hence @xmath387 for all @xmath385 by restricting the tangent vectors @xmath388 to the vectors of the form @xmath389 , it follows immediately a necessary condition for the non - degeneracy of the hessian at local extrema . let @xmath160 and @xmath390 be a local maximizer ( local minimizer ) of @xmath300 if @xmath391 is non - degenerate , then for all @xmath302 the equality @xmath392 holds with @xmath393 and @xmath394 as in @xmath395 . here , @xmath396 denotes the spectrum of @xmath74 . in the case when @xmath169 can be diagonalized by elements in @xmath397 , condition is also sufficient for the nondegeneracy of the hessian of @xmath0 at local extrema . in the remaining part of the section we derive a genericity statement concerning the critical points of the generalized rayleigh - quotient . the result is a straightforward consequence of the parametric transversality theorem @xcite . let @xmath398 , @xmath118 , @xmath399 be smooth manifolds and @xmath400 a smooth map . moreover , let @xmath401 denote the tangent map of @xmath402 at @xmath403 . we say that @xmath402 is _ transversal _ to a submanifold @xmath404 and write @xmath405 if @xmath406 for all @xmath407 . then , the parametric transversality theorem states the following . ( @xcite)[thmtransv ] let @xmath408 be smooth manifolds and @xmath409 a closed submanifold of @xmath399 . let @xmath410 be a smooth map , let @xmath411 and define @xmath412 , @xmath413 . if @xmath405 , then the set @xmath414 is open and dense . now , let @xmath415 be a smooth function depending on a parameter @xmath416 and consider the map @xmath417 where @xmath418 is the cotangent bundle of @xmath118 and @xmath419 denotes the differential of @xmath420 at @xmath142 . with these notations , our genericity result reads as follows . [ f_trans ] let @xmath59 , @xmath82 and @xmath402 be as above and let @xmath409 be the image of the zero section in @xmath418 . if @xmath405 then for a generic @xmath421 the critical points of the smooth function @xmath415 are non - degenerate . fix @xmath411 and define @xmath422 from the transversality theorem [ thmtransv ] it follows that the set @xmath423 is open and dense in @xmath82 if @xmath405 . in the following , we will prove that @xmath424 is equivalent to the fact that the hessian of @xmath420 is non - degenerate in the critical points . this will prove the theorem . first , notice that @xmath425 if and only if @xmath426 is a critical point of @xmath420 . therefore , the transversality condition @xmath427 is equivalent to @xmath428 to rewrite this condition in local coordinates , let @xmath429 be a chart on an open subset @xmath430 around @xmath431 such that @xmath432 and @xmath433 . then define @xmath434 moreover , @xmath111 induces a chart @xmath435 around @xmath436 via @xmath437 here , @xmath438 refers to the natural projection and @xmath439 . thus , for @xmath440 one has @xmath441 . since transversality of @xmath442 to @xmath409 is preserved in local coordinates , is equivalent to @xmath443 then @xmath444 yields that is fulfilled if and only if @xmath445 is nonsingular . finally , the conclusion follows form the identity @xmath446 which is satisfied due to the fact that @xmath431 is a critical point and @xmath433 . here , @xmath447 denotes the hessian form corresponding to the hessian operator via @xmath448 for all @xmath449 . for the generalized rayleigh - quotient , we obtain the following result . [ cor_generic ] the critical points of the generalized rayleigh - quotient are generically non - degenerate . set @xmath450 , @xmath451 . for the simplicity , we will identify the cotangent bundle @xmath418 with the tangent bundle @xmath452 and work with the map @xmath453 instead of , where @xmath454 is the riemannian gradient of @xmath0 at @xmath42 . we will show that @xmath405 , where @xmath409 is now the image of the zero section in @xmath452 , i.e. @xmath455 for all @xmath456 with @xmath457 . as in the proof of theorem [ f_trans ] , we rewrite the transversality condition in local coordinates , i.e. @xmath458 where @xmath459 here , @xmath460 is a chart around @xmath42 with @xmath461 and @xmath433 and @xmath462 is the corresponding induced chart around @xmath463 . with this choice of charts , we obtain @xmath464 where @xmath465 . since @xmath466 is linear , one has @xmath467 thus , condition holds if and only if @xmath468 finally , we will show that @xmath469 which clearly guarantees . let @xmath470 , then we obtain @xmath471 & = { \operatorname{tr}}\biggl(x(\sum\limits_{j=1}^r p_1\otimes \cdots\otimes \xi_j\otimes \cdots\otimes p_r)\biggr ) , \end{split}\ ] ] for all @xmath472 . notice , that the equality @xmath473 follows from @xmath433 . therefore , @xmath474 and this holds if and only if @xmath475 , since alls summands in are orthogonal to each other . thus , we have proved that @xmath476 and hence @xmath405 . from the theorem [ f_trans ] it follows immediately that the critical points of the generalized rayleigh - quotient are generically non - degenerate . unfortunately , for best multilinear rank tensor approximation and subspace clustering problems , we can not conclude from corollary [ cor_generic ] that the critical points of @xmath0 are generically non - degenerate . in these cases , the resulting matrices @xmath6 are restricted to a thin subset of @xmath65 and thus the genericity statement with respect @xmath65 in corollary [ cor_generic ] does not carry over straight - forwardly . exploiting the geometrical structure of the constraint set @xmath164 we develop two numerical methods , a newton - like and a conjugated gradient algorithm , for optimizing the generalized rayleigh - quotient @xmath0 , with @xmath477 the intrinsic riemannian newton algorithm is described by means of the levi - civita connection taking iteration steps along geodesics @xcite . sometimes geodesics are difficult to determine , thus , here we are interested in a more general approach , which introduces the newton iteration via local coordinates , see @xcite . more precisely , we follow the ideas in @xcite and use a pair of local coordinates on @xmath115 , i.e. normal coordinates and qr - coordinates . recall that , a _ _ local parametrization _ _ of @xmath478 around a point @xmath479 is a smooth map @xmath480 satisfying the additional conditions @xmath481 _ riemannian normal coordinates _ are given by the riemannian exponential map @xmath482}p_1e^{[\xi_1,p_1 ] } , \dots , e^{-[\xi_r , p_r]}p_re^{[\xi_r , p_r]}\bigr),\end{array}\ ] ] while _ qr - type coordinates _ are defined by the qr - approximation of the matrix exponential , i.e. @xmath483_q^{\dagger}\;p_1\;[x_1]_q , \dots , [ x_r]_q^{\dagger}\;p_r\;[x_r]_q\bigr ) . \end{array}\ ] ] here @xmath484_q$ ] denotes the @xmath485factor from the unique @xmath486 decomposition of @xmath487.$ ] now , let @xmath488 be a critical point of @xmath0 . choose @xmath489 in a neighborhood of @xmath490 and perform the following newton - like iteration @xmath491 where @xmath492 is a solution of the newton equation @xmath493 replacing the objects in ( @xmath494 ) by their explicit form computed in the previous section , we get the following newton equation : @xmath495 for all @xmath114 as mentioned before , let @xmath496 . solving this system in the embedding space @xmath372 requires a higher number of parameters than necessary . however , exploiting the particular structure of the tangent vectors @xmath497\theta_j^{\dagger } , \ ] ] allows us to solve ( [ sylvester1 ] ) with the minimum number of parameters equal to the dimension of @xmath107 . thus , by multiplying ( @xmath498 ) from the left with @xmath499 and from the right with @xmath500 we obtain an equation in the variables @xmath501 i.e. @xmath502 where the terms @xmath503 and @xmath504 are computed in the following . let @xmath505\in\mathrm{su}_{n_j},\ ] ] where @xmath506 and @xmath507 are @xmath508 and @xmath509 matrices , respectively . then , @xmath510 for expressing @xmath511 with @xmath512 , we introduce the multilinear operators @xmath513 defined in a similar way as @xmath299 by @xmath514 for all @xmath515 and @xmath516 for convenience , we will use the following shortcut @xmath517 [ tabel_n ] furthermore , we partition the matrix @xmath518 into block form @xmath519_{s , t=1}^{n_j},\ ] ] where each @xmath520 is an @xmath521 matrix . then , the linear map @xmath522 is given by @xmath523_{s , t=1}^{n_j } v_j.\ ] ] finally , the complete newton - like algorithm for the optimization of @xmath0 on @xmath115 is given by algorithm 1 . \(a ) for an arbitrary matrix @xmath169 , the computation of @xmath344 and @xmath518 is performed according to formula ( [ tau_delta ] ) . this can be simplified in the case of the applications described in section 3.3 . + * case 1 . * if @xmath524 with @xmath525 @xmath526 , then @xmath527 where @xmath528 and @xmath529 are the @xmath208th mode and respectively @xmath530th mode matrices of the tensors @xmath531 and @xmath532 , respectively . + * case 2 . * if @xmath533 with @xmath534 and @xmath535 , then @xmath536 ( b ) to solve the system ( [ sylvester2 ] ) , one can rewrite it as a linear equation on @xmath537 ( @xmath538 is the dimension of @xmath115 ) using matrix kronecker products and @xmath246operations , then solve this by any linear equation solver . + ( c ) the computation of geodesics on matrix manifolds usually requires the matrix exponential map , which is in general an expensive procedure of order @xmath539 . yet , for the particular case of the grassmann manifold @xmath53 , gallivan et.al . @xcite have developed an efficient method to compute the matrix exponential , reducing the complexity order to @xmath540 ( @xmath541 ) . our approach , however , is based on a first order approximation of the matrix exponential @xmath542}$ ] followed by a qr - decomposition to preserve orthogonality / unitarity . explicitly , it is given by @xmath543_q = w\left [ \begin{array}{ccc } d^{-1 } & \sigma d^{-1 } & 0\\ -\sigma^{\dagger}d^{-1 } & d^{-1 } & 0\\ 0 & 0 & i_{n-2 m } \end{array}\right ] w^{\dagger},\ ] ] where @xmath544 with @xmath545 , @xmath546 and @xmath547 diagonal . furthermore , @xmath548\in\mathrm{su}_{n } , & d:=\sqrt{i_{m}+\sigma^{\dagger}\sigma } , \end{array}\ ] ] where @xmath549\in\mathrm{su}_{n - m}$ ] is an unitary completion of @xmath189 . the computational complexity of this qr - factorization is of order @xmath550 . + ( d ) the convergence of the algorithm is not guaranteed for arbitrary initial conditions @xmath489 and even in the case of convergence the limiting point need not be a local maximizer of the function . to overcome this , one could for example test if the computed direction is ascending , else take the gradient as the new direction . furthermore , one can make an iterative line - search in the ascending direction . in the following theorem we prove that the sequence generated by algorithm 1 converges quadratically to a critical point of the generalized rayleigh - quotient @xmath0 if the sequence starts in a sufficiently small neighborhood of the critical point . let @xmath169 and @xmath551 be a non - degenerate critical point of the generalized rayleigh - quotient @xmath0 , then the sequence generated by the n - like algorithm converges locally quadratically to @xmath490 . for the critical point @xmath552 , the riemannian coordinates ( @xmath553 ) and the qr- coordinates ( @xmath554 ) satisfy the condition @xmath555 . thus , according to theorem 4.1 . from @xcite there exists a neighborhood @xmath556 such that the sequence of iterates generated by the n - like algorithm converges quadratically to @xmath490 when the initial point @xmath42 is in @xmath82 . the quadratic convergence of the newton - like algorithm has the drawback of high computational complexity . solving the newton equation ( [ sylvester2 ] ) yields a cost per iteration of order @xmath557 , where @xmath538 is the dimension of @xmath337 in what follows , we offer as an alternative to reduce the computational costs of the newton - like algorithm by a conjugated gradient method . the linear conjugated gradient ( lcg ) method is used for solving large systems of linear equations with a symmetric positive definite matrix , which is achieved by iteratively minimizing a convex quadratic function @xmath558 . the initial direction @xmath559 is chosen as the steepest descent and every forthcoming direction @xmath560 is required to be conjugated to all the previous ones , i.e. @xmath561 for all @xmath562 . the exact maximum along a direction gives the next iterate . hence , the optimal solution is found in at most @xmath50 steps , where @xmath50 is the dimension of the problem . nonlinear conjugated gradient ( ncg ) methods use the same approach for general functions @xmath563 , not necessarily convex and quadratic . the update in this case reads as @xmath564 where the step - size @xmath565 is obtained by a line search in the direction @xmath538 @xmath566 and @xmath567 is given by one of the formulas : fletcher - reeves , polak - ribiere , hestenes - stiefel , or other . we refer to @xcite for the generalization of the ncg method to a riemannian manifold . for the computation of the step - size along the geodesic in direction @xmath130 , an exact line search as in the classical case is an extremely expensive procedure . therefore , one commonly approximates ( [ stepsize ] ) by an armijo - rule , which ensures at least that the step length decreases the function sufficiently . we , however , have decided to compute the step - size by performing a one - dimensional newton - step along the geodesic , since in the neighborhood of a critical point one newton step can lead very close to the solution . therefore , at @xmath568 the step - size in direction @xmath569 is given by @xmath570 where @xmath571 is the unique geodesic through @xmath42 in direction @xmath130 . let @xmath572 be such that @xmath573 furthermore , let @xmath574 denote the updated point in @xmath575 via the qr - coordinates as in ( [ up ] ) . for the computation of the new direction , a `` transport '' of the old direction @xmath576 from @xmath577 to the tangent space @xmath578 is required . we use the following approximation for the paralle transport of @xmath130 along the geodesic through @xmath42 in direction @xmath130 @xmath579 z_j^{\dagger } & 0 \end{array}\right]\theta_j^{\mathrm{new}^{\dagger } } , & \textrm{where } \theta_j^{\mathrm{new}}=\theta_j\left[\begin{array}{cc } i_{m_j } & -z_j\\[2 mm ] z_j^{\dagger } & i_{n_j - m_j } \end{array}\right]_q , \end{array}\ ] ] for all @xmath114 the complete riemannian conjugated gradient is presented as algorithm 2 . it is recommended to reset the search direction to the steepest descent direction after @xmath538 iterations , i.e. @xmath580 , where @xmath538 refers to the dimension of the manifold . for the maximization of the generalized rayleigh - quotient the initial direction is @xmath581 and the update @xmath582 the convergence properties of the ncg methods are in general difficult to analyze . yet , under moderate supplementary assumptions on the cost function one can guarantee that the ncg converges to a stationary point @xcite . it is expected that the proposed riemannian conjugated gradient method has properties similar to those of the ncg . in this section we run several numerical experiments suitable for the applications mentioned in section 3.2 , i.e. best rank approximation for tensors and subspace clustering , to test the newton - like ( n - like ) and riemannian conjugated gradient ( rcg ) algorithms . the algorithms were implemented in matlab on a personal notebook with 1.8 ghz intel core 2 duo processor . to test the performance of our algorithms we have considered several examples of large size tensors of order @xmath583 and @xmath584 with entries chosen from the standard normal distribution and estimated their best low - rank approximation . we have started with a truncated hosvd ( @xcite ) and performed several hooi iterates before we run our n - like and rcg algorithms . depending on the size of the tensor , the number of hooi iterations necessary to reach the region of attraction of a stationary point @xmath585 , ranges from 10 to 100 . as stopping criterion we have chosen that the relative norm of the gradient @xmath586 is approximately @xmath587 . * computational complexity . * the computational complexity of the n - like method is determined by the computation of the hessian and the solution of the newton equation ( [ sylvester3 ] ) . thus , for the best rank-@xmath588 approximation of a @xmath589 tensor , the computation of the hessian is dominated by tensor - matrix multiplications and is of order @xmath590 . solving the newton equation by gaussian elimination gives a computational complexity of order @xmath591 , i.e. the dimension of the manifold to the power of three . for the computational costs of the rcg method we have to take into discussion only tensor - matrix multiplications , which give a cost per rcg iteration of order @xmath590 . * experimental results and previous work . * the problem of best low - rank tensor approximation has enjoyed a lot of attention recently . apart from the well known higher order orthogonal iterations hooi ( @xcite ) , various algorithms which exploit the manifold structure of the constraint set have been developed . we refer to @xcite for newton methods , to @xcite for quasi - newton methods and to @xcite for conjugated gradient and trust region methods on the grassmann manifold . similar to the newton methods in @xcite , our n - like method converges quadratically to a stationary point of the generalized rayleigh - quotient when starting in its neighborhood . we have compared our algorithms with the existing ones in the literature : quasi - newton with bfgs , riemannian conjugated gradient method which uses the armijo - rule for the computation of the step - size ( cg - armijo ) , and hooi . the algorithms were run on the same platform , identically initialized and with the same stopping criterion . for the bfgs quasi - newton and limited memory quasi - newton ( l - bfgs ) methods we have used the code available in @xcite . [ fig1 ] shows convergence results for two large size tensors @xmath592 and @xmath593 approximated by rank-@xmath594 and rank-@xmath595 tensors , respectively . in fig . [ fig11 ] we plot the convergence behavior of the rcg method for the best rank-@xmath596 approximation of a @xmath597 tensor ( left ) and for the best rank-@xmath598 approximation of a @xmath599 tensor . due to the limited memory space , we were not able to run the n - like and bfgs quasi - newton algorithms for the example on the left . yet it was still possible to run rcg , l - bfgs , cg - armijo and hooi . in table [ tab:1 ] we display the average cpu times necessary to compute a low rank best approximation for tensors of different sizes and orders by n - like , rcg , bfgs and l - bfgs quasi - newton methods . we have performed 100 runs for each example . [ cols="<,^,^,^,^ " , ] [ tab:1 ] * resume . * first we mention that there is no guarantee that the n - like and rcg iterations converge to a local maximizer of the generalized rayleigh - quotient . however , in the examples presented in fig.[fig1 ] and fig.[fig11 ] the limiting points are local maximizers . as the numerical experiments have shown , the n - like method has the advantage of fast convergence . unfortunately , for large scale problems , the n - like algorithm can not be applied , as mentioned before . even when it is possible to apply n - like algorithm , it needs a large amount of time per iteration . as an example , for the best rank-@xmath596 of a @xmath600 tensor , one n - like iteration took three minutes . related algorithms which explicitly compute the hessian and solve the newton equation , such as @xcite , and those which approximately solve the newton equation such as the trust region method @xcite , face the same difficulty for large scale problems . on the other hand , the low cost iterations of the rcg method makes it a good candidate to solve large size problems . the convergence rate is comparative to that of the bfgs quasi - newton method in @xcite , but at much lower computational costs . our experiments exhibit the shortest cpu time for the rcg method . in the examples in which the tensor was a small perturbation of a low - rank tensor , the rcg algorithm exhibits quadratic convergence . the experimental setup consists in choosing @xmath25 subspaces in @xmath601 ( @xmath602 and @xmath584 ) and collections of @xmath603 randomly chosen$ ] . ] points on each subspace . then , the sample points are perturbed by adding zero - mean gaussian noise with standard deviation varying from @xmath604 to @xmath605 in the different experiments . now , the goal is to detect the exact subspaces or to approximate them as good as possible . for this purpose , we apply our n - like and rcg algorithms to solve the associated optimization task , cf . section 3.1 . the error between the exact subspaces and the estimated ones is measured as in @xcite , i.e. @xmath606 where @xmath199 is the orthogonal projector corresponding to the exact subspace and @xmath607 the orthogonal projector corresponding to the estimated one . it can be easily checked that in the case of unperturbed data there is a unique non - degenerate minimizer of @xmath0 , and it yields the exact subspaces . thus , we expect that for noisy data the global minimizer still gives a good approximation . since @xmath0 has many local optima , for an arbitrary starting point our algorithms can converge to stationary points which lead to a significant error between the exact subspaces and their approximation . thus , in what follows , we briefly describe a method ( pda , see below ) for computing a suitable initial point which guarantees the convergence of our algorithms towards a good approximation of the exact subspaces in our numerical experiment : the polynomial differential algorithm ( pda ) was proposed in @xcite . it is a purely algebraic method for recovering a finite number of subspaces from a set of data points belonging to the union of these subspaces . from the data set finitely many homogeneous polynomials are computed such that their zero set coincides with the union of the sought subspaces . then , an evaluation of their derivatives at given data points yields successively a basis of the orthogonal complement of subspaces one is interested in . for noisy data , a slightly modified version of pda @xcite yields an approximation of the unperturbed subspaces . this first " approximation turned out to be a good starting point for our iterative algorithms which significantly improved the approximation quality . for each noise level we perform 500 runs of the n - like and local - cg algorithms for different data sets and compute the mean error between the exact subspaces and the computed approximations . as a preliminary step , we normalize all data points , such that no direction is favored . in fig . [ fig2 ] , @xmath608 randomly chosen data points which lie exactly in the union of two @xmath609-dimensional subspaces of @xmath601 ( left ) and their perturbed standard deviation ] images ( right ) are depicted . moreover , the two plots display the exact subspaces ( left ) as well as the ones computed by our n - like algorithm ( right ) . the error between the exact subspaces and our approximation is ca . @xmath610 , whereas the error for the pda approximation is ca . @xmath611 . in fig . [ fig3 ] , we plot the mean error ( left ) for different noise levels and different number of subspaces . we have included also the mean error for the starting point of our algorithms , i.e. for the pda approximation . on the right we demonstrate the fast convergence rate of the n - like and rcg algorithms for the case of @xmath583 and , respectively , @xmath584 subspaces . * resume . * our numerical experiments have proven that ( i ) the minimization task proposed in section 3 is capable to solve subspace detection problems and ( ii ) our numerical algorithms initialized with the pda starting point yield an effective method for computing a reliable approximation of the perturbed subspaces . how the approximation of the perturbed subspaces varies when the noise in the data follows some law of distribution , is the subject of future investigation . here we provide a proof of proposition @xmath117 , which states that there exists a global riemannian isometry @xmath111 between @xmath1 and @xmath337 the surjectivity of @xmath111 is clear from the definition of @xmath1 . to prove the injectivity of @xmath111 we use induction over @xmath25 . choose @xmath612 such that @xmath613 i.e. @xmath614 where @xmath615 and @xmath616 are the entries of @xmath617 and @xmath618 , respectively . + thus it exists @xmath619 such that @xmath620 . since @xmath621 and @xmath622 have only @xmath623 and @xmath19 as eigenvalues it follows that @xmath624 and @xmath625 . therefore , @xmath626 implies that @xmath627 and the procedure can be repeated until we obtain @xmath628 , for all @xmath302 . thus the injectivity of @xmath111 is proven . so @xmath111 is a continuous bijective map with continuous inverse @xmath629 due to the compactness of @xmath115 . moreover , the map @xmath111 is smooth since the components of @xmath630 are polynomial functions . let @xmath479 and @xmath631 . consider the tangent map of @xmath111 at @xmath42 , i.e. @xmath632 ( \xi_1,\dots , \xi_r)\mapsto \displaystyle\sum\limits_{j=1}^r p_1\otimes\dots\otimes \xi_j\otimes \dots\otimes p_r . \end{array}\ ] ] with the inner products ( [ inner1 ] ) and ( [ innerprod3 ] ) defined on @xmath65 and @xmath372 , respectively , one has @xmath633 this implies that the tangent map @xmath634 is a linear isometry . thus , it is invertible and therefore @xmath111 is a local diffeomorphism . moreover , since @xmath111 is bijective it is a global diffeomorphism , giving thus a global riemannian isometry when the metric on @xmath115 is defined by . this work has been supported by the federal ministry of education and research ( bmbf ) through the project fhprofund 2007 : cooperation program between universities of applied science and industry , `` development and implementation of novel mathematical algorithms for identification and control of technical systems '' .

we present results of an intensive two - month campaign of approximately daily spectrophotometric monitoring of the narrow - line seyfert 1 galaxy akn 564 with the _ hubble space telescope _ ( _ hst _ ) . the fractional variability amplitude of the continuum variations between 13653000is% , about a factor 3 less than that found in typical seyfert 1 galaxies over a similar period of time . however , large amplitude , short time - scale flaring behavior is evident , with trough - to - peak flux changes of about 18% in approximately 3 days . we present evidence for wavelength - dependent continuum time delays , with the variations at 3000 lagging behind those at 1365 by about 1 day . these delays may be interpreted as evidence for a stratified continuum reprocessing region , possibly an accretion - disk structure . the emission - line exhibits flux variations of about 1% amplitude . these variations lag those at 1365 by 3 days , and combining this with the line width yields a virial black - hole mass limit of . we caution , the low amplitude variations may indicate the bulk of the emission region is at larger radii . this scenario affects the veracity of our black hole mass upper limit in an uncertain manner due to the unknown nature of the gas velocity field . our mass estimate is thus unreliable , however , it is consistent with the independent estimate of pounds et al . ( 2001 ) , based on a fluctuation power spectrum analysis of x - ray variability in akn 564 . the black - hole mass and 5100 luminosity of akn 564 are consistent with the hypothesis that , relative to seyfert 1 galaxies , nls1s have lower black hole masses and higher accretion rates . other strong emission lines , e.g. , and , are constrained to vary with amplitudes of 5% . this low - level of emission - line variability is different from most seyfert 1 galaxies , which characteristically display variations of% on similar time scales .

astro-ph0107059

narrow - line seyfert 1 ( nls1 ) galaxies were first classified on the basis of their narrow permitted optical emission lines , with full - width - half - maximum ( fwhm ) 2000(osterbrock & pogge 1985 ) . they also exhibit distinctive x - ray properties ( e.g. , puchnarewicz et al . 1992 ; boller et al . 1996 ; brandt , mathur , & elvis 1997 ; turner 1999 ; leighly 1999a , b ) . these include a steep soft excess with photon index @xmath10 ( photon flux @xmath11 ) , a steep hard power law with @xmath12 , and rapid , short time - scale variations ( e.g. , @xmath13 30% in 1500 sec ) . their uv / optical properties place them at one extreme of the boroson & green ( 1992 ) primary eigenvector that has been identified in a principal component analysis ; specifically , nls1 classification correlates with strong optical and weak 5007 emission . nls1s are thus potentially useful in identifying the underlying physics that defines the primary eigenvector . a number of scenarios have been posited to explain nls1 properties : 1 . nls1s may have larger broad - line region ( blr ) sizes compared to seyfert 1 ( s1 ) galaxies . the steeper soft excesses imply higher ionizing fluxes than s1 galaxies with comparable luminosities . under certain conditions this may lead to larger blr sizes and hence smaller permitted line widths , assuming they reflect virialized motions about the putative black hole ( guilbert , fabian , & mccray 1983 ; wandel & boller 1998 ) . 2 . nls1s may be low - inclination @xmath14 , nearly face - on systems ( osterbrock & pogge 1985 ) . the line widths , attributable to orbital , virialized motions in a common plane are decreased by @xmath15 , and the strong soft x - ray fluxes are reconcilable with accretion disk models ( e.g. , madau 1988 ) . however , this orientation scenario has problems explaining the low luminosities often observed ( e.g. , boroson & green 1992 ) . 3 . nls1s may have relatively low black hole masses with higher than normal accretion rates , compared to s1 galaxies ( pounds , done , & osborne 1995 ) . the smaller line widths are attributable to the reduced gravitational potential in which they form , and higher accretion rates result in luminosities that are large for their masses . balmer ( ) blr sizes and masses have been measured for about 40 agns , including five nls1s ( wandel , peterson , & malkan 1999 ; kaspi et al.2000 ; peterson et al . 2000 ; shemmer et al . these results suggest that nls1s have blr sizes comparable to those of s1s with similar optical luminosities . moreover , nls1 and s1 galaxies delineate a broad mass - luminosity relationship with the former sources populating the low mass extremum of this relationship . this is consistent with the hypothesis that nls1s are undermassive black hole systems with higher accretion rates and/or sources viewed nearly face - on . to distinguish between various models requires systematic measurements of the black hole masses @xmath16 , mass accretion rates @xmath17 , broad - line region sizes @xmath18 , and source inclinations @xmath14 of nls1s and other s1s . these key parameters may all be measured directly or inferred through application of echo mapping ( reverberation ) techniques ( blandford & mckee 1982 ; peterson 2001 ) that use the relative responses of continuum and emission - line components to constrain strongly the nature of the responding emission regions . the responsivity - weighted blr size is given approximately by @xmath19 , with @xmath20 the time lag , measured from cross - correlation analysis , between the continuum and delayed , radiatively driven emission - line variations . combining @xmath21 with the emission - line velocity full - width - half - maximum @xmath22 , assumed to be gravitationally determined and derived from the variable line profile , virial masses @xmath16 follow from @xmath23 , with @xmath24 a factor of order unity that depends on the detailed emission - line gas distribution ( and which may be determined by measuring the velocity - dependent emission - line response ) . the results of wandel et al . ( 1999 ) , kaspi et al . ( 2000 ) , and peterson et al . ( 2000 ) are based on these reverberation techniques . the magnitude and wavelength - dependence of any continuum time delays , when combined with the spectral energy distribution , may be used to constrain the product @xmath25 under certain conditions ( collier et al . 1999 ) , hence @xmath17 follows given @xmath16 from the associated emission - line reverberation measurement . the width of the continuum delay distribution at many wavelengths and the velocity dependent emission - line response ( e.g. , welsh & horne 1991 ) may be used to measure the source inclination . these methods for measuring @xmath17 and @xmath14 have not been applied with much success to existing datasets , since they lack sufficient signal - to - noise and monitoring duration . other methods for measuring @xmath14 include accretion disk fitting to fe k@xmath26 ( e.g. , nandra et al . 1997 ) and combined uv continuum and measurements ( rokaki & boisson 1999 ) . we undertook a program of coordinated multiwavelength observations of the nls1 galaxy akn 564 ( @xmath27 , huchra , vogeley , & geller 1999 ) in the summer of 2000 . akn 564 is the brightest known nls1 galaxy in the 210kev band , with hard x - ray luminosity @xmath28 ( turner et al . it has a steep power - law continuum ( @xmath29 ) , ionized fe k line features , and a steep soft excess , as described by turner et al . a 35 day observation shows evidence for flux variations of @xmath30 36% ( turner et al . 2001 , edelson et al . 2001 ) . optical monitoring by giannuzzo et al . ( 1998 ) reveals variations of about 8% on time scales of about 4 years , rather lower than observed in other s1 galaxies over similar periods ( e.g. , about 25% for ngc 5548 ) . optical spectra reveal the presence of strong [ ] and emission ( van groningen 1993 ) . no program of uv monitoring has been previously undertaken on this source , but 3 presents an archival iue observation . our program included ( turner et al . 2001 hereafter , paper i ) , ( this paper ) , optical ( shemmer et al . 2001 hereafter , paper iii ) , and ( pounds et al . 2001 ) observations , and represents the most comprehensive contemporaneous multiwavelength study of a nls1 galaxy to date . in this paper , we present the uv continuum and emission - line variability results obtained with _ hst _ during the period 2000 may 9 to july 8 . these observations were primarily intended to measure the effective size of the uv blr of a nls1 , and permit measurement of a virial mass @xmath16 based on multiple emission lines , and thereby further constrain the nature of nls1 galaxies . akn 564 also has a rich uv absorption spectrum ( crenshaw et al.1999 ) , and discussion of our absorption - line results will be deferred to a later paper . in 2 , we describe the observations . we discuss the properties of the mean spectra in 3 . the continuum and emission - line variability are presented in 4 and 5 , respectively . our results are discussed and summarized in 6 and 7 , respectively . we observed the nucleus of akn 564 with the space telescope imaging spectrograph ( stis ) on the _ hubble space telescope _ ( _ hst _ ) on 46 occasions during the period 2000 may 9 to july 8 . the first five visits were separated by intervals of @xmath13 5 days . beginning on the fifth visit ( 2000 may 29 ) , the sampling interval was decreased to @xmath13 1 day for the remaining monitoring period . we obtained low - resolution spectra during each visit with the g140l and g230l gratings , thereby providing full uv coverage at a spectral resolution of approximately 1.2 over the range 11501730 and 3.2 over the range 15703150 . we used the @xmath31 slit for the low - resolution spectra to maximize throughput and ensure accurate absolute photometry . as akn 564 is a point source in the stis spectral images , there was no significant resolution degradation by using the wide slit . the exposure times were 1200 and 520 seconds for each g140l and g230l spectrum , respectively , except on 2000 may 29 , when the respective exposure times were 1434 and 816 seconds . to estimate the effects of the intrinsic absorption on the uv emission - line profiles , we obtained high - resolution echelle spectra of the nucleus on 2000 may 29 with the e140 m grating at a spectral resolving power of @xmath32 . the e140 m spectra were obtained through the @xmath33 aperture during four consecutive orbits to yield a combined exposure time of 10,310 sec . the spectra were reduced using the idl software developed at nasa s goddard space flight center for the stis instrument definition team ( lindler 1998 ) . for the low - dispersion spectra , we used standard point - source processing with an extraction height ( perpendicular to the dispersion ) of @xmath34 to obtain flux - calibrated g140l and g230l spectra as a function of wavelength . small wavelength calibration uncertainties exist between spectra of a given grating . to remove these effects , relative wavelength calibration of spectra for each grating is required . the methodology we employ is analogous to that described by korista et al . ( 1995 ) ( except we do not subtract a continuum fit from each spectrum ) . we cross - correlate the spectra near the emission - line peaks of @xmath3 , @xmath5 , and @xmath6 for the g140l grating , and near the peaks of ] @xmath35 and @xmath36 for the g230l grating . hence , we determine the nearest whole pixel shifts for each spectrum relative to its appropriate mean spectrum . the majority of the spectra ( about 90% ) required either a one or zero - pixel shift , and the remaining spectra required a two - pixel shift . the uncertainty in our relative wavelength calibration is of order @xmath37 pixels , i.e. , about 0.6 and 1.7 for the g140l and g230l gratings , respectively . we made no attempt to intercalibrate the g140l and g230l spectra , given their difference in resolution . the observed mean and root mean square ( rms ) of the 46 g140l and g230l spectra are presented in figures 1 and 2 , respectively . the rms spectra have been corrected for the bias due to measurement errors . furthermore , these spectra have been corrected for galactic reddening using @xmath38 mag . ( schlegel et al . 1998 ) . we note narrow line balmer decrement and @xmath6/@xmath39 ratios suggest internal reddening is potentially large for akn 564 ( walter & fink 1993 ; paper iii ) , i.e. , @xmath40mag . determination of the internal reddening in akn 564 will be discussed in a forthcoming paper , crenshaw et al . the mean spectra are rich in both emission and absorption features . the prominent emission features have been identified and labelled in figs . 1 and 2 . many of the unlabelled , poorer contrast features may be associated with emission . intrinsic uv absorption in @xmath3 , @xmath41 , @xmath42 , and @xmath5 is present , along with galactic absorption in lines such as @xmath42 and @xmath36 ( crenshaw et al . we have made no attempt to identify the complex of absorption features shortward of @xmath3 , since this spectral region is affected by time - dependent residuals due to imperfect subtraction of geocoronal emission . we defer discussion of the rms spectra to 5 . in order to quantify the prominent emission - line characteristics , we determine pure emission - line spectra from which appropriate mean spectral measurements are then made . this was done through cubic spline interpolations over the absorption features and subtraction of a power - law fit to the continuum . our spectral measurements ( to be discussed in this section ) assume zero internal reddening . the continuum is defined by four nominally line - free bands chosen by visual inspection of the mean spectrum , fig . 1 ; 11551180 , 13501380 , 14601500 , and 16201660 . these bands only approximate the true continuum level , since balmer recombination continuum , blends and other weak broad emission contaminate them to some extent . the rms spectrum , fig . 1 , indicates these bands are not notably biased by line variability . the best - fit continuum is described by @xmath43 with @xmath44 and @xmath45 . the continuum fit uncertainties are purely statistical and should be treated with caution , since systematic errors due to reddening , for example , are non - negligible . this uv continuum is redder than @xmath46 , typically observed in quasars ( obrien et al . this may be due to , for example , disk irradiation ( assuming the uv emission arises in an accretion disk ) and/or internal reddening . we choose to extrapolate this continuum into the g230l spectral range , since here the continuum level is poorly defined , as the aforementioned contaminants strengthen . we note , in this regime , the extrapolated continuum is about 30% below the local , pseudo - continuum defined largely by the fe ii blends . the validity of the above continuum fit awaits detailed consideration of these fe ii blends and corrections for internal reddening . our emission - line spectral measurements are detailed in table 1 . for each emission line in column ( 1 ) we give the fwhm ( fw@xmath47 ) in column ( 2 ) and flux contained therein ( f@xmath47 ) in column ( 3 ) . similarly , columns ( 4 ) and ( 5 ) give the full width at 20% maximum intensity ( fw@xmath48 ) , and flux therein ( f@xmath48 ) . column ( 6 ) gives the line centroid @xmath49 , defined by f@xmath48 . no attempt has been made to deconvolve blended lines , hence column ( 1 ) refers to the primary emission line only . in table 1 , columns ( 4 ) through ( 6 ) are empty for ] @xmath35 and @xmath36 because f@xmath48 is ill - defined , ] @xmath35 is blended with ] @xmath50 and the estimated , extrapolated continuum near @xmath36 is significantly below the observed level . we note that all values in table 1 are in the observed frame . we chose to measure line fluxes at 20% maximum intensity to minimize the effects of line blending , by measuring the primary emission line flux only . the uncertainties in the full width and flux measurements ( table 1 ) are dominated by systematic errors attributable to uncertainties in the applied absorption ( where applicable ) and reddening corrections . we assess the uncertainty associated with our absorption corrections by using cubic spline and linear interpolations over the @xmath3 absorption feature . figure 3 presents an illustrative example of the ambiguity in the @xmath3 emission line profile due to the different interpolation schemes . we find uncertainties of @xmath13 15% and 8% for our line width and flux measurements , respectively . we do not consider the systematic uncertainty in our applied reddening corrections , given the aforementioned possibility of large internal reddening . for wavelengths 1800 , systematic errors of @xmath51 due to uncertainties in our continuum fit may be important , but are not considered further . the @xmath3 , @xmath41 , and @xmath5 emission lines have fwhms of @xmath52 , compared to typical values of @xmath53 for s1 galaxies . their fw@xmath48 values are @xmath54 . for @xmath6 , the fw@xmath47 and fw@xmath48 estimates are @xmath30 1000 and 3000 kms@xmath55 , respectively ( from the mean g140l spectrum ) . the emission - line profiles appear symmetric about their systemic wavelengths , but this is difficult to quantify due to contaminating emission , such as @xmath56 in the red wing of @xmath6 , and absorption features affecting many of the line profiles . the centroids , @xmath57 , of the observed emission - line fluxes of @xmath3 , @xmath41 , + ] @xmath58 , @xmath5 , ] @xmath59 , and @xmath36 are consistent with those expected based on the systemic redshift @xmath27 , given the 1.2 and 3.2 spectral resolutions of the g140l and g230l gratings . the @xmath60 and @xmath61 emission - line flux centroids are redshifted and blueshifted by 2.5 and 1.6 , respectively . the @xmath60 @xmath57 is possibly biased by @xmath62 emission . the @xmath6 emission - line flux centroid is blueshifted by 1.3 ( @xmath63 ) . the @xmath6 @xmath57 measured from the g230l mean spectrum is most likely biased by @xmath56 contamination . figure 4 presents the mean @xmath3 , @xmath5 , and @xmath36 emission line profiles as a function of velocity , with the above defined continuum fit subtracted . the emission line profile amplitudes have been normalized to unity . the high ionization lines of @xmath3 and @xmath5 and the low ionization line of @xmath36 do not appear notably blueshifted or redshifted relative to systemic velocities . these results are discussed in 6.1 . we find some significant differences between emission - line flux ratios in akn 564 relative to those of more typical seyfert 1 galaxies . from table 1 , measured line ratios for akn 564 are @xmath64 , @xmath65/\lya)}^ * = 0.16\,(0.13)$ ] , @xmath65/\civ)}^*= 0.65\,(0.12)$ ] , @xmath66/\civ ) } = 0.48\,(0.07)$ ] , @xmath67/\lya ) } = 0.13\,(0.08)$ ] , @xmath68 , and @xmath69 . the line ratios in parenthesis are those for the s1 galaxy ngc 5548 , taken from clavel et al . line ratios with an asterisk denote those ( for akn 564 ) estimated using the fluxes at 50% maximum intensity , compared to 20% for the others . this was necessary because , as mentioned previously , f@xmath48 is ill - defined for c iii]@xmath70 and @xmath36 . the uncertainties in these line ratios are dominated by systematic effects , discussed above , and are of order 25% . this error estimate includes an @xmath13 15% bias due to excluding flux in the line - wings . for line ratios that include ] @xmath35 and @xmath36 , the error estimate may be notably underestimated on account of the continuum fit uncertainties described above . our emission - line ratios show akn 564 exhibits weaker @xmath5 and stronger + ] @xmath58 emission compared to ngc 5548 . these results are consistent with earlier results of wilkes et al . ( 1999 ) and kuraszkiewicz et al . we note our / ratio of 0.33 is notably larger than the mean value of 0.05 derived from the small sample of nls1s by kuraszkiewicz et al . this is probably largely attributable to different continuum level estimates near to @xmath36 . finally , we checked archival iue observations of akn 564 to assess whether it was in a comparatively low or high flux state . figure 5 presents the iue swp observation of 1984 jan 17 . for comparison purposes , our mean g140l spectrum of fig . 1 is overlaid as the thicker line . both spectra have been corrected for galactic reddening as detailed above . the continuum and emission - line fluxes are in qualitative agreement . in particular , the @xmath3 and @xmath41 line profiles are very similar ; and suggest negligible emission - line flux differences between the 1984 and 2000 observations . other iue observations from 1981 are also in qualitative agreement with our observations . we select five nominal continuum bands by visual inspection of the mean and rms spectra presented in figs . 1 and 2 : 13501380 , 14601500 , 16201660 , 20702130 , and 29603040 , henceforth referred to by their mean observed wavelengths 1365 , 1480 , 1640 , 2100 , and 3000 , respectively . we re - iterate that each of these continuum bands overestimates the true continuum level on account of various contaminants ( 3 ) . these bands are not notably biased by emission - line variability , and thereby provide good approximations for the continuum variability at these wavelengths . light curves , describing the continuum variability during the 60-day monitoring period , are presented in figure 6 , with the continuum bands as labelled . we note our observed continuum flux measurements are not corrected for reddening . all the continuum regions show the same qualitative behavior . the 1365variations may be described by ; an increase in flux of @xmath13 20% during the first 15 days , followed by a decrease of @xmath13 16% over the next 5 days . these variations are followed by two flares at about @xmath71 and 1711 . the rising trough - to - peak variations are about 18% in @xmath13 3 days , with similar declining peak - to - trough variations occuring over longer time scales of several days . both flare events appear to be asymmetric , the later one more so and of longer duration . between these two events was a relatively quiescent period of about 5 days . thereafter , the variations are less pronounced , at about the 5% level . for the longer wavelength variations , at 2100 and 3000 , the rms variations are of reduced amplitude with broader asymmetric flare profiles . the peaks of the 3000 flares are delayed , relative to those at 1365 , by 1 - 2 days . the contemporaneous asca observations exhibit larger amplitude correlated variations ( including similar but narrower twin flare events ) , possibly delayed by about 0.4 days ( papers i and iii ) . we note the continuum variations may be slightly undersampled during the intensive monitoring period of @xmath721734 , and are definitely undersampled for prior times , given that @xmath73 10% flux amplitude changes on time scales of 1 day are evident ( see below ) . we characterize the 13653000 variations by measuring two common variability parameters : the ratio of the maximum to minimum flux @xmath74 and the amplitude of the intrinsic variability relative to the mean flux @xmath75 . the latter is corrected for the measurement errors @xmath76 , @xmath77 where @xmath78 , @xmath79 , and @xmath80 are respectively the standard mean flux , rms flux , and mean square of the measurement errors , with @xmath81 the number of data points in the light curve ( edelson , krolik , & pike 1990 , rodrguez - pascual et al . the results are presented in table 2 for each of the light curves , as listed in column ( 1 ) . columns ( 2)(5 ) give , respectively , the mean flux @xmath78 , the rms flux @xmath82 , @xmath75 , and @xmath74 . both @xmath83 and @xmath74 are potentially biased by constant flux components , e.g. , emission . however , these effects are likely to be small . on time scales of about 60 days , the fractional amplitude of the intrinsic variations is essentially constant between 1365 and 2100 at about 6% , with the full range of variations @xmath84 . at 3000 the fractional variability amplitude is about 4% , with the full range of variations @xmath85 . this level of intrinsic uv variability is about a factor 3 less than that found in typical s1 galaxies , which typically display fractional flux variations @xmath86 18% on time scales of 60 days ( collier , peterson & horne 2001 ) . the 13651640 continuum variations show evidence for @xmath13 10% flux amplitude changes on time scales of about 1 day , as seen in the events beginning at about @xmath87 , 1710 , and 1707 , and suggest a fraction of the uv continuum emitting region must come from a compact region of 1 light day in size . the 21003000variations show evidence for reduced flux amplitude changes of @xmath13 5% on similar time scales . for comparison , in s1 galaxies far uv variations of about 5% rms occur on similar time scales ( e.g. , korista et al.1995 ; welsh et al . these faster , larger amplitude variations suggest nls1s ( at least akn 564 ) exhibit more variability power on short ( day ) time scales . we confirm this through an autocorrelation analysis . figure 7 presents the akn 564 1365 ( solid line ) and ngc 7469 ( a s1 galaxy ) 1315 ( dotted line ) autocorrelation functions ( acfs ) . the relative steepness of the 1365 acf ( compared to the 1315 acf ) indicates akn 564 s fluctuation power density spectrum is flatter than that for ngc 7469 , and thereby exhibits more power on short time scales . we note pounds et al . ( 2001 ) have shown that the x - ray variations in akn 564 are faster than those of typical s1 galaxies . the full - width - half - maximum of the acfs are 3.27 and 4.93 days for akn 564 and ngc 7469 , respectively , and are indicative of characteristic uv variability time scales . by assuming the mass ratio of the sources is determined by the variability time scale ratio of 0.66 , we estimate the mass of akn 564 to be @xmath88 , given a mass estimate for ngc 7469 of @xmath89 ( wandel et al . 1999 ) . this mass estimate ( for akn 564 ) is in good agreement with that derived from our emission - line reverberation results , discussed in 6.2 . the continuum light curves of fig . 6 clearly display correlated variations between 13653000 . the peaks and troughs of the flare events at @xmath90 and 1711 occur approximately simultaneously ; as mentioned above , the 3000 flare peaks appear delayed by about 1 - 2 days relative to those at shorter wavelengths . differences between the light curves are also evident , for example , decay time scales for the two flares appear comparatively longer at wavelengths @xmath91 2100 , and short ( day ) time scale 1365 variations appear washed out at 3000 . in order to quantify the nature of the correlations between continuum variations we undertook a cross - correlation analysis . two algorithms were employed to compute cross - correlation functions ( ccfs ) , the interpolated ccf ( iccf ) of gaskell & sparke ( 1986 ) as implemented by white & peterson ( 1994 ) , and the @xmath92-transformed discrete correlation function ( zdcf ) algorithm of alexander ( 1997 ) . for our data , the results recovered by the two algorithms are in agreement . figure 8 shows the ccfs obtained by cross - correlating each of the four longer - wavelength continuum light curves with the 1365 light curve . the solid line and data points with error bars show the iccf and zdcf ccfs , respectively , and are in good agreement . the 13653000variations are highly correlated as evidenced by maximum values of the cross - correlation coefficients @xmath93 . the 14802100 ccfs all peak at about zero lag , thereby suggesting these continuum variations occur quasi - simultaneously . the 3000 ccf peaks away from zero at about 0.5 days . the 2100 and 3000 ccfs appear asymmetric , compared to the 1480 and 1640 ccfs , and suggest their responses extend over a larger range of positive delays . figure 9 presents similar ccfs for four optical continuum regions ; 48754915 , 51975237 , 65516622 , and 69357004 , henceforth referred to by the observed wavelengths 4900 , 5200 , 6600 , and 6900 , respectively ( for consistency with paper iii ) . the optical data are from paper iii . these ccfs have been computed as described above . the optical variations are correlated with those at uv wavelengths with @xmath94 . the decreasing maximum correlation coefficient with optical wavelength is likely due to progressively less coherent time delay smeared responses and/or increasing noise dilution ( see table 4 in paper iii ) . the optical ccfs clearly peak away from zero at about 2 days . table 3 summarizes our cross - correlation results . column ( 1 ) lists the light curve that has been cross - correlated with the 1365 continuum light curve . columns ( 2 ) and ( 3 ) give the centroid of the ccf @xmath95 , as determined by the iccf and zdcf algorithms . the iccf centroid is calculated over all points above 0.8 times the maximum cross - correlation coefficient @xmath96 , whereas the zdcf centroid is computed from all points near the peak with a cross - correlation amplitude above half that of the peak . columns 4 and 5 report the time delay @xmath97 , measured from the peak of the ccf . columns 6 and 7 give @xmath96 , for the iccf and zdcf , respectively . column 8 details the fwhm of the iccf . the reported errors on @xmath97 and @xmath98 for the iccf are the 1-@xmath99 uncertainties as determined by the model - independent quasi - bootstrap and flux randomization method of peterson et al . we find no evidence for lags between the 1365 and 1640 light curves , as evidenced by peak and centroid lag measurements consistent with zero . the 2100 and 3000variations , however , lag those at 1365 by @xmath100 and 1.0 day , respectively . here and hereafter , we note the centroid lags since they are a less biased estimator of the size of the responding emission region . these lags are greater than zero at 94% and 99% , respectively . the optical variations at @xmath101 lag behind those at 1365by about 2 days at no less than 99% confidence . the optical variations at wavelengths longer than 6000 appear to lag those at 1365 by a similar amount , although statistically the suggested lags are not significantly different from zero on account of their very low amplitude ( @xmath102% ) variability ; which is comparable to the measurement errors . we investigate these wavelength - dependent continuum lags further by binning the 46 g140l and g230l spectra ( covering 11503140 ) into 40 bins , and forming light curves based on the total flux in each bin . we cross - correlate ( using the iccf ) each of these light curves with the 1365 light curve . the results are presented in figure 10 . in the top panel , the histogram plot shows the centroid lag for each bin as a function of the bin wavelength , thus producing a `` lag spectrum '' . the corresponding values of @xmath103 are shown in the lower panel . again , the error bars for each bin in the upper panel were determined by the method of peterson et al . the solid line represents the best - fit function @xmath104 , with @xmath105 , @xmath106 , and reduced chi - squared @xmath107 for @xmath108 degrees of freedom , to the lag measurements . the small value of @xmath109 suggests the lag uncertainties may be overestimated . there is a clear trend of increasing lag with wavelength throughout the uv : the variations at 3100 lag those at 1365 by about 0.9 days . the lag measurements for @xmath110 are non - zero at no less than 90% confidence . for wavelengths shorter than 2050 , the lag measurements are consistent with zero delay . the 11503140variations are well correlated with @xmath111 . in the region where the g140l and g230l spectra overlap , i.e. , @xmath13 16001700 , the lag measurements are in good agreement . the lag spectrum is reasonably smooth with no clear positive deviations near the emission - line wavelengths , as would be the case if the emission - line response time scales were measurably larger . similarly , there are no negative dips in the maximum correlation coefficient spectrum , with the possible exception near @xmath36 . these results support the negligible presence of emission - line variability to be discussed in 5 . we note the dip in the maximum correlation coefficient at about 1700is due to calibration uncertainties in the g230l spectra . similarly , the slight depression in @xmath112 at 1200 is due to calibration errors in the g140l spectra . we note the two bins with the largest centroid lag errors are those contaminated by @xmath3 and @xmath36 . we use the optical lag measurements ( table 3 ) to extend the uv lag spectrum . the uv / optical lag spectrum is presented in figure 11 . the solid line represents the best - fit function @xmath104 , with @xmath105 , @xmath113 , and @xmath114 for @xmath115 degrees of freedom . the dotted line represents the best - fit function to the uv data alone ( as in fig . 10 ) , i.e. , with @xmath116 . by including the optical data , the lag wavelength relationship appears to flatten . the largely featureless rms spectra of figs . 1 and 2 immediately show that any emission - line variability is of low amplitude . figure 1 shows @xmath3 , @xmath41 , + ] @xmath58 , @xmath5 , and @xmath6 variations are present with amplitudes of @xmath7 4 , 5 , 6 , 5 , and 4% , respectively . figure 2 indicates @xmath50 , ] @xmath35 , and @xmath36 variations with amplitudes of @xmath7 6 , 6 , and 4% , respectively . for cases where emission - line variations are present or suggested , the bulk of the variation occurs in the core of the line . the variations at these wavelengths , for emission lines contaminated by intrinsic absorption , may be a superposition of both emission and absorption - line variability . a discussion of the latter is deferred to a later paper . we note the @xmath41 variations appears to be slightly asymmetric , with a deficit of response in its blue wing . the variations at about 1215 are spurious , and attributable to time - dependent residuals from imperfect subtraction of geocoronal emission . emission - line light curves were constructed by direct integration , summing the flux in specified , absorption - free wavelength regions after subtracting a continuum defined by a linear fit to the nearest continuum regions on either side of the emission - line region . our attempts at validating and possibly improving our direct integration light curves through spectral fitting with the iraf task specfit ( kriss 1994 ) failed . this was because the spectral complexity of the emission - line profiles permitted too much freedom in the fitting process such that artifical emission - line variations were introduced by corresponding changes in the adopted absorption profile . the blanket exclusion of any absorption regions from our variable emission - line flux regions , defined by their respective rms spectra ( figs . 1 and 2 ) , severely inhibited our ability to extract the already low - amplitude variations . we were only able to extract significant emission - line variations for @xmath3 . the compartively smaller fluxes and larger measurement errors for the other emission - lines precluding extraction of significant variations . figure 12 presents the @xmath3 light curves for the monitoring period . the data points with error bars describe the emission - line variations between 12401243 and 12471250 ( `` light curve 1 '' ) , and the dashed line those between 12401250 ( `` light curve 2 '' ) ; the latter includes any possible variable contribution from the intrinsic absorption feature . in both cases the continuum is defined by a linear fit between 11551180 and 13501380 . the 1365light curve , 4.1 , is scaled and vertically shifted to fit light curve 1 , and is shown by the solid line . we note our observed emission - line flux measurements are not corrected for reddening . the @xmath3 variations resemble those of the continuum , but with greatly reduced amplitude and appear delayed by about a few days . in each case , the flux increases by @xmath117% during the first 15 days , and similar flare events occur at similar times , i.e. , @xmath90 and 1711 . the contrast of any features in the light curves is poor on account of the low - amplitude variability , which is comparable to the measurement uncertainties . on time scales of about 60 days , the intrinsic line variations are @xmath102% , with the full range of variation @xmath118% . moreover , the @xmath3 light curves are clearly well correlated and exhibit similar variability patterns . these results suggest any intrinsic absorption - line variability does not notably affect the emission - line variations . cross - correlation of the @xmath3 and 1365 variations quantifies their apparent similarity . our results are presented in figure 13 and table 3 . figure 13 presents @xmath3 ccfs for both line light curves detailed above . the solid line and filled data points represent the iccf and zdcf ccfs , respectively , as described earlier , for light curve 1 , and the dashed line and open data points are for light curve 2 . the @xmath3 variations are correlated with the 1365 variations with a maximum cross - correlation amplitude @xmath119 . the probability of exceeding @xmath120 in a random sample of observations ( of the same number , @xmath121 , as here ) drawn from an uncorrelated parent population is about 0.1% ( bevington & robinson 1992 ) , thereby the @xmath3 variations are correlated with those at 1365 at about 99.9% confidence . the formal significance of this correlation is most likely over - estimated , since not all correlated data points are independent . detailed simulations will be required in order to quantify the magnitude of this bias . by directly cross - correlating light curves 1 and 2 , we confirm they are correlated with essentially zero time delay . the peak and centroid lag measurements for light curves 1 and 2 are consistent to within the large uncertainties . moreover , for light curve 1 the lag measurements are consistent with zero delay . our results only permit an upper limit for the size of the @xmath3 emission region , @xmath122 3 light days . we caution , the low amplitude @xmath3 variations may indicate the bulk of the emission region is at larger radii , as discussed in 6.2 . a complete study of the nature of the uv blr of akn 564 requires detailed photoionization modelling of the spectral energy distribution and is beyond the scope of this paper . here , we offer a preliminary discussion of our results in the context of previous work . the blr of akn 564 is characterised by smaller gas velocity dispersions ( as determined by the emission - line velocity full - width - half - maximum ) than found in typical s1 galaxies ( 3 ) , as noted earlier by rodrguez - pascual et al . ( 1997a ) , i.e. , about 2000compared to 5000 . these authors found evidence for similar magnitude uv line widths in about a dozen nls1s ; see also kuraszkiewicz et al . the full widths of the low and high - ionization lines are comparable , see fig . 4 , and interestingly similar to those of the ] @xmath59 and ] @xmath35 intercombination lines . for example , the full widths of @xmath6 , ] @xmath59 , ] @xmath35 , and @xmath36 are 1800 , 1900 , 1900 , and 1700 kms@xmath55 , respectively ( table 1 ; as measured from the g230l spectra ) . as noted in 3 , the broad emission - line profiles of akn 564 are approximately symmetric about their systemic wavelengths . these results , and those above , contrast with uv / optical observations of quasars ( e.g. , gaskell 1982 ; wilkes 1984 ; corbin 1991 ) and s1 galaxies , showing that high - ionization lines are shifted blueward of low - ionization lines , which are approximately at systemic redshifts , and slightly broader . leighly ( 2000 ) presented observations of two nls1s ( iras 13224 - 3809 and 1h 0707 - 495 ) which also showed the high ionization lines were strongly blueshifted and much broader than the low ionization @xmath36 line , similar to results for the nls1 galaxies 1 zw 1 and ngc 4051 presented by laor et al . ( 1997 ) and peterson et al . ( 2000 ) , respectively . collinge et al . ( 2001 ) find the @xmath5 line in ngc 4051 to be blueshifted by about 100 kms@xmath55 . these prior results are often interpreted as evidence for blr radial motions . the low ionization lines may arise in an accretion disk , with the high ionization lines emitted in a wind launched from this disk ( e.g. , collin - souffrin et al . our results suggest an absence of outflowing high ionization blr gas with velocities 230 kms@xmath55 in akn 564 . moreover , the approximate symmetry of the emission - line profiles may indicate the predominance of optically thin broad line emission , since it is emitted essentially isotropically and traces the full velocity field ( netzer 1976 ; davidson 1977 ; shields , ferland , & peterson 1995 ) . the level of intrinsic line variability is notably less than observed in typical seyfert 1 galaxies over comparable time scales ( 5 ) . for example , ngc 5548 ( korista et al . 1995 ) showed @xmath3 , @xmath5 , and @xmath6 variations of @xmath123% , @xmath124% , and @xmath125% , respectively . similarly , the same lines in ngc 7469 ( wanders et al . 1997 ) showed variations of @xmath8% , @xmath124% , @xmath125% . this difference may simply be a consequence of the reduced level of assumed ionizing continuum variations combined with geometrical dilution due to the finite blr size . to investigate this scenario we convolved the observed continuum variations with thin shell broad line regions at radii of 6 and 2 light days ( i.e. , the predicted sizes of the @xmath3 and @xmath6 emission regions , see 6.2 ) , respectively . in these cases , we find @xmath3 and @xmath6 variations of order 3 and 6% , respectively . these results suggest this type of scenario does not completely explain the reduced level of emission - line variability in akn 564 . on the other hand , the line - emitting gas in akn 564 may be fully ionized in hydrogen , and hence optically thin to the lyman continuum . this would lead to greatly reduced emission - line variability , with detectable variations occurring only for sufficiently large continuum variations that induce significant changes in the ionization structure of the emission region . similarly , this scenario alone is likely insufficient to account for the lack of line variability across the observed range of ionization , as evidenced by the calculations of ferland & mushotzky ( 1982 ) . however , we note these calculations are more appropriate for s1 blrs , and thereby may be misrepresentative of nls1s . some combination of both the above effects appears a reasonable explanation for the reduced level of emission - line variability in akn 564 . the unusual uv spectral properties of akn 564 ( 3 ) , combined with stronger ionizing fluxes relative to other s1 galaxies , suggest nls1s blrs have a physically distinct nature from those found in s1 galaxies and quasars . kuraszkiewicz et al . ( 2000 ) posit nls1 blrs characterized by gas cloud densities @xmath126 , ionization parameters @xmath127 , and blr radii @xmath128 light days . on the other hand , rodrguez - pascual et al . ( 1997a ) posit @xmath129 , @xmath130 , and blr radii @xmath131 light days ; under these conditions the broad line emitting gas becomes fully ionized in hydrogen , and thereby optically thin to the lyman continuum . our emission line profiles and variability results hint at a predominance of optically thin broad line emitting gas in akn 564 . in combination with a small estimated blr size , 5 , of 3 light days , these conditions appear more readily reconcilable with those suggested by rodrguez - pascual et al.(1997a ) , as noted above . the wavelength - dependent continuum time delays of 4.2 suggest a stratified continuum reprocessing region extending a couple of light days from the central black hole , and provide possible evidence for an accretion - disk structure ( collier et al . simple irradiated accretion disk models with radial temperature profiles @xmath132 predict @xmath133 . flatter temperature profiles predict steeper @xmath20@xmath134 relationships , and are realizable with , e.g. , irradiated flared accretion - disk models . these models may be able to explain the steeper uv lag spectrum presented in 4.2 . the fine details are determined by the geometry and structure of the accretion disk , relative prominence of viscous heat dissipation and irradiation effects , and the emission physics . vaughan et al . ( 1999 ) present evidence , based on x - ray data , for a strongly ionised disk in akn 564 . given the close proximity of the uv and x - ray emitting regions ( 4 and papers i and iii ) , our results support their conclusions . in the case of ngc 7469 , the only other source with a clear detection of wavelength - dependent continuum lags ( wanders et al . 1997 , collier et al . 1998 , kriss et al . 2000 ) , the uv / optical lag spectrum was consistent with @xmath135 , and is similar to that observed in 4.2 . on the other hand , these continuum lags may be a result of diffuse continuum emission from broad - line emitting clouds ( korista & goad 2000 ) , or some combination of both these effects . in 5 we estimated an upper limit of 3 light days for the size of the @xmath3 broad line emitting region , @xmath136 . we may compare this estimate with that expected on the basis of results obtained for other agns by using the balmer emission region size @xmath137luminosity relationships of kaspi et al . ( 2000 ) and peterson et al . the luminosity @xmath138 , adopting a mean flux @xmath139 and a distance @xmath140mpc ( assuming @xmath141 ) . this leads to an estimate for the size of the -emitting region @xmath142 light days from the @xmath137luminosity relationship shown in fig . 6 of peterson et al . ( 2000 ) . from previous monitoring programs on s1 galaxies ( netzer & peterson 1997 ) , we can estimate the size of the @xmath3 emitting region to be @xmath143 . similarly , we expect that the highest ionization lines , @xmath41 and @xmath6 , arise in a region of extent @xmath144 . based on the optical luminosity of akn 564 , we thus predict @xmath145light days and @xmath146 light days . the factor of @xmath13 2 difference between our measured and predicted values for @xmath122 is probably not significant , since the intrinsic scatter in the @xmath147luminosity relationship is of order a factor 5 . these results give us confidence our @xmath122 estimate is in the right ballpark . by combining our @xmath122 upper limit of 3 light days with the emission - line velocity full - width - half - maximum @xmath148 , derived from the rms emission - line profile ( and assumed to be gravitationally determined ) , we estimate a virial mass upper limit from @xmath149 where for consistency with wandel et al . ( 1999 ) and kaspi et al . ( 2000 ) we use @xmath150 . this mass estimate is consistent with the independent mass estimate of @xmath4 of pounds et al . ( 2001 ) , based on a fluctuation power spectrum analysis of x - ray variability in akn 564 . and is consistent with our mass estimate based on the variability arguments of 4.1 . furthermore , we note that the position of akn 564 in the agn mass luminosity plane defined by fig . 7 of peterson et al . ( 2000 ) , is consistent with the best - fit regression line based on the nls1 galaxies alone , see figure 14 . this is consistent with the hypothesis that akn 564 harbors a comparatively small black hole accreting with a higher accretion rate and/or is viewed more face - on than typical s1 galaxies ; as found earlier by pounds et al . ( 2001 ) . we conclude with a cautionary note concerning our derived upper limits for the size of akn 564 s @xmath3 blr and black hole mass . the reduced continuum to @xmath3 variations ratio of @xmath13 4 , relative to @xmath13 1 for the s1 galaxies ngc 5548 and ngc 7469 ( references as given in 6.1 ) , suggests our @xmath3 reverberations arise from a region only about 1/4 of the size of the @xmath3 region probed by reverberation in s1 galaxies ; and is therefore biased low . to illustrate the potential importance of this bias , assume that the @xmath3 blr is a thin shell of radius @xmath151 . equal time delay intervals define equal areas on the shell , and if our @xmath3 response from @xmath7 3 light days is from only 1/4 of the area of the shell , we underestimate @xmath151 by a factor of 4 . this would lead to an @xmath122 upper limit of 12 light days . the bias in our black hole mass upper limit depends on the unknown nature of the gas velocity field , and is not readily quantifiable without better data . an intensive two - month monitoring program on the nls1 galaxy akn 564 was undertaken with _ hst _ during 2000 may 9 to july 8 . we summarize our results as follows . 1 . the fractional variability amplitude of the uv continuum variations between 13653000 is about 6% on time scales of 60 days . this level of variability is about a factor of three less than that found in s1 galaxies . we find evidence for fast , large amplitude continuum variations , e.g. , trough - to - peak flux changes of @xmath152% in about 3 days , atypical of those found in s1 galaxies that display similar amplitude variations on longer time scales . 2 . we present evidence for wavelength - dependent continuum time delays . the continuum variations at 3000 lag behind those at 1365 by about 1 day . by combining the uv data with the optical data described in paper iii , we find that the variations at 5200 lag behind those at 1365 by about 2 days . these delays may be interpreted as evidence for a stratified continuum reprocessing region , possibly an accretion - disk structure . however , the delays may be a result of diffuse continuum emission from broad - line emitting clouds , or some combination of both these effects . the @xmath3 emission line exhibits intrinsic rms variations of about 1% on time scales of 60 days . these variations lag those at 1365 by 3 days , and combining this with the line width yields a putative black hole mass limit of @xmath89 . this calculation assumes the line width is determined by the gravity of the black hole . we caution , the low amplitude @xmath3 variations may indicate the bulk of the emission region is at larger radii , and thereby the veracity of our black hole mass limit is questionable . the possible bias in our black hole mass estimate is uncertain given the unknown nature of the gas velocity field . despite the unreliability of our mass estimate , it is consistent with the independent estimate @xmath4 of pounds et al . ( 2001 ) , based on a fluctuation power spectrum analysis of x - ray variability in akn 564 . 4 . the root - mean - square spectra suggest other strong emission line variability , e.g. , of @xmath5 and @xmath6 , occurs with rms amplitudes of @xmath7 5% . the low level of nls1 emission - line variability is in contrast to that found in typical s1 galaxies , which display rms flux variations of @xmath8% on similar time scales . alexander , t. 1997 , in astronomical time series , ed . d. maoz , a. sternberg , & e.m . leibowitz ( dordrecht : kluwer ) , p. 163 baldwin , j. , ferland , g. , korista , k. , & verner , d. 1995 , apj , 455 , 119 bevington , p. , & robinson 1992 , in data reduction and error analysis for the physical sciences ( mcgraw - hill ) , p. 256 blandford , r. , & mckee , c.f . 1982 , apj , 255 , 419 brandt , w. n. , fabian , a. c. , nandra , k. , reynolds , c. s. , & brinkmann , w. 1994 , mnras , 271 , 958 brandt , w. n. , mathur , s. , elvis , m. 1997 , mnras , 285 , 25 boller , t. , brandt , w. n. , & fink , h. 1996 , a&a , 305 , 53 boroson , t.a . , & green , r.f . 1992 , apjs , 80 , 109 collier , s. et al . 1998 , apj , 500 , 162 collier , s. , horne , k. , wanders , i. , peterson , b.m . 1999 , mnras , 302 , 24 collier , s. , peterson , b. m. , & horne , k. 2001 , in probing the physics of agn by multiwavelength monitoring , ed . peterson , b. m. , polidan , r. s. , & pogge , r. w. , asp v224 , 457 collin - souffrin , s. , dyson , j. , mcdowell , j. , & perry , j. 1988 , mnras , 232 , 539 corbin , m. , apj , 1991 , 371 , 51 crenshaw , m. , kraemer , s. , boggess , a. , maran , s. , mushotzky , r. , & wu , c. 1999 , apj , 516 , 750 crenshaw , m. et al . 2001 , in preparation davidson , k. 1977 , apj , 218 , 20 edelson , r. , krolik , j. , & pike , g. 1990 , apj , 359 , 86 . edelson , r. et al . 2001 , in preparation ferland , g. , & mushotzky , r. f. 1982 , apj , 262 , 564 gaskell , c. m. 1982 , apj , 263 , 79 gaskell , c. m. , & sparke , l. 1986 , apj , 305 , 175 giannuzzo , m. e. , mignoli , m. , stirpe , g. , & comastri , a. 1998 , a&a , 330 , 894 goodrich , r. w. 1989 , apj , 342 , 224 guilbert , p. w. , fabian , a. c. , & mccray , r. 1983 , apj , 266 , 466 huchra , j. p. , vogeley , m. s. , geller , m. j. 1999 , apjs , 121 , 287 kaspi , s. , smith , p. s. , netzer , h. , maoz , d. , jannuzi , b. t. , & giveon , u. 2000 , apj , 533 , 631 korista , k. et al . 1995 , apjs , 97 , 285 korista , k. & goad , m. 2000 , apj , 536 , 284 kriss , g. 1994 , pasp conference series , v61 , 437 kriss , g. , peterson , b. m. , crenshaw , d. m. , & zheng , w. 2000 , apj , 535 , 58 kuraszkiewicz , j. , wilkes , b. , czerny , b. , & mathur , s. 2000 , apj , 542 , 692 laor , a. , fiore , f. , elvis , m. , wilkes , b. , & mcdowell , j. c. 1997 , apj , 477 , 93 leighly , k. m. 1999a , apjs , 125 , 297 leighly , k. m. 1999b , apjs , 125 , 317 leighly , k. m. 2000 , newar , 44 , 395 lindler , d. 1998 , calstis reference guide ( calstis version 5.1 ) madau , p. 1988 , apj , 327 , 116 mathur , s. , elvis , m. , & wilkes , b. 1995 , apj , 452 , 230 nandra , k. , george , i. m. , mushotzky , r. f. , turner , t. j. , & yaqoob , t. 1997 , apj , 477 , 602 netzer , h. 1976 , mnras , 177 , 473 netzer , h. , & peterson , b.m . 1997 in astronomical time series , ed . d. maoz , a. sternberg , & e.m . leibowitz ( dordrecht : kluwer ) , p. 85 obrien , p. , wilson , r. , & gondhalekar , p.m. 1988 , mnras , 233 , 801 osterbrock , d.e . , & pogge , r.w . 1985 , apj , 297 , 166 peterson , b.m . , wanders , i. , horne , k. , collier , s. , alexander , t. , kaspi , s. , & maoz , d. 1998 , pasp , 110 , 660 peterson , b.m . 2000 , apj , 542 , 161 peterson , b.m . 2001 , in _ advanced lectures on the starburst agn connection , _ eds . i. aretxaga , d. kunth , & r.mjica , ( world scientific ) , p. 3 pounds , k. , done , c. , & osborne , j. mnras , 277 , l5 pounds , k. , edelson , r. , markowitz , a. , & vaughan , s. 2001 , apj , 550 , l15 puchnarewicz , e.m . , mason , k. , crdova , f. , kartje , j. , branduardi - raymont , g. , mittaz , j. , murdin , p. , & allington - smith , j. 1992 , mnras , 256 , 589 rodrguez - pascual , p. m. , mas - hesse , j. m. , & santos - lle , m. 1997a , a&a , 327 , 72 rodrguez - pascual , p.m. et al . 1997b , apj , 110 , 9 rokaki , e. , & boisson , c. 1999 , mnras , 307 , 41 shields , j.c . , ferland , g.j . , & peterson , b.m . 1995 , apj , 441 , 507 schlegel , d.j . , finkbeiner , d.p . , & davis , m. 1998 , apj , 500 , 525 shemmer , o. et al . 2001 , submitted to apj turner , j. et al . 1999 , in proceedings of the 19th texas symposium on relativistic astrophysics & cosmology , ed paul , t. montmerle , & e. , aubourg ( saclay : cea ) , e441 turner , j. et al . 2001 , submitted to apj vaughan , s. , pounds , k. , reeves , j. , warwick , r. , & edelson , r. 2000 , mnras , 308 , 34 van groningen , e. 1993 , a&a , 272 , 25 walter , r. , & fink , h. 1993 , a&a , 274 , 105 wandel , a. , & boller , t. 1998 , a&a , 331 , 884 wandel , a. , peterson , b. m. , & malkan , m. a. 1999 , apj , 526 , 579 wanders , i. et al . 1997 , apjs , 113 , 69 welsh , w. , & horne , k. 1991 , apj , 379 , 586 welsh , w.f . , peterson , b.m . , koratkar , a.p . , & korista , k.t . 1998 , apj , 509 , 118 white , r.j . , & peterson , b.m . 1994 , pasp , 106 , 879 wilkes , b. 1989 , mnras , 207 , 73 wilkes , b. , kuraszkiewicz , j. , green , p. , mathur , s. , & mcdowell , j. c. 1999 , apj , 513 , 76 lccccc & 2114 & 5.30 & 3553 & 6.51 & 1245.2 & 2809 & 1.98 & 4725 & 2.56 & 1270.5 & 788 & 0.15 & 1563 & 0.20 & 1338.1 + ] & 3270 & 0.70 & 4859 & 0.84 & 1433.9 & 928 & 0.35 & 1487 & 0.44 & 1521.6 & 1934 & 1.33 & 3469 & 1.76 & 1586.3 & 1195 & 0.91 & 2585 & 1.38 & 1679.6 @xmath153 & 1831 & 1.04 & 3875 & 1.44 & 1682.7 ] @xmath153 & 1882 & 0.56 & 3930 & 0.80 & 1794.0 ] @xmath153 & 1920 & 0.86 & & & @xmath153 & 1659 & 1.73 & & & lcccc 1365 & 7.21 & 0.44 & 0.061 & 1.31 1480 & 6.99 & 0.46 & 0.065 & 1.33 1640 & 6.98 & 0.46 & 0.064 & 1.33 2100 & 5.82 & 0.30 & 0.051 & 1.28 3000 & 6.48 & 0.25 & 0.037 & 1.19 l@xmath26@xmath154 & 16.95 & 0.30 & 0.015 & 1.07 l@xmath26@xmath155 & 29.10 & 0.42 & 0.012 & 1.06 lcccccccc 1480 & -0.0@xmath156 & 0.0 & 0.0@xmath1570.0 & 0.0@xmath1570.4 & 1.0 & 1.0 & 3.3 1640 & 0.1@xmath156 & 0.2 & 0.0@xmath1570.1 & 0.0@xmath1570.4 & 0.9 & 1.0 & 3.0 2100 & 0.3@xmath158 & 0.3 & 0.3@xmath159 & 0.0@xmath160 & 0.9 & 0.9 & 3.0 3000 & 1.0@xmath161 & 0.7 & 0.5@xmath162 & 0.0@xmath163 & 0.8 & 0.7 & 3.5 4900 & 1.8@xmath164 & 1.8 & 2.2@xmath165 & 2.7@xmath1571.3 & 0.7 & 0.6 & 3.6 5200 & 1.8@xmath166 & 1.7 & 1.4@xmath167 & 1.7@xmath168 & 0.6 & 0.5 & 3.7 6600 & 2.2@xmath169 & 2.3 & 2.5@xmath170 & 2.1@xmath171 & 0.4 & 0.5 & 3.8 6900 & 2.6@xmath172 & 2.4 & 3.0@xmath173 & 2.7@xmath174 & 0.4 & 0.3 & 3.7

we present abundances for 20 elements for stars in two stellar streams identified by : 18 stars from the arcturus stream and 26 from a new stream , which we call af06 stream , both from the galactic thick disk . results show both streams are metal - poor and very old ( 10 gyrs ) with kinematics and abundances overlapping with the properties of local field thick disk stars . both streams exhibit a range in metallicity but with relative elemental abundances that are identical to those of thick disk stars of the same metallicity . these results show that neither stream can result from dissolution of an open cluster . it is highly unlikely that either stream represents tidal debris from an accreted satellite galaxy . both streams most probably owe their origin to dynamical perturbations within the galaxy . [ firstpage ] stars : abundances stars : moving groups galaxy : kinematics and dynamics galaxy : disk

1207.0767

the phase space of the galaxy s disk , as observed near the sun , contains substructure within the larger structures known as the thin and the thick disk . substructures include open clusters and stellar streams , also often referred to as moving groups . moving groups , as promoted by eggen ( see e.g. , @xcite and references therein ) , are considered to be stars having a common space motion and a common chemical composition , and originating from a disssolving open cluster . while some of the well known moving groups do share a common composition ( e.g. , the hr 1614 group - @xcite ) suggesting they come from a tidally - disrupted open cluster , other groups contain stars having very different compositions , a result demanding an origin more complex than disruption of an open cluster . the term ` stream ' is now applied to some entities in galactic phase space . examples in the thin disk include the hercules stream @xcite and the hyades stream or supercluster ( @xcite , @xcite ) . for streams in the thin disk , a likely explanation involves dynamical interactions of disk stars with the central bar @xcite or spiral density waves @xcite . other possibilities arise for streams belonging to the thick disk including accretion from external galaxies . in this paper , we present chemical compositions for subdwarfs belonging to two streams in the thick disk . @xcite undertook a search for fine structure in the phase space populated by subdwarfs from the large sample of f and g subdwarfs considered by @xcite for which @xcite refined data on stellar distances and kinematics . two clumps in phase space were noted by @xcite . one with @xmath1 @xmath2 km s@xmath3 and @xmath4 km s@xmath3 is referred to as the arcturus stream . an arcturus moving group had been previously identified by @xcite . the second stream af06 with a stronger presence in phase space than the arcturus stream is at @xmath5 km s@xmath3 and @xmath6 km s@xmath3 . here , we report chemical compositions of stars from these two streams . we show that stars in both streams span a range in metallicity but with relative abundances which match closely the ratios reported for field thick disk stars . this result serves to constrain greatly explanations for the origins of the two streams . stars selected for observation came from membership lists of the arcturus and af06 streams given by @xcite : 18 of the 22 stars in former stream and 26 of the 44 stars in the latter stream were observed successfully with the tull coud spectrograph @xcite at the 2.7 meter harlan j. smith telescope of the w.j . mcdonald observatory . spectra at a resolving power of 60,000 were obtained with spectral coverage from about 3800 - 10000 with echelle orders incompletely recorded on the ccd beyond about 5800 . wavelength calibration was provided by an exposure of a th - ar hollow cathode lamp . these two - dimensional data were reduced to one - dimensional relative flux versus wavelength spectra using the image reduction and analysis facility(iraf ) . in a typical spectrum , the s / n ratio at the centre of an order over most of the full spectral range was 100 or greater . atmospheric parameters @xmath0 effective temperature ( t@xmath7 ) , surface gravity ( log @xmath8 ) and metallicity ( [ m / h ] ) have been derived from both photometric and spectroscopic data . in the former case , we relied on published catalogues of photometry and parallaxes , empirical calibrations and the theoretical stellar evolutionary models . and , in the latter case , our high resolution spectra were used to derive the atmospheric parameters including microturbulence ( @xmath9 ) . below , both the procedures are described in brief . the t@xmath7 is derived using ( v - k@xmath10 ) colour and strmgren photometry ( @xmath11 ) calibrations . the k@xmath10 magnitude is taken from 2mass catalogue @xcite . the subscript `` s '' stands for the bandpass of the k filter in the 2mass survey , i.e. , the k@xmath10 filter is narrower than the johnson k filter . the k@xmath10 magnitudes are converted to standard `` k '' magnitudes using relations given in @xcite . the mean difference between the two magnitudes is only k@xmath10 @xmath0 k@xmath12 = @xmath00.001 @xmath13 0.005 and will have no effect when k@xmath10 is used in place of k magnitudes in the calibration between ( v - k ) and t@xmath7 . the v magnitudes for all the stars were adopted from @xcite . the ( v - k ) colour and the empirical relations provided in @xcite are used in deriving t@xmath7 . strmgren colours and indices ( @xmath14 , @xmath15 , @xmath16 ) are available for 26 out of 44 stars in the sample @xcite . values of metallicity and t@xmath7 were obtained using empirical calibrations of strmgren colours and indices given in @xcite and @xcite , respectively . values of metallicity are quite sensitive to reddening as it makes observed @xmath17 more positive and @xmath15 values more negative than their intrinsic colours . however , we expect no significant reddening as the stars are nearby ( d @xmath18130 pc from the sun ) . using the methods given in @xcite , reddening values e(@xmath14 ) have been estimated and , indeed , all reddening estimates are vanishingly small : e(@xmath14 ) @xmath19 0.001@xmath130.006 . temperatures derived using ( v - k@xmath10 ) and strmgren colours are given in table 1 as ( t@xmath7)@xmath20 and ( t@xmath7)@xmath21 , respectively . the mean difference between the two temperatures , ( t@xmath7)@xmath20 @xmath0 ( t@xmath7)@xmath21 = 18 @xmath13 90 k , excluding the outliers hip24030 , hip53070 , hip11952 and g192@xmath021 , for which the difference is large 245 @xmath13 24 k , i.e. , t@xmath7 derived from ( v - k@xmath10 ) colour are hotter than those from @xmath17 . the @xmath17 temperature of hip 24030 and ( v - k@xmath10 ) temperature of g192@xmath021 are much closer to the values obtained using spectroscopy . in the case of the other two , we suspect errors in one of the their colours . the log @xmath8 value is derived from the trignometrical parallax , the ( b - v ) colour , and theoretical isochrones @xcite . errors in the parallax and ( b - v ) colour are taken into account in estimating the uncertainty in the log @xmath8 value . a full set of atmospheric parameters ( t@xmath7 , log @xmath8 , @xmath9 , [ m / h ] ) has been derived from spectral line analysis by standard lte techniques . in this exercise , the lte kurucz grid of atlas9 model atmospheres with the convective overshoot option was adopted @xcite . the rationale for choosing overshoot models for solar type dwarf stars was given in @xcite . since we intend to compare our results with the results from the thin and thick disk studies of reddy et al . , we followed their analysis techniques . the lte line analysis code moog @xcite in its 2009 version was used throughout . the effective temperature was set by the requirement that the fe abundance provided by fei lines be independent of the lower excitation potential ( lep ) of the selected lines . while deriving t@xmath7 , caution was taken to minimise the effect of microturbulence by choosing , initially , very weak lines with a sufficient range in lep . later , microturbulence ( @xmath9 ) was derived by adding fei lines of moderately strong equivalent width ( w@xmath22 @xmath19 120 m ) so that the abundance trend becomes sensitive to changes in @xmath9 . the chosen value of @xmath9 is that for which the abundance is independent of equivalent width . the surface gravity log @xmath8 was obtained by requiring that , for the given t@xmath7 and @xmath9 , fei and feii lines give the same fe abundance . the uncertainties in the derived parameters have been estimated by inspection of dependencies for combinations of models of different sets of parameters . in the case of t@xmath7 , we varied the best representative t@xmath7 in steps of 25 k for given log @xmath8 , @xmath23 and [ m / h ] . for steps of 25 k changes , we found no significant changes in the slope as well as in abundances , however , we see ( see figure 2 ) noticeble changes in abundance trends by increasing or decreasing 50 k from its mean model t@xmath7 . thus , we estimate @xmath1350 k as an uncertainty in the best fit model atmosphere . similarly , we found model uncertaities in log @xmath8 and @xmath23 . in figure 2 , estimation of uncertainties is illustrated for t@xmath7 and @xmath24 . in this way , we found model uncertainties in t@xmath7 , log @xmath8 , and @xmath23 : @xmath13 50 k , @xmath13 0.20 cm s@xmath25 , @xmath13 0.20 km s@xmath3 , respectively . these individual uncertainties translate to an effective error of @xmath13 0.05 dex in metallicity [ fe / h ] . next , photometric and spectroscopic estimates of model atmosphere parameters are compared . temperature , log @xmath8 and metallicity comparisons are given in figure 1 . the mean difference between ( t@xmath7)@xmath20 and spectroscopic t@xmath7 is just -9 @xmath13 87 k. there are a few outliers hip24030 and hip94931 ( for which ( t@xmath7)@xmath21 values are closer to spectroscopic t@xmath7 ) , g10@xmath012 and hip9080 ( for which ( t@xmath7)@xmath21 not available ) with a difference of -77 @xmath13 174 k. excluding the outliers the mean difference becomes -2 @xmath13 73 k. the log @xmath8 values obtained from photometry are in good agreement with the spectroscopic values : the difference between the two methods is only 0.002 @xmath13 0.18 cm s@xmath25 for 23 stars for which photometric gravities could be obtained , a difference within the combined uncertainties . the mean difference between the photometric metallicity from strmgren photometry and the spectroscopic metallicity is @xmath00.03 @xmath13 0.11 dex for 25 stars for which strmgren photometry is available . star g192@xmath021 is an outliers whose spectroscopic metallicity is about 0.7 dex metal richer compared to the corresponding photometric value . difference is probably due to erroneous strmgren photometry . for the two metal - poor stars hip53070 and hip11952 , too few lines are available for a spectroscopic determination of parameters and , therefore , we have adopted photometric ( t@xmath7)@xmath21 and log @xmath8 values . the microturbulent velocity @xmath9 calculated using the relation given in @xcite between @xmath9 , t@xmath7 , and log @xmath8 was adopted for both stars . in the final calculations of abundances we adopt parameters derived from spectroscopy but no conclusions would be changed were the photometric parameters adopted . elemental abundances have been derived using measured equivalent widths and synthetic spectra with lte model atmospheres for the adopted stellar parameters taken from kurucz grid @xcite and the 2009 version of lte line analysis code moog . the line list compiled by @xcite was adopted . solar abundances derived by @xcite have been used as rereference values , except eu abundance . the solar eu abundance ( log @xmath26(eu ) = 0.55 ) has been derived , for this study , using the atlas solar spectrum @xcite and two eu ii lines ( 6645.13 , 4129.72 ) . for transitions with significant hyperfine structure ( hfs ) data were taken from kurucz hfs data base @xcite . the lines of mn , v , cu , and eu with hfs were analyzed by computing synthetic spectra . in the case of vanadium , the line at 6216.36 , one of the six lines used previously , is now judged to be blended in the cooler stars and is omitted from the analysis of such stars . elemental abundances are summarized in tables 2 and 3 . uncertainties in the derived abundances arise primarily from uncertainties in the derived atmospheric parameters , the equivalent widths ( w@xmath22 ) and oscillator strengths ( log @xmath27 ) . the latter two quantities affect primarily the line - to - line scatter . of course , the line - to - line scatter is to some extent also influenced by uncertain model parameters . uncertainty in the log @xmath27 values is minimized as our analysis is a differential one with respect to solar spectrum and photospheric abundances . error in the measured w@xmath22 is estimated following the recipe given in @xcite . for the quality of our data , uncertainty in the measured w@xmath22 varies from 1m to about 2.5m . we have taken on average 2m as the uncertainty in the measured w@xmath22 . in table 4 , we have provided for one representative star ( hip40613 ) , estimated uncertainties for each element in the form of @xmath28[x / fe ] due to @xmath29t@xmath7 , @xmath29log @xmath8 , @xmath29 @xmath9 , @xmath29[m / h ] , and @xmath29w@xmath22 . assuming all five sources of error discussed above are independent , the combined error is obtained by adding them in quadrature . this we call our measured uncertainty or @xmath30 . uncertainties in abundances can also be gauged by comparing results for stars that are common in this and other studies . for seven stars that are common with thick disk sample study by @xcite , differences in the derived abundance ratios [ x / fe ] between the studies are given in table 5 . the differences are quite small and less than measured uncertaintity ( @xmath30 ) . the good agreement between the two studies implies that the results of the two streams in this study can be compared directly with the results of thick disk sample study from @xcite . this is akin to a differential analysis with respect to thick disk . both streams were identified by @xcite as over - densities of stars in phase space . to represent stars in phase space one requires information such as coordinates , radial velocities , distance ( parallaxes ) , and proper motions . in the calculation of galactic motions @xcite used hipparcos data @xcite and radial velocities from @xcite . recently , hipparcos observations were re - reduced by @xcite . consequently , we have rederived the galactic velocities ( u , v , w ) for stream members using the revised hipparcos parallaxes and proper motions and including the radial velocities derived from our mcdonald spectra . radial velocities obtained in this study are in very good agreement with the velocities given in @xcite . the mean difference between the two studies is 0.15 @xmath13 0.50 km s@xmath3 . the u , v and w velocities with respect to the sun were calculated using the method given in @xcite . a right - handed coordinate system is used throughout where u is positive towards the galactic center , v is positive in the direction of galactic rotation and w is positive towards the north galactic pole ( ngp ) . velocities ( u , v , w ) are in good agreement with values given by @xcite ; differences between the two studies are mainly due to the updated parallaxes . none of the stars identified as members by @xcite lost their membership in either stream . derived radial velocities ( r@xmath31 ) and galactic velocities ( u@xmath32 , v@xmath32 , w@xmath32 ) relative to local standard of rest ( lsr ) are given in table 6 . in conversion to the lsr frame , the solar motion of ( u@xmath33 , v@xmath33 , w@xmath33 ) = ( @xmath3410.0 , @xmath345.3 , @xmath347.2 ) km s@xmath3 ) is used @xcite . the mean motion ( u@xmath32 , v@xmath32 , w@xmath32 ) of the arcturus stream based on the 18 member stars observed by us is ( @xmath06.48 @xmath13 49.29 , @xmath0124.79 @xmath13 8.92 , @xmath011.5 @xmath13 49.59 ) and for af06 stream the mean motion is ( @xmath041.55 @xmath13 47.45 , @xmath087.35 @xmath13 7.83 , 3.82 @xmath13 54.34 ) . these values are in good agreement with the streams central values given in @xcite . to compute orbital parameters , the ( u , v , w ) of each star is integrated over the galactic potential provided by d. lin ( private communication ) . orbital parameters : mean of apogalactic and perigalactic distance ( r@xmath35 ) , eccentricity ( e ) , maximum distance star away from the galactic plane ( z@xmath36 ) have been derived and given in table 6 . a distance of 8.5 kpc beween the sun and the galactic center is used in the calculation . the ( u , v , w ) of both the streams , as @xcite appreciated , suggest that the streams are part of the thick disk . probabilities that a particular star belongs to the halo , the thick or the thin disk are calculated using the definitions and recipes given in @xcite . the percentage probability ( p ) that a star belongs to the thick disk is given in table 6 . all the arcturus stream members in our sample are thick disk stars with a probability p@xmath37 @xmath38 80 @xmath39 , while 15 out of the 26 af06 stream members have a probability that would qualify them as thick disk members with the remaining 11 stars having probabilities placing them in either the thin or thick disks . in figure 3 , sample stars of the two streams along with the representative members of the thick disk , the thin disk and the halo are shown in the space of angular momentum per unit mass components j@xmath40 , and j@xmath41= @xmath42 the azimuthal and perpendicular components , respectively . in angular momentum space , stars are clustered in a small region compared to their distribution in the velocity space ( see @xcite ) . values are calculated in the right - handed coordinate system and the lsr velocity is assumed to be 220 km s@xmath3 . the value j@xmath43 = 0 implies no rotational velocity in the direction of the galactic rotation and increasing values mean higher rotational velocities . the j@xmath41 represents extent of tilt of star s orbit with respect to the galactic plane . obviously , for the thin disk it is quite small . arcturus stream and af06 stream have a mean j@xmath44 of @xmath0811 @xmath13 77 kpc km s@xmath3 and @xmath01130 @xmath13 63 kpc km , s@xmath3 , respectively . the presence of the arcturus stream was detected by @xcite in catalogues of metal - poor stars by @xcite and @xcite . they found the arcturus stream lagging the lsr by 120 km s@xmath3 but with a large dispersion of @xmath45 @xmath46 50 km / s . their detected structure has angular momentum ( j@xmath40 ) in the range of ( 700 @xmath0 1100 ) km @xmath47 kpc . it appears to us that this structure detected by @xcite is the combined structure of the arcturus and the af06 stream but this structure was separated into two different streams by @xcite using the wavelet transform technique . by combining our samples of both streams , we get mean values v@xmath32 and j@xmath44 ( @xmath0102.67@xmath1320.34 , @xmath01000@xmath13173 ) very similar to what @xcite found for the structure that they thought to be the single structure labelled the arcturus stream . the arcturus stream has been identified in recent studies @xcite as a significant over density in phase space . @xcite s study is based on more than 7000 stars within a distance of 500 pc taken from rave survey . they have identified stars with high orbital eccentricities having over - densities at about v= @xmath0120 km s@xmath3 and the second one at v=@xmath095 km s@xmath3 which coincide with phase space coordinates of the arcturus and af06 streams as given @xcite . @xcite examined a data set of 16,000 giants from the rave survey and showed pronounced over - density of stars at velocities overlapping with the arcturus stream in the solar circle . determination of stellar ages from location of a star in a colour - magnitude diagram is an uncertain procedure because the majority of the stars in both streams fall very close to the zero age main sequence . for the few stars that appear relatively evolved off the main sequence ages were estimated . mean ages with upper and lower limits are given in table 6 . errors in b - v and parallaxes are used to estimate the limits . in figure 4 , the stars are shown in the hr diagram of m@xmath48 versus ( b - v ) colour along with isochrones of ages 10 gyrs , 12 gyrs and 16 gyrs for the metallicities [ fe / h ] = @xmath00.70 dex and [ fe / h ] = @xmath00.51 dex , for arcturus stream and af06 stream respectively . ages of the stream members range from 10 to 14 gyrs . the stellar ages , therefore , are very similar to the ages of thick disk field stars which are all older than about 10@xmath011 gyr @xcite . the derived ages are in good agreement with previous studies of arcturus group ( cf . @xcite , @xcite , @xcite ) based on different selections of stream members . since their discovery ( cf . @xcite ) , various theories have been put forward to explain the substructures in phase space known variously as moving groups or stellar streams . the most prominent theories identify a stellar stream as either a ) a dissolved open cluster , b ) debris from an accreted satellite galaxy , or c ) the result of dynamical perturbations within the galaxy . in establishing the correct explanation for a particular stream , the importance of the chemical signatures or chemical tagging of stream members has been recognized ( @xcite , @xcite , @xcite , @xcite ) not only to pinpoint their origin but also to understand the formation and evolution of the disk . quantitative abundances of twenty elements has been extracted here for members of the two streams . there are three main groups of elements : iron peak ( v , cr , mn , fe , ni , co ) , @xmath49-elements ( o , mg , si , ca , ti ) , and heavy elements such as @xmath50-process ( ba , nd ) and @xmath51-process ( eu ) elements . for the present ( thin ) disk , iron peak elements mostly come from type ia supernovae explosions ( snia ) and the @xmath49-elements are primarily produced in type ii supernovae ( snii ) . the @xmath50-process elements are known to be produced mainly in evolved stars of low and intermediate mass ( 1 @xmath0 8m@xmath33 ) with the lighter @xmath50-process elements also contributed by snii . eu , an @xmath51-process element , is most proably produced in snii . therefore , an abundance ratio of products of snia and snii helps to track the chemical history of a stellar system . abundances of elements from o to eu in the form of [ x / fe ] ratios , where x is any element , are given in table 2 and 3 and shown in figures 5 , 6 , and 7 as the run of [ x / fe ] against [ fe / h ] . the metallicity distribution of the streams is shown in figure 8 for the arcturus and af06 streams . next , we discuss these abundances for the arcturus and af06 streams in the light of the proposed scenarios for the origin of stellar streams . dispersal of an open cluster as the origin of a stellar stream is one explanation from the suite of potential explanations that is directly testable from photometric and spectroscopic determinations of stellar compositions . given that open clusters exhibit chemical homogeneity , if a stream represents a dissolved cluster , one requires chemical homogeneity also among stream members . chemical homogeneity is not a characteristic of these two streams . results shown in figure 8 and table 2 show that stellar metallicities span a wide range for both streams . for the arcturus stream , [ fe / h ] runs from @xmath01.40 to @xmath00.37 . for af06 stream , the range is from @xmath52 to @xmath53 but the range is @xmath01.69 to @xmath00.17 for those 15 members with a high probability of belonging to the thick disk . for a strictly homogenous populations such as open and globular clusters the degree of chemical homogeneity is quite high and which is about 0.05 dex ( @xcite , @xcite ) for a number of elements . in the case of [ fe / h ] dispersions are found to be in the range of 0.02 @xmath0 0.1 and in some extreme cases dispersions are of the order of 0.2 dex @xcite . the wide range in metallicity clearly shows that the systems from which the stream members originated had a relatively long history with a multiple episodes of star formation . we have also inspected the sample for an evidence of a sub - group with chemical homegenity . results shown in figure 5 , 6 and 7 indicate existence of no such a group among arcturus stream sample , however , for the stream af06 , we find a hint of clustering of stars at [ fe / h ] = @xmath00.4 . about 8 stars ( out of total 26 ) show metallicity [ fe / h ] = @xmath00.4 within the dispersion of 0.04 dex . does it mean that part of the sample stars originated from the disrupted cluster ? probably not , this may be a manifestation of thick disk metallicity distribution which peaks at about @xmath00.6 dex , and to some extent due to a smaller sample size . thus , neither the arcturus nor the af06 stream , as shown by metallicity distribution of member stars of the two streams ( see figure 8) , is a dissolved open cluster . this conclusion about the arcturus stream was reached also by @xcite who selected 134 stream members by selection criteria different from those used by @xcite and applied to stellar catalogues other than that compiled by @xcite . their analyses of high - resolution spectra led to a [ fe / h ] range similar to that quoted above ( see also @xcite ) . among thick disk field stars in the solar neighborhood , there is a strikingly very small dispersion in elemental abundance ratios , i.e. , [ x / fe ] at a given [ fe / h ] . indeed , @xcite found the [ x / fe ] to be gaussian - like with a dispersion @xmath54 of less than 0.10 dex for the most of the elements except for v , y , and zr for which @xmath54 is slightly more than 0.1 dex . furthermore , such dispersions were uncorrected for measurement uncertainties so that the intrinsic or ` cosmic ' dispersion must be very small . in sharp contrast to the very similar [ x / fe ] ratios at a given [ fe / h ] for local field stars , element ratios reflecting different contributions from major processes of stellar nucleosynthesis do vary from stellar system to stellar system . for example , ratios in the galactic bulge are not uniformly identical at a given [ fe / h ] to those among local thick or thin disk field stars . similarly , ratios of @xmath49-elements ( [ @xmath49/fe ] ) at a given [ fe / h ] among stars of dwarf spheroidal galaxies differ appreciably from those of local stars and from galaxy to galaxy ( @xcite , @xcite , @xcite ) . this has been illustrtated in figure 9 where ratio of [ @xmath49-process / fe ] ( mean of @xmath49-process elements mg , si , ca and ti ) is compared with that of thick disk @xcite and a number of dwarf spheroidal galaxies for which data was taken from ( @xcite , @xcite ) . therefore , the chemical signatures or tags in the form of [ x / fe ] for those key elements from the major processes of nucleosynthesis may test some proposed origins for the streams . to address the question ` are these two streams chemically identical to thick disk field stars ? ' , we show plots of [ x / fe ] versus [ fe / h ] in figures 5 to 7 with field stars from @xcite shown as by grey symbols , and the arcturus and af06 stream members shown as black filled circles . to quantify possible systematic offsets between the field thick disk and stream members , we compute mean values and dispersions for the arcturus and af06 streams over the [ fe / h ] interval @xmath00.30 to @xmath01.0 . dispersions about the trends is a combination of both the cosmic scatter as well as errors associated with the model parameters . the values ( @xmath55 , @xmath56 ) are computed as the standard deviation of the residuals from straight line fits to abundance trends of [ x / fe ] against [ fe / h ] . similarly , @xmath57 for the thick disk abundance trends of @xcite are computed over the same interval . all results are provided in table 7 . in figure 10 , we made comparison of @xmath57 with those of arcturus and af06 streams . dispersion values about the abundance trends of arcturus stream show very good agreement with that of the thick disk within about 0.02 . however , for af06 stream , in most cases , dispersions are lower by 0.01 @xmath0 0.03 compared to thick disk dispersions . with the current limited sample it would be far fetching to attribute this to the different chemical evolution for the af06 stream and hence external origin to it . mean abundances of arcturus and af06 streams are quite similar and the differences are within 0.05 dex except three elements ( o , cu , eu ) for which difference is 0.06 - 0.07 dex . dispersions about the trends for all the elements are comparable to the estimated scatter ( @xmath58 ) due to uncertainties in model parameters . diferences between dispersions and @xmath58 are within 0.05 dex . thus , we conclude that the streams are chemically identical to within high precision with the field thick disk stars . given that external galaxies of the local group have different element ratios across the [ fe / h ] range sampled by these two streams , especially for @xmath49-elements ( see , for example , @xcite , figure 11 ) , it seems most unlikely that either stream represents debris from an accreted satellite galaxy . the outstanding results of our abundance analyses are that the subdwarfs comprising the arcturus stream and af06 stream identified as over - densities in phase space by @xcite have ( i ) a considerable spread in metallicity and ( ii ) relative abundance [ x / fe ] identical to those of the galactic thick disk . the metallicity spread excludes the hypothesis that either stream represents a dissolved open cluster . the high - degree of similarity between [ x / fe ] at a given [ fe / h ] for these streams and the field stars of the thick disk greatly strains a proposal that these streams represent the tidal debris of an accreted satellite galaxy . by exclusion , the likely origin of these streams is that they are the product of dynamical interactions within the galaxy . however , chemical identiy with the thick disk and the non - similarity of their chemistry with the satellite galaxies within local group alone may not suffice to rule out the possibility of these streams being a result of disrupted satellites . it is still a matter of debate how perturbation can create streams with such a very high velocity drag and exhibit very tight abundance trends . it would be important to know from the models the extent of regions that get affected due to bar / spiral perturbations and their effect on the abundance trends of clumped stars in phase space . astrometry from gaia will provide unprecented sample size as well as accuracy to map the galaxy precisely which would decipher the galaxy formation and evolution . david lambert thanks the robert a. welch foundation for support through grant f-634 . we thank dr . gajendra pandey for useful comments . we also thank the referee for his / her remarks . kirby e. n. , 2011 , in koleva m. , prugniel p. , vauglin i. , eds , eas publications series vol . 48 of eas publications series , the elemental abundance distributions of milky way satellite galaxies . pp 1925 m. e. k. , freeman k. c. , helmi a. , rave collaboration 2009 , in j. andersen , j. bland - hawthorn , & b. nordstrm ed . , iau symposium vol . 254 of iau symposium , the arcturus moving group : its place in the galaxy . pp 139144 & + star & ( t@xmath7)@xmath20 & ( t@xmath7)@xmath21 & log @xmath8 @xmath13 error & ( [ m / h])@xmath21 & t@xmath7 & log @xmath8 & ( [ fe / h])@xmath59 & @xmath9 & n + & k & k & cm s@xmath3 & dex & k & cm s@xmath3 & dex & km s@xmath3 & + & & & & & & & & + + & & & & & & & & + hip105888 & 5798 & 5665 & 4.05 @xmath13 0.05 & @xmath00.81 & 5790 & 4.30 & @xmath00.55 & 1.08 & ( 42,7 ) + hip36710 & 5301 & ... & ... & ... & 5340 & 4.63 & @xmath00.45 & 0.48 & ( 46,6 ) + hip77637 & 5478 & 5550 & 4.31 @xmath13 0.08 & @xmath00.94 & 5580 & 3.73 & @xmath00.85 & 0.90 & ( 30,7 ) + g103@xmath053 & 5435 & 5340 & 4.16 @xmath13 0.09 & @xmath00.66 & 5290 & 4.40 & @xmath00.65 & 0.52 & ( 42,5 ) + g72@xmath012 & 5041 & 5094 & ... & @xmath00.27 & 5060 & 4.64 & @xmath00.40 & 0.67 & ( 43,5 ) + g4@xmath02 & 5258 & 5238 & ... & @xmath00.61 & 5160 & 4.64 & @xmath00.70 & 0.48 & ( 37,4 ) + hip53070 & 5962 & 5719 & 4.23 @xmath13 0.04 & @xmath01.30 & ... & ... & @xmath01.40 & 1.36 & ( 4,3 ) + g204@xmath030 & 5610 & ... & ... & ... & 5550 & 4.42 & @xmath00.80 & 0.69 & ( 27,5 ) + g139@xmath049 & 5331 & ... & ... & ... & 5380 & 4.12 & @xmath00.75 & 0.47 & ( 30,5 ) + g241@xmath07 & 5446 & ... & ... & ... & 5320 & 4.00 & @xmath00.95 & 0.91 & ( 25,4 ) + hip40613 & 5723 & 5670 & 4.16 @xmath13 0.03 & @xmath00.64 & 5670 & 4.02 & @xmath00.55 & 0.90 & ( 45,6 ) + g42@xmath034 & 4858 & ... & ... & ... & 4920 & 4.22 & @xmath00.60 & 0.44 & ( 47,4 ) + hip36491 & 5741 & 5681 & 4.41 @xmath13 0.05 & @xmath00.96 & 5760 & 4.20 & @xmath00.85 & 1.10 & ( 25,6 ) + hip94931 & 4964 & 5118 & 4.56 @xmath13 0.01 & @xmath00.35 & 5120 & 4.58 & @xmath00 40 & 0.55 & ( 48,6 ) + hip74033 & 5647 & 5574 & 4.02 @xmath13 0.04 & @xmath00.92 & 5690 & 4.04 & @xmath00.70 & 1.06 & ( 42,7 ) + g5@xmath01 & 5562 & ... & 4.37 @xmath13 0.08 & ... & 5470 & 4.25 & @xmath01.05 & 0.45 & ( 24,5 ) + g102@xmath044 & 5253 & ... & ... & ... & 5260 & 4.43 & @xmath00.63 & 0.44 & ( 45,5 ) + hip58253 & 5359 & 5351 & ... & @xmath00.37 & 5280 & 4.38 & @xmath00.35 & 0.52 & ( 41,4 ) + & & & & & & & & & + + & & & & & & & & & + g67@xmath040 & 5473 & 5326 & ... & @xmath00.36 & 5370 & 4.42 & @xmath00.35 & 0.94 & ( 48,6 ) + hip9080 & 5078 & ... & ... & ... & 5250 & 4.45 & @xmath00.25 & 0.53 & ( 51,7 ) + g66@xmath051 & 5420 & ... & ... & ... & 5320 & 4.58 & @xmath00.80 & 0.79 & ( 40,5 ) + g106@xmath08 & 5799 & ... & 4.13 @xmath13 0.10 & ... & 5780 & 4.23 & @xmath00.40 & 1.00 & ( 38,8 ) + hip10652 & 5607 & 5499 & 4.35 @xmath13 0.03 & @xmath00.74 & 5580 & 4.42 & @xmath00.60 & 1.02 & ( 38,5 ) + hip22020 & 5610 & ... & 4.12 @xmath13 0.08 & ... & 5690 & 4.15 & @xmath00.20 & 0.95 & ( 51,8 ) + hip26452 & 5837 & ... & ... & ... & 5830 & 4.14 & @xmath00.68 & 0.70 & ( 28,6 ) + hip31740 & 5293 & 5436 & 4.31 @xmath13 0.12 & @xmath00.32 & 5430 & 4.45 & @xmath00.35 & 0.62 & ( 44,7 ) + hip102923 & 4850 & 4933 & 4.61 @xmath13 0.04 & @xmath00.19 & 4950 & 4.50 & @xmath00.25 & 0.69 & ( 49,6 ) + g146@xmath076 & 5170 & ... & ... & ... & 5090 & 2.67 & @xmath01.60 & 0.82 & ( 16,5 ) + hip34642 & 5775 & ... & 4.07 @xmath13 0.05 & ... & 5800 & 4.07 & @xmath00.40 & 1.00 & ( 41,8 ) + g10@xmath012 & 4957 & ... & ... & ... & 5120 & 3.98 & @xmath00.45 & 0.84 & ( 50,7 ) + hip17147 & 5741 & 5722 & 4.22 @xmath13 0.03 & @xmath00.83 & 5700 & 4.23 & @xmath00.85 & 0.71 & ( 37,6 ) + hip24030 & 5915 & 5697 & 4.16 @xmath13 0.11 & @xmath01.06 & 5730 & 4.20 & @xmath01.05 & 1.16 & ( 15,6 ) + hip11952 & 6029 & 5785 & 4.24 @xmath13 0.10 & @xmath01.57 & ... & ... & ... & 1.38 & ( 4,4 ) + hip29814 & 5160 & 5217 & 4.48 @xmath13 0.01 & @xmath00.37 & 5230 & 4.48 & @xmath00.40 & 0.75 & ( 51,6 ) + g197@xmath045 & 5176 & ... & ... & ... & 5250 & 4.02 & @xmath00.60 & 0.85 & ( 38,6 ) + hip104913 & 5355 & ... & 4.40 @xmath13 0.02 & ... & 5380 & 4.48 & @xmath00.03 & 0.80 & ( 51,8 ) + g192@xmath021 & 5790 & 5513 & 4.32 @xmath13 0.10 & @xmath01.23 & 5820 & 4.20 & @xmath00.50 & 0.88 & ( 36,6 ) + g69@xmath021 & 5562 & 5500 & ... & @xmath00.30 & 5620 & 4.29 & @xmath00.20 & 1.07 & ( 48,7 ) + g68@xmath010 & 5680 & 5589 & 4.36 @xmath13 0.06 & @xmath00.57 & 5570 & 4.22 & @xmath00.50 & 0.73 & ( 47,6 ) + g30@xmath046 & 5076 & ... & ... & ... & 5150 & 4.65 & @xmath340.15 & 0.68 & ( 48,6 ) + hip16169 & 5638 & 5575 & 4.34 @xmath13 0.02 & @xmath00.56 & 5690 & 4.56 & @xmath00.48 & 1.10 & ( 38,7 ) + g6@xmath016 & 5786 & 5655 & 4.16 @xmath13 0.10 & @xmath00.19 & 5800 & 4.30 & @xmath00.05 & 1.10 & ( 48,8 ) + g78@xmath041 & 5411 & 5494 & 4.37 @xmath13 0.12 & @xmath00.38 & 5480 & 4.40 & @xmath00.40 & 0.74 & ( 50,6 ) + g25@xmath05 & 5487 & 5508 & ... & @xmath00.36 & 5560 & 4.50 & @xmath00.35 & 0.93 & ( 50,7 ) + @lcrrrrrrrrr@ star & [ fei / h ] & [ o / fe ] & [ na / fe ] & [ mg / fe ] & [ al / fe ] & [ si / fe ] & [ ca / fe ] & [ sc / fe ] & [ ti / fe ] & [ v / fe ] + & & & & & & & & & & + + & & & & & & & & & & + hip105888 & @xmath00.56 & 0.58 & 0.10 & 0.25 & 0.20 & 0.16 & 0.12 & 0.17 & 0.24 & 0.08 + hip36710 & @xmath00.42 & 0.32 & 0.03 & 0.02 & 0.16 & 0.10 & 0.10 & 0.16 & 0.23 & 0.18 + hip77637 & @xmath00.82 & 0.69 & @xmath00.04 & 0.38 & 0.17 & 0.22 & 0.17 & @xmath00.09 & 0.15 & @xmath00.06 + g103@xmath053 & @xmath00.65 & 0.61 & 0.00 & 0.03 & 0.18 & 0.19 & 0.17 & 0.11 & 0.17 & 0.07 + g72@xmath012 & @xmath00.37 & 0.29 & 0.05 & @xmath00.01 & 0.12 & 0.14 & 0.01 & 0.16 & 0.13 & 0.21 + g4@xmath02 & @xmath00.70 & 0.38 & @xmath00.07 & 0.10 & 0.13 & 0.17 & 0.06 & 0.21 & 0.16 & 0.06 + hip53070 & @xmath01.40 & 0.84 & ... & ... & ... & ... & 0.27 & ... & ... & ... + g204@xmath030 & @xmath00.81 & 0.60 & 0.11 & 0.30 & 0.19 & 0.22 & 0.19 & 0.11 & 0.24 & 0.07 + g139@xmath049 & @xmath00.75 & 0.69 & @xmath00.18 & 0.37 & 0.22 & 0.09 & 0.12 & 0.03 & 0.12 & @xmath00.18 + g241@xmath07 & @xmath00.94 & 0.92 & 0.22 & 0.51 & 0.20 & 0.18 & 0.30 & 0.00 & 0.28 & @xmath00.01 + hip40613 & @xmath00.54 & 0.61 & 0.09 & 0.37 & 0.27 & 0.16 & 0.17 & 0.03 & 0.23 & 0.01 + g42@xmath034 & @xmath00.60 & 0.26 & 0.03 & 0.09 & 0.16 & @xmath00.05 & 0.24 & 0.03 & 0.29 & 0.24 + hip36491 & @xmath00.86 & 0.56 & 0.11 & 0.43 & 0.27 & 0.19 & 0.19 & 0.12 & 0.19 & + hip94931 & @xmath00.42 & 0.39 & @xmath00.01 & 0.12 & 0.14 & 0.11 & 0.10 & 0.18 & 0.21 & 0.23 + hip74033 & @xmath00.70 & 0.60 & 0.09 & 0.35 & 0.20 & 0.16 & 0.13 & 0.11 & 0.19 & @xmath00.03 + g5@xmath01 & @xmath01.04 & 0.62 & @xmath00.05 & 0.31 & 0.30 & 0.00 & 0.11 & @xmath00.03 & 0.19 & @xmath00.20 + g102@xmath044 & @xmath00.61 & 0.55 & @xmath00.03 & 0.03 & 0.17 & 0.08 & 0.16 & @xmath00.06 & 0.25 & 0.03 + hip58253 & @xmath00.38 & 0.61 & @xmath00.08 & 0.09 & 0.16 & 0.19 & 0.12 & @xmath00.05 & 0.15 & 0.18 + & & & & & & & & & & + + & & & & & & & & & & + g67@xmath040 & @xmath00.38 & 0.37 & 0.08 & 0.21 & 0.22 & 0.18 & 0.16 & 0.08 & 0.21 & 0.22 + hip9080 & @xmath00.17 & 0.24 & @xmath00.03 & 0.08 & 0.04 & 0.03 & 0.01 & @xmath00.00 & 0.13 & 0.14 + g66@xmath051 & @xmath00.80 & 0.46 & @xmath00.01 & 0.27 & 0.21 & 0.14 & 0.21 & 0.03 & 0.29 & 0.18 + g106@xmath08 & @xmath00.41 & 0.42 & 0.08 & 0.28 & 0.16 & 0.15 & 0.13 & 0.09 & 0.21 & 0.11 + hip10652 & @xmath00.59 & 0.55 & 0.09 & 0.29 & 0.30 & 0.22 & 0.22 & 0.04 & 0.25 & 0.06 + hip22020 & @xmath00.22 & 0.36 & 0.01 & 0.21 & 0.21 & 0.14 & 0.10 & 0.09 & 0.22 & 0.01 + hip26452 & @xmath00.70 & 0.46 & 0.15 & 0.23 & 0.20 & 0.11 & 0.11 & @xmath00.05 & 0.16 & 0.02 + hip31740 & @xmath00.36 & 0.47 & 0.10 & 0.13 & 0.20 & 0.16 & 0.15 & 0.19 & 0.27 & 0.14 + hic102923 & @xmath00.29 & 0.32 & 0.14 & 0.11 & 0.06 & 0.11 & 0.12 & 0.16 & 0.29 & 0.47 + g146@xmath076 & @xmath01.59 & 0.86 & ... & 0.28 & ... & 0.31 & 0.12 & @xmath00.20 & @xmath00.01 & ... + hip34642 & @xmath00.42 & 0.38 & 0.03 & 0.25 & 0.20 & 0.11 & 0.10 & 0.10 & 0.17 & 0.09 + g10@xmath012 & @xmath00.43 & 0.55 & 0.07 & 0.25 & 0.22 & 0.19 & 0.18 & 0.20 & 0.30 & 0.18 + hip17147 & @xmath00.86 & 0.75 & 0.08 & 0.42 & 0.28 & 0.22 & 0.22 & 0.08 & 0.23 & @xmath00.03 + hip24030 & @xmath01.07 & 0.73 & 0.06 & 0.37 & 0.27 & 0.24 & 0.24 & @xmath00.07 & 0.26 & ... + hip11952 & @xmath01.69 & 1.25 & ... & ... & ... & ... & 0.33 & ... & ... & ... + hip29814 & @xmath00.40 & 0.36 & 0.02 & 0.04 & 0.16 & 0.11 & 0.12 & 0.08 & 0.20 & 0.16 + g197@xmath045 & @xmath00.58 & 0.48 & 0.11 & 0.11 & 0.24 & 0.16 & 0.24 & @xmath00.10 & 0.24 & 0.10 + hic104913 & @xmath00.02 & 0.16 & @xmath00.05 & 0.09 & @xmath00.02 & 0.00 & @xmath00.05 & 0.08 & 0.02 & 0.04 + g192@xmath021 & @xmath00.50 & 0.49 & 0.11 & 0.30 & 0.19 & 0.07 & 0.15 & 0.06 & 0.26 & 0.10 + g69@xmath021 & @xmath00.22 & 0.27 & 0.01 & 0.12 & 0.11 & 0.04 & 0.04 & @xmath00.00 & 0.08 & 0.04 + g68@xmath010 & @xmath00.50 & 0.56 & 0.08 & 0.14 & 0.21 & 0.14 & 0.19 & 0.01 & 0.25 & 0.03 + g30@xmath046 & @xmath340.22 & @xmath00.03 & @xmath00.09 & @xmath00.10 & @xmath00.09 & @xmath00.05 & @xmath00.13 & 0.01 & 0.11 & 0.22 + hip16169 & @xmath00.44 & 0.49 & 0.05 & 0.31 & 0.21 & 0.17 & 0.11 & 0.16 & 0.20 & 0.09 + g6@xmath016 & @xmath00.02 & 0.09 & @xmath00.08 & 0.07 & 0.02 & @xmath00.03 & @xmath00.03 & 0.01 & 0.03 & @xmath00.03 + g78@xmath041 & @xmath00.40 & 0.55 & 0.04 & 0.09 & 0.20 & 0.16 & 0.11 & 0.10 & 0.25 & 0.11 + g25@xmath05 & @xmath00.32 & 0.35 & 0.12 & 0.03 & 0.18 & 0.14 & 0.11 & 0.11 & 0.23 & 0.16 + @lrrrrrrrrrrrr@ star & [ cri / fe ] & [ crii / fe ] & [ mn / fe ] & [ co / fe ] & [ ni / fe ] & [ cu / fe ] & [ zn / fe ] & [ y / fe ] & [ ba / fe ] & [ ce / fe ] & [ nd / fe ] & [ eu / fe ] + & & & & & & & & & & & & + + & & & & & & & & & & & & + hip105888 & @xmath00.10 & 0.02 & @xmath00.34 & 0.13 & @xmath00.01 & @xmath00.05 & 0.30 & 0.02 & @xmath00.17 & @xmath00.02 & 0.16 & 0.36 + hip36710 & @xmath00.17 & @xmath00.04 & @xmath00.28 & 0.09 & @xmath00.04 & 0.07 & 0.21 & @xmath00.08 & @xmath00.22 & 0.15 & ... & 0.41 + hip77637 & @xmath00.17 & @xmath00.06 & @xmath00.37 & @xmath00.01 & @xmath00.09 & @xmath00.20 & 0.26 & 0.09 & 0.01 & @xmath00.11 & 0.18 & 0.21 + g103@xmath053 & @xmath00.14 & @xmath00.10 & @xmath00.35 & 0.16 & @xmath00.05 & 0.13 & 0.26 & @xmath00.11 & @xmath00.19 & 0.14 & 0.26 & 0.10 + g72@xmath012 & @xmath00.11 & @xmath00.04 & @xmath00.24 & 0.10 & @xmath00.06 & 0.09 & 0.17 & @xmath00.01 & @xmath00.19 & 0.10 & 0.38 & 0.16 + g4@xmath02 & @xmath00.14 & 0.04 & @xmath00.31 & 0.09 & @xmath00.07 & 0.06 & 0.12 & @xmath00.15 & @xmath00.24 & ... & ... & 0.49 + hip53070 & ... & ... & ... & ... & @xmath00.06 & ... & 0.12 & 0.19 & 0.03 & ... & ... & ... + g204@xmath030 & @xmath00.17 & ... & @xmath00.27 & 0.08 & 0.05 & @xmath00.08 & 0.14 & 0.05 & @xmath00.09 & ... & ... & 0.37 + g139@xmath049 & @xmath00.14 & ... & @xmath00.37 & 0.05 & @xmath00.14 & 0.05 & 0.36 & @xmath00.15 & @xmath00.15 & ... & ... & ... + g241@xmath07 & @xmath00.05 & 0.06 & @xmath00.48 & 0.07 & 0.01 & 0.01 & 0.43 & 0.05 & @xmath00.20 & 0.02 & ... & ... + hip40613 & @xmath00.18 & 0.00 & @xmath00.32 & 0.06 & @xmath00.04 & @xmath00.01 & 0.28 & @xmath00.07 & @xmath00.15 & @xmath00.26 & 0.03 & 0.24 + g42@xmath034 & @xmath00.11 & ... & @xmath00.31 & 0.21 & @xmath00.10 & 0.16 & 0.02 & @xmath00.14 & @xmath00.23 & 0.33 & 0.05 & 0.47 + hip36491 & @xmath00.14 & @xmath00.06 & @xmath00.37 & 0.24 & 0.02 & @xmath00.12 & 0.20 & @xmath00.03 & @xmath00.17 & 0.15 & 0.46 & 0.34 + hip94931 & @xmath00.14 & 0.04 & @xmath00.20 & 0.06 & @xmath00.05 & 0.17 & 0.16 & 0.05 & @xmath00.21 & 0.17 & 0.16 & 0.43 + hip74033 & @xmath00.12 & 0.01 & @xmath00.30 & 0.10 & @xmath00.03 & @xmath00.02 & 0.32 & 0.11 & @xmath00.08 & @xmath00.04 & 0.16 & 0.33 + g5@xmath01 & @xmath00.18 & @xmath00.05 & @xmath00.43 & 0.07 & @xmath00.07 & @xmath00.40 & 0.08 & @xmath00.08 & @xmath00.22 & 0.13 & ... & 0.29 + g102@xmath044 & @xmath00.10 & 0.02 & @xmath00.35 & 0.04 & @xmath00.07 & @xmath00.05 & 0.20 & @xmath00.09 & @xmath00.14 & 0.11 & 0.40 & @xmath00.01 + hip58253 & @xmath00.14 & @xmath00.01 & @xmath00.31 & 0.23 & @xmath00.00 & 0.01 & 0.13 & @xmath00.02 & @xmath00.09 & ... & ... & 0.13 + & & & & & & & & & & & & + + & & & & & & & & & & & & + g67@xmath040 & @xmath00.11 & @xmath00.06 & @xmath00.28 & 0.08 & 0.04 & 0.14 & 0.15 & @xmath00.02 & @xmath00.23 & 0.17 & ... & 0.36 + hip9080 & @xmath00.07 & @xmath00.06 & @xmath00.18 & 0.03 & @xmath00.02 & 0.23 & 0.08 & 0.07 & @xmath00.10 & 0.29 & 0.22 & @xmath00.01 + g66@xmath051 & @xmath00.02 & ... & @xmath00.26 & 0.12 & @xmath00.02 & 0.05 & 0.05 & 0.08 & @xmath00.17 & 0.25 & ... & ... + g106@xmath08 & @xmath00.11 & @xmath00.04 & @xmath00.25 & 0.13 & 0.00 & 0.06 & 0.20 & @xmath00.07 & @xmath00.08 & 0.14 & 0.12 & 0.18 + hip10652 & @xmath00.07 & @xmath00.01 & @xmath00.30 & 0.08 & 0.07 & 0.05 & 0.12 & 0.01 & @xmath00.20 & 0.03 & 0.22 & 0.22 + hip22020 & @xmath00.03 & 0.02 & @xmath00.23 & 0.12 & 0.03 & 0.16 & 0.21 & @xmath00.05 & @xmath00.07 & 0.04 & 0.17 & 0.14 + hip26452 & @xmath00.09 & 0.01 & @xmath00.29 & @xmath00.04 & @xmath00.02 & @xmath00.19 & 0.14 & @xmath00.07 & @xmath00.05 & 0.22 & 0.48 & 0.09 + hip31740 & @xmath00.05 & 0.03 & @xmath00.26 & 0.04 & 0.04 & 0.21 & 0.25 & 0.07 & @xmath00.13 & 0.02 & 0.15 & 0.45 + hic102923 & @xmath00.02 & 0.01 & @xmath00.08 & 0.17 & 0.02 & 0.27 & 0.22 & @xmath00.11 & @xmath00.21 & 0.19 & ... & 0.39 + g146@xmath076 & @xmath00.25 & ... & @xmath00.56 & ... & @xmath00.06 & ... & 0.02 & @xmath00.23 & 0.05 & ... & ... & 0.23 + hip34642 & @xmath00.10 & @xmath00.06 & @xmath00.19 & 0.16 & 0.05 & 0.07 & 0.16 & @xmath00.04 & 0.00 & @xmath00.09 & 0.28 & 0.22 + g10@xmath012 & @xmath00.05 & 0.10 & @xmath00.26 & 0.15 & 0.05 & 0.17 & 0.25 & 0.06 & @xmath60.12 & @xmath00.01 & 0.24 & 0.41 + hip17147 & @xmath00.12 & 0.02 & @xmath00.41 & 0.08 & 0.01 & @xmath00.18 & 0.27 & 0.14 & 0.09 & 0.03 & 0.10 & 0.30 + hip24030 & @xmath00.03 & ... & @xmath00.42 & 0.12 & @xmath00.01 & @xmath00.26 & 0.28 & 0.01 & @xmath00.03 & ... & ... & @xmath00.02 + hip11952 & ... & ... & ... & ... & ... & ... & 0.21 & 0.10 & @xmath00.15 & ... & ... & ... + hip29814 & @xmath00.06 & @xmath00.20 & @xmath00.18 & 0.08 & 0.01 & 0.15 & 0.19 & @xmath00.05 & @xmath00.14 & 0.18 & 0.22 & ... + g197@xmath045 & 0.03 & 0.02 & @xmath00.26 & @xmath00.03 & 0.02 & 0.15 & 0.09 & @xmath00.23 & @xmath00.28 & 0.14 & ... & 0.10 + hip104913 & @xmath00.14 & @xmath00.13 & @xmath00.17 & 0.00 & @xmath00.06 & 0.13 & 0.10 & @xmath00.03 & @xmath00.11 & 0.16 & 0.27 & 0.18 + g192@xmath021 & @xmath00.08 & @xmath00.17 & @xmath00.38 & 0.12 & 0.05 & 0.10 & 0.15 & @xmath00.03 & @xmath00.08 & ... & ... & 0.12 + g69@xmath021 & @xmath00.08 & 0.01 & @xmath00.16 & 0.05 & @xmath00.01 & 0.13 & 0.09 & @xmath00.02 & @xmath00.08 & 0.09 & 0.29 & 0.15 + g68@xmath010 & @xmath00.08 & @xmath00.06 & @xmath00.31 & 0.11 & 0.02 & 0.05 & 0.29 & @xmath00.05 & @xmath00.16 & @xmath00.16 & @xmath00.07 & @xmath00.03 + g30@xmath046 & @xmath00.10 & @xmath00.03 & @xmath00.13 & @xmath00.02 & @xmath00.10 & 0.15 & @xmath00.08 & @xmath00.09 & @xmath00.26 & 0.08 & 0.12 & 0.07 + hip16169 & @xmath00.09 & @xmath00.02 & @xmath00.23 & 0.14 & 0.03 & 0.06 & 0.07 & 0.01 & @xmath00.19 & 0.16 & 0.04 & 0.07 + g6@xmath016 & @xmath00.04 & 0.03 & @xmath00.14 & @xmath00.10 & @xmath00.05 & 0.11 & @xmath00.02 & @xmath00.12 & @xmath00.07 & 0.00 & 0.10 & @xmath00.01 + g78@xmath041 & @xmath00.09 & @xmath00.03 & @xmath00.29 & 0.08 & 0.01 & 0.21 & 0.20 & @xmath00.04 & @xmath00.18 & 0.07 & 0.18 & 0.28 + g25@xmath05 & @xmath00.04 & 0.03 & @xmath00.19 & 0.09 & 0.03 & 0.12 & 0.12 & @xmath00.04 & @xmath00.14 & 0.05 & 0.36 & 0.29 + stimation of uncertainties in the t@xmath7 ( top ) and @xmath23 ( bottom ) . the solid center line is for the best representative model parameter . broken lines represent models with increaseing or decreasing t@xmath7 and @xmath23 in steps of 25 k and 0.1 km s@xmath3 , respectively .

a microscopic formalism is developed that includes the coupling to two particle - hole phonons in the particle - hole propagator by extending the dressed random phase approximation ( drpa ) equation for a finite system . the resulting formalism is applied to study the low - lying excitation spectrum of . it is observed that the coupling to two - phonon states at low energy generates excited states with quantum numbers that can not be obtained in the drpa approach . nevertheless , the two - phonon states mix weakly with particle - hole configurations and participate only partially in the formation of the lowest - lying positive - parity excited states . the stability of the present calculation is tested vs. the truncation of model space . it is demonstrated that when single - particle strength fragmentation is properly considered , the present formalism exhibits convergence with respect to the chosen model space within the confines of the chosen approximation scheme .

nucl-th0212025

recent interest in the spectroscopic factors for the removal of low - lying @xmath1 and @xmath2 strength from @xmath0 points to a substantial discrepancy between experiment @xcite and theory @xcite . the contribution to the reduction of these spectroscopic factors due to short - range correlations is important and is uniformly calculated to be about 10% in @xmath0 using different realistic interactions and theoretical approaches . the best agreement with the data is obtained when also long - range correlations are included in the calculations @xcite . the results of ref . @xcite demonstrate a particularly strong correlation between the theoretically calculated excitation spectrum of @xmath0 and the resulting one - hole spectral function , which describes the excitations of the residual @xmath3 nucleus . the appearance of an additional @xmath2 fragment at small missing energy when a low - lying excited @xmath4 state is obtained is one such correlation @xcite . another one is the appearance of low - lying @xmath5 and @xmath6 strength when the first @xmath7 state in @xmath0 is correctly described @xcite . both features are in accordance with the experimental data @xcite . the calculation of ref . @xcite employed a random phase approximation ( rpa ) approach to describe the excitation spectrum of @xmath0 . the quality of the spectrum in this approach is , however , quite inadequate . even with realistic @xmath8-matrices as residual particle - hole ( ph ) interaction one generates at most a few low - lying collective isoscalar states of negative parity ( @xmath7 and @xmath9 ) and no low - lying isoscalar positive parity states are obtained @xcite . these considerations suggest that an adequate description of the sp fragmentation , at low energy , requires a better description of the excited states within the same framework of self - consistent green s function ( scgf ) . this improvement of the scgf method is also important because it can be applied to heavier nuclei . this is possible since the method describes the features of the low - energy spectrum in terms of interactions between a relatively small number of quasiparticles and collective excitations . these modes represent the experimentally observed excitations of the system and therefore they include both the effects of short- and long - range correlations . the fragmentation of the single - particle ( sp ) strength is thus self - consistently included . early applications to @xmath10 and @xmath11 have been reported in refs . these calculations did not include self - consistency but nevertheless were able to yield a reasonable description of giant resonances and gamow - teller states as well as the low - lying collective states . moreover a formalism based on the bethe - salpeter equation for ph excitations can be extended to include the effects of the continuum @xcite . although the scgf method has only been applied in doubly closed - shell nuclei , it is feasible that increased computational power will allow applications to open - shell systems . an exploratory calculation for semi - magic isotopes of @xmath12 and @xmath13 was reported in @xcite . in the near future a huge amount of data will become available for unstable nuclei from radioactive beam facilities . for this reason it is important to develop techniques that are flexible enough to describe all the above effects . in this paper , we report a first step toward an extension of the scgf formalism in this direction . we chose to use @xmath0 as a test system for the following reasons . first , a successful calculation for this nucleus will help resolving the aforementioned discrepancies for the spectroscopic factors . second , this system has been studied by various approaches in the past . from the results of these calculations one can infer the relevant physical ingredients to be included in the scgf method . third , since the spectrum of @xmath0 is rather complicated , meeting the challenge of describing this system will virtually guarantee good results when the same method is applied to heavier nuclei . shell - model calculations for this nucleus have been reported in refs . @xcite which indicate the particular importance of four - particle four - hole ( 4p4h ) admixtures to the 0p0h ground state in generating the first excited @xmath4 state . other positive parity states at low energy are dominated by 2p2h components , as indirectly confirmed by inelastic electron scattering @xcite . some of this resulting physics was anticipated in terms of deformation effects which simulate these types of many - particle many - hole admixtures @xcite . a complementary point of view is given by the interacting boson model of refs . @xcite . there , the low - lying positive parity states are understood in terms of the coupling of @xmath7 and @xmath9 isoscalar states . one should note that , due to the predominant 1p1h nature of these excitations , the coupling of different phonons generates the npnh configurations relevant to this problem . in refs . @xcite the @xmath7 and @xmath9 states were used as phenomenological boson degrees of freedom to generate the spectrum of the nucleus . the connection with the underlying fermionic description was indicated but not explored completely . this relation was considered in ref . there it was argued that the microscopic ph interaction contains two - phonon exchange contributions which include the actually observed low - lying states of @xmath0 themselves , thereby generating the correct number of low - lying states observed at low excitation energy . in this work we begin to implement the physical ingredients proposed in refs . @xcite by extending the rpa to include the coupling to two - phonon states . the basic idea of the present approach is illustrated in fig . [ fig:2pierpa ] . this figure depicts the coupling of a ph state to two intermediate phonons that are described by the ph propagator itself . if the two intermediate phonons have been computed using the rpa equations , they will already provide a reasonable description of the low - lying collective isoscalar @xmath7 and @xmath9 excited states . these phonons can be sufficient to generate the quantum numbers of the most important positive parity states . particularly relevant is the coupling of two @xmath7 phonons , @xmath14 , that represent some of the correlated 2p2h states of @xmath0 . in the framework of scgf one employs dressed sp propagator in the construction of the microscopic phonons . this results in a dressed rpa ( drpa ) approach for the ph calculation and in its extension to the coupling to two - phonon states . in the present paper , we employ the self - consistent sp propagator of @xmath0 computed in refs . it should be noted that the incorporation of all the 4p4h effects in the present formalism requires a full four - phonon calculation . for this reason , a complete resolution in terms of a microscopic description of the spectrum may only be partially successful . we note that there has been a tremendous progress in recent years in the microscopic description of _ p _ shell nuclei using green s function monte carlo and no - core shell model methods @xcite . a possible application of the no - core shell model to @xmath0 would properly include such 4p4h effects . however the description of spectroscopic factors would still require the construction of effective operators to include the effects of short - range correlations on these quantities whereas these are automatically included in the scgf method . the paper is organized as follows : sec . [ sec:2pierpa ] introduces the formalism to account for two - phonon coupling in the calculation of the ph propagator . the approach employed here is based on a formalism first introduced by baym and kadanoff for the description of response functions in a many - body system at finite temperature @xcite . this framework provides a procedure to construct the effective interaction of the ph bethe - salpeter equation that generalizes the ( d)rpa approach . this method is described in sec . [ sec : baykad ] and the resulting equations in sec . [ sec : erpa_eq ] . more technical details are left to the appendix . [ sec : results ] describes the results for the spectrum within the current approximation scheme . [ sec : trunc ] is devoted to a study of convergence properties related to the number of two - phonon configurations included , and the role of time - inversion diagrams . in sec . [ sec : instab_0 + ] we discuss the possible appearance of instabilities , in particular for the @xmath4 state , that was also observed in ref . conclusions are drawn in sec . [ sec : concl ] . in this work the central quantity of interest is the two - time polarization propagator , whose lehmann @xcite representation is given by @xmath15 in eq . ( [ eq : pi ] ) , the poles and residues contain the information on the response and excitation energies of the system with @xmath16 particles in terms of the quantities @xmath17 where @xmath18 and @xmath19 are the exact energies and eigenstates of the @xmath16-particle system , the subscript @xmath20 refers to the ground state and @xmath21 ( @xmath22 ) is the creation ( annihilation ) operator of a particle in the state @xmath23 . for a clearer discussion of the formalism employed in this work , it is useful to first give a description of the relevant contributions in terms of feynman diagrams . the exact ph propagator ( [ eq : pi ] ) is a solution of the bethe - salpeter equation , depicted in fig . [ fig : bs_eq ] @xcite . this equation can be written schematically as @xmath24 where @xmath25 represents the free propagation of a quasiparticle and a quasihole in the nuclear medium and the ph kernel @xmath26 is , in general , a four - time quantity . according to the baym - kadanoff procedure , a solution for @xmath27 is obtained by first generating a self - consistent solution of the sp propagator using a given choice of the self - energy . from the functional derivative of this ( self - consistent ) self - energy with respect to the corresponding sp propagator , one then obtains the irreducible ph interaction @xmath26 that generates the corresponding conserving approximation for the ph propagator ( when used in the bethe - salpeter equation ) . the standard rpa approach is derived by applying this procedure to the the hartree - fock ( hf ) sp propagator and self - energy . this corresponds to approximating @xmath26 with the bare interaction @xmath28 and employing bare ( hf ) sp propagators as external lines . ( [ eq : bs_eq ] ) then generates the rpa series of ring diagrams shown in fig . [ fig : rpa_exp ] . the extension of the rpa formalism , proposed and implemented by brand _ et al . _ @xcite , was suggested by the observed fragmentation of the sp strength and the necessity to go beyond a lowest - order self - energy for a commensurate theoretical description . there , the baym - kadanoff procedure is applied to a second - order approximation for the self - energy . this yields contributions to the kernel @xmath26 that automatically include all the terms which couple the ph states to the 2p2h ones , in accordance with the pauli principle for the latter states . thus , this formalism takes into account the mixing with 2p2h configurations in the construction of the ph propagator ( the diagram of fig . [ fig : erpaflip]a gives an example ) . in the work of brand _ _ only a single - pole approximation to the self - consistent propagators was employed . it is the aim of the present work to take the effects of the sp fragmentation more completely into account and therefore a fully dressed sp propagator must be used . the one employed in the present work was obtained in ref . @xcite by means of a faddeev expansion for the nuclear self - energy . when one applies the baym - kadanoff prescription to the latter self - energy , a large set of contributions to @xmath26 is generated @xcite . according to ref @xcite , one may expect that the most important of these terms involve the couplings to two ph phonons as depicted in fig . [ fig:2pierpa ] . it is not difficult to see that different diagrams , similar to those in fig . [ fig:2pierpa ] , can be obtained through pauli exchange of the phonon s external lines . in total there are sixteen such possible contributions , corresponding to all the possibilities of connecting two ph phonons to a ph state by means of a single interaction , both in the upper and lower part of the diagram . it is this approximation to the irreducible interaction that will be pursued in the present work . an additional ingredient entering the 2p2h extended rpa ( erpa ) of refs . @xcite requires further discussion . this involves diagrams similar to the one in fig . [ fig : erpaflip]b that are obtained from the 2p2h contribution by inverting the sense of propagation of either the incoming or the outgoing ph pair . these diagrams involve higher excitations ( at least 3p3h , when combined in the expansion of fig . [ fig : rpa_exp ] ) and are expected to give rather small contributions . nevertheless , they represent corrections to the terms of the ph interactions that control the rpa correlations and they also add pauli corrections to the rpa expansion of fig . [ fig : rpa_exp ] at the 3p3h level . in ref . @xcite it was found that they play a role in stabilizing some particular solutions . the present calculation includes both the direct two - phonon contributions of the type depicted in fig . [ fig:2pierpa ] and the diagrams similar to the one in fig . [ fig : erpaflip]c . obviously , all the 2p2h erpa contributions are incorporated in this approach . moreover , full two - phonon configurations are accouned for and a scgf approach is applied . thus , the present formalism is an extension of ( and goes well beyond ) the calculations of refs . however , to avoid complications in the notation , we will still refer to it as `` extended rpa '' , or as `` two - phonon erpa '' whenever confusion may arise . the baym - kadanoff procedure also generates other two - phonon contributions , for example those coupling a pp and a hh phonon . these could be mixed with the presently considered configurations by means of an all order expansion of the faddeev - yakubovsky type . given our present knowledge , such a massive resummation of diagrams does not appear to be relevant for the understanding of the spectrum of @xmath0 . such a study is in any case beyond the scope of the present paper . the usual drpa equations are obtained from eq . ( [ eq : bs_eq ] ) by choosing @xmath29 , as mentioned above , and by keeping dressed propagators as external lines . in this way , one is left with only two - time quantities and , after fourier transformation , the drpa equation becomes @xmath30 where all the indices and summations are shown explicitly . in eq . ( [ eq : drpa ] ) , the _ free _ polarization propagator @xmath31 is also a two - time quantity , with the following lehmann representation @xmath32 where @xmath33 ( @xmath34 ) are the spectroscopic amplitudes for the excited states of a system with @xmath35 ( @xmath3 ) particles and the associated poles @xmath36 ( @xmath37 ) correspond to the excitation energies with respect to the @xmath16-body ground state . the indices @xmath38 and @xmath39 label the eigenstates of the systems with @xmath40 particles and enumerate the fragments associated with the one - particle and one - hole excitations , respectively . when a dressed propagator is used as input , its one - body overlap functions and quasiparticle energies already contain information about the coupling of sp motion to 2p1h , 2h1p and more complex configurations . as a consequence , contributions beyond the 1p1h case are already included in @xmath31 . methods based on an rpa - like expansion produce an infinite series of diagrams in which the direction of propagation can be reversed from backward to forward and vice versa . in the case of standard ( d)rpa , the interaction kernel is simply given the potential @xmath28 and it is the same for every contribution to the diagrammatic expansion . therefore the usual ( d)rpa equation can be written in compact form , as in eq . ( [ eq : drpa ] ) . this is no longer true when one aims to include additional contributions and at the same time insists on working with two - time quantities . the two - phonon diagrams of figs . [ fig:2pierpa ] and [ fig : erpaflip]c have different analytical expressions . as a consequence one first needs to separate all the four possible time directions of the kernel @xmath26 forward to forward , backward to backward , and the two time - inversion cases before including the relevant diagrams in the bethe - salpeter equation ( [ eq : bs_eq ] ) . this can be achieved by splitting the free ph propagator ( [ eq : pif ] ) into its forward- and backward - going parts , denoted by @xmath41 and @xmath42 respectively , @xmath43 by performing this substitution in eq . ( [ eq : bs_eq ] ) one obtains a similar separation for the complete ph propagator @xmath44 where @xmath41 and @xmath42 now refer to the sense of propagation of the final lines only . suppressing the indices and summations one obtains @xmath45 the last step consists in substituting eq . ( [ eq : split_pi ] ) into ( [ eq : split_bs ] ) and approximating each component of the ph kernel @xmath26 with the sum of the bare interaction and the corresponding two - phonon contributions . the result corresponds to the erpa equations given by @xmath46 in eq . ( [ eq:2pierpa ] ) , @xmath47 represents the contribution of all the sixteen two - phonon diagrams , fig . [ fig:2pierpa ] , in the forward direction . @xmath48 corresponds to the contributions connecting the backward - going terms . analogously @xmath49 represent the sum of the diagrams in which a backward - going hp state is inverted in a forward - going ph one ( fig . [ fig : erpaflip]c illustrates such a case ) while @xmath49 is the time - reversed contribution . the practical implementation of eqs . ( [ eq:2pierpa ] ) requires additional manipulation to treat the freely propagating lines in figs . [ fig:2pierpa ] and [ fig : erpaflip]c . this situation is completely analogous to the one already discussed in ref . @xcite for the 2p1h expansion of the self - energy . specific details for the erpa equation ( [ eq:2pierpa ] ) are given in the sections in the appendix . the next section reports on the application of this formalism to @xmath0 . as in the work of ref . @xcite , the erpa equations where solved within a model space consisting of a finite set of harmonic oscillator states . all the first four major shells ( from @xmath50 to @xmath51 ) plus the @xmath52 where included to account for the sp orbitals that are most relevant for low - lying excitations . the harmonic oscillator parameter was chosen to be @xmath53=1.76 fm . as a consequence of the truncation of the model space a brueckner @xmath8-matrix was used as a microscopic effective interaction . this @xmath8-matrix was derived from a bonn - c potential @xcite and computed according to ref . @xcite . the contributions of two - phonon states were first studied by solving the erpa equation ( [ eq:2pierpa ] ) with an independent - particle model ( ipm ) propagator . this propagator was constructed from a slater determinant composed of the lowest occupied harmonic oscillator ( h.o . ) orbitals in the model space . where the effects of nuclear fragmentation were included , the calculations employed the fully dressed sp propagator computed in ref . this propagator contains no more than two main quasihole ( quasiparticle ) fragments for each state in the @xmath54 ( @xmath55 ) shell . for these orbits , the sp propagator can therefore be well represented by means of one or two principal fragments and an effective pole that accounts for strength far from the fermi energy . for quasiparticle states associated with the @xmath56 shell , instead , the fragmentation is more substantial @xcite and the self - consistent propagator was approximated by including the most important poles ( up to 4 main fragments for the @xmath2 ) plus two effective ones that gather all the background strength . we have checked that the results are not sensitive to the details of the fragmentation of the @xmath57 orbitals . for the @xmath58 orbitals , we found that the strength distribution is sufficiently represented by the adopted prescription and further discuss the sensitivity to the results in sec . [ sec : instab_0 + ] . in both cases ( imp and dressed inputs ) , the standard ( d)rpa equation was solved first , in order to generate the ph phonons that entered the final two - phonon erpa calculation . when coupling different phonons , only the lowest few states of @xmath0 have the right quantum numbers , and energies low enough , to generate two - phonon contributions with unperturbed energy below 20 mev . these are the only states relevant for our purposes @xcite . in practice , the calculations were performed by including , in each channel , all two - phonon configurations up to 30 mev . the stability tests reported in sec . [ sec : furt_stab ] demonstrate that , for the case of a dressed input propagator , higher two - phonon contributions do not change the results for the low - lying states appreciably . the results obtained from the h.o . input propagator are displayed in fig . [ fig:2pierpa_ho ] . the isoscalar eigenvalues obtained for energies below 15 mev are displayed , for both the standard rpa and for the erpa calculation . the erpa calculation produces lower energies for some of the states that were already obtained in rpa . both @xmath7 and @xmath4 shifted down to about 0.5 mev above the experimental energy . as discussed further in sec . [ sec : furt_stab ] , this is to be considered rather fortuitous and one must remember that the dressing of the sp propagator , which has the effect to raise these eigenvalues , has not been taken into account yet . still , this result indicates that correlations between ph and two - phonon states can be substantial and go in the right direction in explaining the experimental results . it is also worth considering the total ph spectral strength of these states obtained by summing the corresponding amplitudes , eq . ( [ eq : piampl_def ] ) , as follows @xmath59 results for both the excitation energy and @xmath60 of the principal levels in fig . [ fig:2pierpa_ho ] are given in table [ tab:2pierpa_ho ] . for the erpa case , also the fraction of quasiparticle - quasihole and two - phonon configuration that appear in the wave function are shown . note that @xmath60 is substantially bigger than one for the @xmath7 and @xmath4 states and that these values increase further for the erpa results . this signals an increase of the collective character of these solutions which may lead to an instability of the rpa equations for interactions that are even more attractive . the two - phonon erpa approach generates a triplet of states at about 14 mev , with quantum numbers @xmath4 , @xmath61 and @xmath62 . however , these levels are almost exclusively composed of two - phonon contributions and contain only small admixtures of ph states , resulting in a small ph spectral strength @xmath60 . the quantum numbers and energies of these states indeed correspond closely to those obtained by coupling two @xmath7 rpa phonons , each at 7.14 mev . we note that a similar triplet is found experimentally at 12.05 , 11.52 and 11.10 mev , which also corresponds to twice the experimental energy of the first @xmath7 phonon . the first experimental @xmath61 is found at lower energy and its spectral strength is known to have relevant ph components @xcite . thus it can not be identified with any of the above two - phonon contribution . for all the lowest states that are not already reproduced by standard rpa a very small ph spectral strength has been found , due to a general lack of mixing between the ph and the two - phonon configurations . of interest is also the @xmath61 that represents the giant quadrupole resonance at 20.7 mev . in this case rpa and erpa give 22.9 and 23.3 mev for the main peak but with a lower @xmath60 in the second case . for this state , part of the ph strength ( about 20% ) is shifted to two - phonon configurations representing the expected fragmentation of the giant resonance . [ fig:2pierpa_3 g ] and table [ tab:2pierpa_3 g ] show the analogous results when dressed sp propagators are used as input . in general , the main effect of fragmentation is to screen the nuclear interaction as a consequence of the quenching of spectroscopic factors for the input sp propagator . for the drpa case , this results in increasing the lowest @xmath7 and @xmath9 solutions by @xmath632 mev . more substantial is the effect on the lowest @xmath4 state with a predominantly ph character which rises to about 17 mev , confirming the sensitivity of this state to details of the fragmentation and the strength of the nuclear interaction . unlike the ipm case , we have chosen to solve the two - phonon erpa by first shifting the lowest solution for both the @xmath4 , @xmath7 and @xmath9 states down to their relative experimental energies . this has the advantage of lowering the most important two - phonon configurations and allows to investigate their interplay with the ph ones . as can be seen in fig . [ fig:2pierpa_3 g ] , the erpa equations still generate a triplet of @xmath4 , @xmath61 and @xmath62 states at twice the energy of the first @xmath7 phonon . due to the screening of the ph interaction , these states mix very little with the ph configurations yielding an almost degenerate triplet . table [ tab:2pierpa_3 g ] also gives a comparison between the total two - phonon content of the states and the individual contributions of the most important configurations . this decomposition demonstrates that this triplet is formed by pure @xmath64 states . this observation is in accordance with the 2p2h character of these states @xcite . since the overall energy of these states approximately agrees with experiment , one can expect that this calculation correctly represents the bulk properties of their wavefunctions . nevertheless , it is clear that an additional interaction is needed in order to split this triplet as observed experimentally . it should be noted that no solution that can be identified with the @xmath61 and @xmath62 levels at 9.8 and 10.4 mev . the @xmath61 level has been interpreted in terms of an @xmath23 particle rotating around an excited @xmath65 core @xcite and therefore involves correlations that may go well beyond the present calculation . with regard to the @xmath61 strength around 20 mev , we note that the sp fragmentation included in the drpa equation already generates a distribution of @xmath61 strength . table [ tab:2pierpa_3 g ] contains a few of these solutions and shows that also for the dressed case a sizable mixing of ph and two - phonon contributions is obtained at this energy . this mixing generates the spreading of ph strength over different solutions , consistent with the finite width of such a resonant state . it should be noted , however , that a discrete basis is used here and therefore it is not possible to properly describe the continuous strength distribution of a giant resonance , for which a continuum or a complex basis should be used @xcite . also interesting are the results for the other isoscalar @xmath4 states . the lowest solution with a predominant ph character , that was found in drpa at @xmath6317 mev , is obtained at a similar energy but is now characterized by a partial mixing with two - phonon configurations . the ph and two - phonon contents of this state given in table [ tab:2pierpa_3 g ] show that this is the result of mixing with the the lowest solution , which ends up at @xmath6311 mev . the latter is predominately a two - phonon state . it is also seen that in both cases the relevant configuration comes from the coupling of two @xmath4 phonons . configurations involving two @xmath7 s make smaller contributions while the other modes not reported in the table were negligible . we note that the wavefunctions for these states contain several relevant ph configurations , obtained from different quasiparticle fragments in the @xmath56 ( @xmath55 ) shells combined with quasihole fragments of the @xmath54 ( @xmath66 ) shells . therefore , the situation is more complicated than the simple picture of only two levels interacting with each other . we observe that the shell - model calculations for the first excited state of @xmath0 give very small contributions from both 0@xmath67 and 2@xmath67 configurations and a strong population of 4@xmath67 states @xcite . on the other hand , inelastic electron scattering experiments @xcite clearly excite this state identifying its partial ph character even though the state is dominated by 4p4h components . from the point of view of the scgf approach , the one - body response is completely described by the ( dressed ) ph propagator , eq . ( [ eq : drpa ] ) , and therefore the total one - body strength must be represented by @xmath60 ( [ eq : z_tot_ph ] ) . this observation points to the necessity of a stronger mixing between ph and two - phonon configurations than obtained in the present calculations . still , the present results indicate that the two - phonon configuration obtained by coupling the two lowest @xmath4 phonons plays a role in the structure of the first @xmath4 state itself . this means that , starting from the dressed ph admixture contained in it , a fully self - consistent calculation would also generate some contributions of 2p2h , 4p4h , and beyond . the calculations of ref . @xcite have shown that the bulk of the 4p4h contributions to the first excited state of @xmath0 may come from the coupling of four different phonons with negative parity ( @xmath7 and @xmath9 ) . however , the self - consistent role of coupling positive parity states was not conidered in that work . both this effect , the rpa correlations and the inclusion of the nuclear fragmentation allow for the at least partial inclusion of configurations beyond the 2p2h case even when no more than two - phonon coupling is considered , as in this paper . a study of the importance of three- and four - phonon configurations within this approach is beyond the scope of the present work . the principal negative parity isoscalar states are still shifted down in energy by erpa over drpa , as expected , but the improvement is less than 0.5 mev . the low - lying @xmath7 and @xmath9 levels remain substantially above the experimental energy at 9.23 and 10.90 mev , respectively , and more correlations will be needed in order to lower their energy . we note that these wavefunctions contain a two - phonon admixture obtained by coupling the low - lying @xmath4 excitation to either the @xmath7 or @xmath9 phonons . according to the above interpretation of the first @xmath4 excited state this corresponds to the inclusion of 3p3h and beyond . at the same time two additional negative parity solutions , with two - phonon character , are found at higher energy in accordance with experiment . the erpa results for an h.o . input give appropriate corrections to several energy levels . however , such corrections tend to become negligible in the successive calculations employing dressed propagators . presumably this happens because a sizable part of the correlations , which in the latter case are introduced by erpa , are already accounted for by the dressing of the sp propagators . in general , the inclusion of fragmentation has also the effect of screening the interaction between ph and two - phonon configurations . however , the results of table [ tab:2pierpa_3 g ] show that a considerable mixing between the two can be obtained if all the relevant configurations are sufficiently low in energy , although no substantial mixing between different two - phonons states is seen . according to these considerations , the biggest deficiency of the present approach is probably the lack of an interaction within the multi - phonon space . refs . @xcite have shown that such correlations are due to pp and hh interactions , neglected here , and that their inclusion has important effects on the final result for the spectrum . these terms may considerably lower most of the eigenstates of the erpa equations . however , a second type of correlation between multi - phonon states , that was not accounted for in refs . @xcite , is the pauli exchange between the fermion lines included in different phonon propagators . these will probably generate the opposite effect , screening part of the pp and hh correlations . although the results obtained with a dressed input propagator for the lowest excited states are not completely successful , it must be recognized that the present erpa calculation does represent a step forward . indeed a proper description of two - phonon contributions , at the present level , had not been accomplished before . these states not only account for configurations at the 2p2h and higher level but appear to have some relevance for the description of states at higher energy . in addition , excitations are generated with quantum numbers that can not be obtained in the standard ( d)rpa approach at these low energies . given the results of the present work , it appears that a calculation including both pp and hh correlations , a more complete treatment of pauli effects and the most relevant configurations up to the four - phonon states can be achieved with the presnt - day computers . such a calculation is planned for the future . the rpa approach may break down for strongly attractive interactions , by genereating an excess of collectivity and lowering the first excited state below the ground - state energy . the inclusion of fragmentation of the sp strength can affect this behavior in several ways . first , both the screening of the nuclear interaction and the splitting of the particle and hole spectral strengths over different fragments tends to stabilize the drpa equations . second , correlations lower most of the sp energies in comparison with the ipm . this results in smaller unperturbed ph energies and therefore has the opposite effect of pushing the drpa approach toward instability . third , the instability can be sensitive to the details of the spectroscopic amplitudes . this last point can be best illustrated by looking at the sp strength with @xmath68 , to which the present calculation is most sensitive . for the model space employed in this work a given particle overlap function is a superposition of two h.o . eigenstates , @xmath69 where we have stressed the fact the two components can sum either constructively or destructively . the fragmentation introduces a non - zero @xmath70 component even for particle states that in the ipm would be described only by the @xmath71 subshell . by considering eqs . ( [ eq : drpa - bmtx ] ) one can see that the ph interaction for the drpa approach can be substantially changed by this new component by virtue of the strong matrix elements of the effective interaction between @xmath72 states . in particular , this also applies to the off - diagonal terms that drive the rpa - like correlations . it is also worth nothing that the sign of this correction changes accordingly to the relative sign of the components in eq . ( [ eq : overlap ] ) , therefore pushing the states either towards stability or instability , respectively . we have observed that choosing a wrong sign for one or more of the @xmath2 particle fragments has drastic effects for the solutions of the isoscalar @xmath4 channel , with the choice of constructive interference in eq . ( [ eq : overlap ] ) leading toward instability of the drpa approach . for the input propagator employed in this work , all of the background fragments have a similar wavefunction and a negative or positive interference sign according to whether their sp energy is lower or higher than 20 mev , respectively @xcite . therefore it is natural to collect such background distributions in two different poles , one for each region of missing energy . a similar behavior applies also to the other quasiparticle states in the @xmath58 and @xmath55 shells . we have checked that the results of sec . [ sec : results ] are stable with respect to the number of main fragments and effective poles included in each subshell , therefore no instability affects the present work . however , in applications in which the number of poles has to be strongly reduced it may not be possible to chose effective poles of the sp propagator that represent the details of the spectral distribution sufficiently accurately . in such a case an instability in the drpa approach may arise . this situation occurred with a previous drpa calculation of the isoscalar @xmath4 channel , reported in ref . @xcite , in which a smaller number of fragments was included . thanks to improvements in the computer code that solves the drpa problem , it is now possible to consider a sufficiently large number of poles in the sp propagator and this problem has been overcome . the contribution of two - phonon terms to the time inversion diagrams of fig . [ fig : erpaflip ] can , in principle , generate other correction to the interaction between ph and hp states that drives the rpa correlations . nevertheless , the effect is very small due to the large energy denominators that appear in eqs . ( [ erpa - bmtx1 ] ) and ( [ erpa - bmtx2 ] ) . we have tested their influence by neglecting the corresponding terms @xmath49 and @xmath73 in eq . ( [ eq:2pierpa ] ) and report the results in table [ tab : b_mtx ] . no appreciable change is generated illustrated by differences with respect to the results of sec . [ sec : results ] of at most 0.1% . it is worth noting that the contribution of @xmath49 and @xmath73 can in principle carry information on the pauli breaking at the level of 3p3h and beyond . therefore , they may become more important for the case of three - phonon calculations . nevertheless , table [ tab : b_mtx ] suggest that they are not likely to play an important role in the description of the spectrum of @xmath0 . [ fig:2phon_trunc ] shows the results for selected solutions of the erpa equations , obtained by employing different sets of two - phonon states . for any given point , only those configuration with energy @xmath74 have been included in the calculation . for @xmath75=70 mev , about 700 to 1000 two - phonon contributions have been included , depending on the channel . this is one order of magnitude larger than the number of ph states that enter the calculation . since no two - phonon configuration has energy lower than @xmath76 mev , the leftmost points in fig . [ fig:2phon_trunc ] correspond to the simple ( d)rpa calculation . as the lowest few two - phonon contributions are included , the solutions for these levels show a sudden jump for all the lowest excited states . after this , the results obtained by using an ipm input still continue to exhibit a dependence on the value of the cut - off , roughly lowering states by @xmath77 mev every time that @xmath75 increases by @xmath78 mev . in particular , the @xmath4 and @xmath7 solutions decrease progressively , eventually heading to rpa instability . this confirms that the values reported in table [ tab:2pierpa_ho ] for these states do not correspond to converged results . the situation is much better for the case of a dressed input propagator , for which the low - lying solutions are approximately stable for @xmath79 20 mev . this confirms that corresponding higher - energy two - phonon excitations do not have strong influence as their effects have already been included by the dressing of the sp motion . in fig . [ fig:2phon_trunc ] we also show the results for the most relevant of the @xmath61 solutions that represent the resonance at about @xmath78 mev . here the variation is more appreciable , in particular for values of the cut - off @xmath75 comparable with the energy of the state itself . also here a more stable behavior is obtained for the case of a dressed input propagator . we conclude that when the nuclear fragmentation is accounted for the choice of @xmath80 mev adopted in sec . [ sec : results ] is a adequate for considering the low - lying spectrum . as a second possible source of uncertainty one may consider the size of the employed model space . the choice of the model space employed in sec . [ sec : results ] allows to take into account excitations up to the @xmath51 shells . this means that ph configurations up to @xmath81 mev are included , consistently with the choice for truncating two - phonon states discussed above . the calculations of ref . @xcite also show that this is adequate and suggest that better convergence is to be expected for a calculation with a dressed input . to check this for the present calculation , we have augmented the model space by adding two more h.o . shells , up to @xmath82=5 , and recomputed the @xmath8-matrix elements accordingly . the input sp propagator used in the calculations was obtained by including an extra particle pole for every subshell added to both the ipm s slater determinant and the fragmented dressed propagator . the sp energies for the new shells were chosen by solving the brueckner - hartree - fock equation within the new model space , in an approach analogous to the one of ref . @xcite . the results are compared to the ones obtained for the smaller space in table [ tab : b_mtx ] . for the ipm , a sizable variation of the @xmath7 excitation energy is found . also , the high values of the ph strength @xmath60 for both the @xmath4 and @xmath7 solutions indicate that these states are approaching instability . as a consequence of the different @xmath7 excitation energy , which also enters the two - phonon calculation , most of the erpa solutions are shifted down in energy . the main conclusions of sec . [ sec : results ] , however , remain unchanged . when dressed sp propagators are employed , no dramatic change occurs for the low - lying solutions and only some states with dominant ph character are shifted by no more that 1 mev . this suggests that the main conclusions of sec . [ sec : results ] are also not affected by the truncation of the model space , whence the bulk of the missing correlations has to be looked for in an extension of the diagrammatic expansion . we note , however , that a more attractive g - matrix is expected to be obtained if the continuum outside the model space were modified by accounting for self - energy corrections . this feature may lead to a further improvement of the present results . in addition , one may explore the inclusion of effective three - body force as due to propagation outside the model space . such effective forces are critical in obtaining the correct binding energy in no - core shell model calculations reported in ref . @xcite . the dressed rpa equations have been extended to account for the coupling of two ph phonons in forming the excited states of a many - body system . the coupling among ph and multi - phonon configurations is conceptually similar to the interacting boson model of ref . @xcite but limited , in this application , to only two - phonon admixtures . nevertheless , the present approach has the added advantage of taking into account both the effects of nuclear fragmentation and the rpa - like correlations , as generated by two - phonon fluctuations in the ground state . the resulting extended rpa formalism has been applied to study the excitation spectrum of @xmath0 . the results suggests a sizable mixing of ph and two - phonon configurations for the case of the @xmath61 quadrupole resonance at 20.7 mev . other solutions , carrying quantum numbers that can not be generated by the simple drpa , are obtained with this method , among which a few isoscalar negative parity states and a triplet with @xmath83=@xmath4 , @xmath61 and @xmath62 near 12 mev , in accord with experiment . in particular , the states in the triplet were seen to be almost pure two - phonon configurations obtained by coupling two @xmath7 phonons . this feature confirms the 2p2h character of these states . the results are less satisfying for the low - lying positive parity states , which are known to require a proper description of 4p4h excitations . the present approach predicts the lowest solution for the isoscalar @xmath4 channel to be at 11.3 mev , considerably above the first experimental excited state . a sizable component of this solution is seen to be generated by the coupling of two of the lowest @xmath4 phonon themselves . this suggests that important contribution of 4p4h excitation may be included already at the two - phonon level in a self - consistent fashion . other correlations beyond the 2p2h level are partially included in the present calculation through the dressing of the propagators and the rpa approach . as expected , though , the present implementation is not sufficient to obtain a complete description of the low - lying excitation spectrum of @xmath0 . the results obtained in this work show that an interaction between multi - phonon configuration is still missing . this can be achieved by including pp and hh correlation between different ph phonons . also the inclusion of up to four - phonon states is expected to be relevant for this system . finally , the stability of the present results has been tested with respect to the truncation of the model space and to the number of two - phonon configurations accounted for . it was found that the effects of nuclear fragmentation acts to ` renormalize ' the sp propagator , making the solutions of the erpa fairly independent of higher energy configurations . as discussed in sec . [ sec : trunc ] , this feature generates stable solutions with respect to the size of the model space and the corresponding @xmath8-matrix interaction used in this work . the four - phonon calculation of refs . @xcite give a very good description of the excitation spectrum of @xmath0 . those findings and the effects of fragmentation discussed in this paper suggest that the present erpa formalism can be suitably extended to generate a satisfactory description of this nucleus , within the framework of scgf . such extensions are presently under consideration . such a calculation may also significantly reduce the discrepancy between the measured and the theoretical sp spectral function for this nucleus @xcite . this work was supported in part by the u.s . national science foundation under grants no . phy-9900713 and phy-0140316 and in part by the natural sciences and engineering research council of canada ( nserc ) . when reducing the kernel of the bethe - salpeter eq . ( [ eq : bs_eq ] ) to a two - time quantity , one has to deal with the fact that some lines in the diagrammatic expansion continue to propagate unperturbed while some interaction occurs between other particles . this is also the case for the diagrams of figs . [ fig:2pierpa ] and [ fig : erpaflip ] . this situation can be overcome by redefining the objects that appear in eq . ( [ eq:2pierpa ] ) and promoting the quantum numbers @xmath84 , that label particle and hole fragments to external indices . the usual form of the ph propagator , eq . ( [ eq : pi ] ) , is obtained only as a last step of the calculation by performing the sum over the @xmath84 indices @xcite . the separation of the propagators @xmath85 and @xmath86 into forward and backward components , outlined in sec . [ sec : erpa_eq ] , is a natural consequence of adopting this prescription . indeed particle and hole external lines turn into each other by time inversion and become quantities that depend on different quantum numbers , i.e. the fragmentation indices @xmath87 . for this reason , eqs . ( [ eq : split_pif ] ) and ( [ eq : split_pi ] ) are only formal relations and the arrow can be substituted by an equal sign only before summing over all the particle and hole fragments . the free polarization propagator ( [ eq : pif ] ) naturally splits in two components that are purely forward and backward - going . the relevant lehmann representations are @xmath88 where no summation is implied and a shorter notation for the unperturbed poles and residues has been introduced . the quantities @xmath89 and @xmath8 can be thought as matrices whose elements contain all the unperturbed poles and residues , respectively . note that in this case @xmath89 is diagonal and depends on the fragmentation indices @xmath90 only , while @xmath8 is rectangular because its column indices depend on the model space orbitals @xmath91 as well . the separation of the complete propagator is a little more complicated . the splitting of eqs . ( [ eq : split_pi ] ) and ( [ eq : split_bs ] ) involves the time direction of the outgoing lines only . since the rpa series contains contributions that can invert several times the sense of propagation of the ph diagrams , both forward and backward poles can appear in each component @xmath92 in eqs . ( [ eq : pi > ] ) and ( [ eq : pi < ] ) the spectroscopic amplitude splits in two contributions @xmath93 and @xmath94 . these appears unchanged in both equations due to the time - inversion symmetries obeyed by the ph states . in terms of these definitions , and applying the summation prescription , eq . ( [ eq : split_pi ] ) can be exactly formulated as follow @xmath95 \ ; , \label{eq : zf+b } \\ \pi^{>}_{\alpha \beta , \gamma \delta}(\omega ) & = & \sum_{n_1 , k_2 } \left [ \pi^{>}_{\alpha n_1 \beta k_2 , \gamma \delta}(\omega ) + \pi^{<}_{\alpha k_2 \beta n_1 , \gamma \delta}(\omega ) \right ] \ ; . \label{eq : pif+b}\end{aligned}\ ] ] the eigenvalue equation for the erpa is obtained in the usual way , by substituting eqs . ( [ eq : pif > ] ) to ( [ eq : pi < ] ) into eq . ( [ eq:2pierpa ] ) and then extracting the residues of the ph poles @xmath100 . the result is an eigenvalue equation in terms of the vectors @xmath93 and @xmath94 . to linearize the problem , we introduce the following components @xmath101 where @xmath102 and @xmath103 represent the ph amplitudes that appear in the standard ( d)rpa equations @xcite and @xmath104 and @xmath105 are the analogous two - phonon amplitudes introduced by the erpa approach . the components @xmath102 and @xmath103 are related to @xmath93 and @xmath94 respectively by @xmath106 eqs . ( [ eq:2pierpa ] ) can be put in the form of a linear eigenvalue equation @xmath107 where the matrix @xmath108 is defined as @xmath109 ( g^*)^\dag & \\ k^ > g^\dag & e & & \\ -g^ * [ v~+~h^ { < , > } ] g^\dag & & - g^ * v ( g^*)^\dag - d & g^ * { k^<}^\dag \\ & & k^ < ( g^*)^\dag & - e \end{array } \right ] \ ; . \label{eq : erpamtx}\ ] ] the off - diagonal 2@xmath1102 blocks in eq . ( [ eq : erpamtx ] ) describe diagrams in which the time direction of propagation is inverted . in the present case the only non vanishing elements are the ones that involve the inversion of a single ph state into a hp one , or vice versa . these correspond to the sum of the first - order term @xmath28 , which represents the kernel of the bare rpa , and the more complex diagrams of fig . [ fig : erpaflip ] . blank spaces would in principle allow to include more complicated contributions , that involve time inversion of two - phonon diagrams into ph or hp configurations . these contributions are not expected to play a relevant role for the present problem . it must be noted that if the terms involving the matrix @xmath111 are discarded in eq . ( [ eq : erpamtx ] ) , the components @xmath102 and @xmath103 decouple from @xmath104 and @xmath105 . in this case , ( [ eq:2pierpa ] ) reduce to the ph - drpa one ( [ eq : drpa ] ) and the matrix ( [ eq : erpamtx ] ) would take the form of the standard rpa matrix @xcite . the normalization condition is derived in the usual way , by extracting the contribution of order zero of the expansion around a given pole and by employing the conjugate equation . one eventually obtains @xcite @xmath112 where the inner product of the vectors @xmath113 and @xmath114 is implied . in the following we give the explicit expression for the matrix elements of eq . ( [ eq : erpamtx ] ) . the contributions originating from the standard ( d)rpa equation are ( here and below , summations over repeated greek indices are understood ) @xmath115 with the corresponding unperturbed ph energies @xmath116 the interaction between two - phonon intermediate states and the ph ones is given by @xmath117 \right . \nonumber \\ \nonumber \\ & & ~ ~ ~ ~ + ~ \left . \left [ n^\pi_a \leftrightarrow n^\pi_b \right ] \right\}\end{aligned}\ ] ] where @xmath118 and @xmath119 are the quantum numbers of the two phonons that form the intermediate state . the quantities @xmath120 are the forward - going amplitudes of the intermediate phonons ( [ eq : defxy12 ] ) or ( [ eq : z_vs_xy ] ) and are obtained from the previous solution of the polarization propagator . note that @xmath121 is symmetric under the exchange of the two indices @xmath118 and @xmath119 , as required by the boson character of the ph phonons . the factor @xmath122 assures that no double counting happens when only two free propagators @xmath85 are coupled . due to the time - inversion properties of npnh states , the corresponding contribution for backward - going propagation is simply related to above one : @xmath123 the analytical expression for the contribution of the time - inversion diagrams is a little more complicated . first , we introduce the following quantity that corresponds to the generation of a two - phonon state by a fluctuation of the nuclear mean field @xmath124 \label{erpa - bmtx1}\ ] ] this is also symmetric under the exchange of the indices @xmath118 and @xmath119 . then , the matrix elements containing the time - inversion diagrams , @xmath49 and @xmath73 , can be written as follows ( a summation over @xmath118 and @xmath119 is also implied ) @xmath125 \right . \nonumber \\ \nonumber \\ & & ~ \left . ~ ~ ~ ~ ~ + ~ \left [ - \left ( n_1 \leftrightarrow n_4 \right ) ~ - \left ( k_2 \leftrightarrow k_3 \right ) ~ + \left(n_1 \leftrightarrow n_4 , k_2 \leftrightarrow k_3 \right ) \right ] \right . \nonumber \\ \nonumber \\ & & ~ \left . + ~ \left [ { \cal x}^{n_1}_{\alpha } { \cal y}^{k_3}_{\gamma } ~ v_{\alpha \epsilon , \gamma \rho } ~ \left ( { \cal x}^{n_\rho}_{\rho } { \cal y}^{k_\epsilon}_{\epsilon } \right)^ * { x^{(1)}}^{~n^\pi_a}_{n_\rho k_\epsilon } ~ { x^{(1)}}^{~n^\pi_b}_{n_4 k_2 } ~ \right ] \right . \nonumber \\ \nonumber \\ & & ~ \left . ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ + ~ \left [ n_1 \leftrightarrow n_4 , k_2 \leftrightarrow k_3 \right ] \right . \nonumber \\ \nonumber \\ & & ~ \left . + ~ { \cal x}^{n_1}_{\alpha } { \cal x}^{n_4}_{\delta } ~ v_{\alpha \delta , \rho \lambda } ~ \left ( { \cal x}^{n_\rho}_{\rho } { \cal x}^{n_\lambda}_{\lambda } \right)^ * { x^{(1)}}^{~n^\pi_a}_{n_\rho k_2 } ~ { x^{(1)}}^{~n^\pi_b}_{n_\lambda k_3 } ~ \right . \nonumber \\ \nonumber \\ & & ~ \left . + ~ { \cal y}^{k_\rho}_{\rho } { \cal y}^{k_\lambda}_{\lambda } ~ v_{\rho \lambda , \beta \gamma } ~ \left ( { \cal y}^{k_2}_{\beta } { \cal y}^{k_3}_{\gamma } \right)^ * { x^{(1)}}^{~n^\pi_a}_{n_1 k_\rho } ~ { x^{(1)}}^{~n^\pi_b}_{n_4 k_\lambda } ~ \right\ } \left ( u_{n^\pi_a n^\pi_b } \right)^ * \ ; . \label{erpa - bmtx2}\end{aligned}\ ] ] g. f. bertsch and s. f. tsai , phys . rep . * 18 * , 125 ( 1975 ) ; r. i d betan , r. j. liotta , n. sandulescu , and t. verste , phys . rev . lett . * 89 * , 42501 ( 2002 ) ; n. michel , w. nazarewicz , m. poszajczak , and k. bennaceur , phys . * 89 * , 42502 ( 2002 ) c. barbieri and w. h. dickhoff , in proc . fifth workshop on _ electromagnetically induced two - hadron emission _ , p. grabmayr _ at al . lund , published on cd , isbn : 91 - 631 - 1612-x , 2001 ) , p. 108 . with an h.o . ipm input propagator , last column . the spectrum in the middle is obtained by solving the standard rpa problem and is employed , as is , to generate two - phonon contributions to the erpa equation . the excited states indicated by dashed lines are those for which the erpa equation predicts a total spectral strength @xmath60 lower than 10% . the first column reports the experimental results taken from ref . [ fig:2pierpa_ho ] , width=432 ] with a dressed input propagator from ref @xcite , middle and last column respectively . in solving the erpa equation , the lowest @xmath7 , @xmath9 and @xmath4 levels of the drpa propagator where shifted to their experimental energies . all other drpa solutions were left unchanged . the excited states indicated by dashed lines are those for which the ( e)rpa equation predicts a total spectral strength @xmath60 lower than 10% . the first column reports the experimental results @xcite . [ fig:2pierpa_3 g ] , width=432 ] ccccccccccc @xmath127 & & & & & + @xmath83 & & @xmath128 & @xmath60 & & @xmath128 & @xmath60 & & @xmath129(% ) & @xmath130(% ) & + @xmath61 & & 22.86 & 1.039 & & 23.13 & 0.823 & & 79 & 21 & + @xmath61 & & & & & 21.30 & 0.133 & & 13 & 87 & + @xmath61 & & & & & 19.11 & 0.118 & & 11 & 89 & + + @xmath4 & & & & & 14.28 & 0.010 & & 1 & 99 & + @xmath61 & & & & & 13.91 & 0.041 & & 4 & 96 & + @xmath62 & & & & & 13.87 & 0.028 & & 3 & 97 & + + @xmath9 & & 9.13 & 1.027 & & 9.20 & 1.075 & & 99.7 & 0.3 & + @xmath7 & & 7.14 & 1.258 & & 6.55 & 1.269 & & 97 & 3 & + @xmath4 & & 9.46 & 1.582 & & 6.52 & 1.820 & & 80 & 20 & + ccccccccccccccc @xmath127 & & & & & @xmath131 & @xmath132 & @xmath133 & @xmath134 & + @xmath83 & & @xmath135 & @xmath60 & & @xmath135 & @xmath60 & & @xmath129(% ) & @xmath130(% ) & ( % ) & ( % ) & ( % ) & ( % ) & + @xmath61 & & 23.77 & 0.468 & & 23.52 & 0.123 & & 26 & 74 & + @xmath61 & & & & & 22.96 & 0.341 & & 78 & 22 & + @xmath61 & & 20.59 & 0.269 & & 20.42 & 0.255 & & 98 & 2 & + + @xmath9 & & & & & 13.37 & 0.148 & & 21 & 79 & & & & 79 & + @xmath7 & & & & & 12.35 & 0.113 & & 16 & 84 & & & 84 + + @xmath4 & & & & & 12.15 & 0.001 & & 1 & 99 & 3 & 96 + @xmath62 & & & & & 12.14 & 0.007 & & 1 & 99 & & 99 + @xmath61 & & & & & 12.12 & 0.008 & & 1 & 99 & & 98 + + @xmath4 & & 16.62 & 0.717 & & 17.21 & 0.633 & & 88 & 12 & 10 & 0.5 + + @xmath4 & & & & & 11.28 & 0.092 & & 12 & 88 & 85 & 2 + + @xmath9 & & 11.19 & 0.720 & & 10.90 & 0.680 & & 94.1 & 5.9 & & & & 5.8 & + @xmath7 & & 9.50 & 0.762 & & 9.23 & 0.735 & & 95.9 & 4.1 & & & 4.0 + cccccccccccccccccccc @xmath127 & & & & & & & & & & & & & + & & & & & & & & & & & & & & & + @xmath83 & & @xmath128 & @xmath60 & & @xmath128 & @xmath60 & & @xmath128 & @xmath60 & & @xmath128 & @xmath60 & & @xmath128 & @xmath60 & & @xmath128 & @xmath60 & + @xmath61 & & 23.13 & 0.823 & & 23.09 & 0.824 & & & & & 22.96 & 0.341 & & 22.96 & 0.342 & & 22.39 & 0.210 & + @xmath61 & & & & & & & & 19.25 & 0.441 & & 21.44 & 0.163 & & 21.43 & 0.161 & & 21.35 & 0.122 & + @xmath61 & & & & & & & & 17.13 & 0.471 & & 20.42 & 0.255 & & 20.41 & 0.258 & & 20.30 & 0.281 & + + @xmath4 & & 14.28 & 0.010 & & 14.28 & 0.010 & & 12.85 & 0.232 & & 12.15 & 0.001 & & 12.15 & 0.001 & & 12.06 & 0.005 & + @xmath61 & & 13.91 & 0.041 & & 13.91 & 0.041 & & 13.88 & 0.101 & & 12.12 & 0.008 & & 12.12 & 0.008 & & 12.05 & 0.013 & + @xmath62 & & 13.87 & 0.028 & & 13.87 & 0.028 & & 11.81 & 0.036 & & 12.14 & 0.007 & & 12.14 & 0.007 & & 12.09 & 0.009 & + + @xmath4 & & & & & & & & & & & 11.28 & 0.092 & & 11.28 & 0.092 & & 10.97 & 0.184 & + + @xmath9 & & 9.20 & 1.075 & & 9.03 & 1.024 & & 9.21 & 1.260 & & 10.90 & 0.680 & & 10.89 & 0.680 & & 10.64 & 0.723 & + @xmath7 & & 6.55 & 1.269 & & 6.54 & 1.266 & & 4.68 & 1.581 & & 9.23 & 0.735 & & 9.22 & 0.735 & & 8.91 & 0.762 & + @xmath4 & & 6.52 & 1.820 & & 6.36 & 1.596 & & 6.74 & 3.055 & & & & & & & & & & +

we propose two novel experiments on the measurement of the casimir force acting between a gold coated sphere and semiconductor plates with markedly different charge carrier densities . in the first of these experiments a patterned si plate is used which consists of two sections of different dopant densities and oscillates in the horizontal direction below a sphere . the measurement scheme in this experiment is differential , i.e. , allows the direct high - precision measurement of the difference of the casimir forces between the sphere and sections of the patterned plate or the difference of the equivalent pressures between au and patterned parallel plates with static and dynamic techniques , respectively . the second experiment proposes to measure the casimir force between the same sphere and a vo film which undergoes the insulator - metal phase transition with the increase of temperature . we report the present status of the interferometer based variable temperature apparatus developed to perform both experiments and present the first results on the calibration and sensitivity . the magnitudes of the casimir forces and pressures in the experimental configurations are calculated using different theoretical approaches to the description of optical and conductivity properties of semiconductors at low frequencies proposed in the literature . it is shown that the suggested experiments will aid in the resolution of theoretical problems arising in the application of the lifshitz theory at nonzero temperature to real materials . they will also open new opportunities in nanotechnology .

0705.1358

the casimir effect @xcite implies that there is a force acting between closely spaced electrically neutral bodies following from the zero - point oscillations of the electromagnetic field . the casimir force can be viewed as an extension of the van der waals force to large separations where the retardation effects come into play . within a decade of casimir s work , lifshitz and collaborators @xcite introduced the role of optical properties of the material into the van der waals and casimir force . in the last few years , the advances following from both fundamental physics and nanotechnology have motivated careful experimental and theoretical investigations of the casimir effect . the first modern experiments were made with metal test bodies in a sphere - plate configuration , and their results are summarized in ref . @xcite . in subsequent experiments the lateral casimir force between corrugated surfaces @xcite and the pressure in the original casimir configuration @xcite have been demonstrated . later experiments @xcite have brought the most precise determination of the casimir pressure between two metal plates . the rapid theoretical progress has raised fundamental questions on our understanding of the casimir force between real metals at nonzero temperature . specifically , the role of conductivity processes and the related optical properties of metals at quasi - static frequencies has become the subject of discussions @xcite . one of the most important applications of the casimir effect is the design , fabrication and function of micro- and nanoelectromechanical systems such as micromirrors , microresonators , nanotweezers and nanoscale actuators @xcite . the separations between the adjacent surfaces in such devices are rapidly falling below a micrometer , i.e. , to a region where the casimir force becomes comparable with typical electrostatic forces . it is important that investigations of the casimir force be done in semiconductors as they are the material of choice for the fabrications of optomechanical , micro- and nanoelectromechanical systems . while the role of conductivity and optical properties of materials can be checked in metals , semiconductors offer better control of the related parameters ( charge carrier density , defect density , size etc . ) and will provide an exhaustive check of the various models . reference @xcite pioneered the measurement of the casimir force acting between a semiconductor surface , single crystal si wafer , and a gold coated sphere . the experimental data obtained for a wafer with the concentration of charge carriers @xmath1 were compared with the lifshitz theory at zero temperature and good agreement was found at a 95% confidence level . at the same time , the theory describing a `` dielectric '' si plate with a concentration @xmath2 was excluded by experiment at 70% confidence . this allows one to conclude that the casimir force is sensitive to the conductivity properties of semiconductors . this conclusion has found direct experimental confirmation in ref . @xcite where the casimir forces between a gold coated sphere and two different si wafers with the concentrations of charge carriers @xmath3 and @xmath4 have been measured . the difference of the mesured forces for the two conductivities was found to be in good agreement with the corresponding difference of the theoretical results computed at zero temperature ( note that the sensitivity of force mesurements in refs . @xcite was not sufficient to detect the thermal corrections predicted in refs . @xcite ) . in the most precise experiment on the casimir force between a metal and a semiconductor @xcite , the density of charge carriers in a si membrane of 4@xmath5 m thickness is changed from @xmath6 to @xmath7 through the absorption of photons from a laser pulse . this is a differential experiment where only the difference of the casimir forces in the presence and in the absence of a laser pulse was measured . this decreases the experimental error to a fraction of a 1pn and allows one to check the role of conductivity processes in semiconductors at the laboratory temperature @xmath8k . the experimental data for the difference casimir force as a function of separation were compared with the lifshitz theory and the outcome was somewhat puzzling . the data were found to be in excellent agreement with the theoretical difference force computed at @xmath8k under the assumption that in the absence of laser light si possesses a finite static dielectric permittivity . by contrast , if theory takes into account the dc conductivity of si in the absence of laser light , it is excluded by the data at a 95% confidence level . this is somewhat analogous to the above - mentioned problems for two real metals where the inclusion of the actual conductivity processes at low frequencies also leads to disagreement with experiment @xcite . the fundamental questions on the role of scattering processes and conductivity at low frequencies in the casimir force have to be clarified for further progress in the field and this calls for new precise experiments . in this paper we propose two experiments on the casimir force between a metal sphere and semiconductor plate which can shed light on the applicability of the lifshitz theory at nonzero temperature to real materials . in the first of these experiments , the patterned si plate with two sections of different dopant densities is oscillated in the horizontal direction below the au coated sphere . as a result , the sphere is subject to the difference casimir force which can be measured using the static and dynamic techniques . this experimental scheme promises a record sensitivity to force differences at the level of 1fn . in the second experiment , we propose to demonstrate the modulation of the casimir force by optically switching the insulator - metal transition in vo@xmath0 films @xcite . the phase transition between the insulator and metal leads to a change in the charge carrier density of order @xmath9 , which is sufficient to bring about a large change in the casimir force . for both experiments the related theory is elaborated and the magnitudes of casimir forces are computed with application to the experimental configurations . the effects from using different theoretical approaches to the description of conductivity processes are carefully analyzed and shown to be observable in the proposed experiments . the present status of developing the apparatus at uc riverside , its calibration and sensitivity is presented . the proposed experiments offer a precision test of the role of conductivity , optical properties and scattering in the lifshitz theory of the van der waals and casimir force at nonzero temperature . they also open up possibilities of radically new nanomechanical devices using the casimir force in imaging applications . the paper is organized as follows . in sec . ii , we describe the proposed experiment on the difference casimir force with the patterned semiconductor plate . the brief description of the experimental apparatus and preliminary results are also provided . section iii contains the calculation of the difference casimir force and equivalent pressure in the patterned geometry using the lifshitz theory at nonzero temperature and different models of conductivity processes at low frequencies . in sec . iv we propose the experiment on the modulation of the casimir force through a metal - insulator transition . the experimental scheme and some preliminary tests are discussed . section v presents theoretical computations of casimir forces in a insulator - metal transition on the basis of the lifshitz theory at nonzero temperature and using different models for the conductivity processes . section vi contains our conclusions and a discussion . the aim of this experiment is to gain a fundamental understanding of the role of carrier density in the casimir force using a nanofabricated patterned semiconductor plate . the proposed design for the experiment is shown schematically in fig . 1 . the gold coated polystyrene sphere of about 100@xmath5 m radius is attached to a cantilever of an atomic force microscope ( afm ) specially adapted for making sensitive force measurements . instead of the simple single crystal si substrate used in the previous experiments @xcite , here a patterned si plate is employed . this plate is composed of single crystal si specifically fabricated to have adjacent sections of two different charge carrier densities @xmath10 and @xmath11 . in this range of doping densities , the plasma frequency @xmath12 will change by a factor of 100 . additional changes in the @xmath12 can be brought about by using both @xmath13(b ) and @xmath14(p ) type dopants as electrons and holes differ in their effective mass by 30% . the preparation of the si sample with the two sections having different conductivities is shown schematically in fig . , one half of the bare si wafer of 0.3 to 0.5 mm thickness @xcite having the lower conductivity shown in ( a ) is masked with a photoresist as shown in ( b ) . next in ( c ) the exposed half of the si wafer is doped with p ions using ion implantation leading to a higher density of electrons in the exposed half in ( d ) . rapid thermal annealing and chemical mechanical polishing of the patterned si plate will be done as the last step . this is to ensure that there are no surface features resulting from the fabrication . similar nanofabrication procedures of semiconductors were used in our previous work @xcite . sharp transition boundaries between the two sections of si plate of width less than 200 nm are possible . the limitation comes from interdiffusion and the resolution of the ion implantation procedure to be used . it might be necessary to further limit interdiffusion by the creation of a narrow 100 nm barrier between the two doped regions . identically prepared but unpatterned samples will be used to measure the properties which are needed for theoretical computations . the carrier concentration will be measured using hall probes . this will yield an independent measurement of the plasma frequencies . a four probe technique will be used to measure the conductivity @xmath15 . from the conductivity and the charge carrier concentration , the scattering time @xmath16 will be found , where @xmath17 is the permittivity of free space . the first important improvement of this experiment , as compared with refs . @xcite , is the direct measurement of difference forces when the patterned plate is oscillated below the sphere . this measurement is performed as follows . the patterned si plate will be mounted on the piezo below a au coated sphere as is shown in fig . 1 . the si plate is positioned such that the boundary is below the vertical diameter of the sphere . the distance between the sphere and si plate @xmath18 will be kept fixed and the si plate will be oscillated in the horizontal direction using the piezo such that the sphere crosses the boundary in the perpendicular direction during each oscillation ( a similar approach was exploited in ref . @xcite for constraining new forces from the oscillations of the au coated sphere over two dissimilar metals , au and ge ) . the casimir force on the sphere changes as the sphere crosses the boundary . this change corresponds to the differential force @xmath19 equal to the difference of the casimir forces due to two different charge carriers densities @xmath20 and @xmath14 , respectively . this causes a difference in the deflection of the cantilever . in order to reduce the random noise by averaging , the periodic horizontal movement of the plate will be of an angular frequency @xmath21hz . the amplitude of the plate oscillations is limited by the piezo characteristics , but will be of order 100@xmath22 m , much larger than typical transition region of 200 nm . the experiment will be repeated for different sphere - plate separations in the region from 100 to 300 nm . the measurement of absolute separations will be performed by the application of voltages to the test bodies as described in ref . @xcite . the second major improvement in this experiment in comparison with all previous measurements of the casimir force is the increased sensitivity . this will be achieved through the use of the interferometer based low temperature afm capable of operating over wide temperature range spanning from 360 to 4k , and the use of two measurement techniques , a static one and a dynamic one . a picture of the newly constructed experimental apparatus of a low - temperature afm is shown in fig . 3 . here the cantilever deflection is measured interferometrically and therefore has much higher sensitivity than photodiodes used in the previous work @xcite . the detection of a difference force @xmath23 will be done by two alternative techniques . the first technique , a static one , reduces to the direct measurement of @xmath23 as described above . we are presently performing the initial tests and calibration trials at 77k . an oil free vacuum with a pressure of around @xmath24torr is used . the instrument is magnetically damped to yield low mechanical coupling to the environment . the temperature can be varied with a precision of 0.2k . we have fabricated special conductive cantilevers with a spring constant @xmath25n / m . the magnitude of @xmath26 is found by applying electrostatic voltages to the plate as discussed in ref . @xcite . to accomplish this , si cantilevers were thermal diffusion doped to achieve the necessary conductivity . note that conductive cantilevers are necessary to reduce electrostatic effects . the cantilever - sphere arrangement has been checked to be stable at 77k . the experimental setup using the static measurement technique allows not only a demonstration but a detailed investigation of the influence of carrier density , conductivity and scattering in semiconductors on the casimir force . according to the results of refs . @xcite and calculations below in sec . iii , the magnitude of the difference force to be measured is about several pn . different theoretical models of conductivity processes at low frequencies lead to predictions differing by approximately 1pn within a wide separation region ( see fig . 5 in sec . the described setup provides excellent opportunity for precise measurement of forces of order and below 1pn . we have measured a resonance frequency of the cantilever of @xmath27hz , a quality factor @xmath28 , and an equivalent noise bandwidth @xmath29hz . the resultant force sensitivity of our cantilever at @xmath30k with the gold coated sphere attached was determined following ref . @xcite to be @xmath31 where @xmath32 is the boltzmann constant . even bearing in mind that the systematic error may be up to an order of magnitude larger , the sensitivity ( [ eq2 ] ) presents considerable possibilities for the precise investigation of the difference casimir force @xmath33 . the second technique for the detection of the difference casimir force is a dynamic one @xcite . here this technique is applied not for a direct measurement of @xmath33 but rather for the experimental determination of the equivalent difference casimir pressure between the two parallel plates ( one made of au and the other one , a patterned si plate ) . the cantilever - sphere system oscillates in the verical direction due its thermal noise with a resonant frequency @xmath34 in the absence of the casimir force , where @xmath35 is the mass of the system . the thermal noise spectrum of the sphere - cantilever system is measured and fit to a lorentzian to identify the peak resonant frequency , @xmath36 . the shift of @xmath36 in the presence of the casimir force when for example the sphere is positioned above a section of the patterned si plate with the density of charge carriers @xmath20 is equal to @xcite @xmath37 next the plate is oscillated in the horizontal direction with a frequency @xmath38 . as a result , the frequency shift @xmath39 between the resonant frequencies above the two different sections of the patterned si plate is measured . using the proximity force approximation @xcite , we determine the difference casimir pressure @xmath40 between the two parallel plates ( the au one and the patterned si ) . note that the systematic error from the use of the proximity force approximation was recently confirmed to be less than @xmath41 @xcite . equations ( [ eq4 ] ) and ( [ eq5 ] ) express the difference casimir pressure through the measured shift of the resonance frequency above the two halfs of the patterned plate . as is shown in the next section , the measurements of the difference casimir pressure using the dynamic technique provides us with one more test of the predictions of the different models of conductivity processes at low frequencies . the experiment on the difference casimir force from a patterned si plate , as described in this section , allows variation of charge carrier density by the preparation of different semiconductor samples . thus , the proposed measurements should provide a comprehensive understanding on the role of conductivity and optical processes in the casimir force for nonmetallic materials and discriminate between competing theoretical approaches the difference casimir force and the equivalent casimir pressure from the oscillation of the patterned si plate below an au coated sphere at @xmath8k in thermal equilibrium are given by the lifshitz theory . in the static technique the data to be compared with theory is the difference of casimir forces acting between the sphere and two sections of the patterned plate . this difference is obtained from the casimir energy between two parallel plates , as given by the lifshitz theory , using the proximity force approximation @xcite @xmath42\,\left[1- r_{te;\tilde{n}}(\xi_l , k_{\bot } ) r_{te}(\xi_l , k_{\bot})e^{-2q_lz}\right]}{\left[1- r_{tm;{n}}(\xi_l , k_{\bot } ) r_{tm}(\xi_l , k_{\bot})e^{-2q_lz}\right]\,\left[1- r_{te;{n}}(\xi_l , k_{\bot } ) r_{te}(\xi_l , k_{\bot})e^{-2q_lz}\right]}\ , . \nonumber\end{aligned}\ ] ] here @xmath43 with @xmath44 are the matsubara frequencies , @xmath45 , and @xmath46 is the projection of the wave vector on the boundary planes . the reflection coefficients on the au plane for the two independent polarizations of the electromagnetic field ( transverse magnetic and transverse electric modes ) are @xmath47 where @xmath48 and @xmath49 is the dielectric permittivity of au along the imaginary frequency axis . in a similar way , the reflection coefficients on the two sections of a patterned si plate with charge carrier densities @xmath20 and @xmath14 are given , respectively , by @xmath50 where @xmath51 and @xmath52 are the dielectric permittivities of si with charge carrier densities @xmath20 and @xmath14 along the imaginary frequency axis . in the dynamic technique the data to be compared with theory is the equivalent difference casimir pressure between two parallel plates , one made of au and the other one a patterned si plate . using the same notations as above , the difference casimir pressure is given by @xmath53^{-1}+\left [ r_{te;\tilde{n}}^{-1}(\xi_l , k_{\bot } ) r_{te}^{-1}(\xi_l , k_{\bot})e^{2q_lz}-1\right]^{-1}\right . \nonumber \\ & & \phantom{aaaa}\left . -\left[r_{tm;{n}}^{-1}(\xi_l , k_{\bot } ) r_{tm}^{-1}(\xi_l , k_{\bot})e^{2q_lz}-1\right]^{-1}- \left[r_{te;{n}}^{-1}(\xi_l , k_{\bot } ) r_{te}^{-1}(\xi_l , k_{\bot})e^{2q_lz}-1\right]^{-1}\right\}\ , . \nonumber\end{aligned}\ ] ] note that in eqs . ( [ eq6 ] ) and ( [ eq9 ] ) we have replaced the 100 nm au coating and the 0.30.5 mm si plate for an au and si semispaces , respectively . using the lifshitz formula for layered structures @xcite it is easy to calculate the force and pressure errors due to this replacement . for example , for an au layer at a typical separation of 100 nm this error is about 0.01% . for si a finite thickness of the plate @xmath54 markedly affects the casimir force when the separation distance @xmath18 exceeds the thickness , i.e. , @xmath55 @xcite . in our case , however , even at the largest separation considered ( @xmath56 nm ) the ratio of the separtion to the plate thickness @xmath57 . this is similar to the case of the experiment @xcite where the finite thickness of si membrane also does not influence the magnitude of the casimir force because at separations @xmath58 nm where statistically meaningful results were obtained @xmath59 . we have performed computations of the difference casimir force ( [ eq6 ] ) and difference casimir pressure ( [ eq9 ] ) for samples with typical values of charge carrier concentrations @xmath20 and @xmath14 as used in experiments @xcite . both sections of the si plate were chosen to have electron conductivity and doped with p. for the section of the plate with higher concentration of charge carriers the values @xmath60 ( such a sample was fabricated in ref . @xcite ) and @xmath61 were used in the computations . the respective dielectric permittivity along the imaginary frequency axis can be represented in the form @xcite @xmath62 here @xmath63 is the permittivity of high - resistivity ( dielectric ) si along the imaginary frequency axis computed in ref . @xcite by means of the dispersion relation using the tabulated optical data for the complex index of refraction @xcite . the values of the plasma frequencies and relaxation parameters are the following @xcite : @xmath64rad / s , @xmath65rad / s , @xmath66rad / s , @xmath67rad / s . in fig . 4 the dielectric permittivities of the samples with high concentrations of charge carriers @xmath68 and @xmath69 are shown as solid lines 1 and 2 , respectively . in the same figure , the dashed line @xmath70 shows the permittivity of high - resistivity , dielectric , si @xcite and the dotted line the permittivity of au computed in ref . @xcite using the tabulated optical data of ref . @xcite . below we will use two models for the permittivity of the section of the si plate with lower concentration of charge carriers @xmath14 . calculations show that for any @xmath71 ( this interval includes the experimental value of @xmath72 in ref . @xcite ) , the obtained values of @xmath73 and , thus , of @xmath23 do not depend on @xmath14 . because of this we use in the computations @xmath74 , the plasma frequency @xmath75rad / s and the relaxation parameter @xmath76rad / s @xcite ( note that for @xmath77 the value of the relaxation parameter does not effect the magnitude of the casimir force ) . then the dielectric permittivity of this section of the si plate along the imaginary frequency axis is given by @xmath78 and is shown as the dashed line @xmath79 in fig . this is one model of si with a lower concentration of charge carriers referred to below as model ( @xmath79 ) . as is seen in fig . 4 , the dashed line @xmath79 , and thereby all respective lines for samples with the concentration of charge carriers smaller than @xmath80 , deviate from the permittivity of dielectric si ( line @xmath70 ) only at frequencies below the first matsubara frequency @xmath81 . because of this , it is common ( see , e.g. , @xcite ) to neglect the small conductivity of high - resistivity materials at low frequencies and describe them in the frequency region below the first matsubara frequency by the static dielectric permittivity . in our case this leads to @xmath82 which is the other model for si with a lower concentration of charge carriers referred to below as model ( @xmath70 ) . from eq . ( [ eq12 ] ) and fig . 4 at all frequencies @xmath83 it follows : @xmath84 . to be exact , at any @xmath85 the density of free charge carriers @xmath14 in semiconductors ( and even in dielectrics ) and thus the conductivity are nonzero ( @xmath86 ) . thus , the model ( [ eq11 ] ) should be considered as more exact than the model ( [ eq12 ] ) . at the same time , if we note that for @xmath77 the conductivity is small , it should be expected that both models should lead to practically identical results . this is , however , not so . in fig . 5 we present the computational results for the difference casimir force using eq . ( [ eq6 ] ) . the solid line @xmath87 demonstrates the values of the difference casimir force versus separation for the patterned si plate with a higher concentration of charge carriers @xmath68 computed under the assumption that the lower concentration section of the plate is described by eq . ( [ eq12 ] ) , i.e. , the conductivity processes at low frequencies are neglected . the dashed line @xmath88 shows the difference casimir force as a function of separation computed with the same @xmath68 but taking into account the conductivity processes at low frequencies in accordance with eq . ( [ eq11 ] ) . as is seen from the comparison of lines @xmath87 and @xmath88 , the difference casimir forces computed using eqs . ( [ eq12 ] ) and ( [ eq11 ] ) differ by 1.2pn at a separation @xmath89 nm and this difference slowly decreases to approximately 0.14pn at a separation @xmath56 nm . the lines @xmath90 and @xmath91 present similar results for the case when the higher charge carrier density is equal to @xmath69 . as is seen from fig . 5 , decreasing the higher concentration by an order of magnitude decreases the predicted magnitude of the difference casimir force by more than two times , but leaves the same gap between the predictions of two different models of the permittivity at low frequencies . importantly , our predictions do not depend on the discussions mentioned in the introduction on the optical properties of metals at quasi - static frequencies @xcite . the resolution of this controversy affects only the value of the au reflection coefficient @xmath92 at zero frequency . the latter , however , does not contribute to the difference casimir force ( [ eq6 ] ) and pressure ( [ eq9 ] ) because for dielectrics and semiconductors @xmath93 regardless of what model ( [ eq11 ] ) or ( [ eq12 ] ) is used for the description of the dielectric permittivity at low frequencies . the obtained difference between the lines @xmath94 and @xmath95 in fig . 5 is completely explained by the different contributions of semiconductor reflection coefficient @xmath96 when one uses eq . ( [ eq11 ] ) or eq . ( [ eq12 ] ) to describe the dielectric permittivity at low frequencies . regarding the semiconductor section with a higher charge carrier density @xmath20 , from eqs . ( [ eq8 ] ) , ( [ eq10 ] ) it is always valid that @xmath97 however , for the section of the plate with a lower charge carrier density @xmath14 it follows from eq . ( [ eq8 ] ) that @xmath98 when eq . ( [ eq11 ] ) or eq . ( [ eq12 ] ) are used , respectively . thus , the difference between the lines @xmath87 and @xmath88 ( and the same difference between the lines @xmath90 and @xmath91 ) can be found analytically . taking only the zero - frequency contribution in eq . ( [ eq6 ] ) and subtracting the difference casimir force calculated using eq . ( [ eq11 ] ) [ model ( @xmath79 ) ] from the difference casimir force calculated using eq . ( [ eq12 ] ) [ model ( @xmath70 ) ] one obtains @xmath99\right\}. \label{eq15}\ ] ] here @xmath100 is the riemann zeta function , and li@xmath101 is the polylogarithm function . the results using the analytic eq . ( [ eq15 ] ) coincide with the differences between the lines @xmath94 and @xmath95 in fig . 5 computed numerically . in the experiment @xcite the difference casimir force between au coated sphere and si plate illuminated with laser pulses was first measured . in the presence of light the charge carrier density was about @xmath7 and in the absence of light of about @xmath6 . the experimental data were shown to be in agreement with model ( @xmath70 ) which uses the finite static dielectric permittivity of si . the model ( @xmath79 ) which includes the dc conductivity of si was excluded at 95% confidence within the separation region from 100 to 200 nm . as was discussed above , in the framework of the lifshitz theory this result is rather unexpected . bearing in mind that illumination with laser pulses leads to several additional sources of errors discussed in ref . @xcite , it is of vital interest to verify the obtained conclusions in a more precise experiment with patterned si plates . the comparison of the experimental sensitivities presented in sec . ii with the magnitudes of the difference casimir forces computed here using different theoretical models demonstrate that the proposed experiment with patterned semiconductor plate will bring decisive results on the discussed problems in the lifshitz theory at nonzero temperature . the calculations of the difference casimir pressure determined in the dynamic mode of the proposed experiment leads to results analogous to those for the difference force . the calculation results using eq . ( [ eq9 ] ) with the same values of parameters as above and two models of lower conductivity si are presented in fig . the difference casimir pressures between an au plate and a patterned si plate with the higher densities of charge carriers @xmath102 ( one section of the plate ) and lower @xmath14 ( another section of the plate ) are shown with solid lines @xmath87 and @xmath90 , respectively , computed under the assumption that si with the lower @xmath14 possesses a finite permittivity ( [ eq12 ] ) at zero frequency . the dashed lines @xmath88 and @xmath91 are obtained under the assumption that si with the lower @xmath14 is described by the permittivity ( [ eq11 ] ) which goes to infinity when the frequency goes to zero . as is seen in fig . 6 , the difference casimir pressure with a patterned plate with charge carrier densities @xmath68 and @xmath14 equals 250mpa at a separation @xmath89 nm [ model ( @xmath70 ) of low conductivity section of the plate ] and the difference in predictions for the two models equals 38.6mpa . the proposed experiment of the difference casimir pressure can reliably discriminate between the solid and dashed lines in fig . 6 thus providing one more test for the lifshitz theory at nonzero temperature . notice that in a similar way to the force , the differences between the lines @xmath94 and @xmath95 in fig . 6 are expressed analytically by taking the zero - frequency contributions in eq . ( [ eq9 ] ) : @xmath103\right\}. \label{eq16}\ ] ] calculations using eq . ( [ eq16 ] ) lead to the same differences between the lines @xmath94 and @xmath95 as were computed numerically in fig . 6 . the exciting possibility for the modulation of the casimir force due to a change of charge carrier density is offered by semiconductor materials that undergo the insulator - metal transition with the increase of temperature . such a transition leads to a change of the carrier density of order @xmath9 . although in literature it is common to speak about insulator - metal transition , this can be considered as a transformation between two semiconductor phases with lower and higher charge carrier densities @xmath14 and @xmath20 , respectively . as was shown above , this is sufficient to bring about a large change in the casimir force . from a fundamental point of view , the modulation of the casimir force due to the phase transition will offer one more precision test of the role of conductivity and optical properties in the lifshitz theory of the casimir force . this experiment suggests some advantages as compared to the difference force measurement with a patterned plate considered in secs . ii and iii . first , because of the large change in the magnitude and bandwidth of the optical properties in a phase transition , the modulation of the casimir force will be larger . second , an insulator - metal transition does not require the special fabrication of patterned plates with one section having a high carrier density , which might not be compatible with robust device design . keeping in mind that the increase of temperature necessary for the phase transition can be induced by laser light , this opens up the possibility of radically new nanomechanical devices using the casimir force in image detection . the phase transition can be also brought about through electrical heating of the material . in this experiment we propose to measure the change of the casimir force acting between an au coated sphere and a vanadium dioxide ( vo@xmath0 ) film deposited on sapphire substrate which undergoes the insulator - metal transition with the increase of temperature . it has been known that vo@xmath0 crystals and thin films undergo an abrupt transition from semiconducting monoclinic phase at room temperature to a metallic tetragonal phase at 68@xmath104c @xcite . the phase transition causes the resistivity of the sample to decrease by a factor of @xmath9 from 10@xmath105 cm to @xmath106 cm ( i.e. , the same change as for two semiconductor half plates in sec . ii with lower and higher charge carrier densities ) . in addition , the optical transmission for a wide region of wavelengths extending from @xmath107 m to greater than @xmath108 m , decreases by more than a factor of 10100 . the schematic of the experimental setup is shown in fig . 7 . in this figure , light from a chopped 980 nm laser will be used to heat the vo@xmath0 film @xcite . about 10100mw power of the 980 nm laser is required to bring about all optical switching of vo@xmath0 films . the same procedure as outlined in sec . ii ( the static technique ) will be used in the measurement of the modulation of the casimir force including the interferometric detection of cantilever deflection . the schematic of the setup is similar to the one used in ref . @xcite in the demonstration of optically modulated dispersion forces . an important point is that in ref . @xcite the absorption of light from a 514 nm ar laser led to an increase of charge carrier density . by contrast , here the wavelength of a laser is selected in such a way that light only leads to heating of a vo@xmath0 film but does not change the number of free charge carriers @xcite . as a first step towards studying the role of the insulator - metal transition in the casimir force , we have recently fabricated thin films of vo@xmath0 on sapphire plates . the preliminary results are shown in fig . it is observed that we have obtained more than a factor of 10 change in the resistivity of the film . these films were prepared by thermal evaporation of vo@xmath0 powder . while films of appropriate thickness approaching 100 nm and roughness of about 2 nm ( shown in fig . 9 ) can be obtained by this procedure , it is not optimal as it leads to the non - stoichimetric formation of the mixed valence states of the vanadium oxide ( vo@xmath109 ) . in the future rf magnetron sputtering will be used to make the films @xcite . vo@xmath0 films using this technique have been shown to have the @xmath9 change in resistivity and a corresponding large change in optical reflectivity and spectrum . the aim of the proposed experiment on the influence of insulator - metal transition on the casimir force is two fold : applications for actuation of nanodevices through a modulation of the casimir force , and to perform fundamental tests on the theory of dispersion forces . to accomplish this , two types of measurements are planned . in the first we plan to demonstrate the modulation of the casimir force through an optical switching of the insulator - metal transition . this modulation will lead to novel microdevices as optical and electrical switches , optical modulators , optical filters and ir detectors that can be actuated optically through the absorption of ir radiation . importantly , such devices can be integrated with si technology which is used in the fabriaction of microelectromechanical systems @xcite . in the second type of measurements , the variable temperature atomic force microscope described in sec . ii will be used to perform precision measurements of the casimir force between a gold coated sphere and vo@xmath0 film . here , the casimir force will be measured at different temperatures from room temperature through 80@xmath104c . this temperature range spans the dielectric ( semiconducting ) and metal regions of vo@xmath0 . careful comparison of the experimental data and the theory ( see the next section ) will be done to understand the role of conductivity and losses in both phases of vo@xmath0 . the casimir force acting between a metal coated sphere and the vo@xmath0 film on a sapphire plate both before and after the phase transition ( i.e. , in the insulating and metal phases or , more exactly , in the semiconductor phases with lower and higher charge carrier densities ) are expressed by the lifshitz formulas in accordance with eqs . ( [ eq1 ] ) and ( [ eq6 ] ) . as above , we label the higher concentration of charge carriers @xmath20 and the lower concentration @xmath14 . to compute the casimir force before and after the phase transition one needs the optical properties of vo@xmath0 on a sapphire plate in a wide frequency region . in ref . @xcite the dielectric permittivity of vo@xmath0 is measured and fitted to the oscillator model for both bulk vo@xmath0 and for a system of 100 nm thick vo@xmath0 film deposited on bulk sapphire plate within the frequency region from 0.25ev to 5ev . this modelling was performed both before and after the phase transition . typical thickness of sapphire substrate is of about 0.3 mm , i.e. , the same as the thickness of patterned si plate in secs . ii and iii . because of this , when calculating the casimir force between gold coated sphere and vo@xmath0 film on sapphire substrate , we can use the lifshitz formala for bulk test bodies ( see sec . iii for details ) . the application region of the models presented in ref . @xcite should be extended in order to perform computations of the casimir force within the separation region from 100 to 300 nm where contribution from optical data up to about 10ev have to be taken into account . for this purpose , we have supplemented equations of ref . @xcite with additional terms taking into account the frequency - dependent electronic transitions at high frequencies @xcite . as a result , the effective dielectric permittivity of the vo@xmath0 film on a sapphire substrate before the phase transition ( at @xmath8k ) is given by @xmath110 here the values of the oscillator frequencies @xmath111 , dimensionless relaxation parameters @xmath112 and of the oscillator strengths @xmath113 taken from fig . 5 in ref . @xcite are presented in table i. the constants related to the contribution of high - frequency electronic transitions [ the last term on the right - hand side of eq . ( [ eq17 ] ) ] are @xmath114 @xcite and @xmath115ev . if we put @xmath116 in the last term on the right - hand side of eq . ( [ eq17 ] ) , this equation is the same as the result in ref . @xcite . after the phase transition we have a phase with increased charge carrier density @xmath20 . similar to eq . ( [ eq10 ] ) the effective dielectric permittivity of the vo@xmath0 film on a sapphire substrate can be described by the dielectric permittivity @xmath117 parameters @xmath118 , @xmath119 and @xmath120 can be found in fig . 6 of ref . @xcite and are listed in table ii . the other parameters are @xmath121 @xmath122ev , @xmath123ev @xcite . setting @xmath116 in the last term on the right - hand side of eq . ( [ eq18 ] ) ( this term takes high - frequency electronic transitions into account ) , returns ( [ eq18 ] ) to the original form suggested in ref . note that the recently suggested model for the dielectric permittivity of vo@xmath0 films @xcite is applicable not only before and after a phase transition but also at intermediate temperatures . this model is , however , restricted to a more narrow frequency region from 0.73 to 3.1ev and uses the simplified description of two oscillators before the phase transition and only one oscillator with nonzero frequency after it . in fig . 10 , the effective dielectric permittivity of vo@xmath0 film of 100 nm thickness on sapphire substrate before and after the phase transition , as given in eqs . ( [ eq17 ] ) and ( [ eq18 ] ) is shown by the solid lines 1 and 2 , respectively . in the same figure , the dielectric permittivity of au versus frequency is shown as dots . the vertical line indicates the position of the first matsubara frequency at @xmath124k ( i.e. , in the region of the phase transition ) . in fig . 11 we present the computational results for the casimir force between the au coated sphere and vo@xmath0 film on sapphire substrate versus separation obtained by the substitution of the dielectric permittivity ( [ eq17 ] ) ( vo@xmath0 before the phase transition in solid line 1 ) and ( [ eq18 ] ) ( vo@xmath0 after the phase transition in solid line 2 ) into the lifshitz formula . as is seen in fig . 11 , after the phase transition the magnitudes of the casimir force increase due to an increase in the charge carrier density . for a comparison with the proposed experiment on the difference casimir force from a patterned si plate , in fig . 12 ( solid line ) we plot the difference of the casimir forces after and before the phase transition , i.e. , the difference of lines 2 and 1 in fig . 11 . it is seen that the difference casimir force from a phase transition changes from 13pn at @xmath89 nm to 1.2pn at @xmath56 nm , i.e. , the magnitudes of the difference from the phase transition are greater than that from the patterned si plate . the difference casimir force in the insulator - metal phase transition provides us with one more test on the proper modelling of the dielectric permittivity in the lifshitz theory of dispersion forces . similar to sec . iii , we arrive at different results for the difference casimir force after and before the phase transition if the conductivity of a dielectric vo@xmath0 at zero frequency is taken into account in our computations . the shift in the values of the difference casimir force is completely determined by the change of the zero - frequency term in the lifshitz formula . by analogy with eq . ( [ eq15 ] ) it follows : @xmath125\right\ } , \label{eq19}\ ] ] where @xmath79 represents the case when the dc conductivity of an insulating vo@xmath0 is taken into account , and @xmath70 represents the case when insulating vo@xmath0 is described by the permittivity ( [ eq17 ] ) . from eq . ( [ eq17 ] ) and table i one obtains @xmath126 in fig . 12 the difference casimir force between an au coated sphere and vo@xmath0 film on sapphire substrate after and before the phase transition computed including the dc conductivity of insulating vo@xmath0 versus separation is plotted with the dashed line . the difference between the solid and dashed lines is determined by eq . ( [ eq19 ] ) . this difference changes from 1.6pn at @xmath89 nm to 0.2pn at @xmath56 nm . thus , in the phase transition experiment the predicted discrepances between the two theoretical approaches to the description of conductivity properties at low frequencies are larger than in the experiment with the patterned semiconductor plate . this will help to experimentally discriminate between the two approaches and deeply probe the role of the material properties in the lifshitz theory at nonzero temperature . in the above we have proposed two experiments on the measurement of the difference casimir force acting between a metal coated sphere and a semiconductor with different charge carrier densities . one of these experiments is based on the formation of a special patterned si plate , two sections of which have charge carrier densities differing by several orders of magnitude . the measurement scheme in this experiment is differential , i.e. , adapted for the direct measurement of the difference in the casimir forces between the sphere and each section of the patterned plate . this allows one to obtain high precision within a wide measurement range . using the dynamic measurement technique , this experiment also permits the measurement of the difference casimir pressure between two parallel plates one of which is coated with gold and the other is patterned and consists of two sections with different charge carrier densities . another proposed experiment directed to the same objective is novel and uses the insulator - metal phase transition brought about by an increase of temperature in the measurements of the casimir force . this transition also leads to the change of charge carrier density by several orders of magnitude while not requiring the formation of special patterned samples . the expected difference in the casimir forces after and before the phase transition is even larger than in the experiment with the patterned si plate . both proposed experiments are motivated by the uncertainties in the application of the theory of dispersion forces at nonzero temperature . as was shown above , different models of the conductivity of semiconductors at low frequencies used in the literature predict variations of the difference casimir force at the level of 1pn . an even greater concern is that the model taking into account the dc conductivity of dielectrics violates the nernst heat theorem @xcite . we have reported an apparatus developed at uc riverside that has the sensitivity of force measurements on the level of 1fn and is well adapted for the systematic investigation of the proposed effects in a wide range of separations . this apparatus includes an interferometer based atomic force microscope operated in high vacuum over a temperature range from 360k to 4k . the proposed experiments are feasible using the developed techniques and will aid in the resolution of theoretical problems on the application of the lifshitz theory at nonzero temperature to real materials . another motivation of the proposed experiments is in the application to nanotechnology . the separations between the adjacent surfaces in micro- and nanoelectromechanical devices are rapidly falling to a region below a micrometer where the casimir force becomes dominant . keeping in mind that semiconducting materials are used for micromachines , the detailed investigation of the dependence of the casimir force on the properties of semiconductors is important . the proposed experiments and related theory clearly demonstrate that it is possible to control the casimir force with semiconductor surfaces by changing the charge carrier density with doping or excitation . this opens new opportunities discussed above for using the casimir force in both the operation and function of novel nanomechanical devices . in addition to the previously performed experiments on the casimir force ( see review @xcite and refs . @xcite ) currently a number of new experiments have been proposed in the literature . thus , ref . @xcite proposes to measure the casimir torque between two parallel birefringent plates with in - plane optical anisotropy separated by either a vacuum or ethanol . in ref . @xcite it is suggested to measure the vacuum torque between corrugated mirrors . references @xcite propose the measurements of the casimir force between metallic surfaces at large separations of a few micrometers as a spherical lens and a plate , a cylinder and a plate or two parallel plates . these experiments are aimed at resolving the theoretical problems arising in the lifshitz theory when it is applied to real metals . in refs . @xcite a proposal to measure the influence of the casimir energy on the value of the critical magnetic field in the transition from a superconductor to a normal state has been made . references @xcite proposed the measurement of the dynamic casimir effect resulting in the creation of photons . the experiments proposed here on the difference casimir force through the use of patterned semiconductor samples or using the insulator - metal phase transition indicate important new promising directions for future investigations in the casimir effect . g.l.k.and v.m.m . are grateful to the department of physics and astronomy of the university of california ( riverside ) for its kind hospitality . the work on the difference casimir force with patterned semiconductor samples and insulator - metal phase transition was supported by the doe grant no . de - fg02 - 04er46131 . the development of the interferometer based low temperature afm was supported by the nsf grant no . g . is grateful for the financial support of ucmexus and conacyt . 99 h. b. g. casimir , proc . akad . wet . * 51 * , 793 ( 1948 ) . e. m. lifshitz , zh . teor . fiz . * 29 * , 94 ( 1956 ) [ sov . jetp * 2 * , 73 ( 1956 ) ] . i. e. dzyaloshinskii , e. m. lifshitz , and l. p. pitaevskii , usp . nauk * 73 * , 381 ( 1961 ) [ sov . ( usa ) * 4 * , 153 ( 1961 ) ] . m. bordag , u. mohideen , and v. m. mostepanenko , phys . rep . * 353 * , 1 ( 2001 ) . f. chen , u. mohideen , g. l. klimchitskaya , and v. m. mostepanenko , phys . rev . lett . * 88 * , 101801 ( 2002 ) ; phys . a * 66 * , 032113 ( 2002 ) . g. bressi , g. carugno , r. onofrio , and g. ruoso , phys . rev . lett . * 88 * , 041804 ( 2002 ) . r. s. decca , e. fischbach , g. l. klimchitskaya , d. e. krause , d. lpez , and v. m. mostepanenko , phys . d * 68 * , 116003 ( 2003 ) . r. s. decca , d. lpez , e. fischbach , g. l. klimchitskaya , d. e. krause , and v. m. mostepanenko , ann . ( n.y . ) * 318 * , 37 ( 2005 ) . r. s. decca , d. lpez , e. fischbach , g. l. klimchitskaya , d. e. krause , and v. m. mostepanenko , phys . d * 75 * , 077101 ( 2007 ) . m. bostrm and b. e. sernelius , phys . * 84 * , 4757 ( 2000 ) . c. genet , a. lambrecht , and s. reynaud , phys . a * 62 * , 012110 ( 2000 ) . m. bordag , b. geyer , g. l. klimchitskaya , and v. m. mostepanenko , phys . lett . * 85 * , 503 ( 2000 ) . b. geyer , g. l. klimchitskaya , and v. m. mostepanenko , phys . a * 67 * , 062102 ( 2003 ) . f. chen , g. l. klimchitskaya , u. mohideen , and v. m. mostepanenko , phys . * 90 * , 160404 ( 2003 ) . i. brevik , j. b. aarseth , j. s. hye , and k. a. milton , phys . e * 71 * , 056101 ( 2005 ) . v. b. bezerra , r. s. decca , e. fischbach , b. geyer , g. l. klimchitskaya , d. e. krause , d. lpez , v. m. mostepanenko , and c. romero , phys . e * 73 * , 028101 ( 2006 ) . j. s. hye , i. brevik , j. b. aarseth , and k. a. milton , j. phys . a : mat . * 39 * , 6031 ( 2006 ) . y. srivastava , a. widom , and m. h. friedman , phys . * 55 * , 2246 ( 1985 ) . e. buks and m. l. roukes , phys . b * 63 * , 033402 ( 2001 ) . g. palasantzas and j. th . m. dehosson , phys . b * 72 * , 1213409 ( 2005 ) . h. b. chan , v. a. aksyuk , r. n. kleiman , d. j. bishop , and f. capasso , science , * 291 * , 1941 ( 2001 ) ; phys . 87 * , 211801 ( 2001 ) . r. e. slusher and c. weisbuch , solid state comm . * 92 * , 149 ( 1994 ) . f. chen , u. mohideen , g. l. klimchitskaya , and v. m. mostepanenko , phys . a * 72 * , 020101(r ) ( 2005 ) ; * 74 * , 022103 ( 2006 ) . f. chen , g. l. klimchitskaya , v. m. mostepanenko , and u. mohideen , phys . lett . * 97 * , 170402 ( 2006 ) . f. chen , g. l. klimchitskaya , v. m. mostepanenko , and u. mohideen , optics express * 15 * , 4823 ( 2007 ) . a. zylbersztejn and n. f. mott , phys . b * 11 * , 4383 ( 1975 ) . u. mohideen , w. s. hobson , s. j. pearton , r. e. slusher , and f. ren , appl . . lett . * 64 * , 1911 ( 1994 ) . u. mohideen , r. e. slusher , f. jahnke and s. koch , phys . * 73 * , 1785 ( 1994 ) . r. s. decca , d. lpez , h. b. chan , e. fischbach , d. e. krause , and c. r. jamell , phys . * 94 * , 240401 ( 2005 ) . h. j. mamin and d. rugar , appl . * 79 * , 3358 ( 2001 ) . j. blocki , j. randrup , w. j. swiatecki , and c. f. tsang , ann . ( n.y . ) * 105 * , 427 ( 1977 ) . t. emig , r. l. jaffe , m. kardar , and a. scardicchio , phys . rev . lett . * 96 * , 080403 ( 2006 ) . a. bulgac , p. magierski , and a. wirzba , phys . d * 73 * , 025007 ( 2006 ) . m. bordag , phys . d * 73 * , 125018 ( 2006 ) . h. gies and k. klingmller , phys . * 96 * , 220401 ( 2006 ) ; phys . d * 74 * , 045002 ( 2006 ) . d. e. krause , r. s. decca , d. lpez , and e. fischbach , phys . lett . * 98 * , 050403 ( 2007 ) . a. lambrecht , i. pirozhenko , l. duraffourg , and p. andreucci , eur . . lett . * 77 * , 44006 ( 2007 ) . j. d. jackson , _ classical electrodynamics _ ( john wiley & sons , new york , 1999 ) . a. o. caride , g. l. klimchitskaya , v. m. mostepanenko , and s. i. zanette phys . rev . a * 71 * 042901 ( 2005 ) . , e. d. palik ( academic , new york , 1985 ) . t. vogel , g. dobel , e. holzhauer , h. salzmann , and a. theurer , appl . opt . * 31 * , 329 ( 1992 ) . n. w. ashcroft and n. d. mermin , _ solid state physics _ ( saunders colledge , philadelphia , 1976 ) . m. soltani , m. chaker , e. haddad , and r. v. kruzelecky , j. vac . a * 22 * , 859 ( 2004 ) . j. y. suh , r. lopez , l. c. feldman , and r. f. haglund , j. appl . phys . * 96 * , 1209 ( 2004 ) . f. y. gan and p. laou , j. vac . a * 22 * , 879 ( 2004 ) . d. h. youn , h. t. kim , b. y. chae , y. j. hwang , j. w. lee , s. l. maeng , and k. y. kang , j. vac . a * 22 * , 719 ( 2004 ) . h. w. verleur , a. s. barker , and c.n . berglund , phys . rev . * 172 * , 788 ( 1968 ) . b. m. gorelov , k. p. kronin , v. v. koval , and v. m. ogenko , techn . * 27 * , 157 ( 2001 ) . v. a. parsegian , _ van der waals forces : a handbook for biologists , chemists , engineers , and physicists _ ( cambridge university press , cambridge , 2005 ) . l. bergstrm , adv . colloid interface sci . * 70 * , 125 ( 1997 ) . h. kakiuchida , p. jin , s. nakao , and m. tazawa , jap . phys . * 46 * , l113 ( 2007 ) . b. geyer , g. l. klimchitskaya , and v. m. mostepanenko , phys . d * 72 * , 085009 ( 2005 ) ; int . a * 21 * , 5007 ( 2006 ) . g. l. klimchitskaya , b. geyer , and v. m. mostepanenko , j. phys . a : mat . gen . * 39 * , 6495 ( 2005 ) . b. geyer , g. l. klimchitskaya , and v. m. mostepanenko , arxiv:0704.3818 ; ann . ( n.y . ) , 2007 , to appear . j. n. munday , d. iannuzzi , yu . barash , and f. capasso , phys . a * 71 * , 042102 ( 2005 ) . r. b. rodrigues , p. a. maia neto , a. lambrecht , and s. reynaud , europhys . lett . * 76 * , 822 ( 2006 ) . s. k. lamoreaux and w. t. buttler , phys . e * 71 * , 036109 ( 2005 ) . m. brown - hayes , d. a. r. dalvit , f. d. mazzitelli , w. j. kim , and r. onofrio , phys . a * 72 * , 052102 ( 2005 ) . antonini , g. bressi , g. carugno , g. galeazzi , g. messineo , and g. ruoso , new j. phys . * 8 * , 239 ( 2006 ) . g. bimonte , e. calloni , g. esposito , l. milano , and l. rosa , phys . 94 * , 180402 ( 2005 ) . g. bimonte , e. calloni , g. esposito , and l. rosa , nucl . b * 726 * , 441 ( 2005 ) . c. braggio , g. bressi , g. carugno , c. del noce , g. galeazzi , a. lombardi , a. palmieri , g. ruoso , and d. zanello , europhys . lett . * 70 * , 754 ( 2005 ) . kim , j. h. brownell , and r. onofrio , phys . * 96 * , 200402 ( 2006 ) . + * fig . 1 . * ( color online ) schematic diagram of the experimental setup for the measurement of the difference casimir force . the patterned si plate with two sections of different dopant densities is mounted on a piezo below the au coated sphere attached to a cantilever of an atomic force microscope . the piezo oscillates in the horizontal direction above different regions of the plate causing the flexing of the cantilever in response to the casimir force . * fig . 2 . * ( color online ) steps in the fabrication of the patterned si plate with patterned doping ( see text for more details ) . * fig . ( color online ) image of the interferometer based variable temperature atomic force microscope with the force sensitivity up to 1fn fabricated at uc , riverside . the critical components are labeled . 4 . * the dielectric permittivity of si along the imaginary frequency axis for samples with high concentration of charge carriers @xmath68 and @xmath69 is shown by the solid lines 1 and 2 , respectively . for the sample with a low concentration of charge carriers @xmath14 the permittivity versus frequency is shown by dashed lines @xmath70 and @xmath79 based on whether the static permittivity is finite or infinitely large . the premittivity of au is indicated by the dotted line . * the difference casimir forces versus separation in the case when the higher concentration of charge carriers is equal to @xmath68 and the sample with a lower concentration , @xmath14 , is described by a finite or infinitely large static permittivity are shown by the solid line 1@xmath70 and dashed line 1@xmath79 , respectively . the analogous difference forces when the higher concentration of charge carriers is equal to @xmath69 are shown by the solid line 2@xmath70 and dashed line 2@xmath79 . * fig . 6 . * the difference casimir pressures versus separation in the case when the higher concentration of charge carriers is equal to @xmath68 and the sample with a lower concentration , @xmath14 , is described by a finite or infinitely large static permittivity are shown by the solid line 1@xmath70 and dashed line 1@xmath79 , respectively . the analogous difference pressures when the higher concentration of charge carriers is equal to @xmath69 are shown by the solid line 2@xmath70 and dashed line 2@xmath79 . * fig . 7 . * schematic of the experimental setup for the observation of modulation of the casimir force in an insulator - metal phase transition . light from a chopped 980 nm laser heats a vo@xmath0 film leading to a phase transition to a state with higher concentration of charge carriers ( sapphire substrate is not shown ) . cooling in between pulses causes the transition to a state with lower concentration of carriers . the cantilever of an atomic force microscope flexes in response to the difference casimir force . * fig . 8 . * ( color online ) preliminary results on the resistance of vo@xmath0 film grown at uc riverside as a function of temperature are shown as black squares ( heating ) and dots ( cooling ) . * fig . 9 . * ( color online ) morphology of the same vo@xmath0 film , as in fig . 8 , grown by thermal evaporation . the heights of roughness peaks are of about 2 nm . * fig . 10 . * the effective dielectric permittivity of vo@xmath0 film on sapphire substrate along the imaginary frequency axis before and after the phase transition are shown by the solid lines 1 and 2 , respectively . the permittivity of au is indicated by the dotted line . * fig . 11 . * the casimir force between an au coated sphere and vo@xmath0 film on sapphire substrate versus separation before and after the phase transition are shown by the solid lines 1 and 2 , respectively . * fig . 12 . * the difference of the casimir forces after and before the phase transition versus separation computed using a finite static dielectric permittivity ( solid line ) and taking into account the dc conductivity of vo@xmath0 in a dielectric state ( dashed line ) . .values of the oscillator resonant frequencies @xmath111 , dimensionless relaxation parameters @xmath112 and oscillator strengths @xmath113 of vo@xmath0 film on sapphire substrate before the phase transition . [ cols="<,<,<,<",options="header " , ]

we show that mixed bipartite and states are geometrically and topologically distinguished in the space of states . they are characterized by non - vanishing euler - poincar characteristics on the topological side and by the existence of symplectic structures on the geometric side .

1305.3660

the existence of quantum correlations for multipartite separable mixed states can be regarded as one of the most interesting quantum information discoveries of the last decade . in 2001 ollivier and urek @xcite ( see also @xcite ) introduced the notion of quantum discord as a measure of the quantumness of correlation . quantum discord is always non - negative @xcite . the states with vanishing quantum discord are called _ pointer states_. they form the boundary between classical and quantum correlations @xcite . the bipartite pointer states can be identified with so - called classical - quantum , @xmath1 states @xcite . an important subclass of cq states are classical - classical , cc states which play an important role in quantification of the quantum correlations @xcite and were recently considered in the context of broadcasting scenarios @xcite . it is known that both classes are of measure zero in the space of density matrices @xcite . in this paper we focus on the symplecto - geometric and topological characterizations of mixed bipartite @xmath0 and @xmath1 states . in [ ] it was shown that pure separable states form the unique symplectic orbit of the local unitary ( lu ) group action . all other lu action orbits are non - symplectic . moreover , the more non - symplectic is a lu action orbit the more entangled are states belonging to it @xcite . as we show in the present work , the symplectic techniques can be also applied to describe mixed states . remarkably , the non - degenerate symplectic structure is present on a generic local unitary orbit through @xmath0 and @xmath1 states rather than separable states . more precisely , for density matrices defined on @xmath2 we show that the closure of all symplectic @xmath3-orbits is exactly the set of @xmath0 states . similarly , the closure of all symplectic @xmath4-orbits gives the set of @xmath1 states . this clearly indicates that symplecticity generically detects non - quantum rather than non - entangled states . for pure states two concepts we also provide the topological characterization of pure seprarable , @xmath0 and @xmath1 states . more precisely , for pure @xmath5-partite and bipartite mixed states we study euler - poincar characteristics , @xmath6 on lu action orbits . using hopf - samelson theorem @xcite we show that for pure states @xmath6 is non - zero exactly on the manifold of separable states . moreover , for bipartite mixed states , @xmath3-orbit has non - vanishing euler - poincar characteristics if and only if the states belonging to it are @xmath0 . similar result is true for @xmath4-orbits and @xmath1 states . as a conclusion separable , @xmath0 , and @xmath1 states are topologically distinguished . the paper is organized as follows . in section [ sec : coadjoint orbits ] we discuss the relevant geometric structures present on the manifold of bipartite isospectral density matrices , @xmath7 . the orbits of @xmath8 and @xmath4 are in a natural way homogenous submanifolds of this manifold . in section [ sec : restrictions - of - geometric ] we discuss the restriction of the geometric structures to arbitrary homogenous submanifolds of @xmath7 . in section [ sec : geometry of mixed states ] we show how symplectic and khler structures distinguish classes of orbits trough @xmath0 and @xmath1 states . the second part of the article deals with the topological characterization of these orbits . section [ sec : topology ] discusses the hopf - samelson theorem for calculation of the euler - poincar characteristic of homogenous spaces . in subsequent section [ sec : topol_results ] we compute the euler - poincar characteristic of orbits of pertinent groups through pure separable , @xmath0 , and @xmath1 states and show that these are the only orbits with non - zero euler - poincar characteristic . by facts we always denote results that are known . we do not present their proofs and refer the reader to the literature . on the other hand , by propositions and corollaries we denote all new results . their proofs are included in the text . a bipartite density matrix is a non - negative , trace - one operator @xmath9 on @xmath2 , i.e. an @xmath10 matrix , @xmath11 , whose spectrum @xmath12 consists of non - negative eigenvalues @xmath13 satisfying @xmath14 . two density matrices are _ isospectral _ if they have the same spectra . in the following we discuss geometric structures present on the set of isospectral density matrices . in particular , we show that this set is a compact khler manifold , that is , there exist mutually compatible symplectic , riemannian and complex structures on it . let @xmath15 be a diagonal bipartite density matrix . the density matrices which are isospectral with @xmath15 form an adjoint orbit through @xmath15 , @xmath16 , of @xmath17 action @xmath18 the lie algebra @xmath19 of group @xmath20 , i.e. the space of anti - hermitian @xmath10 traceless matrices is equipped with the @xmath20-invariant inner product defined by @xmath21 for @xmath22 let @xmath23 be the stabilizer of @xmath9 and @xmath24=0\}}$ ] its lie algebra . the geometric structures we want to discuss are defined on the tangent bundle of @xmath16 , @xmath25 . thus we first need to describe @xmath26 , the tangent space to @xmath16 at any @xmath22 . to this end , for @xmath27 consider the corresponding fundamental vector field @xmath28 @xmath29\,.\label{eq : tangent - vector}\ ] ] as the action of @xmath20 on @xmath16 is transitive , the fundamental vector fields at @xmath22 span @xmath26 . note that for @xmath30 the corresponding fundamental vector field vanishes , @xmath31 . therefore , the tangent space @xmath26 can be identified with @xmath32 , that is , with the orthogonal complement with respect to the inner product ( [ eq : inner ] ) of @xmath33 , @xmath34 @xcite . the symplectic form on @xmath16 is given by the kirillov - kostant - souriau ( kks ) form . this is a @xmath35-form which acts on the tangent space @xmath26 to @xmath16 at any @xmath22 . using @xmath34 it is defined by @xmath36\right)=\left(y\,|\,\left[x,\ , i\rho\right]\right)=\left(x\,|\,\left[i\rho,\ , y\right]\right)\,,\label{eq : kks}\ ] ] where , @xmath37 and @xmath38 ensures that @xmath39 is antihermitian and @xmath40 has real value . clearly when @xmath30 or @xmath41 we have @xmath42 which means that indeed @xmath43 is defined on the tangent space @xmath26 . one can also check that @xmath44 is closed and non - degenerate . therefore @xmath44 defines a symplectic structure on @xmath16 . moreover , group @xmath20 acts on @xmath16 in a symplectic way , i.e. @xmath45 , where @xmath46 denotes the pullback of @xmath44 by the action of @xmath47 . having the kks symplectic form ( [ eq : kks ] ) on @xmath16 there exists a natural almost complex structure associated to it . it is defined as follows . for @xmath22 we compute the polar decomposition of the map @xmath48 , @xmath49}$ ] , restricted to @xmath50 . it is straightforward to see that this restriction is non - degenerate and that it defines a skew - symmetric operator ( with respect to inner product ( [ eq : inner ] ) ) , @xmath51 . therefore the polar decomposition reads @xmath52 where @xmath53 is a positive operator , @xmath54 is orthogonal and skew - symmetric , @xmath55 and @xmath56=0}$ ] . it follows that @xmath57 . therefore @xmath58 can be used to define almost complex structure on @xmath16 . in fact @xmath58 turns out to be integrable and consequently it defines the complex structure on @xmath16 @xcite . the last structure on @xmath16 is the riemannian structure that is compatible with @xmath44 and @xmath59 introduced above . it is given by the following formula @xmath60\right)\,,\label{eq : metric - sympl}\ ] ] for @xmath22 and @xmath61 . one easily checks that so defined @xmath62 is symmetric , positive definite and @xmath20-invariant . moreover , straightforward computation shows that it is compatible with both @xmath44 and @xmath59 , i.e. @xmath63 thus structures @xmath44 , @xmath59 and @xmath62 define khler structure on @xmath16 . due to the positive - definiteness of @xmath64 , @xmath16 is a positive khler manifold @xcite . having defined the relevant geometric structures on @xmath16 we consider the following problem . let @xmath65 be a compact semisimple lie subgroup of @xmath20 , @xmath66 . by restriction of the adjoint action @xmath65 acts on @xmath16 in a symplectic way . we denote @xmath65-orbit through @xmath22 by @xmath67 . obviously @xmath68 . one can thus consider the restriction @xmath69 of the symplectic form ( [ eq : kks ] ) to @xmath67 . the restricted form is still closed , @xmath70 , but it need not to be non - degenerate . as a result , @xmath65-orbits in @xmath7 need not to be symplectic . we want to know which of them are . moreover , we want to know if those which are symplectic are also khler . before we state the relevant theorems we review some background information concerning semisimple lie algebras . for more detailed account of this topic consult @xcite . let @xmath71 be the lie algebra of @xmath65 . as @xmath65 is a compact semisimple lie group , the algebra @xmath71 has the following _ root decomposition _ @xcite @xmath72 where , @xmath73 is a cartan subalgebra of @xmath71 and @xmath74 ranges over all positive roots . the cartan subalgebra @xmath73 is @xmath75,\,\alpha>0\right).\ ] ] moreover , for each positive root @xmath74 the triple @xmath76 is isomorphic with @xmath77 , i.e. @xmath78 & = & 2i\left(e_{\alpha}+e_{-\alpha}\right)\,,\\ \left[ih_{\alpha},\ , i(e_{\alpha}+e_{-\alpha})\right ] & = & -2\left(e_{\alpha}-e_{-\alpha}\right)\,,\\ \left[e_{\alpha}-e_{-\alpha},\ , i(e_{\alpha}+e_{-\alpha})\right ] & = & 2ih_{\alpha}\,.\end{aligned}\ ] ] for @xmath79 the root space decomposition is particularly simple @xmath80 where @xmath81 @xmath82 and @xmath83 . symplectic orbits @xmath84 are characterized by the kostant - sternberg theorem @xcite ( see also @xcite ) . the necessary condition for the orbit @xmath84 to be symplectic is [ torus ] ( the necessary condition ) if the orbit @xmath84 is symplectic ( with respect to the restriction of kks symplectic form ( [ eq : kks ] ) ) then there exists @xmath85 such that @xmath86=0}$ ] for all @xmath87 , where @xmath73 is a cartan subalgebra of @xmath71 . in the following we assume that the necessary condition is satisfied , that is , @xmath88=0}$ ] . using ( [ eq : tangent - vector ] ) and the root decomposition ( [ eq : root - decomposition ] ) we have @xmath89 where the sum is over positive roots and @xmath90,\ , i\left[e_{\alpha}+e_{-\alpha},\,\tilde{\rho}\right]\right)\,.\label{eq : p - alpha}\ ] ] we will need the following fact whose proof can be found in @xcite . [ omega ortogonal]for positive roots @xmath91 , @xmath92 if @xmath93 . thus @xmath94 is non - degenerate if and only if it is non - degenerate on each @xmath95 separately . using ( [ eq : kks ] ) and ( [ eq : p - alpha ] ) it is straightforward to check : [ counting-1 ] assume that @xmath96=0}$ ] . then for any @xmath95 we have exactly three possibilities : ( 1 ) @xmath97 , if and only if @xmath98=0=\left[e_{-\alpha},\,\tilde{\rho}\right]}$ ] , ( 2 ) @xmath99 and @xmath100 , if and only if @xmath101 and @xmath98\neq0}$ ] or @xmath102\neq0}$ ] , ( 3 ) @xmath99 and @xmath103 is non - degenerate , if and only if @xmath104__. _ _ we can now state the kostant - sternberg theorem in its usual form . [ torus-1 ] ( kostant - sternberg theorem @xcite ) the orbit @xmath84 is symplectic if and only if : ( 1 ) there exists @xmath85 such that @xmath86=0}$ ] for all @xmath87 and ( 2 ) for any positive root @xmath74 if @xmath101 then @xmath98=0=\left[e_{-\alpha},\,\tilde{\rho}\right]}$ ] . in order to measure how non - symplectic is an orbit @xmath67 we will , similarly to @xcite , use the notion of degree of degeneracy @xmath69 , @xmath105 . it is given by @xmath106 the dimension of @xmath67 is @xmath107=\left[e_{-\alpha},\,\tilde{\rho}\right]=0\right\ } \right|\right)\right)\,,\label{eq : dimension - of - orbit}\ ] ] where @xmath108 denotes the number of elements of a discrete set @xmath109 . the rank of @xmath69 is @xmath110 in order to characterize orbits @xmath67 that are khler submanifolds of @xmath111 we need the following results [ possitive kahler ] [ ] let @xmath112 be a positive khler manifold . then any complex submanifold @xmath113 is also a khler manifold . [ complex orbits ] an orbit @xmath114 is a complex submanifold of @xmath111 if and only if it is an almost complex sumbanifold of @xmath111 . the only thing one has to check is the integrability is almost - complex of the almost complex structure on @xmath115 that @xmath114 inherits from @xmath111 . we will deal with this in appendix 1 . let us now characterize almost complex orbits of @xmath65 in @xmath111 . by the definition of almost complex structure @xmath116 , an orbit @xmath114 is almost complex submanifold if and only if for @xmath117 ( defined in the fact 1 ) @xmath118 we would like to remark that every simplectic orbit @xmath119 can be equipped with the `` intrinsic '' khler structure . it follows from the fact@xcite that every simplectic orbit @xmath67 is diffeomorphic , via the moment map , with coadjoint orbit of some @xmath65 . the latter possess a standard khler structure , as discussed above . we illustrate this phenomenon on a concrete example . consider the action of the group @xmath120 acting on the projective space @xmath121 in a natural way ( induced from the action of @xmath122 on @xmath123 treated as a carrier space of an irreducible representation of @xmath122 ) . in this case simplectic orbits of are orbits through states corresponding to nonzero eigenvectors of the operator @xmath124 . these orbits are diffeomorphic to two dimensional spheres so they can be equipped with the intrinsic @xmath122 invariant khler structure . nevertheless , only the orbit through the state corresponding to the maximal or minimal eigenvalue of @xmath124 inherits the khler structure from @xmath121 . the exhaustive discussion of the above example can be found in [ ] . in the following we apply the ideas presented in sections [ sec : coadjoint orbits ] and [ sec : restrictions - of - geometric ] to mixed bipartite states . let @xmath125 , where @xmath126 and @xmath127 . we start with definitions of @xmath0 and @xmath1 states . a quantum state @xmath9 defined on @xmath128 is called a @xmath0 state @xcite if it can be written in the form @xmath129 where @xmath130 is an orthonormal basis in @xmath131 and @xmath132 is an orthonormal basis in @xmath133 . a quantum state @xmath9 defined on @xmath128 is called a @xmath1 state if it can be written in the form @xmath134 where @xmath130 is an orthonormal basis in @xmath131 and @xmath135 are density matrices defined on @xmath133 . in order to use the tools presented in section [ sec : restrictions - of - geometric ] we need to choose some subgroup @xmath66 . note that both @xmath0 and @xmath1 are @xmath3-invariant sets . it turns out that for @xmath0 the relevant group is indeed @xmath3 . on the other hand , in order to distinguish geometric properties of @xmath1 states , one has to take @xmath4 . in the following we prove our main results , i.e. we show that the orbits through generic @xmath0 states of the group @xmath136 , @xmath137 , are the only symplectic orbits in the space of density matrices on @xmath128 . we also compute the rank and the dimension of degeneracy of the symplectic form restricted to @xmath65-orbits through @xmath0 states . finally , we illustrate our results by the example two - quibit system . the root decomposition ( [ eq : root - decomposition ] ) of the lie algebra @xmath138 is a direct sum of root decompositions of @xmath139 and @xmath140 which are given by ( [ eq : root - su(n ) ] ) . in the following we use the representation @xmath141 under ( [ eq : rep ] ) we have @xmath142 where @xmath143 @xmath144 @xmath145 where matrices @xmath146 , @xmath147 and @xmath148 are defined as in ( [ eq : hxy ] ) and @xmath149 , @xmath150 . we denote @xmath151 . [ symplectic - necessary]if the @xmath65-orbit is symplectic , then it consists only of @xmath0 states , i.e. @xmath0 states satisfy the necessary condition given in fact [ torus ] . assume @xmath67 is symplectic . by fact [ torus ] , @xmath152=0}$ ] . but @xmath73 consists of matrices @xmath153 , where @xmath154 are traceless diagonal matrices . it is thus clear that @xmath155 . conversely , if @xmath155 then it follows directly from the definition of @xmath73 that @xmath152=0}$ ] . [ sympl cond]the orbit of adjoint action of @xmath136 through cc state @xmath156 is symplectic if and only if the following conditions hold : 1 . for any @xmath157 , @xmath158 , 2 . for any @xmath157 , @xmath159 . let @xmath160 . by fact [ torus-1 ] we first verify when @xmath161 : @xmath162 similarly @xmath163 next , we verify when @xmath164=[e_{-\alpha},\rho]=0}$ ] . @xmath165 & = & \sum_{l=1}^{n_{2}}(p_{jl}e_{ij}-p_{il}e_{ij})\otimes e_{ll}\,,\\ \left[e_{ji}\otimes i_{n_{2}},\rho\right ] & = & \sum_{l=1}^{n_{2}}(p_{il}e_{ji}-p_{jl}e_{ji})\otimes e_{ll}\,,\\ \left[e_{ij}\otimes i_{n_{2}},\rho\right ] & = & \left[e_{ji}\otimes i_{n_{1}},\rho\right]=0\leftrightarrow\forall_{l}\ , p_{jl}=p_{il}\,.\label{eq : p - alpha-0}\end{aligned}\ ] ] analogously @xmath166=\left[i_{n_{1}}\otimes e_{ji},\rho\right]=0\leftrightarrow\forall\ , p_{lj}=p_{li}\,.\label{eq : p - alpha-01}\ ] ] therefore the condition @xmath167=0}$ ] translates to the conditions 1 and 2 above . one can interpret results stated in the proposition [ sympl cond ] in terms of the reduced density matrices of @xmath9 . first , the @xmath65-orbit through a @xmath0 state @xmath9 is symplectic , if spectra of @xmath168 and @xmath169 are non - degenerate . moreover , whenever there is a pair of equal eigenvalues in the spectrum of @xmath170 or @xmath171 , the @xmath65-orbit through @xmath9 is symplectic provided @xmath9 satisfies conditions 1 and 2 stated in proposition [ sympl cond ] . for generic @xmath0 state the spectra of @xmath170 and @xmath171 are non - degenerate . therefore , the set of all @xmath0 states is the closure of all symplectic @xmath65-orbits in the space of quantum states . [ cc - closure ] by proposition [ sympl cond ] the degeneracies in the spectra of @xmath170 and @xmath171 are described by equations for hyperplanes in the set of @xmath0 states for fixed bases @xmath172 and @xmath173 in @xmath131 and @xmath133 respectively , i.e. @xmath174 for some @xmath175 . there is a finite number of them , and thus the complement of the set described by them is dense in the set of all cc states . let us now use to study which symplectic orbits of @xmath176 are actually khler ( they inherit the khler structure form @xmath111 ) . straightforward computations based on and show that the tangent space to the orbit of @xmath176 at the state @xmath9 is spanned by the following vectors ( @xmath177 ) . @xmath178 & = & \sum_{l=1}^{n_{2}}(p_{il}-p_{jl})y_{ij}\otimes e_{ll}\,,\\ \left[y_{ij}\otimes i_{n_{2}},i\rho\right ] & = & \sum_{l=1}^{n_{2 } } ( p_{jl}-p_{il } ) i x_{ij}\otimes e_{ll}\,,\\ \left[i_{n_1}\otimes x_{ij},i\rho\right ] & = & \sum_{k=1}^{n_{1 } } ( p_{ki}-p_{kj } ) e_{kk}\otimes y_{ij}\,,\\ \left[i_{n_1}\otimes y_{ij},i\rho\right ] & = & \sum_{k=1}^{n_{1 } } ( p_{kj}-p_{ki } ) e_{kk}\otimes i x_{ij}\ .\ \label{eq : tangent space}\end{aligned}\ ] ] the action of @xmath179 on each @xmath180 ( see for the convention used to describe roots of @xmath181 ) @xmath182,i\rho\right ] & = & - \sum_{l=1}^{n_{2}}(p_{jl}-p_{il})^2 i x_{ij}\otimes e_{ll}\ , \\ \left[\left[y_{ij}\otimes i_{n_{2}},i\rho\right],i\rho\right ] & = & \sum_{l=1}^{n_{2}}(p_{il}-p_{jl})^2 y_{ij}\otimes e_{ll}\ , \\ \left[\left[i_{n_1}\otimes x_{ij},i\rho\right ] , i\rho\right ] & = & - \sum_{l=1}^{n_{1}}(p_{kj}-p_{ki})^2 i e_{kk } \otimes x_{ij } \ , \\ \left[\left[i_{n_1}\otimes y_{ij},i\rho\right],i\rho\right ] & = & \sum_{k=1}^{n_{1 } } ( p_{ki}-p_{kj})^2 e_{kk } \otimes y_{ij}\ . \label{eq : complex local}\end{aligned}\ ] ] we have @xmath183 for @xmath184 . direct inspection shows that in order for the orbit through the state @xmath9 to be khler the following conditions have to be satisfied , @xmath185 where @xmath186 and @xmath187 are real paremetres depending only on indices @xmath188 and @xmath189 . [ kahlerlu ] the solutions to conditions and are the following : 1 . @xmath190 and @xmath191 for @xmath192 . @xmath193 for all pairs of indices @xmath194 . one easily checks that the above satisfy and . assuming that there exist two pairs of indices @xmath194 and @xmath195 such that @xmath196 leads to a contradiction . the above reasoning reproduces results from [ ] where the author showed that pure separable states form the unique khler orbit of the complexification of @xmath65 , @xmath197 , in @xmath198 . however , we would like to point out that in general there might be more khler orbits of @xmath136 in the set of density matrices . proposition [ kahlerlu ] treats this problem . we now turn to a detailed description of symplectic properties of orbits through @xmath0 states . we compute dimensions of orbits , @xmath199 , rank of the symplectic form restricted to orbits @xmath200 , and its degree of degeneracy , @xmath201 . we first introduce the notation which will be used in formulas for these quantities . for fixed @xmath202 we consider the coefficients @xmath203 of @xmath9 as entries of the @xmath204 matrix @xmath205 . let @xmath206 be its @xmath188-th row and @xmath207 be its @xmath189-th column . let @xmath208 denote the sum of elements of @xmath209 , for @xmath209 being either a row or a column . define @xmath210 that is , @xmath211 and @xmath212 consist of all numbers one can get by summing elements in rows and columns of @xmath205 respectively . for each @xmath213 and @xmath214 , let @xmath215 be sets consisting of indices that label rows and columns of @xmath205 whose sums of elements are equal to @xmath216 and @xmath217 respectively . of course , @xmath218 and @xmath219 , where @xmath220 denotes the union of sets . moreover , for each @xmath213 and @xmath214 , let @xmath221 be the sets consisting of rows and columns of @xmath205 whose sums equal @xmath216 and @xmath217 respectively . let @xmath222 be the sets consisting of all rows and columns of @xmath205 . for each @xmath223 let @xmath224 be the sets consisting of indices that label rows and columns of @xmath205 that equal @xmath225 and @xmath226 respectively . of course for each row @xmath225 we have @xmath227 and for each column @xmath226 we have @xmath228 . moreover for each @xmath213 , @xmath229 and for each @xmath230 , @xmath231 . finally , let us denote by @xmath108 the number of elements of a discrete set @xmath109 and by @xmath232 the binomial coefficient . in what follows we assume the convention @xmath233 for @xmath234 . by ( [ eq : dimension - of - orbit ] ) to calculate @xmath199 it is enough to determine when @xmath235=0=\left[e_{-\alpha},\,\rho\right]}$ ] . using ( [ eq : p - alpha-0 ] ) and ( [ eq : p - alpha-01 ] ) we have @xmath236 & = & 0\longleftrightarrow i , j\in\mathcal{i}_{r}\,\,\mathrm{for}\,\mathrm{some}\ , r\in\mathcal{r}\,,\\ \left[i_{n_{1}}\otimes e_{ij},\,\rho\right ] & = & 0\longleftrightarrow i , j\in\mathcal{j}_{c}\,\,\mathrm{for}\,\mathrm{some}\ , c\in\mathcal{c}.\end{aligned}\ ] ] hence @xmath237 by ( [ eq : rank ] ) to calculate @xmath238 it is enough to determine when @xmath101 . by ( [ eq : rank1 ] ) and ( [ eq : rank2 ] ) we have @xmath239 thus @xmath240 having established formulas for @xmath199 and @xmath238 we arrive at our final result : [ maincc ] the dimension of the degeneracy of the symplectic form on the orbit through the @xmath0 state @xmath160 is equal to @xmath241 @xmath0 states , for which the corresponding @xmath65-orbits have the maximal degree of degeneracy @xmath242 , correspond to so - called magic rectangles @xcite . to be more precise , the maximum in equation ( [ eq : gegeneracy cc ] ) is obtained when @xmath243 , @xmath244 and both @xmath211 and @xmath212 have precisely one element . translating these conditions to the properties of @xmath204 matrix @xmath203 encoding a given @xmath0 state @xmath9 , one arrives at the following conditions : 1 . each row and colum of @xmath203 have to consist of different elements sums of elemenst in each row are the same . the same concerns sums of elements in each column . let @xmath245 . we will now use proposition [ maincc ] to calculate dimensions of orbits , ranks of the form @xmath43 , and its degrees of degeneration . therefore , let us consider the matrix @xmath246 of course , @xmath247 . there are four possibilities : [ [ distinct - sums - both - in - columns - and - in - rows ] ] distinct sums both in columns and in rows + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + consider the case @xmath248 , @xmath249 . it follows that no two columns or two rows are identical , therefore due to proposition [ maincc ] , @xmath250 . the rank of @xmath251 is also @xmath252 , so it is non - degenerate and the orbit is symplectic . [ [ equal - sums - in - columns - distinct - sums - in - rows ] ] equal sums in columns , distinct sums in rows + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + consider the case @xmath248 , @xmath253 . because @xmath254 , it follows that @xmath255 , and setting @xmath256 gives us @xmath257 , @xmath258 and @xmath91 , that is @xmath259 due to proposition [ maincc ] , the rank of @xmath43 is equal to @xmath260 , because there is a pair of columns with equal sums . now , if these columns are equal , then @xmath261 , meaning @xmath262 . condition @xmath91 implies @xmath263 . from proposition [ maincc ] it follows that @xmath264 , as there are two identical columns and no identical rows . the orbit is symplectic of dimension 2 . for any @xmath265 and @xmath266 , we get @xmath267 , so the degeneracy of @xmath94 is equal to @xmath268 . [ [ distinct - sums - in - columns - equal - sums - in - rows ] ] distinct sums in columns , equal sums in rows + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + similarly to the case above , we consider @xmath9 such that @xmath269 and @xmath266 . we get that if @xmath261 and @xmath263 , then the orbit is symplectic of dimension 2 , and if @xmath265 and @xmath266 , then the orbit is of dimension @xmath252 and @xmath251 has degeneracy @xmath35 . [ [ equal - sums - in - columns - and - rows ] ] equal sums in columns and rows + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + what remains is the case @xmath270 , @xmath253 . with @xmath271 , simple calculations lead us to @xmath272 . now , if @xmath273 , then both columns and both rows are equal , and the orbit is of dimension @xmath274 . in fact , it is just a point @xmath275 . for any @xmath263 , both of the columns are distinct , and so are the rows . therefore , due to proposition [ maincc ] , @xmath267 and @xmath276 , that is the degeneracy of @xmath251 is equal to @xmath277 , and is maximal possible for a @xmath0 state . because in the case considered @xmath0 states in a fixed computational basis @xmath278 , @xmath279 , @xmath280 and @xmath281 form a three - dimensional simplex , it is possible to draw pictures illustrating the results discussed above . we denote vertices of the simplex we denote by @xmath282 , @xmath283 , @xmath284 and @xmath285 . figures [ fig1]-[fig3 ] show dimension of @xmath286-orbits as well as the rank and degree of degeneracy for different points in this simplex . dimensions of orbits through @xmath0 states of two qbits . large dot : @xmath287 , dotted lines : @xmath288 , elsewhere : @xmath289,width=226 ] ranks of @xmath251 for orbits through @xmath0 states of two qbits . thick dashed line : @xmath290 , lined surfaces : @xmath291 , elsewhere : @xmath292.,width=226 ] degrees of degeneracy of @xmath251 for orbits through @xmath0 states of two qbits . thick dashed line : @xmath277 , lined surfaces : @xmath268 , dotted lines and elsewhere : @xmath293.,width=226 ] setting @xmath294 give us analogous results for @xmath1 states as we got for @xmath0 states . in this case the lie algebra of @xmath65 , @xmath295 is represented on @xmath128 _ via _ the mapping @xmath296 throughout the computations we will use the notation analogous to the one used for @xmath0 states : @xmath297 @xmath298 where matrices @xmath146 , @xmath147 and @xmath148 are defined as in ( [ eq : hxy ] ) and @xmath299 . if the orbit of adjoint action of @xmath300 is symplectic , then it consists only of cq states.[torus cq ] let @xmath67 be symplectic . by fact [ torus ] , we may assume that @xmath152=0}$ ] . thus @xmath9 and elements from @xmath73 must have common eigenvalues . therefore , @xmath9 has the form of the cq state . the orbit of the coadjoint action of @xmath301 through a @xmath1 state @xmath302 is symplectic if and only if for any @xmath157 , @xmath303.[sympl cq ] let @xmath304 . following fact [ torus ] , we check for which @xmath74 we have @xmath161 . @xmath305 @xmath306 we also check the condition @xmath164=[e_{-\alpha},\rho]=0}$ ] . @xmath307=p_{j}e_{ij}\otimes\rho_{j}-p_{i}e_{ij}\otimes\rho_{i}\label{orbit - dim-1}\ ] ] @xmath308=p_{i}e_{ji}\otimes\rho_{i}-p_{j}e_{ji}\otimes\rho_{j}\ ] ] @xmath307=[e_{ji}\otimes i_{n_{2}},\rho]=0\iff p_{j}\rho_{j}=p_{i}\rho_{i}\ ] ] so @xmath309=0}$ ] for all positive roots @xmath310 translate to the conditions given above . a generic orbit of the adjoint action of @xmath301 through a @xmath1 state is symplectic and the set of all cq states is equal to the closure of the sum of all symplectic orbits of @xmath4 in the space of quantum states . analogous to the proof of corollary [ cc - closure ] . we now give more detailed description of @xmath65-orbits through @xmath1 states . we compute dimensions of orbits , @xmath199 , rank of the symplectic form restricted to orbits @xmath200 , and its degree of degeneracy , @xmath201 . we first introduce necessary notation which we then use to compute numbers we are interested in . let us fix @xmath311 and let @xmath312 be the set of probabilities that appear in ( [ eq : cq def ] ) . for each @xmath313 , let @xmath314 be the set of indices that have the same value of @xmath315 . of course @xmath316 . for @xmath313 , let @xmath317 let @xmath318 be the set consisting of all @xmath319 that appear in the decomposition ( [ eq : cq def ] ) . for each @xmath320 , let @xmath321 of course for each @xmath320 we have @xmath322 . moreover for each @xmath313 , @xmath323 . as before , we assume the convention @xmath233 for @xmath324 . we essentially repeat arguments that were given to justify proposition [ maincc ] . the only difference now is the structure of the group @xmath65 . one should introduce minor corrections in the argumentation . in particular , sets @xmath325 and @xmath326 should be replaced by the set @xmath327 . also , the set @xmath328 should be taken instead of the sets @xmath211 and @xmath212 . by the analogy we obtain the following formulas : @xmath329 @xmath330 we can now state the final result . [ main cq]the dimension of the degeneracy on the @xmath65-orbit through a @xmath1 state @xmath9 is equal to @xmath331 having discussed the geometric properties of @xmath0 and @xmath1 states we focus now on their topological features . in the following we recall the notion of the euler - poincar characteristics for compact manifolds and homogenous spaces . in particular we invoke the hopf - samelson theorem which will be used in the section [ sec : topol_results ] for caclulation of euler - poincar characteristics of @xmath65-orbits through @xmath0 and @xmath1 states . let @xmath65 be a compact connected lie group . assume that @xmath65 acts on a compact manifold @xmath112 . it is known that each orbit @xmath332 of @xmath65-action on @xmath112 is a homogenous space , i.e. @xmath333 , where @xmath334 is the isotropy subgroup of @xmath335 , that is , @xmath336 . in the following we analyze the topological structure of orbits @xmath337 . in particular we invoke the hopf - samelson theorem @xcite which says that the euler - poincar characteristics , @xmath338 is always non - negative and positive if and only if the ranks of @xmath65 and @xmath334 are the same . the most fundamental topological invariant of a topological space @xmath209 is the euler - poincar characteristics . for a compact manifold @xmath209 it is defined as @xmath339 where @xmath340 denotes the space of closed @xmath315-forms . recall that the @xmath315-th cohomology group of @xmath209 is a quotient space of the closed @xmath315-forms @xmath341 by the exact @xmath315-forms @xmath342 @xmath343 @xmath344 @xmath345 the beauty of the euler - poincar characteristics is manifested by the following equality @xmath346 here we state the hopf - samelson theorem for the euler - poincar e characteristics of a homogenous space . we first recall concepts of a maximal torus and of a weyl group . a maximal torus of @xmath65 , denoted by @xmath347 , is defined as a maximal abelian compact and connected subgroup of @xmath65 . a maximal torus is in general not unique . yet , all maximal tori are conjugate , i.e. they are of the form @xmath348 for @xmath349 . the normalizer , @xmath350 is defined by @xmath351 . analogously , the centralizer , @xmath352 , is given by @xmath353 . both @xmath350 and @xmath354 are compact . centralizer @xmath352 is clearly a normal subgroup of @xmath350 . therefore the quotient group @xmath355 is well defined and finite @xcite . the group @xmath356 is called a weyl group of @xmath65 . for example , when @xmath357 , the maximal torus @xmath347 may be chosen to consists of diagonal matrices in @xmath358 . the corresponding @xmath359 is given by these matrices in @xmath65 which permute diagonal elements of @xmath360 and hence @xmath361 . we can now state the hopf - samelson theorem . ( hopf - samelson theorem ) [ ( hopf - samelson - theorem)]let @xmath65 be a compact connected lie group and @xmath334 be a closed subgroup of @xmath65 . if @xmath334 contains some maximal torus of @xmath65 , ie . @xmath362 , the euler - poincar characteristic is given by @xmath363 . when @xmath334 does not contain a maximal torus @xmath364 . in this section we compute euler - poincar characteristic for orbits through pure separable , @xmath0 , and @xmath1 states . we show that the euler - poincar characteristic of the orbit of the relevant group distinguishes classes of pure separable , @xmath0 , and @xmath1 states . we start with the discussion of pure separable states . while discussing @xmath0 and @xmath1 states we will use the notation from section [ sec : geometry of mixed states ] . consider the system consisting of @xmath5 distinguishable particles described by @xmath365 manifold of pure states , @xmath366 , consist of all rank - one orthogonal projectors defined on @xmath128 . the group of local unitary operations @xmath367 acts on @xmath366 in a natural manner , @xmath368 the manifold of separable states , @xmath369 , consists of states of the form @xmath370 for some normalized @xmath371 . separable states form the @xmath65-orbit through @xmath372 , where @xmath373 is a simple tensor . in what follows we will prove that the manifold of separable states is the only @xmath65-orbit that have non - vanishing euler - poincar characteristic . any maximal torus of @xmath65 is given by @xmath374 , where each @xmath375 is some maximal torus of @xmath376 . it is a matter of straightforward calculation to see that @xmath377 one easily checks that if a stabilizer of a given state @xmath378 contains some @xmath347 , then @xmath379 is separable ( proof is analogous to the proof of proposition [ torus ] ) . moreover , it is known that for the separable @xmath379 we have @xcite @xmath380 straightforward calculation gives @xmath381 taking into account the above discussion and fact [ ( hopf - samelson - theorem ) ] we arrive at the following proposition . among all orbits of the local unitary group @xmath382 in the manifold of pure states only the orbit of separable states have the the non - vanishing euler - poincar characteristic . it is given by @xmath383 in order to compute the euler - poincar characteristic of orbits through @xmath0 and @xmath1 states we need to compute the stabilizers of the action of relevant groups acting on the state of interest . we first show that @xmath0 and @xmath1 states are uniquely characterized by the non - vanishing euler - poincar characteristic of orbits @xmath3 and @xmath4 respectively . [ eurel cc char]let @xmath22 be a mixed state of a bipartite system @xmath384 . let @xmath385 be a group of the local unitary operations . the @xmath65-orbit through @xmath9 , @xmath386 , has non - vanishing euler - poincar characteristic if and only if @xmath9 is a @xmath0 state . every maximal torus of @xmath65 is of the form @xmath387 , where @xmath388 and @xmath389 are maximal tori of @xmath390 and @xmath391 respectively . by proposition [ torus ] , if @xmath9 is stabilized by some @xmath347 , then @xmath9 is a @xmath0 state . conversely , straightforward calculation shows that every @xmath0 state is stabilized by some maximal torus @xmath392 . by the fact [ ( hopf - samelson - theorem ) ] we get that @xmath393 . let @xmath22 be a mixed state of a bipartite system @xmath384 . let @xmath300 . the @xmath65-orbit through @xmath9 , @xmath386 , has the non - vanishing euler - poincar characteristic if and only if @xmath9 is a @xmath1 state . the proof is completely analogous to the proof of proposition [ eurel cc char ] . we next compute the euler - poincar characteristic . we start with the case of @xmath1 states as the computation for @xmath0 will follow from the former . [ euler cq ] for a @xmath1 state @xmath9 and @xmath301 , under the notation used in proposition [ main cq ] , the euler characteristic of @xmath386 is equal to @xmath394 due to fact [ ( hopf - samelson - theorem ) ] , @xmath395 . obviously @xmath396 , so it suffices to find @xmath397 . because @xmath398 is connected ( see appendix 2 for the proof ) , in order to find @xmath399 it is enough to find the lie algebra of the stabilizer , @xmath400 . it is precisely the annihilator of @xmath9 with respect to the adjoint action of @xmath71 . since all non - zero elements of @xmath401\}$ ] are linearly independent ( it follows directly from ( [ orbit - dim-1 ] ) ) , the annihilator of @xmath9 is spanned by all @xmath402 and all @xmath403 for which @xmath164=0}$ ] . recall that @xmath404=0}$ ] if and only if @xmath405 for some @xmath320 . therefore we have @xmath406 where @xmath407 is spanned by elements of the cartan algebra that are not included in the first part of the above expression . note that @xmath408 for some @xmath409 . passing to the lie group picture we get @xmath410 where @xmath411 is a @xmath409-dimensional torus . it follows that @xmath412 we will use parts of the above reasoning in the computation of @xmath413 for @xmath0 states . [ euler cc]for a @xmath0 state @xmath9 and @xmath414 , under the notation used proposition [ maincc ] , the euler - poincar characteristic of @xmath386 is equal to @xmath415 due to fact [ ( hopf - samelson - theorem ) ] , @xmath416 . obviously @xmath417 . in order to compute @xmath418 we find @xmath399 . just like in the @xmath1 case @xmath399 turns out to be connected ( see appendix 2 for the proof ) and in order to find @xmath399 it is enough to find its lie algebra , that is , the annihilator of @xmath9 with respect to the adjoint action of @xmath71 . we have @xmath138 . for @xmath419 and @xmath420 the non - zero elements of the form @xmath421}$ ] and @xmath422}$ ] are linearly independent ( see ( [ eq : p - alpha-0 ] ) and ( [ eq : p - alpha-01 ] ) ) . therefore , the annihilator of @xmath9 with respect to the adjoint action of @xmath71 is a direct sum of annihilators with respect to actions of @xmath139 and @xmath140 taken separately : @xmath423\,,\label{eq : rep left}\\ \mathfrak{su}(n_{2 } ) & \ni & y\rightarrow[i_{n_{1}}\otimes y,\,\rho]\,.\label{eq : rep right}\end{aligned}\ ] ] from the perspective of the representations ( [ eq : rep left ] ) and ( [ eq : rep right ] ) the state @xmath9 can be considered separately as @xmath1 or @xmath424 state . the definition of a @xmath424 state is analogous to the definition of @xmath1 state . annihilators of @xmath9 with respect to the action of @xmath139 or @xmath140 are thus annihilators of @xmath9 treated as a @xmath1 or @xmath424 state . let @xmath425 and @xmath426 be sets @xmath327 and @xmath427 ( see ( [ eq : def q ] ) and ( [ eq : def i sigma ] ) ) when @xmath9 is treated as a @xmath1 state . analogously , let @xmath428 and @xmath429 be sets @xmath327 and @xmath427 when @xmath9 is treated as a @xmath424 state . repeating the reasoning from the proof of proposition [ euler cq ] we get ( see ( [ eq : stab cq ] ) ) @xmath430 for some @xmath409 . therefore we have @xmath431 @xmath432 we now consider different presentations of a @xmath0 state @xmath160 that are suitable when it is treated as a @xmath1 or @xmath424 state . we define the marginal distributions @xmath433 and @xmath434 . we have @xmath435 where @xmath436 . expression ( [ eq : cq pres ] ) is useful when @xmath9 is treated as a @xmath1 state . analogously we have @xmath437 where @xmath438 . on the other hand , expression ( [ eq : cq pres ] ) is useful when @xmath9 is treated as a @xmath424 state . closer examination of ( [ eq : cq pres ] ) shows that the set @xmath425 is in the bijection with the @xmath326 . that is each @xmath439 correspond to the unique @xmath440 . for such a pair we have @xmath441 . similarly , we have the bijection between @xmath428 and @xmath325 . each @xmath442 correspond to the unique @xmath443 and we have @xmath444 . these two observations together with ( [ eq : final ] ) conclude the proof . we have showed that geometric and topological methods can be applied to distinguish interesting classes of mixed states of composite quantum systems . on the geometrical side we proved that for bipartite system @xmath384 the generic @xmath0 states are distinguished as symplectic orbits of @xmath3 in the manifold of isospectral density matrices . similarly , the generic @xmath1 states are distinguished as symplectic orbits of @xmath4 in the same manifold . on the topological side we studied the euler - poincar characteristic of orbits of relevant groups through arbitrary multipartite pure , @xmath0 , and @xmath1 states . we proved that non - zero euler - poincar characteristic of orbits of the local unitary group through pure multipartite and bipartite mixed states characterizes pure separable and @xmath0 states . analogously , non - vanishing euler - poincar characteristic of orbits @xmath4 on bipartite mixed states detects @xmath1 states . above results can be easily generalized to mixed states of multipartite systems . for example in the tripartite case , geometric and topological features of the orbits of the suitably chosen groups should distinguish classes of @xmath445 or @xmath446 states . another interesting generalization would involve the usage of the same geometric and topological methods to study mixed states of fermionic and bosonic systems . we would like to thank marek ku and jarosaw korbicz for fruitful discussions . we are grateful to the anonymous reviewer for pointing out mistakes in our original treatment of the problem of induced khler structures . we acknowledge the support of polish mnisw iuventus plus grant no . ip2011048471,sfb / tr12 symmetries and universality in mesoscopic systems program of the deutsche forschungsgemeischaft and a grant of the polish national science centre under the contract number dec-2011/01/m / st2/00379 . references h.ollivier and w. h. zurek , `` quantum discord : a measure of the quantumness of correlations '' , _ phys . lett . _ * 88 * 017901 ( 2001 ) l. henderson l and v. vedral , `` classical , quantum and total correlations '' , _ j. phys a _ * 34 * 6899 ( 2001 ) a. datta , `` condition for the nullity of quantum discord '' , ( _ preprint _ axiv:1003.5256 ) j. oppenheim , m. horodecki , p. horodecki , and r. horodecki , `` thermodynamical approach to quantifying quantum correlations '' _ phys . * 89 * 180402 ( 2002 ) m. piani , p. horodecki , and r. horodecki , `` no - local - broadcasting theorem for multipartite quantum correlations '' , _ phys . lett . _ * 100 * 090502 ( 2008 ) j. k. korbicz , p. horodecki , r. horodecki , `` quantum - correlation breaking channels , broadcasting scenarios , and finite markov chains '' , _ phys . rev . a _ * 86 * 042319 ( 2012 ) d. chruciski , `` quantum - correlation breaking channels , quantum conditional probability and perronfrobenius theory '' , _ phys . a _ * 377 * 606 ( 2013 ) a. ferraro a , l. aolita , d. cavalcanti , f. m. cucchietti , and a. acin , `` almost all quantum states have nonclassical correlations '' , _ phys . a _ * 81 * 052318 ( 2010 ) a. sawicki , a. huckleberry , and m. ku , `` symplectic geometry of entanglement '' , _ comm . _ * 305 * 441 ( 2011 ) a. huckleberry a , m. ku and a. sawicki , `` bipartite entanglement , spherical actions , and geometry of local unitary orbits '' , _ j. math . phys . _ * 54 * 022202 ( 2013 ) h. hopf and h. samelson , `` ein satz uber die wirkungsraume geschlossener liescher gruppen '' , _ comment . * 13 * 240 ( 1941 , in german ) m. a. rieffel , `` dirac operators for coadjoint orbits of compact lie groups '' , _ munster j. of math . _ * 2 * 265 ( 2009 ) v. guillemin v and s. sternberg , _ symplectic techinques in physics _ , ( cambridge : cambridge university press 1984 ) b. c. hall , _ lie groups , lie algebras , and representations : an elementary introduction _ ( new york : springer 2003 ) d. chrucinski , `` symplectic orbits in quantum state space '' , _ j. math . phys . _ * 31 * 1587 ( 1990 ) a. baecklund , i. bengsson , four remarks on spin coherent states , arxiv:1312.2427 t. hagedorn , `` on the existence of magic n - dimensional rectangles '' , _ discrete math . _ * 207 * 53 ( 1999 ) i. bengtsson and k. yczkowski _ geometry of quantum states _ ( cambridge : cambridge university press 2006 ) k. f. hofmann and s. a morris , _ the structure of compact groups : a primer for students - a handbook for the expert _ ( berlin : de gruyter studies in mathematics 2006 ) mikio nakahara , _ geometry , topology and physics , second edition _ ( iop publishing 2003 ) in this appendix we show that the almost integrable structure on the symplectic orbit @xmath447 introduced in the proof of proposition [ complex orbits ] is integrable . in order to do it we recall ( see [ ] for a more detailed treatment of complex and almost complex structures ) one of the equivalent definitions of integrability of the complex structure @xmath59 . let @xmath448 be the complexified tangent boundle of @xmath449 . for each @xmath450 the complex structure @xmath59 defines the decomposition of @xmath451 , @xmath452 where @xmath453 in the above expression @xmath454 denotes the identity operator on the complexified tangent space at @xmath335 , @xmath451 . vectors belonging to @xmath455 and @xmath456 are called holonomic and respectively anti - holonomic . introducting the notation . @xmath457 we get the decomposition of the complexified tangent bundle , @xmath458 sections of bundles @xmath459 and @xmath460 are called holonomic and respectively anti - holonomic vector fields . we denote the ( complex ) vector spaces of holonomic and anti - holonomic vector fields by @xmath461 and @xmath462 respectively . we are now ready to give the definition of integrability of the almost complex structure @xmath59 . almost complex structure @xmath59 is integrable@xcite if and only if for every @xmath463 we have @xmath464_{\mathrm{lie}}\in\mathcal{x}\left(\mathcal{m}\right)^+ \ , \ ] ] where @xmath465_{\mathrm{lie}}}$ ] denotes the lie bracket of vector fields . we now apply the notions introduced above to prove the integrability of the complex structure introduced in proposition [ complex orbits ] . we have @xmath466 . we assume @xmath386 is symplectic i.e. that it inherits the symplectic structure @xmath44 from @xmath198 . as before , the symplectic structure on @xmath386 is denoted by @xmath251 . by @xmath467 we denote the associated almost complex structure . because symplectic structure @xmath251 is @xmath65 invariant is is enough the check the integrability of the associated almost complex structure in the neighbourhood of @xmath117 such that @xmath468}$ ] for @xmath87 . let @xmath469 denote such a neighbourhood . because @xmath386 is symplectic for @xmath470 , @xmath471 we have the decomposition @xmath472 where @xmath473 k^{-1},\ , i k\left[e_{\alpha}-e_{-\alpha},\,\tilde{\rho}\right ] k^{-1}\right)\ , \label{eq : decop p(x)}\ ] ] and @xmath474 direct computation ( see ) shows that the decomposition of the complexified tangent space at @xmath335,@xmath475 reads as @xmath476 k^{-1 } |\ \alpha\in\mathcal{a}\right ) , \ t_x^-k.\rho^\mathbb{c}= \mathrm{span}_{\mathbb{c}}\left ( k\left [ e_{-\alpha},\,\tilde{\rho}\right ] k^{-1 } |\ \alpha\in\mathcal{a}\right ) \ .\ ] ] we observe that gives the local trivialistation of @xmath477 , @xmath478 . using this and the commutation relations in the lie algebra @xmath479 we conclude that for the above defined almost complex structure condition is satisfied and therefore it is integrable . in this part we show that the stabilizers of the action of the relevant groups through @xmath0 and @xmath1 states are connected . this result was used in the proofs of propositions [ euler cq ] and [ euler cc ] . let us first recall that that a ( not necessary maximal ) torus @xmath347 of a lie group @xmath20 is defined as a compact connected abelian subgroup of @xmath20 . we have the following fact . @xcite [ stab tori]let @xmath392 be a torus in a connected compact lie group @xmath65 . let @xmath480 be a centralizer of @xmath347 in @xmath65 . then the centralizer @xmath354 is a connected compact subgroup of @xmath481 where summation is over index labeling distinct eigenvalues @xmath484 and @xmath485 are projectors onto eigenspaces of @xmath9 . note that @xmath486 are formed from rank 1 projectors onto separable tensors . therefore , to the decomposition ( [ eq : cc eigen ] ) we can associate a unique torus @xmath392 where @xmath490 . using the fact that @xmath491 stabilizes @xmath9 if and only if @xmath409 preserves eigenspaces of @xmath9 and repeating the argument from the proof of proposition [ connected cc ] we get that @xmath492 where @xmath347 is a torus in @xmath65 given by

the non - vanishing of the natural orbital occupation numbers of the one - particle density matrix of many - body systems has important consequences for the existence of a density matrix - potential mapping for nonlocal potentials in reduced density matrix functional theory and for the validity of the extended koopmans theorem . on the basis of weyl s theorem we give a connection between the differentiability properties of the ground state wave function and the rate at which the natural occupations approach zero when ordered as a descending series . we show , in particular , that the presence of a coulomb cusp in the wave function leads , in general , to a power law decay of the natural occupations , whereas infinitely differentiable wave - functions typically have natural occupations that decay exponentially . we analyze for a number of explicit examples of two - particle systems that in case the wave function is non - analytic at its spatial diagonal ( for instance , due to the presence of a coulomb cusp ) the natural orbital occupations are non - vanishing . we further derive a more general criterium for the non - vanishing of no occupations for two - particle wave functions with a certain separability structure . on the basis of this criterium we show that for a two - particle system of harmonically confined electrons with a coulombic interaction ( the so - called hookium ) the natural orbital occupations never vanish .

1306.4142

the fractional occupation numbers @xmath0 of the correlated one - body reduced density matrix ( 1rdm ) have intrigued many scientists in the past decades . they are defined by the eigenvalue equation @xmath1 where the 1rdm itself is defined in terms of the usual creation and annihilation field operators as @xmath2 for a state @xmath3 where @xmath4 is space - spin coordinate . the one - particle orbitals @xmath5 in eq . are denoted as the natural orbitals ( no ) whereas the eigenvalues @xmath0 are called the no occupation numbers . as an integral kernel the 1rdm is a bounded linear hermitian operator with an infinite but countable eigenvalue spectrum and the set of all nos form a basis in the set of quadratically integrable functions . if the state @xmath3 is fermionic it is not difficult to prove that @xmath6 @xcite . in the following we will restrict ourselves to electronic systems such that this property holds . the fact that the occupation numbers can also have non - integer values between zero and one is one of the most distinct features of interacting systems compared to non - interacting systems which can only have integer occupation numbers typically . therefore the occupation numbers reflect strongly the electronic correlations present in the system under consideration . a system is considered weakly correlating when the occupation numbers differ only slightly from zero or one , in which case the full many - electron wavefunction can well be approximated by a single slater determinant ( non - interacting wavefunction ) . if a system is strongly correlated , the occupation numbers deviate strongly from integer values and multiple determinants are required to obtain a sufficiently accurate approximation to the many - body wavefunction which captures the physics of the system . the ability of the 1rdm occupation numbers to signal strong correlation has encouraged people to develop 1rdm functional theory as an alternative to traditional density function theory ( dft ) to handle strongly correlated systems such as dissociating molecules @xcite , mott insulators @xcite and quantum hall systems @xcite , for which the current approximate density functionals fail miserably . the sum of the occupation numbers equals the number of electrons in the system . therefore , if we order the occupation numbers , @xmath7 , from the highest to the lowest one , their values need to decay to zero sufficiently fast for @xmath8 , i.e. @xmath9 or even become zero after some point @xmath10 . the question whether they actually do become zero or only approach zero for @xmath8 is not only an academic question , but is also of practical interest for methods that try to build an accurate approximation to the wavefunction by making an expansion in terms of slater determinants , e.g. configuration interactions ( ci ) . this question has recently been addressed for the dissociating hydrogen molecule @xcite . one would expect that an optimal set of orbitals exists which leads to the fastest convergence of the expansion of the wavefunction in terms of slater determinants @xcite . one can prove that if all determinants are taken into account ( full ci ) , that the highest occupied nos are the orbitals which give the fastest convergence towards the exact one in the @xmath11-norm @xcite . the nos become even more interesting if the occupation numbers become all zero for @xmath12 sufficiently large , since this would imply that only a finite set of nos would already be sufficient to expand the full many - electron wavefunction . the question if zero occupation numbers exist in coulomb systems is maybe even more important for 1rdm functional theory . basic theorems in 1rdm functional theory @xcite follow similar arguments as the famous hohenberg kohn theorem @xcite of density functional which establishes a one - to - one correspondence between densities , potentials and non - degenerate ground states . the main difference between 1rdm functional theory and density functional theory is that the natural conjugate variable to the 1rdm is a non - local external potential of the form @xmath13 rather than the local potential of density - functional theory . it therefore immediately follows that the energy contribution of the nonlocal external field to the total energy is given by @xmath14 with this expression the hohenberg kohn proof can be followed exactly as in density - functional theory and gilbert @xcite in fact did this to establish that there is a one - to - one correspondence between non - degenerate ground states @xmath3 and their corresponding 1rdm @xmath15 . this is already sufficient to establish 1rdm functional theory , since the ground state energy can be written as a functional of the 1rdm @xmath16 $ ] . in density - functional theory one can further prove that two different ( up to a gauge ) potentials can not have the same non - degenerate ground state . the analogous proof fails in 1rdm theory since there can exist nonlocal potentials @xmath17 with the property that @xmath18 for a given ground state @xmath3 of some hamiltonian @xmath19 . such a potential can therefore always be added to this hamiltonian without affecting the ground state ( it could be that @xmath3 is now an excited state but by multiplying @xmath17 by a small enough number we can ensure that @xmath3 is still the ground state ) . let us now see how eq . can come about . let us first define the annihilation operator @xmath20 which annihilates the no @xmath21 from any many - body quantum state . suppose now that @xmath22 for some of the labels @xmath23 , which means that the orbital @xmath21 does not appear in any slater determinant of a ci expansion of @xmath3 . this implies that @xmath24 such that the corresponding no occupation number vanishes . we can then construct the following one - body potential @xmath25 where @xmath26 is an arbitrary hermitian matrix and where we sum only over the labels for which @xmath27 for the state @xmath3 . it is clear that this potential exactly has the property @xmath28 . in real space this corresponds to a nonlocal spatial potential of the form of eq . where @xmath29 we therefore see that eq . can be satisfied whenever the state @xmath3 has vanishing no occupations . the non - vanishing of the no occupation numbers for electronic ground states is therefore a necessary condition for the existence of a one - to - one mapping between nonlocal potentials and 1rdms . to the best of our knowledge the answer to the question whether the necessary condition is also a sufficient one is unknown . the one - to - one mapping between non - local potentials and 1rdms would be relevant for the foundations of linear response 1rdm functional theory and also its time - dependent extension would greatly benefit from the resulting simplifications . other consequences of vanishing occupation numbers arise in the extended koopmans theorem @xcite . the extended koopmans theorem is an extension to arbitrary wavefunctions of the well known theorem by koopmans that the occupied hartree fock orbital energies provide approximations to the ionization energies @xcite . if the exact wavefunction is used in the extended koopmans procedure , even the exact ionization energies should result , provided the set of partially occupied nos is complete , i.e. none of the occupation numbers vanishes . a less restrictive condition has been derived by pernal and cioslowski @xcite , though in practice it simply implies that none of occupation numbers should vanish . for systems with coulombic interactions the extended koopmans theorem is found to hold to very high numerical accuracy @xcite although this does not prove its validity . we have therefore seen that the possible vanishing of no occupation numbers has important consequences for ci expansions as well as for the validity of fundamental theorems in many - body theory . this then immediately raises the question in which cases the no occupation numbers vanish . if none of the occupation numbers vanishes , then every no is needed in an expansion of the ground state wave function . one general observation that one can make is that infinite expansions are typically required when expanding non - smooth functions in terms of smooth ones . in the case of electronic ground states the coulomb interaction requires the wavefunction to have a cusp at the positions where the electrons come together of the form @xmath30 where @xmath31 . this cusp gives an infinite kinetic energy which exactly compensates the infinity from the coulomb interaction between the electrons @xcite . due to this non - analytic behavior of the wavefunction , a full expansion of the wavefunction in one - electron functions requires in general all functions to be present . hence , one may expect in the particular case of an expansion in nos , none of the nos should have an occupation number equal to zero , since that would imply that the no is not required in the expansion . although this argument sounds very reasonable , it is certainly not a proof that zero occupations do not occur in coulomb systems . though infinitely many occupation numbers are required to be non - zero , it might be that some of them are still zero in some special situations . in the case of the homogeneous electron gas ( heg ) , however , this argument can be turned into a proof . since the nos of the heg are simply plane waves , the occupation numbers are then given by the momentum distribution , @xmath32 . kimball has shown that the momentum distribution is required to decay as @xmath33 due to the inter - electronic cusp condition @xcite , so the occupation numbers never become exactly zero . in the case of the heg we were in the fortunate situation that the nos are plane waves and , so that their occupation numbers are simply given by the momentum distribution . for general systems we are not in such a convenient position , because a straightforward expansion in a finite basis set effectively smoothens the electron - electron cusp and the argument does not apply anymore . for two - electron systems we are in a more fortunate situation , however , since for singlet two - electron systems there is a strong connection between the nos and the wavefunction . the spatial part of the singlet two - electron wavefunction is symmetric and can therefore be diagonalized @xmath34 by calculating the corresponding spin - integrated 1rdm , one readily finds that the eigenfunctions are nos and that the coefficients are related to the occupation numbers as @xmath35 . though we are not in such a good position as the for the heg , this connection is quite useful , since it allows us to connect the behavior of the occupation numbers directly to the analytic properties of the wavefunction , instead of going via the 1rdm in which much of the analytic properties are integrated out . therefore , we will focus our attention in this paper mainly to singlet two - electron systems to demonstrate how the form of the interaction determines the analytic properties of the wavefunction , which in turn dictates the asymptotic decay of the occupation numbers . before we present a general treatment for simple explicitly correlated wavefunctions , we first consider some specific examples for which we can solve the nos and coefficients explicitly or at least prove that none of the nos have a vanishing coefficient ( occupation number ) . let us first consider the simplest wavefunction with a cusp in one dimension ( 1d ) @xmath36 where @xmath37 is a normalization constant and @xmath38 is an arbitrary orbital apart from the fact that it is positive , @xmath39 . to calculate the nos and the coefficients in the spectral expansion of the wavefunction , we need to solve the following eigenvalue equation @xmath40 introducing the following function @xmath41 the eigenvalue equation can be written as @xmath42 now differentiating this equation twice with respect to @xmath43 , we obtain the following differential equation for @xmath44 @xmath45 where we used that @xmath46 . from this equation we see that @xmath47 if @xmath48 , since otherwise we would have @xmath49 which is not an eigenfunction . hence , for @xmath48 we can divide by @xmath50 and write the equation as @xmath51 where @xmath52 in the case that a simple slater function is used for the orbital , @xmath53 , the differential equation can be cast into a bessel s differential equation . the full construction of all the nos and their coefficients is rather technical and has been deferred to appendix [ ap : modelatom ] . the result of the calculation is that none of the occupation numbers is zero and that the occupation numbers behave asymptotically as @xmath54 where @xmath55 is a constant . we will see later that such a power law behavior is typical for wave functions that are at most a finite number of times differentiable ( in our case zero times ) . let us now try to extent the approach of the previous example to a three dimensional case , so that we have a wavefunction in three dimensions ( 3d ) that satisfies the cusp condition of the form @xmath56 where @xmath57 . since the nos are eigenfunctions of the wavefunction , they satisfy the eigenvalue equation @xmath58 now introducing similar function as in the 1d case , @xmath59 , the eigenvalue equation can be written as @xmath60 unfortunately , just taking the laplacian does not work , since @xmath61 . taking the laplacian a second time , however , we obtain the sought after delta function and our equation becomes @xmath62 we can now use the same argument as in the 1d case . since the orbital @xmath63 will in general not vanish on some open set , @xmath64 would imply that @xmath65 for unoccupied nos . since such orbitals are not normalizable , our simple explicitly correlated wavefunction does not have any nos with occupation numbers equal to zero . since the differential equation for the functions @xmath66 is now fourth order and additionally in 3d , it becomes quite hard to obtain the nos explicitly from this equation . in any case , it is exactly the cusp behavior that allows us to conclude that none of the no occupation numbers vanish . although we have found that the exact treatment of the cusp in the simple hylleraas wavefunction prevented the nos to have a zero occupation number , one can wonder if the electron - electron cusp is actually essential to have only non - zero occupation numbers . this is actually not the case as one can show explicitly for a system with harmonic interactions in one dimension . the hamiltonian for such a system can be written as @xmath67 where @xmath68 . the coordinates can be decoupled by making a transformation to center - of - mass coordinates , which gives the following expression for the hamiltonian @xmath69 where @xmath70 , @xmath71 and @xmath72 . since this is the hamiltonian for two independent harmonic oscillators , we can immediately write spatial part of the singlet ground state wavefunction as @xmath73{\frac{\omega\,\tilde{\omega}}{\pi^2}}{\textrm{e}}^{-{\frac{1}{2}}(\omega s^2 + \tilde{\omega}t^2 ) } \notag \\ & = \sqrt[4]{\frac{\omega\,\tilde{\omega}}{\pi^2}}{\textrm{e}}^{-{\frac{1}{2}}\omega x_1 ^ 2}{\textrm{e}}^{-{\frac{1}{2}}\omega x_2 ^ 2 } \;{\textrm{e}}^{-{\frac{1}{2}}(\tilde{\omega}-\omega)t^2}.\end{aligned}\ ] ] the structure of this solution is sufficiently simple to determine the nos and the corresponding wavefunction coefficients explicitly . due to the purely harmonic nature of our system , one might suspect that the nos also have the form of harmonic oscillator solutions . indeed , one can actually calculate the nos to be ( see appendix [ ap : harmnos ] ) @xmath74{\omega\tilde{\omega}}\,x\bigr ) { \textrm{e}}^{-{\frac{1}{2}}\sqrt{\omega\tilde{\omega}}x^2},\end{aligned}\ ] ] where @xmath75 are the hermite polynomials and @xmath76 normalization constants satisfying @xmath77{\omega\tilde{\omega}}}{2^kk!\sqrt{\pi}}.\end{aligned}\ ] ] the corresponding wavefunction coefficients are @xmath78{\omega\tilde{\omega}}}{\sqrt{\omega}+\sqrt{\tilde{\omega}}}\ , ( -1)^k \biggl(\frac{\sqrt{\omega}-\sqrt{\tilde{\omega}}}{\sqrt{\omega}+\sqrt{\tilde{\omega}}}\biggr)^k.\end{aligned}\ ] ] we find that the wavefunction coefficients are alternating and only vanish in the limit @xmath8 . only when @xmath79 ( no interactions ) , all the coefficients become zero , except for the first one @xmath80 . an important conclusion is now that instead of a power law behavior we now have an exponential decay of the form @xmath81 with @xmath55 and @xmath82 constants . we will show below that such a behavior is typical in the limit @xmath83 for infinitely differentiable wave functions . our examples seem to indicate that there is a connection between the differentiability properties of the wave functions and the asymptotic behavior of the no occupations . in the next section we make this connection more precise . the eigenvalue equation for the nos is a fredholm integral equation , where the two - body wavefunction @xmath84 is the kernel and @xmath50 are the eigenvalues . ( often one considers the characteristic values , @xmath85 , instead of the eigenvalues . ) in the theory of fredholm integral equations lower bounds on the decay rate of the eigenvalues have been established on the differentiability and analyticity of the integral kernel @xcite . although only lower bounds have been found , these bounds are a nice illustration of the strong link between the differentiability of the kernel and the decay rate of the kernel . the first result of interest is by hille and tamarkin , who showed that the eigenvalues of operators with analytic kernels decay exponentially . more precisely @xmath86 where @xmath87 are the eigenvalues and the constant @xmath88 is related to the size of the region where the kernel is analytic @xcite . indeed , the kernel of the double harmomium is analytic and the corresponding coefficients decay exponentially . a result for finitely differentiable integral kernels by weyl @xcite is of particular interest for wavefunctions with a cusp . weyl showed that the eigenvalues of a finitely differentiable kernel only need to decay polynomially . more precisely , if the partial derivatives of a symmetric kernel are continuous up to order @xmath89 , the decay rate of its eigenvalues is bounded by @xmath90 where @xmath91 is the dimension of the integration variable . a derivation of weyl s result can be found in appendix [ ap : weyltheorem ] . let us compare weyl s theorem with the results for the simple 1d hylleraas atom ( sec . [ sec : hylleraas1d ] ) . the first derivative of the wavefunction is already discontinuous in this case , so @xmath92 . this gives us a rather modest lower bound on the decay rate of the coefficients , comparing to the actual quadratic decay - rate @xmath93 of the coefficients ( see appendix [ ap : modelatom ] ) . the bound by weyl can be tightened by using more regularity properties of the discontinuous derivatives of the wavefunction @xcite . since the discontinuity is bound , the two - electron wavefunction can be put in @xmath94 @xcite which gives the lower bound @xmath95 on the decay - rate of the coefficients . though weyl s theorem does not put a very stringent constraint on the decay of the occupation numbers , it clearly demonstrates that wavefunctions with an interelectronic cusp will typically have only a polynomial decaying occupation number spectrum , whereas the occupation numbers of a wavefunction without a cusp will decay exponentially . since in a finite basis set the non - analyticity of the cusp can not be fully represented and thereby effectively smoothened , the calculated occupation numbers will always decay exponentially or faster , i.e. too fast compared to the typical polynomial decay . this is actually not surprising , since if only a finite number of orbitals is included in the calculations , all the orbitals in the complement are automatically nos with zero occupation number . additionally , since a finite basis set representation effectively removes the cusp from the wavefunction , the original argument that nos with vanishing occupation numbers do not exist in many - body coulomb systems , does not even apply anymore . the example with the harmonic interaction shows that the cusp actually might not be essential to prevent occupation numbers becoming zero . it seems that any correlation that requires some @xmath96 behavior that can not be expressed in a finite number of simple orbital products will probably imply the absence of unoccupied nos . we can make this idea more precise in the case of a singlet two - electron system which is limited to the form of a simple orbital product times an arbitrary correlation function depending only on @xmath97 , i.e. a wavefunction of the form @xmath98 where @xmath99 . considering the situation that an no , @xmath100 , has a zero occupation number ( so also @xmath101 ) , the eigenvalue equation then simplifies to @xmath102 where @xmath103 . since this condition has the form of a convolution product , we can deconvolute it by taking the fourier transform @xmath104 provided that @xmath105 almost everywhere , it follows that @xmath106 . since @xmath107 almost everywhere , this implies that the no @xmath108 . because this is not a normalizable function , we can conclude that no eigenvalues @xmath101 exist , if @xmath105 almost everywhere . the converse also holds . if @xmath109 on some finite interval , an no with zero occupancy exists by construction . to construct this no , define a @xmath110 which is non - zero on this interval where @xmath111 vanishes . by fourier transforming back , we can construct the corresponding no ( @xmath112 ) . since this will be a function in the null - space of the ( linear ) 1rdm - operator , it can not be expressed linear combination of nos with finite occupancy . let us show this more explicitly . suppose that such a linear combination exists , then we write the new no as @xmath113 since the 1rdm - operator is linear , we can write its action on @xmath114 as @xmath115{\sum_{i\in \{i : n_i \neq 0\}}}{\int\!\!\!\:{\mathrm{d}}{{{\mathbf{r}}}'}\:}\gamma({{\mathbf{r}}},{{\mathbf{r}}}')\phi_i({{\mathbf{r}}}')b_i.\end{aligned}\ ] ] if we now take the inner product of this result with @xmath114 , we find @xmath116 hence , we have a contradiction , so the constructed no with zero occupation number can not be expressed as a linear combination of the nos with a finite occupancy . now let us check if our proof indeed recovers the result that our previous examples do not have vanishing occupation numbers . in the case of the harmonium the correlation function is simply a gaussian . since the fourier transform of a gaussian is simply again a gaussian , the fourier transform is non - zero every where . hence , no unoccupied nos should exist , which is in agreement with our explicit construction . the situation is more complicate in the case of the simple hylleraas wavefunction . the correlation function is not an @xmath11 function anymore , so we can expect distributions to appear in its fourier transform . ] . note that a divergent correlation function is allowed , provided the divergence of the correlation function is compensated by a stronger decay of the orbital @xmath63 , to make the wavefunction normalizable . the fourier transform in 1d becomes @xmath117(k ) = 2\pi\delta(k ) - \frac{2\eta}{k^2}\end{aligned}\ ] ] and in 3d we find @xmath118(k ) = -\frac{4\pi}{k}\left(\frac{2\eta}{k^3 } + \pi\delta'(k)\right).\end{aligned}\ ] ] the calculation of these fourier transforms has been worked out in more detail in appendix [ ap : fouriers ] . since we find in both cases that the fourier transform is non - zero everywhere , we recover the result that no unoccupied nos exists as we have found before . now let us apply the theorem to some other systems . first we will consider a system with inverse harmonic interactions which has been considered before by morrison et al . the inverse harmonic interaction is also sometimes referred to as the calogero interaction @xcite . the full hamiltonian we will consider here is given as @xmath119 the eigenstates of this hamiltonian can be solved exactly by making a transformation to the centre - of - mass coordinates , which decouples the coordinates . the hamiltonian for the centre - of - mass coordinate is simply the hamiltonian of a harmonic oscillator , so is readily solved . the hamiltonian for the relative coordinate is more involved , but can still be solved in terms of confluent hypergeometric functions @xcite . fortunately , the ground state reduces to the following particularly simple form @xmath120 where @xmath121 is the gamma function and @xmath122 in the exponent seems to be missing in ref . @xcite . this error propagates throughout the article . ] . since the ground state is simply the product of an orbital , @xmath123 , times a correlation function , @xmath124 , our theorem can be applied to this case . for the fourier transform of the correlation function we find ( see appendix [ ap : fouriers ] for details ) @xmath125(k ) \\ = -\frac{4\pi}{k } \times \begin{cases } \pi(-1)^{\alpha/2}\delta^{(1+\alpha)}(k ) & \text{even $ \alpha$ } \\ \dfrac{\gamma(2 + \alpha)\sin(\pi\alpha/2)}{k^{2+\alpha } } & \text{otherwise } , \end{cases}\end{gathered}\ ] ] where @xmath126 denotes the @xmath127 order derivative of the delta - function . therefore , we find that zero occupation numbers can only exist for even @xmath128 , which is in agreement with the findings of morisson et al . actually , it is not surprising that there are only a finite number of unoccupied nos for even @xmath128 , since in that case the correlation function becomes exactly separable in @xmath129 and @xmath130 , so the wavefunction can be represented by a finite number of orbitals . for example in the simplest non - trivial case of @xmath131 the wavefunction can be written as @xmath132,\end{gathered}\ ] ] where we used @xmath133 as a compact notation for the various one - particle functions . since only five orbitals are required to represent this wavefunction , only five nos with non - zero occupation number exist . we see that the contribution from the @xmath89 orbitals to the wavefunction is already diagonal , so to obtain the nos , we only need to normalize them @xmath134 and we have similar expressions for @xmath135 and @xmath136 of course . their coefficient in the spectral expansion is readily obtained as @xmath137 . the contribution from the other two orbitals @xmath138 and @xmath139 is not diagonal , so has to be diagonalized . this is readily achieved by the following linear combinations @xmath140 where @xmath141 . the orbitals @xmath142 are constructed such that they are orthonormal , so they are the required nos . their corresponding expansion coefficients can be calculated to be @xmath143 . this example clearly demonstrates that the cusp is actually not essential for the absence of unoccupied nos . the absence of zero occupation numbers is caused by correlation in the full many - body wavefunction , which can not be expanded in a finite series of one - electron functions . the cusp actually causes the exact wavefunction to have such a form , hence there will be an infinite amount of non - zero occupation numbers . in this section we will apply our theorem to hooke s atom . the interaction is now the coulomb interaction , the interaction of interest for electrons . however , the confining potential is still harmonic to allow for the separation of variables by changing the coordinates to the centre - of - mass frame . the hamiltonian is given as @xmath144 the ground state has the following form @xmath145 where @xmath146 is a normalization constant . the function @xmath147 satisfies the following differential equation @xcite for the ground state @xmath148 where @xmath149 with @xmath150 as the contribution to the energy from the relative coordinate ( the total energy is @xmath151 ) . unfortunately , this differential equation does not allow for an explicit solution , though a series solution can be constructed which even has only a finite number of terms for specific ratios @xmath152 . more importantly , by neglecting the second order derivative and the last term , we readily find that the solution has to behave asymptotically for large @xmath153 as @xmath154 so at least its fourier transform exists as a distribution and in particular its laplace transform @xmath155(s ) { \coloneqq}{\int^{\mathrlap{\infty}}_{\mathrlap{0}}{\mathrm{d}}{\rho}\,}{\textrm{e}}^{-s\rho}t(\rho)\end{aligned}\ ] ] exists . the differential equation for @xmath147 can be transformed into the following differential equation for the laplace transform @xmath156 where we used that @xmath157 . the fourier transform of the correlation function can now directly be obtained from @xmath158 as @xmath159(k ) & = \frac{2\pi { \textrm{i}}}{k}\lim_{\sigma \to 0^+}\left(\tilde{t}(\sigma+{\textrm{i}}k ) - \tilde{t}(\sigma-{\textrm{i}}k)\right),\end{aligned}\ ] ] where the limits are important to include possible poles at the origin . since the laplace transform is a solution of a second order differential equation , it has to be an analytic function . this analyticity is carried over to the fourier transform except for some possible irregularities located at the origin ( @xmath160 ) . since an analytic function can only be zero in an open set if it vanishes everywhere , the fourier transform of the correlation function of the hookium atom does not vanish in a finite region and hence , the no coefficients ( occupation numbers ) do not vanish for the hookium atom . -channel . in the weak correlation regime ( large @xmath161 ) only the highest occupied no has a positive coefficient and all the other nos have a negative coefficient . in the strongly correlated regime ( small @xmath161 ) the signs of the no are alternating . in the color version the sign of the corresponding coefficient has been coded as black for positive and red for negative . ] this result is orthogonal to the claim made by cioslowski and pernal @xcite . they have studied the behavior of the wavefunction coefficients of hooke s atom numerically and found that the coefficients become very small and that the sign pattern of the most significant coefficients around these points change . therefore , they concluded that the expansion coefficients have to become zero to change their sign . as a courtesy to the reader , we have repeated their calculations to obtain an accurate expression for the wavefunction and calculated the no coefficients by diagonalizing the wavefunction directly . the results for the occupation numbers in the @xmath23-channel are shown in fig . [ fig : sharmonium ] and are identical to the ones reported in ref . indeed , the most significant nos have a different sign in the large and small @xmath161 limit . however , upon closer inspection of the plot in fig . [ fig : sharmonium ] , one readily sees that coefficients of _ different _ nos actually gain in amplitude when making the transition between the weakly and strongly correlated regime , so there is actually no evidence that the no coefficients do cross zero . the numerical results of cioslowski and pernal are therefore in agreement with our findings . the question of the existence of nos with vanishing occupation numbers in many - body coulomb systems is important for a number of practical applications . for the direct expansion of wavefunctions in finite orbital basis sets ( ci expansions ) it would be beneficial if only a finite number of nos has a finite occupancy . however , the presence of unoccupied nos will cause complications in the formal developments of 1rdm functional theory . for the extended koopmans theorem the existence of vanishing natural occupation numbers could even be catastrophic , since the ionization energies do not necessarily converge to the exact ones when the approximate wavefunction converges to the exact many - body state . the divergence of the coulomb interaction between the electrons requires the wavefunction to have a cusp at the coalescence points of the electrons . this non - analytic behavior can only be represented by including an infinite amount of orbitals ( nos ) in the expansion of the wavefunction , so there are an infinite amount of nos with a non - zero occupation number . however , this argument does not provide a proof that natural occupation numbers equal to zero do not exist , i.e. that the nos with a non - zero occupation numbers form a complete set . however , we have been able to show that wavefunctions of the form @xmath162 do not have any vanishing occupation number , if and only if the fourier transform of the correlation function , @xmath163 , does not vanish on an open set . the fourier transform of the correlation function only seems to disappear if the wavefunction is separable , i.e. representable in a finite basis . applying our theorem to the harmonium atom , we have shown that the occupation numbers do not vanish in contradiction with earlier assertions by cioslowski and pernal based on numerical calculations @xcite . however , a more careful inspection of their results showed that only their interpretation was incorrect and that their results actually agree with our proof up to numerical accuracy . further , we have demonstrated that a discontinuity in the wavefunction is not required for the absence of unoccupied nos . even the perfectly smooth ground state of the double harmonium has no vanishing occupation numbers . more essential is the non - separability of the wavefunction , which is caused by the discontinuity of the cusp . the discontinuity of cusp does have an effect on the decay - rate of the occupation numbers . it causes the occupation numbers to decay merely polynomially compared to an exponential decay of the occupation numbers for wavefunctions without a cusp . the authors acknowledge the academy of finland for research funding under grant no . kjhg also gratefully acknowledges a veni grant by the netherlands foundation for research nwo ( 722.012.013 ) . though the proof of is `` '' according to weyl @xcite , it might be worthwhile to expose its derivation . weyl starts with an alternative derivation of a theorem by e. schmidt @xcite to prove that @xmath164 where @xmath87 are the eigenvalues of the integral kernel @xmath165 ordered in descending order and @xmath166 is a hermitian finite rank operator @xmath167 with @xmath168 and @xmath169 . this theorem is easily understood by using the spectral representation of the integral kernel . the minimum value of the integral is achieved by using the largest eigenvalues of @xmath165 at the diagonal , @xmath170 and the corresponding eigenfunctions for the functions @xmath171 . this choice exactly eliminates the largest eigenvalues of @xmath165 and only the smaller @xmath172 eigenvalues will contribute to the integral . any other choice for @xmath166 will give a larger value of the integral . for more rigor , consider the following proof . to cover the infinite dimensional case we use the rayleigh quotient to define the eigenvalues . the first eigenvalue ( and largest in magnitude ) is defined as @xmath173 where @xmath174 is the usual @xmath11 norm and the operator @xmath175 is defined be the action of the integral kernel @xmath165 on a function @xmath176 as @xmath177 the function that achieves this maximum , @xmath178 , is the corresponding eigenfunction . the other eigenvalues are defined ( found ) by searching over a subspace where the previously found eigenfunctions have been projected out @xmath179 and the function that achieves this maximum , @xmath180 , is the corresponding eigenfunction . using this definition we readily find for the eigenvalue of the the sum of two linear operator @xmath181 and @xmath182 @xcite @xmath183 since the rank of @xmath184 is only @xmath185 , it has only @xmath185 non - zero eigenvalues at maximum . therefore , we find that @xmath186 applying this inequality for all eigenvalues of @xmath37 , we readily recover schmidt s inequality . schmidt s inequality gives a lower bound for the integral . an upper bound can be obtained from taylor s theorem . this procedure is probably most clearly explained at the end of ref . @xcite . to simplify the analysis , we assume without loss of generality that we integrate over a finite block with sides of length @xmath187 , so that we can divide it in @xmath188 smaller blocks , where @xmath91 is the dimension of our integration variable . in each of these regions we can make a taylor expansion around its centre @xmath189 @xmath190 where @xmath191 denotes a multi - index @xmath192 and the remainder satisfies @xmath193 by choosing all possible powers of @xmath194 up to order @xmath89 as basis functions for @xmath184 , we can create @xmath195 linearly independent functions per block , so @xmath196 functions @xmath197 in total . using these basis functions , we can set the kernel @xmath184 with @xmath198 equal to the taylor expansions of @xmath37 in these blocks . the error of the taylor expansion can now be approximated as @xmath199 where @xmath200 as @xmath201 . now combining this inequality from the taylor expansion of the kernel with schmidt s inequality we have @xmath202 so for @xmath203 we need @xmath204 now setting @xmath205 and cleaning up the limit , weyl s inequality for the asymptotic behavior of the eigenvalues readily follows . tighter bounds on the asymptotic decay of the eigenvalues can be found by using additional properties of the integral kernel @xcite . first we need to determine the boundary condition to be imposed on the solutions , which will give the required quantization of the expansion coefficients @xmath50 . to find the boundary conditions , we first rewrite the integral equation as @xmath206 now considering the limit @xmath207 , the last integral vanishes and we find that the solutions have to behave asymptotically as @xmath208 since the orbital should decay exponentially for the wavefunction to be normalizable . because we are dealing with a symmetric orbital , the solutions can be separated in gerade and ungerade functions . for the even solutions only the first contribution survives , so we have the following boundary condition for @xmath207 for the gerade solutions @xmath209 likewise , for the ungerade solutions only the last term survives and we find @xmath210 note that this analysis of the boundary conditions is completely general for symmetric @xmath211 . now we have found the boundary conditions for the various solutions , we turn our attention to the solution of the differential equation @xmath212 where the normalization constant is calculated to be @xmath213 since we are looking for symmetry adapted solutions , we only need to solve this differential equation for @xmath214 . if we define ( for @xmath214 ) @xmath215 then for a function @xmath216 we have @xmath217 and the differential equation in term of @xmath218 becomes @xmath219 this is the bessel differential equation for the zeroth order bessel functions and the general solution to this equation is @xmath220 with @xmath221 and @xmath222 constants and @xmath223 and @xmath224 bessel functions of the first and second kind respectively and @xmath225 and @xmath226 their modified counterparts . now going back to the original function @xmath227 we find for @xmath228 @xmath229 and in the case of @xmath230 we find @xmath231 now we construct the even and odd solutions by imposing the corresponding boundary conditions . first we impose the boundary conditions at @xmath232 . the odd solutions need to vanish at the origin , so we find @xmath233 , \\ \varphi^+_{u}(x ) & = c\bigl[\operatorname{\mathrm{k}}_0\bigl(\tilde{\lambda}\bigr)\operatorname{\mathrm{i}}_0\bigl(s(x)\bigr ) - \operatorname{\mathrm{i}}_0\bigl(\tilde{\lambda}\bigr)\operatorname{\mathrm{k}}_0\bigl(s(x)\bigr)\bigr],\end{aligned}\ ] ] where @xmath234 and @xmath55 is a normalization constant . for the even solutions the first order derivative at @xmath232 needs to vanish , so for the even solutions we find @xmath235 , \\ \varphi^+_{g}(x ) & = c\bigl[\operatorname{\mathrm{k}}_1\bigl(\tilde{\lambda}\bigr)\operatorname{\mathrm{i}}_0\bigl(s(x)\bigr ) + \operatorname{\mathrm{i}}_1\bigl(\tilde{\lambda}\bigr)\operatorname{\mathrm{k}}_0\bigl(s(x)\bigr)\bigr].\end{aligned}\ ] ] to obtain the proper quantization of the eigenvalue @xmath236 , we need to impose the proper boundary conditions for @xmath207 , i.e. @xmath237 . from the asymptotic behavior of the bessel functions for small @xmath23 , we find that our ungerade solutions behave asymptotically for @xmath207 as @xmath238 , \\ \varphi^+_{u}(x ) & \sim c\left[\operatorname{\mathrm{k}}_0\bigl(\tilde{\lambda}\bigr ) + \operatorname{\mathrm{i}}_0\bigl(\tilde{\lambda}\bigr ) \bigl(\gamma + \ln\bigl(\tilde{\lambda}/2\bigr ) - zx\bigr)\right],\end{aligned}\ ] ] where @xmath15 is the euler mascheroni constant . we found before , however , that the odd solutions do not have a linear term , so we must have @xmath239 since @xmath240 does not have any zero , only solutions for @xmath228 exist , which are related to the zeros of the the zeroth order bessel function of the first kind , @xmath241 , as @xmath242 using the coefficients of the odd nos are readily determined to be @xmath243 the zero s of the zeroth order bessel function behave asymptotically as @xmath244 so asymptotically , the coefficients decay quadratically , @xmath245 . to construct the corresponding nos , we use that @xmath246 , so @xmath247 the normalization constant of the ungerade nos is readily calculated by using that the @xmath127 order bessel functions satisfy @xmath248 where @xmath249 is the @xmath250 zero of @xmath251 . making the substitution @xmath252 and taking @xmath253 , we find that the normalization constant for the odd nos is given as @xmath254 now we will construct the even solutions . first we consider the large @xmath255 behavior of the even solutions . from the asymptotic behavior of the bessel functions for small @xmath23 we find that for @xmath207 @xmath256 , \\ \varphi^+_{g}(x ) & \sim c\bigl[\operatorname{\mathrm{k}}_1\bigl(\tilde{\lambda}\bigr ) - \operatorname{\mathrm{i}}_1\bigl(\tilde{\lambda}\bigr ) \bigl(\gamma + \ln\bigl(\tilde{\lambda}/2\bigr ) - zx\bigr)\bigr].\end{aligned}\ ] ] since we know that the the even solutions have to behave for @xmath207 as given by the asymptotic relation in , we must have that the the ratio between the linear and constant term must be equal to @xmath257 , i.e. @xmath258 , where @xmath259 , \\ f^-(y ) & { \coloneqq}\operatorname{\mathrm{j}}_1(y ) - \frac{\eta}{z}\left[\frac{\pi}{2}\operatorname{\mathrm{y}}_1(y ) - \operatorname{\mathrm{j}}_1(y)\bigl(\gamma + \ln(y/2)\bigr)\right].\end{aligned}\ ] ] note that in @xmath260 we could divide by @xmath261 , since this is an exponentially growing function and has no zeros apart from @xmath262 . for four different values of @xmath263 . ] let us now first consider the positive coefficients , @xmath230 . the function @xmath264 is constructed out of modified bessel functions of the first and second kind , which are monotonically decreasing and increasing functions respectively . since the logarithm is a monotonically increasing function , the function @xmath264 is also a monotonically increasing function , so @xmath264 can have only one zero at most . because the ratio of the modified bessel functions @xmath265 diverges at @xmath266 as @xmath267 , we have @xmath268 . further , due to the logarithm @xmath264 diverges logarithmically for @xmath269 , so @xmath264 has always exactly one zero ( see fig . [ fig : fplus ] ) . hence we find that there is exactly one no with a positive coefficient of gerade symmetry . the corresponding no , @xmath270 , is readily constructed as @xmath271{\textrm{e}}^{-z{\lvertx\rvert}},\end{gathered}\ ] ] where @xmath272 is a normalization constant and @xmath273 is the zero of the function @xmath264 . the corresponding attains the simple expression @xmath274 the negative solutions , @xmath228 , can be obtained in a similar manner . the only difference is that the function @xmath275 has an infinite amount of zeros , @xmath276 , due to the infinite amount of oscillations of the usual bessel functions . following the same steps as for the positive orbital , we find that the gerade nos with negative coefficients are @xmath277{\textrm{e}}^{-z{\lvertx\rvert}}\end{gathered}\ ] ] where @xmath278 are normalization constants and @xmath276 are the zeros of the function @xmath275 defined . the corresponding ( negative ) coefficients are @xmath279 for large @xmath12 , @xmath276 is a large number , so asymptotically the condition @xmath280 becomes @xmath281 for very large @xmath282 the logarithm on the right - hand side will diverge and the tangent on the left - hand side is only large when @xmath283 for an integer @xmath185 . so we find that the the zeros behave asymptotically as @xmath284 so the negative coefficients of the even nos decay quadratically , @xmath285 , exactly as the coefficients of the odd nos . in this appendix we will show how to find the nos for the ground state of a system with only harmonic interactions ( sec . [ sec : harmint ] ) . the nos satisfy the following integral equation @xmath286 where the integral kernel is simply the wavefunction which is for the current purpose most conveniently written as @xmath287{\frac{\omega\tilde{\omega}}{\pi^2}}{\textrm{e}}^{-\alpha(x^2+y^2 ) - 2\beta x y},\end{aligned}\ ] ] where @xmath288 and @xmath289 . since all the interactions are harmonic , we expect that a gaussian , @xmath290 , might be a solution . working out the integral gives @xmath291{\frac{\omega\tilde{\omega}}{\pi^2 } } \sqrt{\frac{\pi}{\alpha+\nu}}{\textrm{e}}^{-\left(\alpha - \frac{\beta^2}{\alpha + \nu}\right)x^2},\end{aligned}\ ] ] so we find that a gaussian is indeed an eigenfunction if the exponent is set to @xmath292 the corresponding eigenvalue can be worked out to be @xmath293{\omega\tilde{\omega}}}{\bigl(\sqrt{\omega } + \sqrt{\tilde{\omega}}\bigr)}.\end{aligned}\ ] ] the other solutions will now be of the form @xmath294 , where @xmath295 is a polynomial of order @xmath12 . from the eigenvalue equation , it follows that these polynomials satisfy @xmath296{\frac{\omega\tilde{\omega}}{\pi^2}}{\int\!\!\!\:{\mathrm{d}}{y}\:}p_k(y){\textrm{e}}^{-\frac{1}{4}\left((\sqrt{\omega}+\sqrt{\tilde{\omega}})y + ( \sqrt{\omega}-\sqrt{\tilde{\omega}})x\right)^2 } = c_kp_k(x).\end{aligned}\ ] ] now we make the following coordinate transformation @xmath297 and @xmath298 , which simplifies the equation for the polynomials to @xmath299 where @xmath300{\frac{\pi^2}{\omega\tilde{\omega}}}c_k , \\ q_k(x ) & { \coloneqq}p_k\biggl(\frac{2x}{\sqrt{\omega}+\sqrt{\tilde{\omega}}}\biggr).\end{aligned}\ ] ] solving the equation for the polynomials of the first four orders gives @xmath301 we see that the eigenvalues are simply related by a factor @xmath302 , so we should be able to recover this relation . taking the derivative @xmath185 times from the integral equation for @xmath303 , we find @xmath304 where @xmath305 denotes the highest order coefficients of the polynomial @xmath303 . hence we find that the eigenvalues are simply @xmath306 , which can be worked out to give . due to the integration gaussian as integration weight , we expect that the polynomials are related to the hermite polynomials . using the following definition for the hermite polynomials , @xmath307 it follows from our results for the polynomials @xmath308 @xmath309 that they should be related to the hermite polynomials as @xmath310 . since the hermite polynomials satisfy the recursion relation @xmath311 the polynomials @xmath312 should satisfy @xmath313 inserting the recursion relation in the integral equation for @xmath314 , we find that @xmath315 is indeed also a solution if @xmath312 is a solution , with an eigenvalue @xmath316 . hence , we have found all nos in terms of the hermite polynomials . working out them out in the original quantities , one recovers the expression for the nos for the harmonic interaction . in this appendix we show in more detail how the fourier transforms of the correlation functions of the various model systems were calculated . let us first consider the correlation functions of the 1d systems . in the case of the double - harmonium the correlation function is a gaussian , so the fourier transform should be again a gaussian . indeed , when we work it out , we find @xmath317(k ) & = { \int^{\mathrlap{\infty}}_{\mathrlap{-\infty}}{\mathrm{d}}{x}\,}{\textrm{e}}^{{\textrm{i}}k x}{\textrm{e}}^{-{\frac{1}{2}}(\tilde{\omega } - \omega)x^2 } \\ & = { \textrm{e}}^{-{\frac{1}{2}}\frac{k^2}{\tilde{\omega}-\omega}}{\int^{\mathrlap{\infty}}_{\mathrlap{-\infty}}{\mathrm{d}}{x}\ , } { \textrm{e}}^{-{\frac{1}{2}}(\tilde{\omega } - \omega)\left(x - \frac{{\textrm{i}}k}{\tilde{\omega}-\omega}\right)^2 } \\ & = { \textrm{e}}^{-{\frac{1}{2}}\frac{k^2}{\tilde{\omega}-\omega}}{\int^{\mathrlap{\infty}}_{\mathrlap{-\infty}}{\mathrm{d}}{y}\,}{\textrm{e}}^{-{\frac{1}{2}}(\tilde{\omega } - \omega)y^2 } \\ & = \sqrt{\frac{2\pi}{\tilde{\omega}-\omega}}{\textrm{e}}^{-{\frac{1}{2}}\frac{k^2}{\tilde{\omega}-\omega}},\end{aligned}\ ] ] where we used that the integration contour could be shifted through the complex plane , since there are no poles . for the fourier transform of the 1d hylleraas wavefunction we have @xmath318(k ) = { \int^{\mathrlap{\infty}}_{\mathrlap{-\infty}}{\mathrm{d}}{x}\,}{\textrm{e}}^{{\textrm{i}}k x}\bigl(1+\eta{\lvertx\rvert}\bigr ) \\ = 2\pi\delta(x ) + \eta\lim_{\alpha \to 0^+}{\int^{\mathrlap{\infty}}_{\mathrlap{0}}{\mathrm{d}}{x}\,}x\bigl({\textrm{e}}^{{\textrm{i}}k x } + { \textrm{e}}^{-{\textrm{i}}k x}\bigr){\textrm{e}}^{-\alpha x},\end{gathered}\ ] ] where we introduced the @xmath128-limit to make the integral convergent . this integral is readily worked out by differentiating under the integral sign @xmath319 now taking the limit @xmath320 , we find for the full fourier transform as in . the other systems are three dimensional and have in common that their correlation function only depends on the length , @xmath96 , so we can already do the integration over the angles @xmath321(k ) & = { \int\!\!\!\:{\mathrm{d}}{{{\mathbf{r}}}}\:}{\textrm{e}}^{{\textrm{i}}{{\mathbf{k}}}\cdot{{\mathbf{r}}}}f(r ) \\ & = 2\pi{\int^{\mathrlap{\infty}}_{\mathrlap{0}}{\mathrm{d}}{r}\,}r^2f(r){\int^{\mathrlap{1}}_{\mathrlap{-1}}{\mathrm{d}}{s}\,}{\textrm{e}}^{{\textrm{i}}k r s } \\ & = \frac{4\pi}{k}{\int^{\mathrlap{\infty}}_{\mathrlap{0}}{\mathrm{d}}{r}\,}r \sin(kr)\ , f(r).\end{aligned}\ ] ] let us first consider the fourier transform of the correlation function of the 3d hylleraas wavefunction @xmath322(k ) & = \frac{4\pi}{k}{\int^{\mathrlap{\infty}}_{\mathrlap{0}}{\mathrm{d}}{r}\,}r \sin(kr)\,\bigl(1 + \eta r\bigr).\end{aligned}\ ] ] for the first integral we have @xmath323 for the second integral we need to introduce the convergence factor again , so we evaluate @xmath324 using this result in the limit @xmath320 together with , we find the full fourier transform of the 3d hylleraas correlation as in . finally let us consider the fourier transform for the correlation function of the system with inverse harmonic interactions . for general @xmath325 we have @xmath326(k ) & = \frac{4\pi}{k}\lim_{\eta\to0^+}{\int^{\mathrlap{\infty}}_{\mathrlap{0}}{\mathrm{d}}{r}\,}r^{1 + \alpha } \sin(kr){\textrm{e}}^{-\eta r } \\ & = \frac{2\pi}{{\textrm{i}}k}\lim_{\eta\to0^+}{\int^{\mathrlap{\infty}}_{\mathrlap{0}}{\mathrm{d}}{r}\,}r^{1 + \alpha } \bigl({\textrm{e}}^{-(\eta - { \textrm{i}}k)r } - { \textrm{e}}^{-(\eta + { \textrm{i}}k ) r}\bigr ) \\ & = \frac{2\pi}{{\textrm{i}}k}\lim_{\eta\to0^+ } \biggl(\frac{1}{(\eta - { \textrm{i}}k)^{2 + \alpha}}{\int^{\mathrlap{(\eta - { \textrm{i}}k)\infty}}_{\mathrlap{0}}{\mathrm{d}}{t}\,}\quad t^{\alpha + 1}{\textrm{e}}^{-t } - { } \\ * & { \hphantom{{}={}}}\hphantom{\frac{2\pi}{{\textrm{i}}k}\lim_{\eta\to0^+}\biggl ( } \frac{1}{(\eta + { \textrm{i}}k)^{2 + \alpha}}{\int^{\mathrlap{(\eta + { \textrm{i}}k)\infty}}_{\mathrlap{0}}{\mathrm{d}}{t}\,}\quad t^{\alpha + 1}{\textrm{e}}^{-t}\biggr ) \\ & = \frac{2\pi}{{\textrm{i}}k^{3+\alpha}}\left({\textrm{e}}^{(2+\alpha)\pi{\textrm{i}}/2 } - { \textrm{e}}^{-(2 + \alpha)\pi{\textrm{i}}/2}\right){\int^{\mathrlap{\infty}}_{\mathrlap{0}}{\mathrm{d}}{t}\ , } t^{\alpha + 1}{\textrm{e}}^{-t } \\ & = \frac{4\pi}{k^{3+\alpha}}\sin\left(\frac{2+\alpha}{2}\pi\right)\gamma(2+\alpha ) \\ & = -\frac{4\pi}{k^{3+\alpha}}\sin(\pi\alpha/2)\gamma(2+\alpha),\end{aligned}\ ] ] where we used that the integration interval could be deformed without difficulty , since the integrant does not have any poles , and that the definition of the gamma function @xmath327 we find that the fourier transform diverges for @xmath328 . in the case of even @xmath128 , however , it is not clear what happens , since we obtain a division of zero by zero . to asses what the real answer should be , we have to calculate these cases separately . for even @xmath128 , we can work out the fourier transform by taking successive derivatives under the integral @xmath326(k ) & = \frac{4\pi}{k}{\int^{\mathrlap{\infty}}_{\mathrlap{0}}{\mathrm{d}}{r}\,}r^{1 + \alpha } \sin(kr ) \\ & = \frac{4\pi}{k}\bigl(-{\partial}^2_k\bigr)^{\alpha/2}{\int^{\mathrlap{\infty}}_{\mathrlap{0}}{\mathrm{d}}{r}\,}r \sin(kr ) \\ & = \frac{4\pi^2}{k}(-1)^{\alpha/2}\delta^{(1+\alpha)}(k)\end{aligned}\ ] ] where @xmath126 denotes the @xmath127 order derivative of the delta - function and we used the previous result for the 3d fourier transform of a constant function . combining these results , we find for general @xmath325 the fourier transform as reported . note that this result for general @xmath329(k)$ ] can also be used to obtain the fourier transform of the correlation function of the 3d hylleraas wavefunction and give the same result . here we will consider a series solution for the differential equation of the laplace transform of the correlation function of the hookium atom . to construct the series , first observe that a physical solution requires , @xmath330 , otherwise the inverse laplace transform does not exist @xcite . hence , the solution should be expanded in powers of @xmath331 rather than in powers of @xmath23 . using the frobenius trick , we find that the solution for @xmath158 can be expanded as @xmath332 where the coefficients @xmath333 satisfy the following recursion relation @xmath334 the second boundary condition is effectively given by the condition that the full wavefunction is normalizable @xcite . this additional condition serves as a quantization condition for the internal energy contribution , @xmath150 . in general , the power series will not terminate . however , there are special ratios @xmath152 at which the series terminates after a finite amount of terms . since the series truncates after a finite amount of terms at these ratios , the wavefunction is automatically normalizable @xcite . the series solution for @xmath158 could also have been obtained by laplace transforming the series solution for @xmath147 by taut @xcite term - by - term using that @xmath335(s ) = { \int^{\mathrlap{\infty}}_{\mathrlap{0}}{\mathrm{d}}{r}\,}r^{m+1 } = \frac{(m+1)!}{s^{m+2}}.\end{aligned}\ ] ] similar to the series solution of @xmath147 @xcite , the series solution of @xmath158 has at least two terms due to the cusp condition . as a check consider the solution for @xmath336 , which has only two terms @xmath337 using , we can construct the fourier transform of the correlation function , @xmath176 . from the first term we get up to the factor @xmath338 @xmath339 where we used that the delta function @xmath340 the second term is easier , since we can drop the limit immediately to give @xmath341 combing these results , we find @xmath342(k),\end{aligned}\ ] ] where the identification with the fourier transform of @xmath343 was readily made by comparing with the fourier transform of the correlation function of the 3d hylleraas wavefunction . 40ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1063/1.4801431 [ * * , ( ) ] _ _ , @noop ph.d . thesis , , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1063/1.430253 [ * * , ( ) ] link:\doibase 10.1063/1.430254 [ * * , ( ) ] link:\doibase 10.1063/1.430509 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1063/1.1336543 [ * * , ( ) ] link:\doibase 10.1063/1.464028 [ * * , ( ) ] link:\doibase 10.1016/s0009 - 2614(98)00302 - 9 [ * * , ( ) ] link:\doibase 10.1103/physrev.94.1600 [ * * , ( ) ] link:\doibase 10.1002/cpa.3160100201 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1088/0305 - 4470/8/9/021 [ * * , ( ) ] http://gdz.sub.uni-goettingen.de/dms/load/img/?iddoc=63048 [ * * , ( ) ] link:\doibase 10.1007/bf02403043 [ * * , ( ) ] _ _ , @noop master s thesis , , ( ) @noop * * , ( ) link:\doibase 10.1063/1.1664820 [ * * , ( ) ] link:\doibase 10.1063/1.1664821 [ * * , ( ) ] link:\doibase 10.1063/1.1665604 [ * * , ( ) ] `` , '' ( , , ) pp . , ed . link:\doibase 10.1103/physrev.128.2687 [ * * , ( ) ] link:\doibase 10.1103/physreva.48.3561 [ * * , ( ) ] link:\doibase 10.1063/1.1318767 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop _ _ , ed . ( , , )

a list of 4,131 dwarfs , subgiants , and giants located , or suspected to be located , within 33 parsecs of the sun is presented . all the stars are drawn from the new lspm - north catalog of 61,976 stars with annual proper motions larger than yr . trigonometric parallax measurements are found in the literature for 1,676 of the stars in the sample ; photometric and spectroscopic distance moduli are found for another 783 objects . the remaining 1,672 objects are reported here as nearby star candidates for the first time . photometric distance moduli are calculated for the new stars based on the ( m,v - j ) relationship , calibrated with the subsample of stars which have trigonometric parallaxes . the list of new candidates includes 539 stars which are suspected to be within 25 parsecs of the sun , including 63 stars estimated to be within only 15 parsecs . the current completeness of the census of nearby stars in the northern sky is discussed in light of the new candidates presented here . it is estimated that ( ) of nuclear burning stars within 33 parsecs ( 25 parsecs ) of the sun remain to be located . the missing systems are expected to have proper motions below the yr limit of the lspm catalog .

astro-ph0506152

proper motion catalogs have historically been the primary sources for the identification of stellar systems in the vicinity of the sun . while star distances are ultimately determined from measurements of annual parallactic motions , these require a substantial investment of time and effort . parallax programs are thus most successful when they rely on input lists of objects already suspected to be nearby . since a large proper motion is an indicator of proximity , high proper motion ( hpm ) stars have always been prime targets for parallax programs . it is a historical fact that the vast majority of the stars now known to be in the `` solar neighborhood '' ( @xmath5 pc ) , and recorded in compendiums such as the _ third catalog of nearby stars _ @xcite hereafter the cns3 , or the nasa _ nstars database _ , have first been identified as hpm stars . perhaps the most important feature of the nearby star census is the fact that it is dominated by low - luminosity red dwarfs , with absolute magnitudes @xmath6 . the vast majority of the `` solar neighborhood '' red dwarfs are therefore not listed in the hipparcos catalog @xcite which has a limiting magnitude @xmath7 . on the other hand , most nearby red dwarfs are expected to be detected on the photographic plates of the large schmidt telescope surveys ( e.g. poss - i , poss - ii , ses , aao ) which typically reach @xmath8 . because nearby red dwarfs have optical colors similar to those of distant giants or dust - reddened stars , they are most easily and reliably identified by their large proper motions . in fact , the main strength of proper motion surveys is that they can locate nearby stars independently of their physical appearance ( color , luminosity ) . one caveat is that a high proper motion is neither a necessary nor a sufficient condition for a star to be nearby . nearby stars can have small proper motions if their motion vector is pointing toward , or away from the sun @xcite . conversely , distant stars can have large proper motions if their transverse motion relative to the sun is also large , such as is typically the case for stars on galactic halo orbits . hence the dilemma : the completeness of a nearby star census based on proper motion surveys increases as the proper motion limit is set to smaller values , but decreasing the proper motion limit also greatly increases the number of distant stars that make it into the survey , which can be viewed as `` contaminants '' for parallax programs aimed at finding the nearest stars . the earlier proper motion catalogs of @xcite and @xcite had relatively high proper motion limits ( @xmath9 yr@xmath1 - 0.5@xmath10 yr@xmath1 ) and together contained a little over 2500 stars , a manageable size for parallax programs . however , later catalogs became increasingly larger , bringing the number of known high proper motion stars to sizes difficult to manage . the catalogs by @xcite and @xcite contain 11,000 + stars with @xmath11 yr@xmath1 , while the famous `` nltt catalog '' of @xcite , the main reference in the field for 25 years , contains over 58,000 stars with @xmath12 yr@xmath1 , down to a magnitude @xmath13 . these large catalogs provide several times more targets than can be handled by existing ground - based parallax programs @xcite . to illustrate this , one need only consider the yale catalog of trigonometric parallaxes @xcite , which compiles all published ground - based parallax measurements prior to 1995 , and which contains data for only 1501 stars fainter than @xmath14 . given these limitations , most efforts at finding nearby stars have focused on a subsample of the nltt , the `` lhs catalog '' @xcite , which basically lists the 4470 nltt stars with the largest proper motions ( @xmath15 yr@xmath1 ) . it is however possible to mine large proper motion catalogs for nearby stars , by using secondary distance estimators such as photometric or spectroscopic distance moduli . photometry and spectroscopy are more readily obtained for large samples of objects than parallax measurements , although they provide much less reliable distance estimates . but even approximate distance moduli can be used to trim a large proper motion sample down to a more manageable list of probable nearby stars . luyten himself devoted substantial efforts to obtaining photographic magnitudes and colors for all the stars in his catalogs , though these were sometimes only approximate . while his @xmath16,@xmath17 photographic magnitude system was used to identify some nearby objects @xcite , its usefulness in estimating distance moduli for the ubiquitous red dwarfs was found to be very limited in practice @xcite . more photometric / spectroscopic data was needed , and the follow - up vri photometric survey of @xmath18 nltt stars with @xmath19 yr@xmath1 @xcite proved successful in identifying 295 new solar neighborhood candidates . a major advance came with the availability of accurate infra - red jhk@xmath20 photometry from the 2 micron all - sky survey ( 2mass ) . in a series of papers ( `` meeting the cool neighbors . i .- ix . '' ) , i. n. reid and colleagues used data from the 2mass _ second incremental release _ to systematically mine the nltt for nearby stars . in the first paper of their series , a combination of luyten s estimated photographic red magnitude ( @xmath17 ) with accurate @xmath21 magnitudes from the _ 2mass second incremental release _ was succesfully used to identify a subsample of candidate red dwarfs within 20 pc of the sun @xcite . distance moduli were calculated based on the 2mass magnitudes and follow - up bvri photometry @xcite , and spectroscopy @xcite of the nltt candidates . unfortunately , positions quoted in the nltt catalog often contain large errors @xcite ; this proved a major impediment , as 2mass counterparts of nltt stars sometimes could not be found , or were mismatched . these problems were mitigated by the use of the _ revised nltt _ ( or rnltt ) catalog of @xcite , which provide re - estimates of nltt positions , and through the use of additional selection criteria ( e.g. color - color cuts ) to identify and reject mismatches @xcite . with these successes , the nearby star census appears now to be only limited by the nltt catalog itself . apart from the issues of the accuracy of its photometry and astrometry , the nltt has long been known to be significantly incomplete in some parts of the sky @xcite . the nearby star census would benefit from a more complete proper motion survey , but also from an expansion to fainter magnitudes , and especially from an expansion to smaller proper motions , the nltt itself being limited to stars with @xmath22 yr@xmath1 . this is why the new lspm - north catalog , as a replacement of the old nltt , promises to improve significantly on nearby stars surveys . the lspm - north catalog @xcite is the product of a massive data mining of the digitized sky surveys ( dss ) with a specialized software ( superblink ) that uses image subtraction algorithms to identify moving and variable objects . the lspm - north lists 61,976 stars with proper motions @xmath23 down to v=21.0 , and is estimated to be @xmath24 complete down to @xmath25 . besides being marginally deeper , and significantly more complete than the nltt , the lspm is also much more accurate , with positions within @xmath26 at the 2000.0 epoch , and optical photographic magnitudes to @xmath27mags obtained from the usno - b1.0 catalog of @xcite . the lspm - north catalog also provides counterparts from the 2mass _ all sky point source catalog _ @xcite when they exist . by being deeper , more accurate , and more complete , the lspm - north catalog supersedes the nltt in all respects ( though only at northern declinations for now ) . the use of the 2mass _ all sky point source catalog _ also offers the opportunity to expand significantly upon the reid _ analysis , which was restricted to the fraction of the sky covered by the 2mass _ second incremental release_. the first paper of this series presents our initial efforts at identifying nearby stars in the lspm - north catalog , specifically the identification of new candidate main - sequence dwarfs within 33pc of the sun . data are found in the literature for all lspm stars that have published trigonometric parallaxes or photometric / spectroscopic distance moduli . a calibration of the @xmath28 $ ] relationship in the @xmath29 system of the lspm - north catalog provides the identification of 1672 new candidate nearby stars with distances estimated to be within 33pc of the sun . identification of nearby white dwarfs from the lspm - north catalog will be detailed in the second paper of this series , while a subsequent paper will address the problem of identifying nearby metal - pool subdwarfs ( lpine , _ in preparation _ ) . this paper is organized as follows : 2 describes the calibration of the @xmath30 color magnitude relationship for main sequence dwarfs , which is used to estimate photometric distance moduli of lspm stars . 3 gives complete lists of nearby dwarfs listed in the lspm catalog for which there exists published trigonometric parallaxes or photometric / spectroscopic distance moduli , and presents the selection of new candidate stars within 33pc . 4 analyzes the current completeness level of the nearby star census in terms of limiting magnitude and proper motion selection . the main results are summarized in the conclusion ( 5 ) . photometric distance moduli ( m - m)@xmath31 for all main sequence lspm - north stars are estimated based on a color - magnitude relation in the ( v , v - j ) color - magnitude system used in the lspm - north catalog . an absolute magnitude - color calibration @xmath32 is calibrated with a subsample of 3,104 lspm stars for which there exist reliable trigonometric parallax ( @xmath33 ) measurements . most calibration stars are selected from the hipparcos and tycho catalog @xcite , and from the yale catalog of trigonometric parallaxes @xcite hereafter ypc . calibration stars are selected based on the following criteria : [ 1 ] the star has a measured @xmath33 with an estimated accuracy better than 10% , [ 2 ] the star has @xmath34 , placing it within 40pc of the sun . [ 3 ] the star does not have a companion within @xmath35 . the first constraint ensures that absolute magnitudes can be calculated from apparent magnitudes with an accuracy of @xmath36mag . the second constraint is used because our calibration is to be applied in determining distances for nearby stars , and we wish to compare objects of a given color class over a similar distance range in order to avoid or limit biases such as dust reddening or possible systematic errors in our usno - b1.0 optical magnitudes . the third constraint excludes proper motion doubles with small separations , for which v and j magnitudes are often affected by large errors , particularly for the faint companions . the hipparcos and tycho catalog provides accurate space - based parallaxes for 1530 lspm stars . the hipparcos parallaxes generally have errors under 5 mas . unfortunately , these are limited to relatively bright ( @xmath37 ) objects , which are only effective in mapping the brighter half of the main sequence ( from spectral types ob to spectral type k ) . for the low - luminosity red dwarfs ( spectral type m ) , one has to rely on ground - based parallaxes . the ypc catalog compiles most ground - based parallaxes published in the literature before 1995 . these , however , are not as homogeneous or accurate as the hipparcos parallaxes , and sometimes have errors as large as 40 - 50 mas . after applying the criteria above , we find reasonably accurate ypc parallaxes for a total of 695 additional lspm stars . hipparcos parallaxes can also be used for a limited number of fainter stars that happen to be common proper motion companions of hipparcos objects ( and are thus physically related and at the same distance ) . buy cross - correlating lspm positions with the hipparcos and tycho catalog , we find 60 lspm stars to be common proper motion companions of @xmath34 hipparcos stars , with angular separations @xmath38 between the components . the secondaries range in absolute magnitude from @xmath39 to @xmath40 . despite these additions , the combined sample of hipparcos and ypc stars remains deficient in objects at the bottom of the main sequence ( @xmath41 ) . accurate parallaxes of very low - mass m and l dwarfs have however been published in @xcite and in @xcite . these papers provide parallaxes for 13 faint lspm stars , of which 11 have @xmath41 . finally , a search of the nstars database provides reliable parallaxes ( compiled from a variety of sources ) for 52 additional lspm stars , including 16 with @xmath41 . the final tally is 1,951 objects to be used as color - magnitude calibrators . absolute magnitudes @xmath42 for stars in the calibration subsample are calculated from their apparent @xmath43 magnitudes and observed trigonometric parallaxes @xmath33 . these are plotted against their @xmath44 colors in figure 1 . the main sequence is clearly visible , as is the white dwarf sequence on the lower left of the diagram . a few stars are found in between , and are most probably subdwarfs . the top of the main sequence ( @xmath45 ) is more densely populated than the bottom ( @xmath46 ) , which reflects the systematic availability of accurate @xmath33 for bright stars ( from the hipparcos catalog ) . at this time , only limited samples of fainter stars have accurate ( ground - based ) parallaxes . for the main sequence ( ms ) stars , we find a monotonic relationship between the @xmath44 color index and the absolute magnitude @xmath42 . the relationship has several inflection points . we subdivide it into 5 segments . for each , we calculate the relationship from a linear regression of the data points ( least squares fitting ) . we map the main sequence as follows : @xmath47\ ] ] @xmath48\ ] ] @xmath49\ ] ] @xmath50\ ] ] @xmath51\ ] ] for any ms star , we can then calculate a photometric distance modulus @xmath52 with : @xmath53 where @xmath54 follows the relationships defined above . photometric parallaxes @xmath55 can then be estimated with : @xmath56 cccc 1.0 - 2.0 & -0.001 & 0.344 & 647 + 2.0 - 3.0 & -0.042 & 0.331 & 372 + 3.0 - 4.0 & 0.034 & 0.573 & 259 + 4.0 - 5.0 & -0.027 & 0.690 & 196 + 5.0 - 6.0 & -0.059 & 0.557 & 69 + 6.0 - 9.0 & -0.038 & 0.631 & 32 we estimate the accuracy of our photometric distance moduli by calculating @xmath52 for all our calibration stars , and comparing with the trigonometric distance moduli @xmath57 , where @xmath58 for each star , we calculate the difference between the trigonometric and photometric distance moduli @xmath59 . these are plotted in figure 2 as a function of @xmath44 for our calibration stars . we find a dispersion that increases from @xmath60 mag at @xmath61 to @xmath62 mag at @xmath63 , and down to @xmath64 mag at @xmath65 . the values for the scatter in the color - magnitude relationship are listed in table 1 , for given ranges in @xmath44 . these provide standard errors for the @xmath52 distance moduli estimates . the smaller ( larger ) errors on @xmath55 for the bluer ( redder ) stars is largely consistent with the expected accuracy of the @xmath43-band magnitudes in the lspm . the bluer calibration stars happen to be among the brighter stars in the lspm catalog ( @xmath37 ) for which the @xmath43 magnitudes are quoted from the tycho-2 catalog @xcite , and are accurate to @xmath36mag . most of the redder calibration stars , on the other hand , are drawn from the group of fainter lspm stars , which have their @xmath43 magnitudes estimated from photographic plate measurements , mostly based on magnitudes from the usno - b1.0 catalog of @xcite which are only accurate to @xmath27mag . one must note that at a given ( @xmath44 ) , the error in the distance modulus is generally not equivalent to the error in the @xmath43 magnitude . rather , it depends on the slope @xmath66 of the color - magnitude relationship @xmath67 . from equation 3 , we see that a @xmath68 error in @xmath43 propagates as @xmath69 . hence for @xmath70 , a steeper slope means a larger error on the photometric distance modulus for a given @xmath68 . hence between @xmath71 , the @xmath72 mag error on the photographic magnitudes is expected to yield an @xmath73 mag error on the distance modulus , which is consistent with our results . in any case , the dispersion in @xmath74 can not be attributed only to errors in the @xmath43-band magnitude measurements . there is an _ intrinsic _ dispersion in the color - magnitude relationship caused by a variety of factors such as multiplicity , variability , and metallicity . this has been known historically as the `` cosmic scatter '' . this scatter is clear from the @xmath75 mag dispersion observed for @xmath76 stars , whose @xmath43 magnitudes are those determined from hipparcos measurements , and are generally accurate to better than @xmath77 mag . intrinsic factors are also probably responsible for the existence of the several outliers ( @xmath78 from the mean ) . these can be unresolved doubles , variables , or even metal - poor subdwarfs that are naturally expected to fall below the metal rich main sequence of the local galactic disk stars . the direct calibration of absolute magnitudes using parallax measurements is generally susceptible to a type of systematic errors , known as the lutz - kelker bias . a nice review of the topic can be found in @xcite . this bias generally results in absolute magnitudes being overestimated , and is dependent of the ratio of the parallax measurement errors over the measured parallaxes @xmath79 . given that our calibrators were selected to have @xmath80 , and given the stars were selected based on proper motion , we need to determine whether the lutz - kelker is significant , and whether corrections should be applied . generally , the lutz - kelker bias occurs in a subsample of objects with measured parallaxes @xmath81 and affected by measurement errors @xmath82 , when the distribution of true parallaxes @xmath83 in the underlying population decreases monotonically with @xmath83 ( as when the number of calibrators increases with distance ) . the measurement errors are symmetric about @xmath81 , but the probability distribution @xmath84 is not because of the asymmetry in the distribution of @xmath83 values around @xmath81 . simply put , because there are more distant stars than nearer ones , a star with measured parallax @xmath81 is more likely to be a distant star whose parallax is overestimated ( @xmath85 ) than a nearby one whose parallax is underestimated ( @xmath86 ) . on average , the mean parallax of a group of stars will thus be overestimated , which means the derived absolute magnitude will be overestimated too . the classical case investigated by @xcite is the one where the calibrators are uniformly distributed in space , for which the distribution in true parallax follows @xmath87 . an analytical derivation indicates a bias in the absolute magnitude calibration of + 0.11 mag for @xmath88 , down to + 0.02 mag for @xmath89 , and + 0.01 mag for @xmath90 . all our calibrators have @xmath91 , following our selection criteria . the hipparcos stars , which populate the upper half of the main sequence ( @xmath45 ) , tend to have significantly more accurate parallax measurements , with @xmath92 . this value corresponds to a very small classical lutz - kelker bias of only + 0.014 mag . for the calibrators that define the lower half of the main sequence ( @xmath46 ) , our sample has @xmath93 , which corresponds to a bias of + 0.019 mag . however , the above values most probably _ overestimate _ the actual lutz - kelker bias in our data . the reason is that our sample of calibrators is not drawn from the general supersample of stars in the solar neighborhood ( whose density is uniform ) but rather from a restricted supersample that includes only stars with proper motions above a certain value . this proper motion limit is @xmath94 for the hipparcos objects ( the lower limit of the lspm - north catalog ) . for the fainter calibration , it is much larger because faint parallax targets have historically been selected among subsamples of stars with very large proper motions ( e.g. the lhs catalog , with a proper motion limit @xmath95 ) . the effect of drawing the calibrators from a proper - motion selected supersample tends to reduce the lutz - kelker bias , because the supersample will be increasingly incomplete at larger distances ( see 4 below ) , and the distribution function @xmath96 will be shallower than in the classical case . estimates of the lutz - kelker bias for shallower distribution has been investigated by e.g. @xcite , who showed that for power law distributions @xmath97 with @xmath98 , the magnitude of the lutz - kelker correction is reduced compared with the classical case ( @xmath99 ) . given the uncertainty in the underlying bias associated with the supersample ( i.e. uncertainty in the form of @xmath96 ) , the application of any specific lutz - kelker correction to our calibration can not be justified . in any case , the lutz - kelker bias is most likely to be very small ( @xmath100 mag ) . our photometric distances may thus systematically underestimate the true distances of the candidate nearby stars , but only at the @xmath101 level . for any given star , this bias is negligible compared with e.g. the intrinsic `` cosmic scatter '' of @xmath102 mag , or to the @xmath103 mag photometric errors . our goal is to identify the largest possible number of dwarf stars within the traditional confines of the solar neighborhood ( @xmath5 pc ) . since our photometric distances are accurate only to @xmath104 , an extension to 33pc ensures that the large majority of 25pc stars will fall into the sample . a limiting distance of 33pc is thus adopted . the lspm - north catalog does allow one to identify objects at much larger distances , but the objective of this paper is to provide a list of high priority objects of a size that can be managed by ground - based parallax programs . the lspm - north contains thousands of new hpm stars , but lists all previously known hpm stars as well , including those from the nltt and lhs catalogs . it is thus not surprising that a large number of lspm entries are stars that have have long been suspected ( or confirmed ) to be nearby . since the lspm is virtually complete for all stars with @xmath105 yr@xmath1 down to a magnitude @xmath25 , the vast majority of known nearby stars from the north celestial hemisphere are expected to have counterparts in the lspm - north . the only exceptions are known nearby stars with very small proper motions ( @xmath106 yr@xmath1 ) or objects with very faint optical magnitudes ( @xmath107 ) . the latter group includes most known nearby l and t dwarfs @xcite ; these are generally too faint in the optical bands to be detected on the poss - i / poss - ii red plates , on which the lspm - north is based . in any case , most of them are not stars but brown dwarfs . we searched the literature for parallax measurements of lspm sources which would place the star within 33 pc of the sun . our search turned up parallax measurements @xmath108 for 1676 lspm stars . this excludes stars whose positions in the color - magnitude diagram suggest they are either white dwarfs or subdwarfs . the complete list of objects is provided in table 2 ; most of the stars are main sequence objects ( dwarfs ) but there are a few subgiants and giants . parallax data are primarily from the hipparcos and tycho catalogue , including data for 1207 stars that are formal hipparcos objects , and for 54 more stars that are close companions or common proper motion companions of hipparcos objects . another major source of parallax data is the ypc , which provides parallaxes for 361 additional lspm stars . additional @xmath33 are found in the nstars database for 40 more lspm stars . parallax data are found in @xcite for 13 faint , red lspm stars . finally , a parallax for lsr j1835 + 3259 is published in @xcite . table 2 gives the name under which the star is generally known along with the lspm designation . the table quotes the lspm position and proper motion at the 2000.0 epoch , and gives the @xmath43 magnitude and @xmath44 color , also from the lspm . the trigonometric parallax is given along with its estimated error , and the bibliographical source . the distance modulus @xmath52 , as estimated from the calibration described in 2 , is shown for comparison . note that many stars do not have @xmath52 estimates , because their blue color ( @xmath109 ) falls outside the range of of calibration system . the final column in table 2 gives the distance from the sun ( in parsecs ) as derived from the trigonometric parallax . figure 3 compares the derived distance modulus @xmath52 with the distance modulus calculated from the measured trigonometric parallax @xmath110 . the expected locus of single stars is shown in figure 3 ( full line ) . the difference between the two estimates , @xmath111 , has a mean value of + 0.04 mag , and a dispersion 0.50 mag , after removal of the 3@xmath112 outliers . this is consistent with our calibration of @xmath52 using a subset of these stars ( see 2 above ) . there appears to be a small number of stars that have @xmath113 ( see figure 3 ) . such an underestimate of the distance , by photometric means , would be consistent with those stars being unresolved doubles . the expected locus of unresolved binaries with equal - luminosity components in the @xmath57,@xmath52 diagram is drawn in figure 3 ( dashed line ) . there are also a few objects whose photometric distance appear to be significantly overestimated , with @xmath114 . in all likelihood , these objects are metal - poor subdwarfs , for which the color - magnitude calibration of 2 is invalid . there exist several hundred high proper motion stars that do not have trigonometric parallaxes , but that have been previously cited in the literature as candidate nearby stars . we compile a list of the main sequence dwarfs ( i.e. excluding known white dwarfs and subdwarfs ) that have been cited as candidate nearby dwarfs with estimated distance within 33 pc , and that have a counterpart in the lspm - north . the list , comprising 783 stars , is provided in table 3 . some 468 stars have published photometric distance estimates , while another 315 have spectroscopic distance estimates . some 359 stars have photometric distances quoted in the cns3 but no formal trigonometric parallax . most of these stars have their distance modulus calculated from the photometric survey of @xcite . most of the other nearby star candidates are from the study of the nltt catalog conducted by reid _ _ we find photometric distance estimates for 56 stars in @xcite and for 29 more in @xcite . @xcite give spectroscopic distance estimates for 28 stars that have counterparts in the lspm - north , and @xcite give spectroscopic distances for 169 more , and @xcite for 91 . spectroscopic distance estimates for high proper motion discovered with superblink were also obtained for 104 stars with @xmath115 yr@xmath1 by @xcite . the list comprises 19 stars that are @xmath116pc candidates ; these are included in table3 . there now also exists spectroscopic / photometric distance moduli for numerous cool and ultra - cool dwarfs discovered with the 2mass and denis surveys . photometric distance moduli are given for 52 ultra - cool 2mass dwarfs in @xcite ; we find matches in that list for 15 of the reddest lspm stars . spectroscopic distance moduli are calculated for 251 late m dwarfs and l dwarfs in @xcite . the majority of these objects are either in the southern sky or are too faint in the optical to be included in the lspm catalog , but we nevertheless find matches to 5 additional red lspm stars . two nearby ultra - cool dwarfs discovered in the denis infrared survey are also found among the reddest lspm entries : the l4 dwarf denis - p j104842.8 + 011158 has a spectroscopic distance modulus estimated by @xcite , while the denis - p j031225.1 + 002158 has a photometric distance modulus cited in @xcite . a photometric distance for lp 584 - 4 ( denis j000206.1 + 011536 ) is also given in @xcite . the lspm - north catalog also includes the very nearby star lhs 2090 ( @xmath117pc ) , whose photometric distance was first reported by @xcite . a spectrophotometric parallax for the very high proper motion ( @xmath118 yr@xmath1 ) star so 025300.5 + 165258 is given in @xcite . finally , a photometric / spectroscopic distance modulus for lsr j0602 + 3910 , the brightest l dwarf in the sky ( v=20.1 ) , is given by @xcite . figure 4 compares the photometric and spectroscopic distance moduli cited in the literature to the distance modulus @xmath52 calculated from the v magnitude and v - j color cited in the lspm catalog . the published @xmath119 agree very well with the lspm distance modulus estimate . values for @xmath120 have a mean of 0.03 mag and a dispersion of 0.43 mag , after removal of 3@xmath112 outliers . the good match between the two distance moduli probably only attests to the fact that the two calibrations are essentially based on the same sample of objects with measured @xmath33 . on the other hand , there is a slight disagreement between the published spectroscopic distance moduli and @xmath52 , with @xmath121 having a mean value of + 0.15 mag . the excellent agreement observed between @xmath52 and both @xmath57 and @xmath119 argues against any systematic error in @xmath52 . this means that the @xmath122 underestimate distances by @xmath123% on average . stars with @xmath122 are overwhelmingly calibrated with the spectral - indices / absolute - magnitude relationships defined by @xcite . the 0.15 mag systematic error possibly results from the limited number of stars used to calibrate the relationships . note that the dispersion of 0.47 mag is only marginally larger than the dispersion on @xmath120 , which suggests that @xmath122 is , in principle , just about as accurate as @xmath119 , provided the zero point is accurately calibrated . using the color - magnitude calibration in 2 , it is straightforward to identify new candidate nearby red dwarfs from the lspm - north catalog . photometric distance moduli are simply estimated from @xmath43 and @xmath44 ( equations 1,2 ) . the only caution is that since the relationship is valid only for metal - rich ( disk ) ms stars , one must avoid applying it to other types of objects . while the bulk of the stars in the lspm - north catalog are disk ms stars , the catalog also includes significant numbers of metal - poor halo subdwarfs and white dwarfs , along with smaller numbers of giants and subgiants . relative to the dwarf ms stars , giant stars are _ overluminous _ for a given @xmath44 color . applying the dwarf color - magnitude relationship thus yields underestimated distances . giant stars are quite rare in the solar neighborhood , but they can significantly contaminate samples of color - selected nearby star candidates since they tend to be detected over a much larger volume . fortunately , proper motion selection filters out the majority of more distant objects : the 0.15@xmath10 yr@xmath1 limit of the lspm - north catalog eliminates most disk stars beyond @xmath124pc ( assuming disk stars to have transverse velocities @xmath12570km / s ) , and most halo stars beyond @xmath126pc ( assuming @xmath125400km / s ) . giant stars that made it into the lspm - north catalog are bright enough to be hipparcos objects , and since there already is good parallaxes for them , no confusion is possible . the more problematic objects might be the subgiants , which are only moderately overluminous relative to the ms stars . the lspm - north potentially contains subgiant stars that are too faint ( @xmath127 ) to be hipparcos objects . however , these are expected to be relatively blue ( @xmath128 ) . according to our calibration ( eq.1 ) , main sequence stars with @xmath128 have absolute magnitudes @xmath129 , which means they have apparent magnitude @xmath130 at 33pc . since the current goal is to identify ms stars within 33pc of the sun , the search can be restricted to lspm stars with @xmath131 . any bluer object within 33pc should have already been found and measured by hipparcos . this constraint also effectively eliminates subgiants as a source of contaminants . white dwarf stars , on the other hand , are _ underluminous _ for any given @xmath44 color . as can be seen from fig.1 , the white dwarf sequence falls @xmath1325 magnitudes below the main sequence in the standard color - magnitude diagram . assuming the color - magnitude relationship for ms stars ( eq.1 ) were applied to a white dwarf , the object would have its distance overestimated by a factor of 10 . restricting the search to stars within 33pc of the sun virtually eliminates the possibility of a contamination of the sample with white dwarfs , because to make it into the 33pc sample of candidate stars , the white dwarf would actually have to be within 3.3pc of the sun . at this time , there is no known isolated white dwarf star within 3.3pc of the sun . the probability of white dwarfs `` contaminating '' the 33 pc sample is very small . subdwarf stars are also _ underluminous _ in @xmath42 for a given @xmath44 color , although not as much as are the white dwarfs . most nearby subdwarfs are local member of the galactic halo population . one way to separate disc dwarfs from halo subdwarfs is to use an optical - to - infrared reduced proper motion diagram @xcite . because the reduced proper motion is proportional to both the absolute magnitude of a star and its transverse velocity , halo subdwarfs , with their larger systemic velocities , fall well below the disc dwarf sequence . but while this helps in separating halo dwarfs from disc subdwarfs , the method only works , statistically , on large groups of objects , and is unreliable for individual stars . in any case , halo subdwarfs are rare in the solar neighborhood ( about 1 star in 200 ) ; if they are found in significant numbers in proper motion catalogs it is because of their larger mean velocity relative to the sun . again , restricting the search to ms stars within 33 pc of the sun minimizes contamination of the nearby stars sample . the @xmath0 yr@xmath1 limit includes all stars at 33 pc with transverse velocities @xmath133 km s@xmath1 , which still overwhelmingly consists of disk stars . increasing the distance would , however , increase the ratio of subdwarfs to dwarfs in the census . for example , lspm stars at 100 pc have @xmath134 km s@xmath1 and exclude most of the disk dwarfs , leaving a much larger fraction of halo subdwarfs . our nearby ms stars selection box , in the @xmath135 diagram , is depicted in figure 5 , overlaid on the distribution of lspm - north stars . the selection box marks the locus of ms stars within 3 pc and 33 pc from the sun . note how the selection box avoids regions on the graph that are mainly populated with giants , white dwarfs , and subdwarfs , as explained above . any star that falls in the box , and that is not a known nearby star ( i.e. it is not listed in tables 2 - 3 ) is flagged as a new nearby star candidate . llrrrrrrrr & lspm j0001 + 0659 & 0.315899 & 6.993230 & -0.447 & -0.081 & 16.79 & 5.50 & 2.09@xmath1360.56 & 26.2@xmath136 8.3 + lp 584 - 4 & lspm j0002 + 0115 & 0.525909 & 1.260014 & 0.468 & 0.071 & 19.48 & 7.31 & 1.30@xmath1360.63 & 18.2@xmath136 5.7 + & lspm j0005 + 0209 & 1.411041 & 2.165013 & 0.261 & -0.270 & 17.73 & 5.74 & 2.57@xmath1360.56 & 32.6@xmath13610.3 + & lspm j0005 + 4129 & 1.480111 & 41.491386 & 0.208 & 0.012 & 12.96 & 3.56 & 2.55@xmath1360.57 & 32.3@xmath136 9.8 + & lspm j0006 + 0439 & 1.609240 & 4.653243 & 0.316 & -0.253 & 20.35 & 7.39 & 2.02@xmath1360.63 & 25.4@xmath136 8.0 + g 217 - 32 & lspm j0007 + 6022 & 1.927582 & 60.381760 & 0.342 & -0.029 & 13.42 & 4.51 & 0.82@xmath1360.69 & 14.6@xmath136 4.8 + & lspm j0009 + 0603 & 2.323657 & 6.062851 & 0.094 & -0.210 & 15.37 & 4.73 & 2.30@xmath1360.69 & 28.9@xmath136 9.5 + & lspm j0011 + 0227 & 2.785660 & 2.466367 & 0.097 & -0.134 & 15.63 & 4.78 & 2.46@xmath1360.69 & 31.0@xmath13610.2 + & lspm j0011 + 5736 & 2.779359 & 57.614510 & 0.192 & -0.063 & 14.66 & 4.54 & 2.00@xmath1360.69 & 25.1@xmath136 8.3 + & lspm j0012 + 0206 & 3.011115 & 2.106582 & 0.292 & 0.074 & 15.40 & 4.81 & 2.16@xmath1360.69 & 27.1@xmath136 8.9 + & lspm j0014 + 0213 & 3.742652 & 2.218896 & 0.317 & -0.113 & 16.81 & 5.40 & 2.32@xmath1360.56 & 29.1@xmath136 9.6 + & lspm j0015 + 4344 & 3.828451 & 43.742985 & 0.232 & 0.038 & 16.69 & 5.47 & 2.05@xmath1360.56 & 25.7@xmath136 8.5 + & lspm j0015 + 5829 & 3.980497 & 58.494160 & 0.257 & -0.003 & 14.86 & 4.71 & 1.83@xmath1360.69 & 23.3@xmath136 7.7 + g 242 - 49 & lspm j0015 + 7217 & 3.902023 & 72.283562 & 0.319 & 0.185 & 12.70 & 3.86 & 1.53@xmath1360.57 & 20.3@xmath136 6.1 + & lspm j0016 + 3000 & 4.168521 & 30.016624 & 0.225 & 0.028 & 12.63 & 3.52 & 2.32@xmath1360.57 & 29.1@xmath136 8.8 table 4 provides the complete list of new candidate nearby stars . the lists includes 1672 stars with @xmath137 and @xmath138 . a total of 730 stars are objects listed in the literature or in previous proper motion catalogs , but which have never before been formally identified as nearby star contenders . the other 942 candidates are high proper motion stars identified with superblink , and first listed in the lspm - north catalog . our color - magnitude calibration puts this bright star at only @xmath139 pc from the sun . while it is a @xmath140 magnitude star , hd 29271 is not a hipparcos star , and there is no published parallax for it . the star is listed with a spectral type k0 in the henry draper catalog , and is also listed in the tycho-2 catalog with tycho magnitudes @xmath141 and @xmath142 . it is unclear whether this star is an actual nearby k dwarf or a more distant k giant . with @xmath143 and @xmath144 ( from 2mass ) , this object has very unusual ir colors which are consistent with neither a k dwarf or a k giant . because it has a proper motion of only 0.165@xmath10 yr@xmath1 , it is more likely to be a more distant object , although it is kept on the list of candidate nearby stars at this time , pending clarification of its status . we estimate for this giclas object a photometric distance for only @xmath145 pc . that no follow - up observation has been performed on this object since its identification comes as a surprise , since it does have a large proper motion @xmath146 yr@xmath1 , making it an obvious candidate for a nearby star search . this object should be a high priority target for parallax programs . this faint ( @xmath147 ) high proper motion star was first identified by superblink , and is one of the new stars listed in the lspm - north catalog . with @xmath148 and @xmath149 , the object is most probably a very low - mass star , of spectral type m7 or later . the photometric distance modulus places it at a distance of @xmath150 pc from the sun . this moderately faint ( @xmath151 ) star has a photometric distance estimate of only @xmath152 pc . it is another one of the new high proper motion stars found with superblink , and also one of the nearest . as with the one above , it should be a high priority target for parallax programs . this @xmath153 star has a photometric distance estimate of only @xmath154 pc . it is also a newly discovered high proper motion star . its moderately small proper motion ( 0.17@xmath10 yr@xmath1 ) and location at low galactic latitudes ( b=+8.33 ) would have made it difficult to find in previous proper motion surveys . even more interesting , this star was previously found to be a counterpart of the rosat point source rx j1935.4 + 3746 by @xcite who determined a spectral type m4ve . the spectral type is consistent with the star being at a distance @xmath155 pc from the sun . a histogram of the distribution of nearby and candidate nearby ms stars as a function of distance is presented in figure 6 . stars that were known before this study are indicated by the shaded area . the present study makes a very significant contribution for stars beyond 20 pc ; only few new additions are made within 15 pc . assuming the current census to be complete up to a distance of 12 pc , we estimate the local stellar density of ms stars to be 0.77 stars pc@xmath156 . the expected distribution of objects as a function of distance , for a uniform distribution of stars in the local volume , is shown in figure 6 for comparison . the current census clearly shows signs of increasing incompleteness beyond 15pc . this is despite the fact that the present study is contributing a significant number of additions beyond this limit , and especially beyond 25pc , where the new candidates account for more than half of the known nearby objects . the source of the incompleteness lies not in the limiting magnitude of the lspm - north catalog . the lspm - north is estimated to 99.5% complete down to @xmath25 , with a limiting magnitude @xmath157 @xcite . the vast majority of the ms disk dwarfs have absolute magnitudes @xmath158 , and at 33 pc are all bright enough to be in the lspm - north . figure 7 plots the distribution of stars as a function of absolute magnitude and distance . it is clear that the bulk of the red dwarf stars are located between @xmath159 . for stars within 33 pc , the majority are too faint to be in the hipparcos catalog , but they are well within the magnitude limit of the lspm - north . the 33 pc census built from the lspm catalog is expected to be deficient only in stars that lie at the very bottom of the main sequence ( @xmath160 ) . the main source of incompleteness in the 33 pc census remains the high proper motion selection . at 33 pc , the @xmath105 yr@xmath1 limit of the lspm selects for stars with transverse velocities @xmath161 km s@xmath1 , which excludes a significant number of disk stars with low space motions relative to the sun . figure 8 plots the distribution of nearby and candidate nearby stars as a function of distance and transverse velocity @xmath162 in km s@xmath1 , where the latter is calculated from the proper motion and estimated distance ( @xmath163 ) . northern nearby stars that are in the lspm catalog are represented by black dots . the incompleteness due to proper motion selection is clear . the very abrupt drop across the @xmath164 yr@xmath1 points to the existence of a significant population of @xmath116pc stars with proper motions @xmath106 yr@xmath1 . the detection ranges of catalogs with various proper motion limits are noted in figure 8 . the hipparcos catalog lists 265 stars in the northern sky with @xmath116 pc and @xmath106 yr@xmath1 . the ypc lists 37 more @xmath116 pc stars which are not in the lspm - north either , because their proper motion is too small . these 302 low proper motion nearby stars are plotted as open circles in figure 8 . it is interesting to note that the census of known nearby stars with _ small _ proper motions is disproportionately biased toward hipparcos ( @xmath37 ) objects , whereas the census of nearby stars with _ large _ proper motions is dominated by non - hipparcos ( @xmath165 ) stars . this is consistent with the fact that ground - based parallax programs ( which the ypc is a compilation of ) have been largely dependent on lists of candidates selected on the basis of large proper motions . the proper motion limit results in an increasing incompleteness in the nearby star census as a function of distance . it is possible to estimate how many low - velocity stars are missing at larger distances by mapping the distribution of low - velocity stars that actually get detected at smaller distances . figure 9 compares the distribution as a function of transverse velocity ( @xmath166 ) for stars within 33 pc that are actually detected by the lspm - north , with a estimate of the expected distribution of objects . the estimate is based on a correction of the actual number of stars found within 33 pc at a given @xmath166 to the effective volume sampled by the lspm - north for stars having that value of @xmath166 . no correction is necessary for velocity bins @xmath167 km s@xmath1 , since the @xmath0 yr@xmath1 limit of the lspm - north is sensitive to star with such velocities out to at least 33 pc . for bins with @xmath168 km s@xmath1 , the bins are normalized by a factor @xmath169 , where @xmath170 is the maximum distance a star with transverse velocity @xmath166 can be detected by a proper motion survey with @xmath164 yr@xmath1 limit . the `` predicted '' velocity distribution at small @xmath166 ( see figure 9 ) shows fluctuations because of small number statistics , but it is clear that at least a thousand stars must be missing from the census because of their small proper motion . a simple estimate suggests that there should be @xmath171 stars within 33 pc of the sun . the current census includes only 4434 stars , the 4,131 ones listed in the lspm - north , plus the 302 known nearby low proper motion objects ( see above ) . the current , overall completeness of the 33pc census is thus only @xmath172 . this indicates that @xmath173 northern , nuclear - burning stars remain to be found within a distance of about 33 pc , and that these have proper motions @xmath106 yr@xmath1 . the situation in the 25pc volume is slightly better , with @xmath174 nuclear burning stars expected in the northen sky , and 2,304 currently located ( including 114 with @xmath106 yr@xmath1 ) . the current completeness of the 25pc census , the traditional extent of the `` solar neighborhood '' , is thus @xmath175 . the 500 or so missing stars are expected to have proper motions below the limit of the lspm - north . it is clear that much can be gained by expanding our proper motion survey to reach smaller proper motion limits . figure 10 plots the expected rate of detection as a function of distance , for surveys with lower proper motion limits @xmath176 yr@xmath1 , 0.2@xmath10 yr@xmath1 , 0.1@xmath10 yr@xmath1 , and 0.05@xmath10 yr@xmath1 . these detection rates are based on the transverse velocity distribution shown in figure 9 . one sees that to achieve a @xmath177 detection rate at a distance of 25 pc would require to assemble a complete survey of stars with @xmath178 yr@xmath1 . this paper has presented a list of nearby and candidate nearby main sequence stars , subgiants , and giants within 33 pc of the sun . the list is based on an analysis of the lspm - north catalog of stars with proper motions @xmath105 yr@xmath1 . the census contains a total of 4221 stars in the half of the sky north of the j2000 celestial equator . the list includes 1676 previously known nearby stars with measured trigonometric parallaxes , 783 previously suspected nearby objects , and 1762 new candidate nearby stars . a relationship is determined between the absolute v magnitude ( @xmath42 ) and the @xmath44 color index . this relationship can be used to calculate photometric distance moduli @xmath179 in the @xmath135 magnitude system . this is convenient for estimating distances of all the stars in the lspm - north ( for which @xmath43 and @xmath180 magnitude estimates exist ) to the extent that the relationship is applied to main sequence(h - burning ) stars of near - solar metallicity . for single stars , the relationship is accurate to @xmath181 mag , which means that it provides distance estimates accurate to @xmath182 . the relationship breaks down for white dwarfs and subdwarfs , whose cases will be addressed in upcoming papers of this series . it is found , however , that the sample of main sequence dwarfs selected based on the color - magnitude relationship should not be significantly contaminated by mis - identified white dwarfs and subdwarfs , as long as the search is limited to stars within 33 pc of the sun . the main purpose of the present list of candidate objects is to provide targets for upcoming parallax programs . a fraction of the 783 previously suspected nearby ms stars are probably already on parallax programs . the list of 1762 new candidate nearby stars should be examined to sort out the most promising candidates to be added to the parallax programs . note that current distance estimates based on photometry are only approximate , and should be used with caution . a large spectroscopic follow - up program is under way , which will soon provide spectral types and radial velocities for most of the new candidate nearby stars presented here . results will be presented in upcoming papers in this series . while these will also yield spectroscopic distance moduli , these are not expected to be significantly more accurate than the photometric distance moduli presented here . accurate distances will ultimately come from parallax measurements . the census of nuclear - burning stars within 33 pc of the sun , based on the present study , appears to be only @xmath172 for the northern sky . the census of nuclear - burning stars within 25 pc , the traditional confines of the so - called `` solar neighborhood '' is estimated to be @xmath175 complete . it is revealed that the main source of incompleteness lies in the low proper motion limit ( @xmath105 yr@xmath1 ) of the lspm - north catalog . expanding our survey to reach a smaller proper motion limit appears to be the best way to significantly increase the completeness of the nearby star census . an extension of the superblink survey down to a proper motion limit of @xmath183 yr@xmath1 is now being completed , and a second extension down to @xmath184 yr@xmath1 is being planned . an analysis of the @xmath184 yr@xmath1 catalog should ultimately yield the identification of @xmath185 of the nuclear burning stars up to and at a distance of 25 pc . this work has been made possible by the generous support of hilary lifsitz and of the american museum of natural history . the author is indebted to michael m. shara ( amnh ) and r. michael rich ( ucla ) for their constant support and encouragement . cutri , r. m. , _ et al . _ 2003 , the 2mass all - sky catalog of point sources university of massachusetts and infrared processing and analysis center ( ipac / california institute of technology _ cds - vizier catalog number ii/246 _ ) giclas , h. l. , burnham , r. , & thomas , n. g. 1971 , lowell proper motion survey northern hemisphere . the g numbered stars . 8991 stars fainter than magnitude 8 with motions @xmath186 0".26/year , flagstaff , arizona : lowell observatory _ cds - vizier catalog number i/79 _ ) hog e. , fabricius c. , makarov v.v . , urban s. , corbin t. , wycoff g. , bastian u. , schwekendiek p. , & wicenec a. 2000 , the tycho-2 catalogue of the 2.5 million brightest stars , , 355 , 27 ( _ cds - vizier catalog number i/259 _ ) perryman , m. a. c. 1997 , the hipparcos and tycho catalogues . astrometric and photometric star catalogues derived from the esa hipparcos space astrometry mission , publisher : noordwijk , netherlands : esa publication ( _ cds - vizier catalog number i/239 _ ) reid , i. n. , cruz , k. l. , laurie , s. p. , liebert , j. , dahn , c. c. , harris , h. c. , guetter , h. h. , stone , r. c. , canzian , b. , luginbuhl , c. b. , levine , s. e. , monet , a. k. b. , monet , d. g. 2003b , , 125 , 354 llrrrrrrrrrr g 129 - 55 & lspm j0000 + 1659 & 0.200457 & 16.988232 & -0.069 & -0.305 & 8.80 & 1.85 & 0.0318@xmath1360.0012 & hip & & 31.4@xmath136 1.2 + lhs 101 & lspm j0002 + 2704 & 0.542444 & 27.082132 & 0.830 & -0.989 & 5.77 & 1.07 & 0.0806@xmath1360.0030 & hip & & 12.4@xmath136 0.5 + g 130 - 40 & lspm j0004 + 2316 & 1.234669 & 23.269632 & 0.382 & -0.007 & 7.82 & 1.42 & 0.0391@xmath1360.0009 & hip & & 25.6@xmath136 0.6 + lhs 1014 & lspm j0005 + 4547 & 1.295397 & 45.786568 & 0.870 & -0.151 & 10.05 & 3.35 & 0.0870@xmath1360.0014 & hip & 0.17@xmath1360.57 & 11.5@xmath136 0.2 + lhs 1016 & lspm j0005 + 4548s & 1.420823 & 45.810383 & 0.839 & -0.162 & 8.97 & 2.83 & 0.0851@xmath1360.0027 & hip & 0.33@xmath1360.33 & 11.8@xmath136 0.4 + lhs 1017 & lspm j0005 + 4548n & 1.420888 & 45.812080 & 0.879 & -0.154 & 8.83 & 2.73 & 0.0851@xmath1360.0027 & hip & 0.39@xmath1360.33 & 11.8@xmath136 0.4 + ltt 10023 & lspm j0006 + 2901 & 1.653266 & 29.021517 & 0.381 & -0.178 & 6.07 & 1.34 & 0.0730@xmath1360.0008 & hip & & 13.7@xmath136 0.1 + g 243 - 13 & lspm j0006 + 5826 & 1.565996 & 58.436764 & 0.271 & 0.030 & 6.36 & 1.49 & 0.0493@xmath1360.0010 & hip & & 20.3@xmath136 0.4 + lhs 1022 & lspm j0007 + 0800 & 1.996295 & 8.005391 & -0.349 & -0.413 & 13.33 & 3.94 & 0.0440@xmath1360.0063 & ypc & 1.96@xmath1360.57 & 22.7@xmath136 3.8 + v * alf and & lspm j0008 + 2905 & 2.096912 & 29.090433 & 0.136 & -0.162 & 2.05 & -0.08 & 0.0336@xmath1360.0007 & hip & & 29.8@xmath136 0.7 + v * v740 cas & lspm j0008 + 6627 & 2.238476 & 66.456650 & 0.178 & 0.002 & 8.62 & 1.59 & 0.0313@xmath1360.0021 & hip & & 31.9@xmath136 2.3 + lhs 1027 & lspm j0009 + 5908 & 2.294539 & 59.149780 & 0.527 & -0.180 & 2.27 & 0.56 & 0.0599@xmath1360.0006 & hip & & 16.7@xmath136 0.2 + ross 310 & lspm j0011 + 5820 & 2.838105 & 58.349915 & 0.235 & 0.020 & 9.48 & 2.29 & 0.0357@xmath1360.0051 & hip & & 28.0@xmath136 4.7 + ccdm j00126 + 214 & lspm j0012 + 2142n & 3.139525 & 21.713451 & 0.183 & -0.289 & 11.91 & 3.07 & 0.0345@xmath1360.0117 & hip & 2.73@xmath1360.57 & 29.0@xmath13614.9 + lp 348 - 42 & lspm j0012 + 2142s & 3.143580 & 21.706341 & 0.187 & -0.294 & 13.65 & 3.99 & 0.0345@xmath1360.0117 & hid & 2.16@xmath1360.57 & 29.0@xmath13614.9 llrrrrrrrrrrrrr lp 404 - 33 & lspm j0008 + 2050 & 2.224675 & 20.840403 & -0.061 & -0.255 & 13.90 & 5.03 & 0.13@xmath1360.35 & cns & & & 0.20@xmath1360.56 & 10.6@xmath1361.9 + lp 191 - 43 & lspm j0008 + 4918 & 2.229864 & 49.315651 & 0.347 & 0.205 & 16.53 & 5.67 & & & 0.99@xmath1360.17 & r03 & 1.50@xmath1360.56 & 15.8@xmath1361.3 + & lspm j0011 + 2259 & 2.970996 & 22.984573 & 0.146 & -0.229 & 11.98 & 3.12 & & & 1.14@xmath1360.33 & r03 & 2.68@xmath1360.57 & 16.9@xmath1362.8 + lhs 1037 & lspm j0011 + 3303 & 2.985193 & 33.054703 & -0.544 & -0.395 & 12.98 & 3.91 & 1.55@xmath1360.30 & r04 & & & 1.69@xmath1360.57 & 20.4@xmath1363.0 + lp 149 - 35 & lspm j0012 + 5059 & 3.238347 & 50.988163 & 0.295 & 0.032 & 17.24 & 5.83 & & & 1.95@xmath1360.18 & r03 & 1.91@xmath1360.56 & 24.5@xmath1362.1 + lp 404 - 66 & lspm j0016 + 2003 & 4.236689 & 20.065327 & 0.228 & 0.024 & 14.11 & 4.43 & 1.73@xmath1360.43 & cns & & & 1.68@xmath1360.69 & 22.2@xmath1364.8 + lp 404 - 80 & lspm j0017 + 2057w & 4.494271 & 20.955217 & -0.272 & -0.384 & 12.16 & 3.47 & 2.42@xmath1360.30 & r04 & & & 1.98@xmath1360.57 & 30.5@xmath1364.5 + lp 404 - 81 & lspm j0017 + 2057e & 4.496541 & 20.956783 & -0.272 & -0.384 & 10.72 & 2.47 & 2.30@xmath1360.30 & r04 & & & & 28.8@xmath1364.3 + & lspm j0017 + 3028 & 4.491442 & 30.469551 & 0.287 & 0.052 & 16.28 & 4.61 & & & 2.14@xmath1360.18 & r04 & 3.47@xmath1360.69 & 26.8@xmath1362.3 + lp 292 - 66 & lspm j0018 + 2748 & 4.723306 & 27.813824 & 0.387 & -0.101 & 13.80 & 4.27 & 1.42@xmath1360.41 & cns & & & 1.71@xmath1360.69 & 19.2@xmath1364.0 + lhs 1060 & lspm j0021 + 1843 & 5.318830 & 18.732157 & 0.682 & -0.084 & 17.17 & 5.85 & & & 1.69@xmath1360.19 & r03 & 1.80@xmath1360.56 & 21.8@xmath1362.0 + lp 149 - 56 & lspm j0021 + 4912 & 5.490994 & 49.210533 & 0.204 & -0.031 & 12.74 & 3.60 & 2.02@xmath1360.24 & cns & & & 2.23@xmath1360.57 & 25.4@xmath1363.0 + lhs 1073 & lspm j0025 + 2253 & 6.335831 & 22.886417 & -0.241 & -0.457 & 14.71 & 4.99 & 1.00@xmath1360.19 & cns & & & 1.09@xmath1360.69 & 15.8@xmath1361.4 + lp 193 - 488 & lspm j0026 + 3947 & 6.510745 & 39.789875 & 0.224 & 0.021 & 16.33 & 5.34 & & & 2.44@xmath1360.34 & r03 & 1.97@xmath1360.56 & 30.8@xmath1365.2 + lp 349 - 25 & lspm j0027 + 2219 & 6.983301 & 22.325680 & 0.408 & -0.174 & 17.40 & 6.79 & & & -0.35@xmath1360.33 & r03 & 0.22@xmath1360.63 & 8.5@xmath1361.4

we investigate the details of the bulk - boundary correspondence in lorentzian signature anti - de sitter space . operators in the boundary theory couple to sources identified with the boundary values of non - normalizable bulk modes . such modes do not fluctuate and provide classical backgrounds on which bulk excitations propagate . normalizable modes in the bulk arise as a set of saddlepoints of the action for a fixed boundary condition . they fluctuate and describe the hilbert space of physical states . we provide an explicit , complete set of both types of modes for free scalar fields in global and poincar coordinates . for , the normalizable and non - normalizable modes originate in the possible representations of the isometry group for a field of given mass . we discuss the group properties of mode solutions in both global and poincar coordinates and their relation to different expansions of operators on the cylinder and on the plane . finally , we discuss the extent to which the boundary theory is a useful description of the bulk spacetime . # 1([#1 ] ) * bulk vs. boundary dynamics * * in anti - de sitter spacetime * * vijay balasubramanian , per kraus and albion lawrence * .5 cm __lyman laboratory of physics , harvard university _ _ _ cambridge , ma 02138 , usa _ .5 cm _ california institute of technology _ _ pasadena , ca 91125 , usa _ # 1#2

hep-th9805171

the description of certain charged black holes as d - branes in string theory implies a connection between the low - energy gauge dynamics on the brane and the low energy supergravity in spacetime . recently , maldacena has proposed decoupling limits in which the brane gauge dynamics is dual to string theory on the near - horizon anti - de sitter geometry of the corresponding black hole @xcite . a more precise definition of this duality was developed in @xcite . we associate to the string compactification on @xmath6 a conformal field theory living on a space conformal to the @xmath7-dimensional boundary @xmath8 of the ads factor . to each field @xmath9 there is a corresponding local operator @xmath10 in the conformal field theory . the relation between string theory in the bulk and field theory on the boundary is : @xmath11 here @xmath12 is the effective action in the bulk , @xmath13 is the field @xmath9 restricted to the boundary , and @xmath14 is the time - ordering symbol in the field theory on @xmath8 . the expectation value on the right hand side is taken in the boundary field theory , with @xmath13 treated as a source term . in the classical supergravity limit , given a boundary field we solve for the corresponding bulk field and use it to relate the bulk effective action to boundary correlation functions . in euclidean ads this proposal used the absence of normalizable solutions to the field equations , and the resulting unique extension of a boundary field @xmath13 into the bulk @xcite . we are interested in issues of spacetime causal structure and dynamics , so we would also like a hamiltonian formulation of the bulk theory in lorentzian signature ads spaces . the standard construction of quantum field theory depends on the existence of a complete set of normalizable modes , which is in tension with the unique extendibility of ads boundary conditions into the bulk . indeed , several consistent quantizations have been found @xcite , involving particular choices of boundary conditions for ads spacetimes and the resulting set of normalizable modes . on the other hand , the bulk - boundary correspondence demands the ability to tune the boundary conditions in order to describe the appropriate boundary correlation functions . in this paper we resolve these tensions by arguing that the bulk - boundary correspondence as formulated in @xcite demands the inclusion of both normalizable and non - normalizable modes . the former propagate in the bulk and correspond to physical states while the latter serve as classical , non - fluctuating backgrounds and encode the choice of operator insertions in the boundary theory . we begin in sec . [ sec : review ] by reviewing the computation of boundary correlation functions , and providing a prescription for computing the bulk effective action in a hamiltonian formulation . we will argue that specifying the boundary conditions involves turning on non - normalizable modes which do not fluctuate . including such non - fluctuating modes may seem strange , but several well - known examples exist in other contexts : a classic example is a field theory which undergoes spontaneous symmetry breaking , and more recent related examples arise in @xcite . normalizable solutions to the wave equation are then used in the mode expansion of operators and the construction of a fock space . the resulting hilbert space of states is identified with the hilbert space of the boundary theory @xcite . in sec . [ sec : norm ] we make this more explicit by studying solutions to the wave equation for free scalars of arbitrary mass . we find that for general masses , the field equations have both normalizable and non - normalizable solutions and the latter couple to the boundary . in sec . [ sec : ads3 ] we specialize to @xmath0 , whose bulk isometry group and boundary conformal group are both @xmath15 . we show that the normalizable modes transform in unitary irreducible representations of @xmath16 while the non - normalizable modes transform in non - unitary reducible representations which contain a highest weight module . the normalizable and non - normalizable highest weight representations are built on states with @xmath1 weights @xmath17 . is the inverse of the cosmological constant of @xmath0 . ] interesting subtleties arise for small masses and for integral @xmath18 . we carry out the analysis in both global and poincar coordinates in order to discuss conformal field theories both on the cylinder and the plane . the latter case has some curious features because poincar coordinates only cover a patch of the global spacetime . we conclude the paper with a discussion of the utility and limitations of the bulk - boundary correspondence for describing bulk spacetime physics via the boundary gauge theory . [ [ euclidean - formulation ] ] euclidean formulation : + + + + + + + + + + + + + + + + + + + + + + + specifying the boundary behaviour of a field @xmath19 in euclidean ads leads to a unique solution to the equations of motion , given some regularity conditions ( see @xcite for discussion and references ) . so ( [ eq : bbrel ] ) is unambiguously interpreted by evaluating the @xmath20 effective action on the unique bulk extension of the boundary field . for a free scalar @xmath19 with boundary value @xmath21 , @xmath22 we relate this to the right hand side of ( [ eq : bbrel ] ) by integrating by parts . since @xmath19 is a solution to the equations of motion , the bulk contribution vanishes . there is , however , a boundary term which is non - vanishing for the relevant solutions to the euclidean wave equation : @xmath23 ( here @xmath24 is the boundary of spacetime in the poincar coordinates defined in the appendix . ) as @xmath25 this is quadratic in @xmath21 and can be thought of as a quantity in the boundary theory . in order to compare to the right - hand side of ( [ eq : bbrel ] ) , we must identify the boundary operator coupling to @xmath19 . the dimension of the operator is determined by the growth of @xmath19 at the boundary of ads @xcite . in particular , suppose that the boundary behavior of @xmath19 is @xmath26 then the corresponding operator has mass dimension @xmath27 . in fact , this procedure is not completely well - defined : in general , ( [ eq : boundpart ] ) blows up and some regularization is required in order to extract the correlator @xcite . for interacting theories , we can calculate @xmath12 perturbatively by integrating over fluctuations away from the classical solution . these fluctuations should have finite action and vanish appropriately at the boundary . [ [ lorentzian - formulation ] ] lorentzian formulation : + + + + + + + + + + + + + + + + + + + + + + + + in lorentzian signature ads spacetime the relation in ( [ eq : bbrel ] ) is more subtle , because there are normalizable solutions to the wave equation @xcite which do not affect the leading boundary behavior ( [ eq : boundlim ] ) . so the boundary value @xmath21 does not uniquely specify the bulk field . furthermore , the normalizable modes can not be projected out since they are needed to expand quantum operators , build a fock space , and compute green s functions in the bulk . the bulk effective action appearing in ( [ eq : bbrel ] ) can be written as : @xmath28 where the path integral is taken over fields @xmath9 that take the boundary value @xmath13 . the measure @xmath29 can be normalized by requiring that @xmath30 when the boundary condition is trivial ( @xmath31 ) . in general there is a set of saddle points that must be summed over ; these correspond to the normalizable modes that are solutions to the equations of motion with a fixed boundary condition . for a free theory in minkowski space , the normalizable modes indicate that there is a manifold of flat directions that must be integrated over for fixed boundary behavior . for a free theory the calculations in refs . @xcite are not affected as the flat directions lead to an overall volume factor that is divided out when we appropriately normalize the path integral . in order to define the path integral in ( [ eq : path1 ] ) we must specify the behaviour of fields at the ads boundary . we will see that boundary conditions can be specified by including certain modes that are generically not normalizable or at least perturb the asymptotic geometry too violently . in fact , these are precisely the modes that are identified in @xcite with sources coupling to boundary operators . as such , these modes should be locked or non - fluctuating . such non - fluctuating modes arise in a variety of contexts . one example is the homogeneous mode of a scalar field that takes a nonzero value through spontaneous symmetry breaking . another example is euclidean field theory on spaces with constant negative curvature ; the authors of ref . @xcite note that the large volume at infinity keeps modes with certain boundary conditions from fluctuating . our discussion is particularly reminiscent of liouville theory , where local operators create non - normalizable wavefunctions when inserted in a disc amplitude @xcite . this analogy is especially attractive after the work in @xciteand was a motivation for the present form of this paper ; indeed , for @xmath0 the gravity sector of the bulk has a chern - simons action which reduces to a boundary liouville theory @xcite . in particular ref . @xcite relates the asymptotic behavior of solutions to the wave equation to the gravitational dressing of chiral operators in the boundary cft . for an interacting theory , our prescription provides the analog of an s - matrix for field theories in ads space , an issue raised in ref . @xcite . in standard flat - space field theory , one specifies boundary conditions at infinity by turning off the interactions and physically separating the initial excitations , so that we can sensibly discuss asymptotic states . in our picture , even though the bulk excitations may not be separable and asymptotic states may not exist , we can sensibly discuss the dependence of the partition function on boundary conditons for locked fields at infinity . [ [ hamiltonian - quantization ] ] hamiltonian quantization : + + + + + + + + + + + + + + + + + + + + + + + + + + once we have specified the boundary conditions , we must ask what fluctuating modes to keep in a hamiltonian formalism . the point is that quantization requires a complete , normalizable set of field modes , which in ads space requires some sort of boundary condition at infinity . several consistent choices have been suggested in the past ; for the bulk - boundary correspondence to make sense the choice should be unambiguously determined by the physics of the situation . our prescription is heavily motivated by the work of breitenlohner and freedman @xcite ( which covers @xmath32 ) , so we will review their discussion for scalars of arbitrary mass @xmath33 in @xmath20 ( also see @xcite ) . fix a spacelike slice @xmath34 with coordinates @xmath35 and an orthogonal , timelike coordinate @xmath36 . given two solutions @xmath37 to the scalar wave equation , define the inner product : @xmath38 if @xmath39 , this is the integral of the time component of the current @xmath40 here @xmath41 is only time - independent up to boundary terms coming from the timelike boundary of anti - de sitter space ; it is easy to see , using the equations of motion , that @xmath42 where @xmath43 is the boundary of the spacelike slice . the authors of @xcite find complete sets of modes by demanding , first , that the flux normal to the boundary vanish at infinity . they also demand that no energy is exchanged with the boundary : @xmath44 the authors of @xcite considered the relative merits of the canonical and improved stress tensors in this equation . this potential ambiguity in the stress tensor arises because the curvature is constant , and its possible coupling to a scalar field would appear as an effective mass term . in our case the choice is dictated by the underlying string theory and the supergravity effective action that descends from it . we will simply treat scalar fields of mass squared @xmath45 with the understanding that the mass term may contain a contribution from a curvature coupling . we will find that for masses @xmath46 , simply requiring normalizability is sufficient to isolate a suitable class of fluctuating solutions , and the more refined discussion in ref . @xcite is unnecessary . for @xmath47 ( stability requires @xmath48 ) , there are two sets of normalizable solutions and some criterion is needed to distinguish them . refs . @xcite show that for any given field propagating on @xmath49 , conservation of the inner product ( [ eq : adsnorm ] ) and the vanishing of @xmath50 at the boundary requires either but not both sets . furthermore , in the case of @xmath32 , the authors of @xcite show that the two sets of modes are built on representations of the conformal algebra with different lowest energy states and that supersymmetry generically requires one to take both types of mode into account . the clearest example is that of the scalar and pseudoscalar in the gravity supermultiplet of gauged @xmath51 supergravity @xcite . both are conformally coupled , and supersymmetry requires that the modes of the scalar lie in one representation while the modes of the pseudoscalar lie in the other . further criteria are needed , however , to decide which assignment is realized . in this particular case , hawking @xcite imposed the requirement that the metric be asymptotically anti - de sitter ( in a sense defined in that work ) to find particular boundary conditions on the linearized metric perturbations . supersymmetry then requires the scalar mode to live in the representation whose highest - weight state has the lowest energy . it would be nice if similar criteria could be applied to general modes in the range @xmath47 , independently of supersymmetry ; we will not investigate this point , however . [ [ summary ] ] summary : + + + + + + + + + the lesson is that we need to keep both normalizable and non - normalizable modes in ads spacetimes . the non - normalizable modes correspond to operator insertions in the boundary gauge theory ; from the ads point of view they provide non - trivial boundary conditions . the normalizable modes fluctuate in the bulk ; quanta occupying such modes have a dual description in the boundary hilbert space . in path integral language , ( [ eq : bbrel ] ) can be understood via the background field expansion : we compute the effective action by expanding the path integral of the bulk theory as @xmath52 @xmath53 is a classical , nonnormalizable solution to the equations of motion , corresponding to an operator insertion at infinity and a particular choice of boundary conditions . then @xmath54 is the fluctuating piece over which we integrate to get the partition function . the normalizable modes appear as stationary points of the action given the background @xmath55 . in this section we show that a scalar of arbitary mass in ads spacetime has both fluctuating and non - fluctuating solutions that implement the bulk - boundary correspondence as advocated above . generically , the non - fluctuating solutions are not normalizable in the norm ( [ eq : adsnorm ] ) . we begin with solutions in poincar coordinates , which allows for a direct comparison with refs . it is easy to separate variables in these coordinates by writing the scalar of mass squared @xmath56 as : @xmath57 @xmath58 then satisfies the equation @xmath59 for @xmath60 , the solution is @xcite : @xmath61 where @xmath62 . this solution is non - normalizable at the boundary at infinity but well - behaved in the interior . the second , independent solution @xmath63 is very badly behaved ( i.e. blows up exponentially ) in the interior and is therefore eliminated . if we consider anti - de sitter space as the near - horizon geometry of a brane , then in the asymptotically flat region @xmath64 would have imaginary momentum perpendicular to the brane . for @xmath65 , there are two possible solutions which are regular in the interior . if @xmath66 is not integral , @xmath67 are two independent solutions . if @xmath66 is integral , @xmath68 are equivalent , and the two independent solutions are @xmath69 in ( [ eq : poinnorm ] ) and @xmath70 ( note that many of the scalar modes arising from the kaluza - klein reduction of string theory down to @xmath71 have integral @xmath66 @xcite , so this is a `` special case '' of particular importance . ) the boundary term ( [ eq : boundpart ] ) will vanish for @xmath69 ; for @xmath72 it will either go to a constant ( for @xmath73 ) or blow up . @xmath64 and @xmath72 both behave as @xmath74 as discussed in @xcite , where @xmath75 for @xmath76 , @xmath72 behaves as @xmath77 as in @xcite , @xmath78 couple to operators of dimension @xmath79 . the asymptotic behavior of bessel functions is well - known , so criteria for selecting fluctuating and non - fluctuating modes are easy to apply . for @xmath80 , only @xmath69 is normalizable and @xmath72 must therefore act as the non - fluctuating source term in ( [ eq : bbrel ] ) . for @xmath81 the story is more complicated as both of @xmath82 are normalizable and , as pointed out in refs . @xcite , both kinds of modes are necessary for supersymmetry . indeed , for compactifications of m - theory on @xmath83 , a conformally coupled scalar and pseudoscalar appear in the kk spectrum @xcite . case in general . ] both analysis of the kk spectrum and its relation to operators in the dual 3d cft @xcite , and the arguments in @xcite , show that the fluctuating modes for the scalar are of the form @xmath72 and the fluctuating modes for the pseudoscalar are of the form @xmath69 . furthermore , the `` conjugate '' solutions @xmath69 and @xmath72 couple to operators with dimension @xmath84 and @xmath85 , respectively @xcite . contradicts the statement made in ref . @xcite for euclidean theories that the dimension of operators in the cft is bounded below by @xmath86 . it however satisfies the unitarity condition @xmath87 for scalar operators in the dual cft @xcite . ] one may worry legitimately that for the scalar mode , interpreting ( [ eq : bbrel ] ) is problematic because the surface term ( [ eq : boundpart ] ) diverges for this mode and not for the `` source '' modes coupling to the boundary . perhaps the answer is that when computing correlation functions , one does not perturb the background with classical normalizable modes . at present we will leave the resolution of this issue for future work . it is also worth noting that the borderline non - normalizable modes , @xmath72 for @xmath88 , contain the singleton @xcite . let us examine a scalar of mass @xmath56 . make the substitution : @xmath89 here @xmath90 are the @xmath91 spherical harmonics on @xmath92 , for which : @xmath93 with @xmath94 . the wave equation in global coordinates ( see the appendix ) is : @xmath95 \chi = 0\ .\ ] ] we have chosen @xmath19 with which we can build an arbitrary configuration on @xmath96 for a given @xmath97 ; ( [ eq : globalwave ] ) then gives us the @xmath97 dependence . for general @xmath45 , @xmath98 the equation of motion is easily converted into a hypergeometric equation which can naturally be expanded in variables that vanish either at the origin ( @xmath99 ) or the boundary ( @xmath100 ) . for maintaining regularity at the origin , it will be easier to examine the solutions as functions of @xmath101 . for examining the boundary behavior , solutions as functions of @xmath102 will be more convenient . we will display both sets of solutions explicitly . of course one can pass between them using standard formulae relating hypergeometric functions as arguments of @xmath103 and @xmath104 @xcite ; these permit the enforcement of regularity conditions at the origin and the boundary , thereby imposing quantization conditions on the spectrum of the theory . begin by substituting : @xmath105 let @xmath106 . it is then easy to see that : @xmath107 \partial_y f - \left [ ( h + b)^2 - \frac{\omega^2}{4 } \right ] f = 0\ , \label{eq : hypersinewave}\ ] ] where @xmath108 the equations for @xmath109 each have two solutions : be what ref . @xcite calls @xmath110 and @xmath111 be what ref . @xcite calls @xmath110 . then @xmath112 is related to these as @xmath113 @xmath114 the hypergeometric equation will have two independent solutions , corresponding to the two solutions of the indicial equation for @xmath115 . one solution will be logarithmic if @xmath116 is an integer , if @xmath7 is even . choosing instead @xmath117 , one gets the hypergeometric equation : @xmath118 \p_x f + \left [ \left(h + b\right)^2 - \frac{\omega^2}{4 } \right ] f = 0\ , \label{eq : hyperwave}\ ] ] with @xmath109 as before . again , the hypergeometric equation has two independent solutions ; this time they correspond to the two solutions of the indicial equation for @xmath112 . one solution will be logarithmic if is an integer . in fact , if we wish to transform solutions as functions of @xmath119 to solutions as functions of @xmath102 , the relevant formulae are modified in the case of integral @xmath66 ; thus we will examine the two cases separately . [ [ behaviour - at - the - origin ] ] behaviour at the origin : + + + + + + + + + + + + + + + + + + + + + + + + + the behaviour at the origin is conveniently analyzed by studying solutions as a function of @xmath119 . choose , without loss of generality , @xmath120 . the first solution as a function of @xmath119 is : @xmath121 the second solution depends on whether @xmath7 is even or odd . if @xmath7 is odd , then : @xmath122 if @xmath7 is even the second solution is logarithmic : @xmath123 \nonumber \\ & & - \sum_{k=1}^{\ell + \half d - 1 } \frac{(k-1 ) ! ( 1 - \ell - \frac{d}{2})_k}{(1 - h_+ - \half(\ell + \omega))_k ( 1 - h_+ - \half(\ell + \omega))_k } \sin^{-2k}\rho\ , \end{aligned}\ ] ] where @xmath124 + k \right ) } \nonumber \\ & & + \psi\left(\half[\half e+ f+ g-\omega ] + k\right ) - \psi(1 + f + k ) - \psi(1+k)\ , \label{eq : hdef}\end{aligned}\ ] ] following ref . @xcite , and @xmath125 we must impose a regularity condition at the origin because in order for ( [ eq : bbrel ] ) to make sense , we should not have contributions to correlation functions coming from the interior . so we will only keep solutions for which the boundary term of the classical action vanishes at the origin @xmath99 : @xmath126 it is easy to show that this means that only the first solution @xmath127 is allowed . [ [ behaviour - at - the - boundary ] ] behaviour at the boundary : + + + + + + + + + + + + + + + + + + + + + + + + + + + to study the behaviour at the boundary it is most convenient to work with solutions as a function of @xmath102 : @xmath128 and @xmath129 in general the regular solution at the origin ( @xmath127 ) is a linear combination of @xmath82 . we can see directly that the leading behavior at the boundary of @xmath82 is @xmath130 . for @xmath80 the norm ( [ eq : adsnorm ] ) of @xmath72 diverges at @xmath131 , while the norm of @xmath69 converges . thus , up to regularity at the origin , @xmath69 is our candidate normalizable mode . similarly , a combination of @xmath69 and @xmath72 is a candidate non - normalizable mode . again , its behavior at infinity indicates that it will couple to operators of dimension @xmath85 in ( [ eq : bbrel ] ) . for @xmath81 both modes are well - behaved at infinity and further examination is required to select the relevant fluctuating modes , just as in our discussion of the solutions in poincar coordinates . again , fields which have the solutions @xmath82 as their fluctuating modes will have solutions @xmath132 which act as source terms in ( [ eq : bbrel ] ) for the related operator of dimension @xmath133 in the dual cft . [ [ quantization - condition - for - normalizable - modes ] ] quantization condition for normalizable modes : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as we have discussed , regularity at the origin requires the choice of @xmath127 . this solution can be written as a linear combination of @xmath82 : @xmath134 where @xmath135 ( note that @xmath66 is assumed to be non - integral here . ) for @xmath80 , @xmath136 must vanish for a fluctuating solution because the norm of @xmath72 diverges at the boundary . this will happen if one of the gamma functions in the denominator has zero or a negative integer as its argument , if @xmath137 so this is the spectrum of normalizable modes @xmath69 for @xmath80 . for the same range of @xmath66 , non - fluctuating modes do not have a quantization condition . for a special set of frequencies @xmath138 the non - fluctating modes are purely of the @xmath72 type . we will see in the next section that such modes are in a highest weight representation . in general , however , the non - normalizable modes are simply the linear combination appearing in ( [ eq : lincomb ] ) . type . in fact , there is no need to impose such a condition . since the ads / cft correspondence should work when time is noncompact , we should be allowed sources with arbitrary time dependence . we would like to thank j. maldacena for emphasizing this to us . ] for @xmath81 , both @xmath69 and @xmath72 are potentially normalizable because both norms are well - behaved at the boundary . the quantization condition ( [ eq : hplusquant ] ) gives the spectrum of normalizable modes of the form @xmath69 . normalizable modes of the form @xmath72 are obtained by imposing instead : @xmath139 after picking one of these towers as the fluctuating modes , the remaining modes that are regular at the origin should be locked and mediate the bulk - boundary correspondence . as we have already discussed , the case when @xmath81 and @xmath72 is the fluctuating mode is confusing since the locked mode falls off faster at the boundary than the fluctuating mode . as a result , for the non - fluctuating mode to couple as source to an operator in the boundary theory with dimension @xmath84 , it must not contain any terms which behave as @xmath140 at the boundary . thus , its frequency must be quantized according to ( [ eq : hplusquant ] ) ; it seems that in this case one can not write a classical source with arbitrary behavior in global time . if we simply interpret the left - hand side of ( [ eq : bbrel ] ) as a path integral over field configurations which fall off as @xmath141 , then there will be no saddle points in the path integral when the time dependence of the boundary configuration is arbitrary . we leave the resolution of this conundrum for future work . upon imposing the conditions ( [ eq : hplusquant ] ) and ( [ eq : hminusquant ] ) for @xmath69 and @xmath72 respectively , the solutions may be written in terms of jacobi polynomials : @xmath142 it is easy to show that these quantized @xmath69 can be made orthonormal under the norm ( [ eq : adsnorm ] ) . for @xmath81 these quantized @xmath72 can be made orthonormal as well . as functions of @xmath119 the solutions @xmath144 are the same as before . because @xmath66 is the difference of the roots of the indicial equation , however , the solutions expanded near infinity are a little different than those listed above . @xmath69 is the same as before : @xmath145 again , the norm ( [ eq : adsnorm ] ) is well - behaved at infinity . for @xmath76 @xmath72 becomes : @xmath146\right)\ . \label{eq : nu - eq - zero}\end{aligned}\ ] ] here @xmath147 is defined in ( [ eq : hdef ] ) . note that this has a well - behaved norm at the boundary . for @xmath148 the second solution is : @xmath149 the norm of these solutions blows up at the boundary . [ [ quantization - condition ] ] quantization condition : + + + + + + + + + + + + + + + + + + + + + + + + once again we can start with the solution @xmath127 which is regular at the origin and examine its behavior at the boundary . the transformation laws are modified when @xmath66 is integral , so that : @xmath150 \right)\ . \label{eq : int - nu - psi}\end{aligned}\ ] ] once again , in order to isolate normalizable modes which fall off as @xmath141 at the boundary we must impose ( [ eq : hplusquant ] ) . at these frequencies , the gamma functions in the denominator of the coefficient of the final sum have negative integer argument ; so only terms in the sum which have compensating poles will survive . such poles will come from one of the final two @xmath151 functions ; thus , the logarithmic term drops out and the solution falls off at the boundary as desired , giving a series of modes built from @xmath69 . equivalently , one can easily show that when ( [ eq : hplusquant ] ) holds , @xmath69 is regular at the origin . as before , for general @xmath98 we have non - fluctuating modes which are well - behaved at the origin and which have a logarithmic part at infinity . a particularly interesting set of non - normalizable solutions occurs when @xmath152 for such frequencies the final sum in ( [ eq : int - nu - psi ] ) vanishes and the result is , as in the case of non - integral @xmath66 , a rational function in @xmath153 . we will see in the next section that these solutions are part of a special highest weight representation that exists for integral @xmath66 . the zoo of solutions that we have described in global coordinates should fall in various representations of the spacetime isometry / boundary conformal group . the fluctuating modes should clearly fall in unitary representations , and the boundary operators should create states ( in sectors for which the state - operator map is one - to - one ) which fall in such unitary representations as well . the non - normalizable modes do not have to fall in unitary representations of the conformal group , but we will see that they do lie in linear representations . since they couple to primary boundary operators and their conformal descendants , such representations are also important ; in order to understand the coupling of descendants we need to understand how the various representations combine . we will discuss mode solutions in @xmath0 , for which the representations of the isometry group are well - known ; see especially @xcite for a discussion and references . the highest - weight unitary representations in global coordinates and the continuous representations in poincar coordinates were discussed in @xcite . as we will see , the results of the previous section provide explicit expressions for the wavefunctions for all of the linear representations . structure in solutions to the wave equation in black hole backgrounds , see the review @xcite and the references therein . ] the irreducible representations of the @xmath160 algebra are well known @xcite . barut and fronsdal @xcite derived a set of linear representations which contain them ; we will follow their discussion as we will find that this more general set is important for our purposes . and @xmath176 , and the casimir is called @xmath177 . in this notation our generators are @xmath178 and @xmath179 ; our expression for the casimir is @xmath180 . ] the representations are indexed by 2 invariants . the first is @xmath112 ( called @xmath181 in @xcite ) which is related to the quadratic casimir : @xmath182 we will see that this is the same @xmath112 as defined in the previous section ; @xmath133 are the two solutions for a given value of the casimir . the second invariant is the fractional part @xmath183 of the spectrum of @xmath171 for a given representation ( here we use the same notation as in @xcite . ) representations are filled out by starting with a given vector in the representation and acting on it an arbitrary number of times with @xmath184 . the resulting representations are : * @xmath185 . this an irreducible , infinite - dimensional representation but does not have a highest or lowest weight state . @xmath112 and @xmath183 are not related ; the only condition is that @xmath186 is not an integer . @xmath172 does not even have to be real , but since we wish to describe stable modes in spacetime we will not consider complex values . we can define @xmath187 without losing generality ; the spectrum is @xmath188 for @xmath189 an arbitrary integer . for fixed @xmath190 and @xmath183 , the representations for each branch of ( [ eq : casimir ] ) are equivalent . the non - unitary representations will correspond to non - normalizable modes for which the energies are not related to the mass , i.e. for which @xmath98 is not quantized in even integers above @xmath191 . imposing unitarity restricts @xmath185 to two types of representations : 1 . @xmath192 the `` principal series '' occurs for @xmath193 ; thus @xmath194 . @xmath195 will correspond to unstable modes in spacetime as noted in @xcite . @xmath196 the `` supplementary series '' occurs for @xmath197 . here @xmath112 is real and @xmath198 . this occurs in the range @xmath199 discussed in the previous section . * this is an irreducible , infinite - dimensional highest weight representation and exists for @xmath201 . @xmath202 and the spectrum is @xmath188 for integral @xmath203 ; the highest weight @xmath204 state is annihilated by @xmath173 . the representation will be unitary if @xmath205 . these states correspond to the solutions in global coordinates quantized according to ( [ eq : hplusquant ] ) for general @xmath66 . for non - integral @xmath66 solutions quantized according to ( [ eq : hminusquant ] ) also transform in this representation . the solutions @xmath69 have @xmath206 and live in a unitary positive energy representation . the solutions @xmath72 have @xmath207 ; they live in a non - unitary representation for @xmath80 and in a unitary representation for @xmath81 . the derivation of this representation in @xcite shows that it can be imbedded in a reducible , nondecomposable representation where @xmath189 is an arbitrary integer . in this representation we can reach @xmath208 by starting with negative @xmath189 states and acting on them repeatedly with @xmath174 ; however , once we examine states in @xmath208 we can not leave the irrep with actions of @xmath184 because of the highest weight state . * @xmath209 . this is an irreducible , infinite - dimensional lowest weight representation and again exists for @xmath210 . @xmath211 and the representation is unitary for @xmath205 . the spectrum is @xmath212 for integral @xmath203 . the lowest weight @xmath204 state is annihilated by @xmath174 . these are the negative energy modes corresponding to @xmath208 . @xciteimbeds this in a reducible , nondecomposable representation which contains energies larger than that of the lowest weight ( highest energy ) state . * this is an irreducible , finite - dimensional representation . it occurs when @xmath214 and @xmath215 . its spectrum is @xmath216 for integral @xmath217 . the representation is only unitary in the case @xmath218 , i.e. for the identity representation , also known as the singleton @xcite . it is contained in a reducible nondecomposable representation for which @xmath189 is arbitrary . @xmath213 arises in @xmath0 as @xmath72 for integral @xmath66 . in global coordinates , the case @xmath219 , @xmath220 for integral @xmath98 corresponds to a tensor product @xmath221 transforming under @xmath222 . for arbitrary @xmath223 nondecomposable representations containing this irrep can occur . in the next subsection we will explicitly discuss how these representations are realized as solutions to the wave equation in @xmath0 in global coordinates . in global coordinates the @xmath0 metric is : @xmath224 \label{eq : metglob}\ ] ] and a scalar field of mass @xmath33 has a wave equation @xmath225 where : @xmath226 in these coordinates , a convenient basis for @xmath168 is : @xmath227 starting with generators in ( [ eq:4dgens ] ) and using the coordinate patch in appendix [ app:3coords ] that yields the global metric , it is easy to work out the explicit representation : @xmath228 \label{eq : glob-}\\ l_+ & = & i e^{iw } \left [ { \cosh(2\mu ) \over \sinh(2\mu ) } \partial_w - { 1 \over \sinh(2\mu ) } \partial_{{\bar{w } } } - { i \over 2 } \partial_\mu \right ] \label{eq : glob+}\end{aligned}\ ] ] where @xmath229 and @xmath230 . the generators @xmath231 , @xmath232 of @xmath233 are obtained by exchanging @xmath234 and @xmath235 in ( [ eq : glob0][eq : glob+ ] ) . clearly @xmath236 generate time translations and rotations . [ [ discussion - of - explicit - solutions ] ] discussion of explicit solutions : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the dalembertian for scalar fields is given in terms of the left and right casimirs as : @xmath237 the highest weight states @xmath238 and @xmath239 are simplest to describe . we require that @xmath240 , and this imposes @xmath241 . using this , the dalembertian in ( [ eq : caswave ] ) acting on highest weight states reduces to : @xmath242 explicit solutions of the equation @xmath243 give @xcite : @xmath244 so that @xmath112 in this section and in the previous section are the same . for @xmath245 non - integral , @xmath246 are precisely the minimum - energy normalizable modes found in sec . [ sec : globsoln ] ( here @xmath247 as in appendix [ app:3coords ] ) . @xmath248 are the non - normalizable modes of lowest energy in the spectrum ( [ eq : nonnormquant1 ] ) . other non - fluctuating modes will live in non - highest - weight representations . descendant states are constructed on the primary @xmath249 by the action of @xmath250 and have weights @xmath251 and @xmath252 . examining the differential operator @xmath174 shows that all of these solutions have the same boundary behaviour as the primary states and therefore share their normalizability properties . finally , @xmath253 is the generator of time translations , and @xmath254 is the generator of rotations , so that the frequency and angular momentum are given by @xmath255 and @xmath256 . the spectra of the two towers of states are given by : @xmath257 this matches the spectra for the @xmath258 solutions found in sec . [ sec : globsoln ] . for integral @xmath66 , the situation is the same for representations built on @xmath85 ; the modes living in this representation are again the normalizable solutions @xmath69 . the non - fluctuating states will fall into two types . if @xmath98 is not quantized according to ( [ eq : intnonnorm ] ) , the solutions will live in the representations @xmath259 . if the energies are quantized according to ( [ eq : intnonnorm ] ) , then the solutions @xmath72 will live in nondecomposable representations containing the finite - dimensional representations @xmath260 . the lowest energy ( highest weight ) state in @xmath261 has @xmath262 and is given by @xmath263 the highest energy ( lowest weight ) state in @xmath261 has @xmath264 and is given by @xmath265 which is annihilated by @xmath174 and @xmath266 . one may also find these modes in the manner described in sec . [ sec : intmodes ] . the case @xmath267 is quite straightforward and interesting . in addition to the ( unitary ) identity representation at @xmath219 there are solutions with @xmath268 : @xmath269 one may apply @xmath184 to reach other such solutions . for general @xmath223 , applying @xmath184 to @xmath270 gives us the state @xmath271 . applying @xmath173 to @xmath272 gives the irreducible identity representation @xmath273 . applying @xmath274 annihilates all states @xmath275 . applying @xmath276 to such states brings one to the normalizable set of solutions forming the irrep @xmath277 . there are similar solutions for @xmath278 and states generated from these by applying the raising and lowering operators . the full representation in this example is a nondecomposable representation and was discussed in ref . @xcite . such representations were discussed for @xmath279 in @xmath20 in refs @xcitefrom a somewhat different point of view . in summary , we have found that every solution to the wave equation which is well - behaved in the bulk lies in a representation of the bulk isometry / boundary conformal group @xmath280 . the energy is simply the @xmath171 eigenvalue , and the singularity or zero of the wavefunction at infinity is related to the invariant @xmath112 of the representation . the map is quite natural . normalizable modes ( plus the singleton ) correspond to unitary representations of @xmath16 ; it makes sense to quantize these modes . non - normalizable modes correspond to non - unitary representations of @xmath16 . these are the modes we wish to keep as non - fluctuating backgrounds . [ [ relation - to - conformal - field - theory - on - the - cylinder ] ] relation to conformal field theory on the cylinder : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as @xmath281 , the generators @xmath168 acting on surfaces of fixed @xmath282 become : @xmath283 these generators and their @xmath233 companions which replace @xmath234 by @xmath235 are the standard conformal symmetries of the cylinder . acting on descendants of a primary state @xmath284 with weight @xmath112 , the generators at the boundary become : @xmath285 the shift of @xmath184 by @xmath286 arises from the radial derivatives in the bulk generators . in poincar coordinates the @xmath0 metric and dalembertian are : @xmath287 as discussed in appendix [ app:3coords ] these coordinates only cover a region of @xmath0 and there is a horizon at @xmath288 . a convenient basis for @xmath168 in these coordinates is : @xmath289 starting with the explicit generators in ( [ eq:4dgens ] ) and implementing the poincar coordinates in appendix [ app:3coords ] yields the generators : @xmath290 \label{eq : p+}\end{aligned}\ ] ] here @xmath291 and @xmath292 . the generators @xmath164 of @xmath233 simply exchange @xmath103 and @xmath293 . [ [ translations - and - cft - on - the - plane ] ] translations and cft on the plane : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + it is instructiive to examine the action of the generators ( [ eq : p0 ] ) ( [ eq : p+ ] ) on surfaces of constant @xmath294 at the boundary @xmath24 : @xmath295 these are the standard generators of conformal transformations on the plane . ( the factors of @xmath296 and @xmath297 arise because we are in minkowski space and @xmath103 is a dimensionful coordinate . ) in the case of the cft on the cylinder , @xmath171 generated translations . @xmath174 generates translations on the plane , while @xmath171 generates dilatations . furthermore , the basis ( [ eq : plin ] ) for @xmath160 is different from the basis ( [ eq : globbasis ] ) we used for global coordinates . as discussed in the appendix , the boundary of the patch of spacetime covered by poincar coordinates is conformal to minkowski space and we have chosen the corresponding natural basis for @xmath171 and @xmath184 . it is important to emphasize that we are _ not _ dealing with the standard bijective map between cfts on the cylinder and the plane . the plane that appears at the poincar boundary is merely a patch of the cylinder and can only be expected to describe the theory on the cylinder in a thermal sense , after tracing over some degrees of freedom . the poincar mode solutions of sec.[sec : poinsoln ] with fixed frequency @xmath98 and momentum @xmath298 are eigenstates of @xmath174 and @xmath266 with eigenvalues @xmath299 ( see also @xcite ) . @xmath174 generates a non - compact subgroup of @xmath160 @xcite ; thus the wavefunctions we get by diagonalizing this operator should be continuously moded , as we have found . [ [ representing - the - isometries ] ] representing the isometries : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we can also construct a set of scalar fields in poincar coordinates that carry well - defined weights under @xmath171 and @xmath300 . again , the dalembertian is the sum of left and right casimirs : @xmath301 where we have used the definition in ( [ eq : casdef ] ) of @xmath160 casimirs . states of weight @xmath302 under @xmath171 and @xmath300 must satify the equations : @xmath303 since @xmath171 and @xmath300 are linear in derivatives , a product of eigenstates of these operators is still an eigenfunction . a general class of eigenstates is given by : @xmath304 the left and right weights are then : @xmath305 , -[c + ( a + d + e)/2 ] \right)\ ] ] in global coordinates the eigenstates of @xmath171 essentially provided a fourier basis for mode expansions . here the @xmath171 eigenstates provide an expansion in a power series of functions . to find highest weight ( primary ) states we want to additionally solve the equations @xmath243 . simultaneous solution of these conditions requires that @xmath306 implying that @xmath307 and imposing symmetry between @xmath103 and @xmath293 in primary states . using @xmath306 and the equation for @xmath308 in @xmath309 gives : @xmath310 which is easily solved to give : @xmath311 it can be explicitly checked that this is a primary state with weights @xmath312 . requiring that the dalembertian acting on this solution has eigenvalue @xmath313 yields the condition : @xmath314 exactly as in the case of global coordinates . for @xmath315 and @xmath316 the solutions with @xmath317 vanish at the boundary ( @xmath318 ) and at the horizon ( @xmath319 ) while the the @xmath84 solutions diverge in both locations . this agrees with the claim in sec . [ sec : norm ] that the @xmath84 representation alters the boundary conditions but does not fluctuate . furthermore , it suggests that in poincar coordinates the horizon should also be treated as a boundary with which flux can be exchanged . indeed , it is well known that quantum field theory in a spacetime with horizon requires the specification of horizon boundary conditions . for @xmath320 the situation is more disturbing - the @xmath85 solution is singular in the bulk of spacetime at @xmath321 . this appears to be a pathology that arises because the surface @xmath322 is a fixed point of @xmath171 , the generator of dilatations . the eigenstates of @xmath171 are accordingly singular on this surface . the original motivation for this work was the desire to study quantum gravity in the bulk spacetime from the perspective of the boundary gauge theory . in particular , we would like to study the appearance of spacetime singularities and horizons ( see @xcite , for example ) which we expect to be related to nonperturbative issues in the gauge theory . a preliminary step is to determine whether we can say anything about local spacetime physics from the boundary perspective . the existence of the normalizable modes displayed in this article implies that there is a natural hilbert space of small fluctuations around the bulk ads background . these fluctuating modes are the probes of bulk causal structure . since we can map such states into the boundary hilbert space , we might expect that there is an analysis of local bulk processes from the boundary point of view . on the other hand it seems that we can not reconstruct position space correlation functions in terms of the correspondence as written in ( [ eq : bbrel ] ) . to see this let us expand both sides of ( [ eq : bbrel ] ) in a formal series in the boundary field @xmath21 . for the moment we will work in euclidean space where this is a well - defined procedure since the bulk field @xmath19 is uniquely determined by @xmath21 . ( we will return to the lorentzian version below . ) as pointed out in @xcite , @xmath19 can be expressed in terms of @xmath21 via the equation ( here , in poincar coordinates ) : @xmath323 where @xmath324 is a solution to the wave equation behaving as a delta function times a given power of @xmath325 on the boundary . the quadratic piece of the action can be written as : @xmath326 @xmath327 is the inverse spacetime propagator for @xmath19 , which we would like to extract from the boundary theory . this piece can be written as a quadratic expression in @xmath21 via ( [ solution ] ) . expanding ( formally ) ( [ eq : bbrel ] ) to quadratic order , we find : @xmath328 extracting @xmath327 from this would require `` inverting '' @xmath324 , which seems impossible . the integration over @xmath329 washes out any information about localization in the direction perpendicular to the boundary . essentially , this point was made in ref . @xcite the correspondence ( [ eq : bbrel ] ) relates off - shell operators in the boundary to on - shell fields in the bulk . since the @xmath294 dependence of the latter depends on the @xmath330 dependence , full localization is not possible . the point is that instead of being able to reconstruct arbitrary off - shell correlators , we must be content with a description of on - shell quantities in the bulk . indeed , our knowledge of the mapping between the bulk and boundary hilbert spaces and their hamiltonians implies that transition amplitudes between arbitrary _ physical _ states are computable in either of the dual descriptions . off - shell physics can be probed to the extent that on - shell correlators receive contributions from off - shell modes in intermediate states . this situation will force the boundary analysis of the bulk geometry to be somewhat subtle and indirect , but still possible in principle . in addition to correlation functions , we want to be able to ask how the vacuum in the bulk string theory relates to the boundary field theory . this question really concerns the global causal structure the existence of different natural vacua in a spacetime often reflects the presence of horizons , for example . understanding this is also relevant to the study of possible singularities in black hole backgrounds from the dual gauge theory point of view . it may be that although local bulk physics is quite difficult to examine via the boundary field theory , the global causal structure may somewhat easier to address . the presence of normalizable modes in lorentzian ads might appear to render the correspondence ( [ eq : bbrel ] ) ambiguous , since there is no preferred solution corresponding to a given boundary value . however , there is a natural prescription - we must sum the effective action over all spacetime backgrounds with the same boundary behavior , with a weighting which depends upon the state of the system . we can generalize the calculations in @xcitethat we outlined above , by appropriately modifying ( [ solution ] ) : @xmath331 where @xmath332 is an arbitrary normalizable solution of the field equations . here @xmath324 is a particular solution we have chosen which solves the wave equation and has the same behaviour at infinity as in the euclidean case . @xmath12 can be written as : @xmath333 ^ 2 + m^2 \phi^2\ . \label{msolution}\ ] ] upon integrating by parts , the bulk term vanishes . it is easy to see that upon including the correct measure factors , the non - vanishing boundary terms are : @xmath334\ .\ ] ] when summing over field histories in the lorentzian path integral , one must also specify a state at early and late times in the form of wavefunctionals @xmath335 $ ] . a path integral obtained by continuation from euclidean signature will pick out vacuum wavefunctionals , since the @xmath336 prescription damps out all excited states . in such a case , the normalizable mode @xmath332 is set to zero and the boundary contribution to the action in ( [ msolution ] ) vanishes . more generally , one could choose to study excited states at early and late times by explicitly including the appropriate wavefunctionals ; in this case the boundary terms become physically relevant and give contributions to correlation functions evaluated in excited states . for calculations in interacting theories , the effects of the ambiguity can be more pronounced . in euclidean ads , a dramatic example of the effect of multiple saddlepoints with same boundary behaviour is the high - temperature transition between ads space and the ads black hole discussed in sec . ( 3.2 ) of @xcite . our normalizable modes represent a large class of saddlepoints of the bulk action in lorentzian ads and , in an interacting theory , they encode the non - trivial dynamics of the bulk . in this work we have developed a lorentzian signature version of the bulk - boundary correspondence . this required understanding the respective roles played by normalizable and non - normalizable modes . the two sets of modes emerge naturally , either from direct solution of the field equations or from the field representations of the ads isometry group . the non - normalizable modes act as backgrounds and couple to local operators in the boundary description , while normalizable modes describe fluctuations in the bulk . the picture we have presented suggests several avenues for the study of black hole spacetimes from the boundary perspective . black holes can be constructed in ads spaces @xcite by making discrete identifications of the geometry . unlike pure ads , the resulting bulk spacetime has global horizons and singularities in the classical supergravity approximation . the question is whether and how the boundary theory describes the interior of the black hole . one would hope that the hilbert space of states within the black hole is identified with a sector of states in the boundary theory ; this would realize a form of the black hole complementarity advocated by t hooft and susskind . in our picture , this issue would be studied by considering the roles of the normalizable and non - normalizable solutions to the wave equation in the black hole spacetime . we have enjoyed discussions with e. gimon , s. kachru , e. keski - vakkuri , r. leigh , e. martinec , s. mathur , j. schwarz , e. silverstein , a. strominger and c. vafa . we are also grateful to o. aharony , j. maldacena , r. helling , r. myers , and s. ross for useful comments on the original preprint version of this work . thanks stanford university for their hospitality during the course of this work . is supported by the harvard society of fellows and by the nsf under grant nsf - phy-92 - 18167 . is supported by in part by doe grant de - fg03 - 92-er40701 and by the dubridge foundation . is supported by nsf grant nsf - phy-92 - 18167 . global coordinates for @xmath0 are defined by : @xmath339 these yield the metric : @xmath224 \label{eq : globalmet}\ ] ] here @xmath340 , @xmath341 and @xmath342 . we unwrap @xmath36 to have range @xmath343 to @xmath344 in order to work on @xmath345 , the universal cover of @xmath0 . from the above we see that @xmath0 has the topology of a disk times a line . the boundary of spacetime at @xmath100 is a cylinder @xmath349 . ( see fig . [ fig : one ] ) the bulk - boundary correspondence asserts that a conformal field theory on this cylinder is dual to quantum gravity in the bulk . a penrose diagram illustrating the causal structure can be drawn by drawn by considering a two dimensional cross section of the hyperboloid . we choose to display a @xmath350 slice , and obtain the diagram in fig . [ fig : two]a . poincar coordinates are defined by : @xmath351 giving the metric : @xmath352 here @xmath36 and @xmath35 range between @xmath343 and @xmath344 , and @xmath353 . poincar coordinates only cover one half of @xmath0 , as shown in fig . [ fig : one ] . there is a horizon in these coordinates at @xmath319 . the boundary at @xmath354 is clearly conformal to flat minkowski space @xmath355 . the bulk - boundary correspondence asserts that a conformal field theory on this boundary plane is dual to quantum gravity in the bulk . note however , that the plane only covers half of the cylindrical boundary of global @xmath0 . the second half of @xmath0 displayed in fig . [ fig : one ] can be covered by labelling the hyperboloid as in ( [ poincare ] ) but now letting @xmath356 . these two patches together cover @xmath0 , while to cover @xmath345 one assembles a vertical tower of such patches . the penrose diagram is displayed in fig . [ fig : two]b . for completeness , we consider a third coordinate system which is useful for constructing the btz black hole . divide the hyperboloid into three regions : @xmath357 note that one can not take both @xmath358 and @xmath359 to be positive . we then cover each region with four coordinate patches . @xmath360 in all three regions , @xmath363 and @xmath364 range between @xmath343 and @xmath344 . @xmath365 has range @xmath366 in region 1 , @xmath367 in region 2 , and @xmath368 in region 3 . so altogether there are twelve patches covering @xmath0 . to draw a penrose diagram we again consider the slice @xmath350 . note that on this slice region 3 is a dimension one submanifold , while regions 1 and 2 are dimension two submanifolds . thus only regions 1 and 2 will appear in the penrose diagram , and we label the various patches as @xmath369 , @xmath370 in an obvious notation . the penrose diagram then appears as in fig . [ fig : two]c . the various coordinate systems defined above generalize straightforwardly to arbitrary dimensions . global coordinates for @xmath20 give the metric : @xmath374\ ] ] with @xmath347 , @xmath375 . thus @xmath20 is globally a d - dimensional disk times a line and the boundary at @xmath376 is a cylinder @xmath377 . the dalembertian operator in global coordinates is @xmath378 poincar coordinates yield a metric : @xmath379 here @xmath380 is the flat metric on @xmath381 and @xmath382 . there is a horizon at @xmath288 and the boundary at @xmath24 is the plane @xmath383 . the dalembertian operator in poincar coordinates is @xmath384 s.r . das and s.d . mathur , `` comparing decay rates for black holes and d - branes '' , * b478 * , 561 - 576 , hep - th/9606185 . j. maldacena and a. strominger , `` universal low - energy dynamics for rotating black holes '' , * d56 * , 4975 - 4983 , hep - th/9702015 . klebanov , `` worldvolume approach to absorption by nondilatonic branes '' , * b496 * , 231 - 242 , hep - th/9702076 . douglas , j. polchinski and a. strominger , `` probing five - dimensional black holes with d - branes '' , electronic journal jhep 12 , hep - th/9703031 . gubser , i.r . klebanov and a.a . tseytlin , `` string theory and classical absorption by three - branes '' , * b499 * , 217 - 240 , hep - th/9703040 . m. cvetic and f. larsen , `` black hole horizons and the thermodynamics of strings '' , nucl . suppl * 62 * , 443 - 453 , hep - th/9708090 . p. breitenlohner and d.z . freedman , `` positive energy in anti - de sitter backgrounds and gauged extended supergravity '' , * b115 * ( 1982 ) 197 ; `` stability in gauged extended supergravity '' , * 144 * ( 1982 ) 249 . n. seiberg , `` notes on quantum liouville theory and quantum gravity '' , prog . theor . phys . suppl . * 102 * , 319 ; lectures given at the 1990 cargese meeting on random surfaces , quantum gravity , and strings ; and at the yukawa insitute seminar on common trends in mathematics and quantum field theories . t. banks and m.b . green , `` non - perturbative effects in @xmath386 string theory and @xmath387 susy yang - mills '' , rutgers preprint ru-98 - 17 , cambridge preprint dampt-98 - 25 , ucsb / itp preprint nsf - itp-98 - 054 ; hep - th/9804170 . m. banados , c. teitelboim , and j. zanelli , `` the black hole in three - dimensional space - time '' , phys . lett . * 69 * 1849 , hep - th/9204099 ; m. banados , m. henneaux , c. teitelboim , and j. zanelli , `` geometry of the ( 2 + 1 ) black hole '' , * d*48 1506 , gr - qc/9302012

this paper presents a physical mapping tool for quantum circuits , which generates the optimal universal logic block ( ulb ) that can , on average , perform any logical fault - tolerant ( ft ) quantum operations with the minimum latency . the operation scheduling , placement , and qubit routing problems tackled by the quantum physical mapper are highly dependent on one another . more precisely , the scheduling solution affects the quality of the achievable placement solution due to resource pressures that may be created as a result of operation scheduling whereas the operation placement and qubit routing solutions influence the scheduling solution due to resulting distances between predecessor and current operations , which in turn determines routing latencies . the proposed flow for the quantum physical mapper captures these dependencies by applying ( i ) a loose scheduling step , which transforms an initial quantum data flow graph into one that explicitly captures the no - cloning theorem of the quantum computing and then performs instruction scheduling based on a modified force - directed scheduling approach to minimize the resource contention and quantum circuit latency , ( ii ) a placement step , which uses timing - driven instruction placement to minimize the approximate routing latencies while making iterative calls to the aforesaid force - directed scheduler to correct scheduling levels of quantum operations as needed , and ( iii ) a routing step that finds dynamic values of routing latencies for the qubits . in addition to the quantum physical mapper , an approach is presented to determine the single best ulb size for a target quantum circuit by examining the latency of different ft quantum operations mapped onto different ulb sizes and using information about the occurrence frequency of operations on critical paths of the target quantum algorithm to weigh these latencies . experimental results show an average latency reduction of about 40% compared to previous work .

1501.02524

it has been speculated that quantum computers , which harness physical phenomenon unique to quantum mechanics ( especially quantum interference and superposition ) to realize a fundamentally new mode of information processing , can overcome the limitation of classical computers in efficiently solving hard computational problems . in particular , polynomial - time algorithms have been offered to solve several np - intermediate problems by using quantum computers @xcite . a key challenge in building large - scale quantum computers is the environmental noise - more precisely , loss of quantum information due to unwanted interactions between the quantum system and its environment . this phenomenon is called _ quantum _ _ decoherence_. it has been shown that arbitrarily accurate quantum computation is possible , provided that the error per operation is below a _ threshold _ value @xcite and @xcite . in other words , quantum error correction works if the level of imperfections and noise in quantum gate operations and measurements are below a certain _ error threshold _ and if corrections can be applied repeatedly @xcite . based on the error threshold for each quantum circuit fabric technology , an appropriate encoding technique ( such as m - level concatenated steane code @xcite ) is selected to achieve fault tolerant realization of the quantum circuit . a typical quantum circuit fabric consists of a two - dimensional array of identical primitive structures ( called _ primitive cells _ in this paper ) , each structure containing some sites for generating / initializing qubits , measuring them , performing operations on one or two qubits , and channels for moving qubits or swapping their information . [ fig : fig3 ] for an example of such an array of primitive cells . unfortunately , dealing directly with this primitive cell array is very cumbersome and unwieldy . so in practice another 2-d array of super - templates ( which we call _ tiles _ ) is built . each tile comprises a number of primitive cells . instead of mapping a quantum circuit to the quantum fabric , the quantum circuit is mapped to this tiled architecture ( see below ) . a quantum logic synthesis tool ( surveyed in reference @xcite ) generates a quantum circuit . every qubit in the output is called a _ logical qubit _ , which is subsequently encoded into several _ physical qubits _ to detect and correct potential errors . to prevent the propagation of errors in the quantum circuit , the ( reversible ) _ logic gates _ in the synthesized circuit ( which are typically , not , cnot and toffoli gates ) must be converted into ft _ quantum operations_. a possible universal ( but redundant ) set of ft quantum operations includes cnot , h ( hadamard ) , t ( @xmath0/4 rotation ) , s ( phase ) , x , y and z gates . implementation of these ft quantum operations depend on the chosen error correction method . note that the set \{cnot , h , t } constitutes a universal basis for quantum circuit realization the other instructions are included to allow logic simplification during the process of converting the logic synthesis output to the ft quantum operation realization . a given quantum circuit fabric is natively capable of performing a universal set of one and two - qubit instructions ( also called _ native quantum instructions _ ) . this set differs among various quantum technologies ( e.g. , ion - trap vs. superconducting qubit vs. quantum dots ) . each ft quantum operation can be implemented by using a composition of these native quantum instructions . the transformation from logical gates ( which are the result of the quantum logic synthesis ) to the ft quantum operations and from the ft quantum operations to the native quantum instructions are called _ ft quantum synthesis _ and _ native quantum synthesis _ , respectively . ft and native quantum syntheses are outside of the scope of the present paper . each of the ft quantum operations performs a desired function on one or two logical qubits as the input producing one or two logical qubits as the output ; each of the input qubits is encoded with some number of physical qubits . the output qubits will also be encoded . moreover , each of these ft quantum operations requires subsequent _ syndrome extraction _ circuitry to detect and correct any errors ( up to a certain number ) that may have been introduced by the quantum operation itself . based on the adopted encoding scheme , implementation of each of the ft quantum operations may require hundreds to tens of thousands of native quantum instructions . different works in the literature e.g. , @xcite and @xcite have suggested using the _ tiled quantum architecture _ ( tqa ) , comprised of a regular two - dimensional array of _ universal logic blocks _ ( ulbs ) to avoid dealing with this complexity . notice that each ulb in the tqa is capable of performing _ any _ ft quantum operation . ulbs are separated by routing channels , which are used to move logical qubits ( or information about these qubits ) from some source ulbs to a target ulb in the tqa . a pictorial representation of the tqa is shown in fig . [ fig : fig1 ] . the quantum structures placed at the junctions of routing channels may be thought of as _ quantum crossbars _ ( possibly with some _ qubit purification _ @xcite capability ) . routing channels and quantum crossbars are also built from primitive cells . a ulb is analogous to a _ configurable logic block _ ( clb ) in a field programmable logic array ( fpga ) device , in that it can implement any of a set of desired functions . moreover , the same ulb ( as identified by its unique row and column indices in the ulb array ) can be configured to perform different ft quantum operations at different times as needed . this is analogous to an _ on - the - fly - reconfigurable clb_. trapped - ion technology is a very promising technology for implementing quantum circuits @xcite . hence , it is selected as the underlying technology for our quantum physical mapping tool . see section [ qcrep ] for more details of this technology . a trapped - ion quantum circuit fabric comprises of small quantum cells that are organized in a two - dimensional grid , similar to the tqa . although each primitive cell has the complete set of capabilities provided by the trapped - ion technology , it is not large enough to realize a complete ft operation with an acceptable latency . that is why we resort to the ulb abstraction . notice that each ulb itself is simply a fixed - size 2-d array of primitive cells , which can realize any ft operation with an acceptable latency . after appropriate high - level transformations , a quantum algorithm may be represented by a _ quantum operation dependency graph _ ( qodg ) , where nodes represent ft quantum operations and edges capture data dependencies . based on the target quantum fabric and error threshold , a particular quantum coding scheme is selected , and subsequently , a high - level tool maps the qodg into a tqa , where each ulb ( tile ) in this architecture can implement any desired one or two - input quantum operation in a fault - tolerant way . the latency of the quantum algorithm mapped to the tqa can be calculated as the length of the longest path in the qodg , where the length of an input - output path in the qodg is the summation of operation latencies plus routing latencies . the operation latencies are dependent on how precisely the required operation is realized within a ulb . this realization comprises static provisioning ( setting the size of the ulb in terms of an array of primitive structures in the target quantum fabric ) and mapping of the target operation into the ulb . on the other hand , the routing latencies are dependent on how long it takes for the logical qubits to move from source ulbs to a target ulb to perform the ft quantum operation . the routing latency , which depends on the actual move distances and the latency of moving a qubit through some link in the tqa , is highly sensitive to ulb size , operation placement , and qubit routing on the tqa . the focus of this paper is on finding the optimal ulb size for the tqa in the trapped - ion technology . the most important part of ulb design is to minimize the operation latency for a fixed ulb size by optimally mapping the quantum instructions to the underlying quantum fabric . this action is called _ quantum physical mapping_. the quantum physical mapping procedure is decomposed into three sub - procedures : ( a ) quantum instruction scheduling , ( b ) quantum instruction placement , and ( c ) dynamic qubit routing . next , based on operation usage frequencies in the target quantum algorithm and the average movement of logical qubits per operation in the tqa , the optimal size of the ulb is determined . each quantum fault - tolerant operation and its required syndrome extraction circuit can be modeled by a _ quantum instruction dependency graph _ ( or qidg ) . a qidg for an ft quantum operation is what a qodg is to a quantum algorithm . as an example , a cnot operation between two physical qubits may be implemented by using more than one native quantum instructions in the trapped - ion technology . moreover , this number of native quantum instructions is multiplied by a factor of seven if 7-qubit steane code [ [ 7,1,3 ] ] is used to implement the cnot in a fault - tolerant manner . adding syndrome extraction circuit for logical qubits increases the number of native quantum instructions even more . these numbers are far larger in the case of more complex ( i.e. , non - transversal ) ft quantum operations such as t operation in the [ [ 7,1,3 ] ] steane code , or more complex quantum encoding schemes such as the _ concatenated _ 7-qubit steane code . different solutions for quantum physical mapping are introduced in the literature see for example , @xcite . in this work , an iterative approach is proposed to schedule and place the instructions on quantum fabric to capture the relation between scheduling level of the instructions , instruction placement solution and qubit routing latencies . the proposed algorithms in each step are based on the state - of - the - art scheduling and placement solutions adding the unique characteristics of the quantum instructions scheduling and placement . to capture the quantum no - cloning theorem that forbids fan - out in quantum circuits , a pre - processing step is implemented on the qidg to resolve the multiple read dependencies between instructions by adding auxiliary edges between them considering the criticality factor of each instruction . a force - directed instruction scheduling approach is used as the scheduling algorithm to minimize the contention , i.e. the number of concurrent quantum instructions , and hence reduces the routing latency due to congestion among qubits . quantum instructions can only be executed at specific locations on the quantum circuit fabric called _ interaction wells_. the placement of the instructions in these interaction wells affects their start time due to the qubit routing latencies . hence , we present a placement approach based on a state - of - the - art placement engine with adaptive re - levelization of the operations . more precisely , the placement algorithm calls the quantum instruction scheduler several times to update the scheduling levels of instructions based on the current quantum placement solution . these algorithms are implemented as part of a quantum cad tool called _ quantum ulb factory designer _ ( qufd ) . to evaluate the performance of the proposed quantum instruction scheduling and placement algorithms , a quantum fabric emulator including a greedy router is developed . the result of the proposed algorithms with respect to previous work is evaluated . on average , 40% improvement with respect to the state - of - the - art quantum mapper is observed . moreover , considering toffoli profiling information , the best aspect ratio for a ulb to achieve the best latency is calculated and explained in the experimental section . the remainder of this paper is organized as follows . reviews of the most relevant prior work are presented in section [ prior ] . background on the quantum circuit representation and trapped - ion fabric is presented in section [ qcrep ] . the proposed flow for ulb design is presented in section [ cadflow ] . quantum physical mapping tool is discussed in section [ physicalmapping ] . quantum instruction scheduling and placement problem formulation and proposed algorithms are presented in sections [ sched ] and [ placement ] . experimental results are presented in section [ results ] and the paper is concluded in section [ concl ] . qufd is not the first tool of its kind , but it dramatically improves the performance of existing tools . the rest of this section gives a survey on previous quantum physical design tools and summarizes the contributions and unique features of qufd compared to the prior art . balensiefer et al . @xcite developed quale , a tool for designing microarchitectures for trapped - ion quantum computers . this tool employs as late as possible scheduling on qodg , greedy qubit placement , and pathfinder @xcite based routing . the key disadvantage of the proposed placement in @xcite is that it is independent of the structure of the given qodg . hence , two qubits that have a lot of interactions may be placed far from each other . this increases the routing cost of bringing the two qubits together to perform the required gate level operations . moreover , it increases the congestion in the routing channels . qpos @xcite uses a similar flow as quale , but distinguishes between the source and destination operands of a two - qubit instruction during the routing step . this tool employs a combination of list scheduling and as soon as possible scheduling methods for quantum instruction scheduling and a priority - based routing algorithm for qubit routing . reference @xcite tweaks qpos by creating a priority queue for instructions based on the total delay of their dependent instructions . reference @xcite develops a quantum mapping tool and compares different architectures for quantum fabrics . the main focus of @xcite is to improve the quality of high - level mapper . however , it lacks the aggressive optimization aspect of tile ( ulb ) design . we believe that this optimization dramatically affects the total latency of the final quantum circuit . reference @xcite discusses the notion of the classical and quantum placement including differences and similarities . the authors propose heuristics to place the quantum instructions and route the qubits by applying the swap instruction to minimize the total latency of the quantum circuit . the proposed heuristics in @xcite consider fixed instruction scheduling solution and does not consider the global instruction placement solution . fixed instruction scheduling means the order of instructions are determined before determining the placement and routing solutions which limits the effect of scheduling algorithm solution on the circuit s latency . moreover , considering the global placement solution reduces the possibility of being trapped in a local optimum . reference @xcite proposes a physical synthesis flow for ancillary qubit selection to reduce the total latency of the quantum physical mapping . this work is focused on changing the design of the fabric to decrease the circuit latency ( similar to @xcite ) . a mapping tool for trapped - ion technology , called qspr , is proposed in @xcite . this tool supports ion multiplexing in the trapped - ion technology ; it offers a more optimized global placement of the qubits on the quantum circuit fabric by using an iterative placement approach based on forward and backward computations in a qidg ; and it improves the routing solution by simultaneously moving the source and destination qubits toward a designated site . it provides about 41% improvement over quale . we consider this mapping tool the - state - of - the art technique and compare our results with it . the proposed greedy method for instruction placement in this approach can not decrease the latency in large quantum circuits because the impact of initial optimized qubit placement fades after running of few instructions ; in other words , qubit placement can only enhance the routing delay for some of the first instructions and has no effect on later instructions . the proposed quantum physical mapper and ulb designer in our work aim to automate ulb design in quantum mapping tools to decrease the total latency of the quantum circuit realization as much as possible . the proposed quantum physical mapper considers the effect of the routing time on quantum instruction scheduling and placement and presents a cross - layer optimization between these solutions to minimize the latency . in addition , the proposed offline method to optimize the ulb size considering the logical qubit routing times and operation repetition frequency on the critical path of the quantum circuit is not discussed in the prior art . quantum circuits can be composed in quantum assembly ( qasm ) language @xcite , which provides a compact way of representing quantum circuits . qasm is a low - level quantum language that does not support complex control and data structures like arrays , loops , conditionals , etc . as an example of a quantum circuit , fig . [ fig : fig2a ] shows a circuit for preparing logical zero using steane code [ [ 7,1,3 ] ] ( the output is shown as @xmath1 ) . note that in this circuit , qubits are initialized to @xmath2 . each qubit participates in several one- and two - qubit operations . seven physical qubits at the output of the circuit represent a bundled logical qubit . [ fig : fig2b ] shows the qasm description of this encoding circuit . the first part of the qasm file describes the complete set of qubits involved in the circuit and their initial state , e.g. @xmath2 , @xmath3 , @xmath4 or @xmath5 . the second part of the qasm file describes the quantum operations based on their execution order . operations having no dependency , e.g. h q0 and h q1 , can be added to the file with any ordering . in each line , the operation type , e.g. cnot , is followed by the qubit(s ) involved in the operation . if two qubits are involved in the operation , the first mentioned qubit determines the _ control qubit _ and the second one determines the _ target qubit_. the qidg for the circuit is built from this qasm description after applying quantum fabric synthesis . 0.4 ] , ( b ) qasm program description of the logical zero preparation circuit , title="fig : " ] 0.2 [ cols=">,<",options="header " , ] [ table : toffolitable ] note that operations 7.b , 9.a , 10.a and 10.c are not part of the critical path of this circuit . as can be seen , the total latency of the critical path of the circuit is @xmath6 . in section [ results ] , after finding the latency for each ft operation as a function of the ulb size , the value of the toffoli gate latency is reported as a function @xmath7 and the best ulb size is chosen . [ fig : fig5 ] shows the qufd ulb sizing flow . in this flow , the smallest ulb size is set by the size required for the most complex operation circuit that needs to be mapped to the quantum fabric ( this is typically a non - transversal gate operation ) . as the ulb size increases , the resource contention decreases ( i.e. , there is less routing resource contention and thus potentially shorter waiting times to access routing channels during the dynamic routing step ) ; however , the intra - ulb routes become longer , and hence , routing latencies may actually become higher . from another perspective , increasing the ulb size increases the inter - ulb routing latency . in practice , there is an optimum ulb size that minimizes the aforesaid cost function in [ formula1 ] . the proposed ulb sizing flow is an offline process to generate the tiles that comprise the tqa and the gate library that is used by the quantum physical mapper . considering this fact , long run - time for this process does not raise a problem . moreover , the number of ulb sizes explored to find the optimal @xmath7 is finite considering the fact that beyond a certain ulb size , increasing the ulb size does not positively affect the mapped circuit latency . the quantum physical mapper converts a quantum circuit to a list of low - level control commands by specifying the exact qubit movements and instruction execution times . quantum physical mapping problem includes scheduling , placement and routing sub - problems . the scheduling step determines the order of instruction execution , the placement step determines locations of qubits at the beginning and end of each operation as well as the location for each instruction execution , while the routing step determines the qubit location at movement times . qubit movement latency is a big part of the quantum circuit latency . therefore , the routing latency ( which is a function of instruction placement and channel congestion ) must be included in the decision making procedure for scheduling and placement problems . to account for routing latency in scheduling and placement steps , routing latency estimates based on the physical distance between interaction wells are used . physical distance is different from manhattan distance because it is the shortest path between two wells going through movement channels . hence , it is equal or longer than manhattan distance . to explain the concurrent scheduling and placement problem in quantum physical mapper and the similarities and differences between this problem and scheduling and placement problem in the traditional vlsi cad field , we consider two scenarios . in the first scenario , each instruction is executed in a unique interaction well and in the second scenario , more than one instruction can be executed in one interaction well as long as their execution times do not overlap . * first scenario * : interaction wells in this scenario are similar to gates or transistors in vlsi circuits with the difference that there are shared connectors ( channels ) between interaction wells in the quantum fabric in place of dedicated connectors ( point - to - point wires ) in the vlsi circuits . considering the large channel capacity in the quantum fabric , we can ignore the effect of routing resource contention in this scenario . under this assumption , the best scheduling policy is the _ as soon as possible schedule_. moreover , the placement problem in this scenario is similar to vlsi timing - driven placement problems . the goal of the optimization is to minimize the total added latency to the critical path of the quantum circuit . * second scenario * : sharing interaction wells between instructions complicates the scheduling and placement problems . the scheduling and placement problems can thus be seen as a 3-dimensional placement problem with time as the third dimension . in addition to the constraint of no - overlap between execution times of instructions placed at the same interaction well , placing instructions along the time axis must satisfy the instruction dependency constraints . the resulting circuit latency and area in this scenario are much smaller than those of the first scenario . to simplify the placement problem , we consider discrete start times for instruction executions . an example of the scheduling and placement solution for this scenario is shown in fig . [ fig : fig6a ] , where _ es@xmath8 _ , _ ef@xmath8 _ and _ f@xmath8 _ denote the expected start time , expected finish time , and actual finish time of the instructions in time slot _ t _ , respectively . moreover , _ i@xmath9 _ to _ i@xmath10 _ indicate the interaction wells on the quantum fabric ; numbers in rectangles show the routing times of the operand qubits , and horizontal arrows represent the execution latencies of instructions . instructions with dark color belong to the first scheduling level whereas instructions with white background belong to the second scheduling level . as can be seen in fig . [ fig : fig6a ] , considering synchronized start time for qubit routing in each scheduling level ( _ es@xmath9 _ for level 1 and _ f@xmath9 _ for level 2 ) leads to inefficient use of resources and increases the circuit latency . to solve this problem , if instructions are started as soon as their qubits and the host interaction well become ready , the total latency of the circuit is lowered as shown in fig . [ fig : fig6b ] . this scheme has commonly been used in previous work such as @xcite . unfortunately , this modification decreases the predictability of start times for instruction executions . more precisely , this scheduling contradicts the assumption about separation of scheduling and placement steps because it makes it possible for two instructions that have statically been scheduled in two different time slots to overlap in the actual schedule of the circuit due to changes to routing latencies . furthermore , placing instructions on the circuit fabric based on the instruction level has an underlying assumption about the stop time of the previous - level instructions . this assumption does not hold if the instructions are scheduled as soon as their qubits and interaction wells are ready . an example of this case is shown in fig . [ fig : fig7a ] . as can be seen , the second scheduling level instructions that are placed in the interaction wells hosting instructions 2 and 6 in the first scheduling level ( @xmath11 and @xmath12 ) are deferred . if the placement module had been aware of this situation , instructions 8 and 10 would have been placed on other interaction wells to lower their finish times . to be able to predict the actual instruction execution times in the scheduling and placement step , our approach calculates the estimated start time of each instruction and if the estimated start time exceeds a start time threshold ( shown by @xmath13 in fig . [ fig : fig7a ] and calculated as the middle point of @xmath14 and @xmath15 ) , the instruction will be deferred and the scheduling and placement solutions are modified . an example of this modification is shown fig . [ fig : fig7b ] . as can be seen , this modification reduces the total latency of these ten instructions . details of the proposed scheduling and placement solutions are presented in sections [ sched ] and [ placement ] , respectively . the first step of the quantum physical mapper is _ loose scheduling_. the goal of this step is to assign each instruction to a scheduling level ( or time slot ) to minimize the total latency of the circuit subject to dependency and resource availability constraints . all instructions assigned to one scheduling level can be executed simultaneously . this step is called loose because it does not consider the routing latencies when scheduling the instructions . to modify the scheduling solution based on placement decisions , a part of this scheduling step is iteratively called during the placement procedure . the input to the scheduling procedure is a qidg , which is a _ directed acyclic graph _ , capturing the dependencies between instructions specified in the qasm file . each node of this graph is a native quantum instruction , i.e. , it can be performed in one step in an interaction well in the trapped - ion quantum circuit fabric . two instructions are dependent if any of these conditions hold : 1 . the target qubit of instruction a is used as the control qubit of instruction b ( read after write ( raw ) dependency ) 2 . the target qubit of instruction a is used as the target qubit of instruction b ( write after write ( waw ) dependency ) 3 . the control qubit of instruction a is used as the target qubit of instruction b ( write after read ( war ) dependency ) . each dependency is shown using a directed edge from the vertex representing instruction a to the vertex representing instruction b. fig . [ fig : fig8 ] shows an example of qidg for the quantum circuit presented in fig . [ fig : fig2a ] . , scaledwidth=40.0% ] each instruction and each qubit in the qidg are identified by unique identifiers , represented by index @xmath16 and @xmath17 respectively . in the qidg , nodes with the same parent instruction are called _ sibling instructions_. these instructions are two - qubit instructions that share the same control qubit . due to the _ no - cloning theorem _ @xcite , only one copy of each qubit can exist in the circuit . more precisely , only one instruction can access a qubit at each point in time . so , if one qubit needs to be accessed by sibling instructions , e.g. qubit q0 in instructions 4 and 7 in fig . [ fig : fig8 ] , those instructions can not be executed at the same time . based on the assumption that all instructions in one scheduling level can be executed simultaneously , at most one instruction from each sibling instruction set can be placed in one scheduling level . therefore , an ordering should be imposed among sibling instructions to be able to schedule the instructions without additional restrictions . to order the sibling instructions , we add _ auxiliary edges _ between them . note that adding these auxiliary edges may increase the critical path length . hence , this process should be done carefully . from another perspective , the number of instructions at each scheduling level is limited by the number of available interaction wells in the ulb . the minimum latency qidg scheduling problem ( called _ quantum instruction scheduling problem _ or qisp ) with the assumption of the same execution length for every instruction may be formulated as follows : @xmath18 where @xmath19 and @xmath20 denote the sets of _ parent _ and _ sibling instructions _ for instruction @xmath16 , respectively ; @xmath21 denotes the start time of instruction @xmath16 ; _ l _ , which stands for latency , is defined as the time when the last instruction starts its execution ; @xmath22 denotes the maximum number of concurrent instructions at each time . parameter @xmath23 is the scheduling parameter , which is equal to one if instruction @xmath16 starts from level _ l _ ; otherwise it is _ _ zero . equation relates the start time of an instruction to the scheduling parameter ( @xmath23 ) . constraint forces an instruction to have only one starting time . constraint captures the dependency constraints in the qidg . constraint captures the fact that sibling instructions can not be scheduled at the same time . constraint enforces the limit on the number of concurrent instructions at each scheduling level . qisp is an integer linear programming ( ilp ) problem . in fact , qisp is a _ resource - constrained scheduling _ problem @xcite with one important difference . in qisp , in addition to the limit on the total number of processing elements in each scheduling level ( cf . constraint ) , the number of processing elements that can be used for sibling instructions is limited to one ( cf . constraint ) . our solution to qisp decomposes the problem into two steps : ( i ) a pre - processing step based on the _ list scheduling _ algorithm @xcite is applied to the qidg to _ totally order _ the sibling instructions ; ( ii ) an enhanced version of the well - known _ force - directed scheduling _ algorithm @xcite is applied to the modified qidg so as to minimize a cost function which is a function of the number of scheduling levels and the number of concurrent instructions per scheduling level . details are provided below . the goal of the pre - processing step is to impose a total ordering on the sibling instructions to avoid any violations of constraint during the scheduling process . a well - known approach for solving the resource - constrained scheduling problem is the list scheduling algorithm @xcite , which is reviewed in the next paragraph . first , _ as soon as possible _ ( asap ) and _ as late as possible _ ( alap ) schedules for a given task dependency graph are constructed . the difference between levels of an instruction in the asap and alap schedules , which is called _ mobility _ , sets the priority of an instruction in scheduling with respect to the other instructions ( that is , less mobile instructions have higher priority to be scheduled first ) . to schedule the instructions based on this approach , starting from the first level of instructions in the asap schedule , if a resource constraint is violated , an instruction with the highest mobility ( least priority ) will be picked and delayed by one scheduling level . while the resource constraint is violated , this process continues . this process is repeated for the next level and continues until all conflicts are resolved . the motivation behind this greedy heuristic is that the instruction with higher mobility can be delayed without adversely affecting the overall execution latency . algorithm [ alg : pre ] provides the pseudo - code for the pre - processing step of the scheduling solution , which is based on list scheduling algorithm . one of the difficulties in qisp is that the length of the critical path that determines the instruction mobilities is not known _ a priori _ because the sibling instructions have not been ordered yet . therefore , new asap and alap schedules are determined in each step based on the modified qidg ( namely , current partial ordering of the sibling instructions ) . let @xmath24 , @xmath25 , and @xmath26 denote the scheduling level , asap level , and alap level of instruction _ i_. asap and alap levels of each instruction along with the priority values ( i.e. , @xmath27 ) are calculated in the first iteration . starting from the set of instructions with @xmath28 , an instruction with the highest priority among siblings that have an asap level equal to zero is selected . this instruction is added as a parent for its sibling instructions with the same or higher asap level . note that this selection fixes the instruction with the least mobility and moves its sibling instructions to the next levels in contrast to the list scheduling that repeatedly defers the instruction with the highest mobility until the resource constraint is satisfied . in this special problem , these approaches have the same results ( with lower complexity for our algorithm ) , because the number of allowed sibling instructions in each scheduling level is only one . after these modifications , asap and alap and priority metrics are recalculated . if constraint is no longer violated for the selected set of instructions , the set of instructions with @xmath29 will be selected and the process of adding edges will be repeated . this process is continued for all the asap levels or until all required edges are added to the qidg . compute @xmath25 , @xmath26 and @xmath30 + @xmath31 and @xmath32 , @xmath33 //target asap @xmath34 @xmath35 @xmath36 , @xmath37 @xmath38 @xmath39 * break * @xmath40 ; compute @xmath25 , @xmath26 and @xmath41 + @xmath31 and @xmath32 applying this pre - processing step to the qidg of fig . [ fig : fig8 ] results in the qidg of fig . [ fig : fig9a ] . dashed ( red ) edges in this figure show the added auxiliary edges . 0.3 after preprocessing , ( b ) example of scheduled instructions , title="fig:",scaledwidth=90.0% ] 0.3 after preprocessing , ( b ) example of scheduled instructions , title="fig:",scaledwidth=90.0% ] the number of quantum instructions being executed at the same time is upper bounded by the number of available interaction wells in the ulb . this bound is captured in constraint of the qisp formulation . although increasing the number of concurrent instructions in each scheduling level tends to decrease the number of scheduling levels , it may increase the total latency of the circuit as explained next . recall that the number of concurrent instructions in each scheduling level greatly affects the placement solution , routing delays , and thus , the latency of the quantum circuit . an instruction can not be executed until its qubit operands arrive at the host interaction well . the routing time of a qubit is defined as the difference between the arrival time of the qubit in the case of one - qubit instructions ( or that of the last arriving qubit in the case of two - qubit instructions ) and the time when the instruction was scheduled to be performed . the routing time is a function of the specific route selected for the operand qubit(s ) and the wait ( stall ) time of the qubit during the routing due to the congestion ( shared resource contention ) . increasing the number of concurrent instructions forces some instructions to be placed in interaction wells that are far from the interaction wells that are hosting their parent instructions . moreover , increasing the number of concurrent instructions results in an increase in the number of qubits that need to be simultaneously routed . in addition to the effect of longer routes on the routing time , higher number of routed qubits increases congestion , and thus , increases the wait ( stall ) time . evidently , increasing the number of concurrent instructions may delay the start of instruction execution times , and thereby , adversely affect the total latency of the quantum circuit . therefore , to reduce the total latency of a quantum circuit , we must minimize the number of scheduling levels and evenly distribute concurrent instructions among all scheduling levels . this is a multi - objective optimization problem . to solve it , a modified version of the well - known force - directed scheduling ( fds ) technique @xcite is used as detailed below . fds is a technique used to schedule directed acyclic graphs so as to minimize the resource usage under a latency constraint in a shared resource system . the algorithm consists of three steps : ( i ) determine the range of available time slots for scheduling an instruction ; ( ii ) create a distribution graph , which captures the resource pressures on each time slot , by assuming that an instruction is equally likely to be scheduled to start in any feasible time slot ; ( iii ) a metric called _ force _ is defined to minimize the resource utilization . by repeatedly assigning instructions to all possible time slots and calculating the force associated with the assignment , several force values will be calculated . the instruction - to - time slot assignment with the lowest force value is chosen , which also balances the instruction concurrencies ( i.e. , reduces resource pressure ) . the first goal of the quantum instruction scheduling is to minimize the number of concurrent instructions at each scheduling level . the second goal is to minimize the number of scheduling levels . to capture both of these goals in our scheduling method , we use a latency - constrained fds technique which tries to evenly distribute the instructions with fixed circuit latency with the addition of a fixed bound ( @xmath42 ) on the number of concurrent instructions at each scheduling level . for this reason , after assigning an instruction to a scheduling level , the number of instructions assigned to that scheduling level is incremented by one . when this number reaches @xmath42 , no new instruction is assigned to the scheduling level . based this fact , assignment probabilities in the fds technique are updated . if an instruction @xmath16 is found to have no possible scheduling level assignment ( @xmath43 ) , the number of scheduling levels will be incremented by one and the assignment probabilities for the unassigned instructions will be re - calculated . this process continues until all instructions are scheduled . notice that there is a hard global limit on the number of concurrent instructions in each scheduling level , which is @xmath22 . our simulation results show that considering a limit less than this global value may decrease the total circuit latency due to a decrease in the routing latency in each scheduling level . therefore , the scheduling process must be repeated for a few different @xmath44 values and the best result ( after final routing ) chosen . in this paper , five different values for @xmath42 are considered : @xmath45 where @xmath46 . as mentioned in previous sections , the placement solution affects routing latencies , and hence , the start times of instructions . this fact motivates the development of a cross - layer optimization approach spanning the scheduling and placement steps in the quantum physical mapper . the proposed placement algorithm in this section is based on the state - of - the - art placement algorithm for placing logic gates in 2-d layouts . modifications are made to account for the cross - later optimization issue . the placement solution is an iterative solution , which targets instructions in some scheduling level in each of its iterations . more precisely , at iteration @xmath47 , considering placement solution for instructions in scheduling levels @xmath48 to @xmath49 has already been obtained , a global instruction placement problem is formulated as a quadratic programming problem and solved by using a quadratic program solver . after this step , a host interaction well is determined for each instruction . next , the _ approximate start time _ for each instruction in scheduling level @xmath47 is calculated based on the fixed placement of its parent instructions and static routing latencies . if an instruction can not satisfy the start time requirement of scheduling level @xmath47 , it will be deferred by one scheduling level . then , the force - directed scheduling algorithm is applied to determine the best scheduling solution based on the new deferrals . after applying the scheduling algorithm , the quadratic program solver is applied to the placement problem with new scheduling levels for instructions to obtain ( and freeze ) the placement solution for instructions in scheduling level @xmath47 . in this section , details of this procedure are presented . first , a mathematical formulation of the placement problem is given . next , a solution without considering the effect of the placement on the scheduling solution is discussed , and , in the end , revisions to the algorithm in order to consider the effect of the routing latencies and placement solution on the scheduling solution are described . the ratio of the qubit movement latency to the instruction latency shows the importance of the routing latency in a quantum fabric realization . for example , in fig . [ fig : fig3 ] the minimum manhattan distance between two interaction wells is 10 . considering 1/10 as the ratio of single move latency to the two - qubit instruction latency according to @xcite , the minimum routing latency is as large as the latency of performing one instruction , which underlines the importance of the routing latency . similar to routing steps , routing latency can be divided into two components : 1 ) _ static routing latency _ , and , 2 ) _ dynamic routing latency_. the static routing latency is related to the manhattan distance between two interaction wells whereas the dynamic routing latency is related to stall times of the qubit in channels due to transient saturation of the channel capacities . the dynamic behavior of the routing latency is hard to predict in a fully analytical way . in other words , the quantum circuit should in fact be simulated in order to determine the dynamic routing latency . based on this line of reasoning , only static routing latency is considered in the quantum instruction placement formulation . there are two types of qubits in the circuit : 1 ) _ i / o qubits _ , and , 2 ) _ ancillary qubits_. i / o qubits enter and exit ulbs from specified locations whereas ancillary qubits are created , consumed and then purified to be used again for other instructions . let @xmath50 , @xmath51 and @xmath52 denote the set of ancillary , i / o and all qubits in the qidg , respectively . similarly , @xmath53 , @xmath54 and @xmath55 denote the set of operand ancillary , i / o and all qubits of instruction @xmath16 , respectively . the scheduling level and the parent set for each quantum instruction are determined in the quantum instruction scheduling step . in the quantum instruction placement step , each instruction should be assigned to some interaction well considering its scheduling level . recall that the placement solution determines the routing latency for each qubit movement , which can change the scheduling level of instructions , and subsequently , affects the start time of the quantum instructions . the _ quantum instruction placement problem _ ( qipp ) can be formulated as follows : @xmath56 where @xmath57 denotes the time when qubit _ q _ arrives at its output location after finishing the last instruction , @xmath58 , @xmath59 and @xmath60 denote the set of locations of the interaction , creation wells and input / output ports in the circuit , respectively , @xmath61 is the latency of the @xmath62 instruction , @xmath63 is the static routing latency between the interaction well @xmath64 and @xmath65 , @xmath66 denotes the static routing time from the creation well ( or input port ) @xmath67 to the interaction well @xmath68 , and @xmath69 is an assignment parameter that determines if the instruction is assigned to an interaction well ( when it is equal to 1 ) or not ( when 0 ) . similarly , @xmath70 denotes an assignment parameter that determines if the qubit is assigned to a creation well ( or input port ) or not . for input qubits , this parameter is fixed . parameters @xmath69 , @xmath70 and @xmath21 are the optimization parameters in the qipp . note that the result of the quantum instruction scheduling can not determine the exact timing of the instructions and can only determine the order of dependent instructions . the gaol in qipp is to minimize the total latency in the quantum circuit . constraint determines the earliest start time of each instruction based on the start time , static routing time and latency of its parent instructions . constraint determines the earliest start time of the quantum instruction that accesses qubit(s ) for the first time . constraint determines the time when qubits arrive at their output locations after finishing the last instructions . constraint limits the number of concurrent instructions that can be executed at each interaction well to one . constraints and define the binary assignment parameters . qipp is a non - linear , mixed - integer problem . from another perspective , qipp is a timing - driven 3d placement problem . the first two dimensions correspond to the x and y dimensions of the quantum fabric whereas the third dimension represents time . in a regular placement problem , the overlap between different placement objects is not acceptable , but in qipp , the physical overlap is acceptable as long as the execution times of different instructions do not coincide . a sample instruction placement solution for the qidg shown in fig . [ fig : fig9b ] is depicted in fig . [ fig : fig10 ] . each rectangular cube represents an instruction placed at an interaction well on the ulb . longer rectangular cubes represent two instructions placed at an interaction well in two consecutive scheduling times . , scaledwidth=70.0% ] in this paper , to manage the problem complexity , time is discretized . the duration of each time slot ( _ d _ ) is considered to be the latency of the fastest instruction such as a rotation operation . also to solve the timing - driven placement , a net - weighting technique is employed @xcite . details of the problem formulation and the algorithm are presented next . considering fixed instruction levels based on the result of the quantum instruction scheduling step , the qipp can be simplified to placing instructions on interaction wells such that there are no overlaps among instructions with the same scheduling level . note that the routing latency of instructions may delay instructions from their original scheduled time . thus , minimizing the routing latencies can decrease the time shifts and eventually decreases the total latency of the circuit . however , not all routing latencies have the same importance . an increase in the routing latency of qubits related to an instruction that lies in the critical path of the qidg will increase the total latency of the circuit , while adding to the routing latency of qubits related to an instruction in a non - critical path may only cause a time shift on the start times of the remaining instructions in that path without changing the total latency of the circuit . to account for this fact , the goal of the placement solution is simplified to minimizing the weighted sum of the routing latencies . the weight for the qubit routing latency between the _ _ i__@xmath71 ( child ) and the _ _ j__@xmath71 ( parent ) instructions is called @xmath72 and is found from equation . @xmath73 where , @xmath74 denotes the maximum weight for cases with infinity weight . this weight is in inverse proportion to the difference between alap and asap levels of the child instruction and also the difference between levels of the child and parent instructions . the higher the difference between the child and parent instruction levels , the more time the qubit has to be routed without delaying the child instruction . moreover , the difference between alap and asap levels of the child instruction shows the criticality of this instruction to the rest of the circuit . to model the routing latency in this problem , we use the manhattan distance between the parent and child instruction locations on the circuit fabric . using the aforementioned simplifications , the quantum instruction placement problem with _ fixed instruction level _ ( qipp - fil ) is to minimize the weighted summation of the routing latencies in the quantum circuit . moreover , no more than one instruction with a specific level can be placed on any interaction well . the effect of initial qubit routing latencies ( between the initial qubit locations and the location of the first instructions that use them ) and final routing latencies ( between the last operation that uses i / o qubits and their final locations ) is considered with weights calculated from equation when @xmath75 . qipp - fil consists of an instruction placement problem with per - level overlap avoidance constraints plus a qubit placement problem . to solve this problem , an instruction placement technique is used , which is followed by placing the ancillary qubits on the creation sites based on the instruction placement solution . to place the instructions on the quantum circuit fabric , a modified version of the state - of - the - art force - directed placement tool , simpl @xcite , simpl works best when placing 10@xmath76 to 10@xmath77 objects on a circuit fabric . this range complies with the number of instructions that we have for the placement step considering multi - level concatenated error correction scheme for transversal and non - transversal operations . simpl is reviewed next . to minimize the half perimeter wire length in vlsi circuits , without any overlap avoidance constraints , an iterative optimization procedure is used . starting from an initial solution , a quadratic optimization problem is formulated and solved . distances between placed components are used to formulate a new quadratic optimization problem to be solved . after convergence of the solution , the placement result is the optimal solution . to remove the overlaps , an algorithm called _ rough legalization _ , which uses a _ non - linear scaling _ method is proposed in @xcite . based on this approach , over - utilized areas in the fabric are examined and components that are placed in those areas are distributed evenly in the surrounding area . to avoid components from reverting back to their original locations , a _ pseudo - net _ is added in the new location of any moved component and a wire is added between that pseudo - node and the original component . weight of the added wire is increased as the number of iterations increases . note that adding pseudo - nets does not change the problem size because it does not add to the number of variables or equations and only changes the objective function related to the original node . after this modification , the half perimeter wire length minimization problem is formed and solved again . this process continues until all overlaps are removed . finally , a simple approach is used for final legalization . simpl improves the speed of prior placement tools and has similar quality to the analytical placement solutions . due to the difference between the quantum instruction placement problem and the vlsi placement problem , some modifications must be made to simpl as explained next . instead of minimizing the total routing latencies , a weighted summation of the routing latencies is minimized . moreover , the rough legalization step in this solution is applied to the instructions in each scheduling level . in particular , only overlaps between instructions on the same scheduling level are removed with the non - linear scaling approach . note that changing the locations of instructions during the rough legalization in one level affects the placement solution for other scheduling levels . therefore , to consider this effect , the quadratic optimization solution is repeated after finalizing the rough legalization for each scheduling level . to find the final placement solution ( assignment of instructions to interaction wells ) , similar to rough legalization , each scheduling level is processed separately . for each scheduling level , the interaction wells are sorted based on their distance to the instructions , and then the best interaction well for each instruction is chosen . after finalizing the placement solution for all instructions , the best creation well for each ancillary qubit is determined . each ancillary qubit is assigned to the free creation well that is closest to the location of the instruction that uses it for the first time . the proposed method based on classical placement tool plus net - weighting relies on a fixed scheduling level for each instruction . this assumption is violated with the addition of the routing latencies . more precisely , adding the routing latencies may significantly vary the start time of each instruction from its nominal start time . considering fixed levels for instructions in the placement problem has two negative effects on the placement solution . first , there may be a large number of instructions at some scheduling level , and hence , distributing them on the circuit fabric may cause some of the instructions to be placed far from the interaction wells hosting their parent instructions . however , if the scheduling level of those instructions is changed , they can be placed near the interaction wells hosting their parent instructions . this may even reduce the finish time for the deferred instructions . second , placing instructions on the circuit fabric based on the fixed scheduling level relies on an assumption about the start and stop times of the previous level instructions . this assumption typically does not , which results in the deferral of every instruction that is placed on the same interaction well in the next scheduling levels , cf . [ fig : fig7b ] . in order to avoid big differences between nominal and actual start times for an instruction , a rule can be added to the placement solution as follows . the start time of an instruction assigned to scheduling level @xmath47 should be less than a threshold time ( @xmath78 ) . this threshold start time can be the middle time in time slot @xmath47 . this rule results in higher accuracy of the execution time prediction for instructions in different scheduling levels . to modify the placement solution based on this rule , we propose an iterative approach . in each iteration , instructions in one scheduling level are fixed . in iteration @xmath47 : ( i ) considering fixed placement solution for instructions in scheduling level @xmath48 to @xmath49 , the proposed solution for qipp - fil is applied ; ( ii ) an approximate start time for each instruction ( @xmath79 ) in scheduling level @xmath47 is calculated based on the distance between the location of the interaction well assigned to it and locations of the interaction wells hosting the parent instructions . if an instruction can not satisfy the start time condition ( @xmath80 ) , it will be deferred by one scheduling level ( @xmath81 ) ; ( iii ) all of the finalized instructions that do not satisfy the timing constraints are removed from the quantum circuit fabric ; ( iv ) the force - directed scheduling step is applied to determine the best scheduling solution for all floating instructions based on the new deferrals . the instruction scheduling step is augmented to support the new parameter ( @xmath82 ) ; ( v ) the proposed solution for qipp - fil is applied to the placement problem with new scheduling levels for instructions to fix the placement solution for the remaining instructions in the scheduling level @xmath47 . the proposed approach may increase the number of scheduling levels , but it significantly decreases the overall latency of the circuit due to considering a more realistic scheduling solution . the flowchart for finalizing the placement for each scheduling level is provided in fig . [ fig : fig11 ] . notice that repeating the scheduling and global placement after each scheduling level with at least one deferred instruction is costly in case of large circuits . instead of this approach , the algorithm can be changed to repeat the scheduling and global placement algorithms after the number of deferred instructions reaches a threshold value or the summation of ( @xmath83 ) for deferred instructions reaches a limit . using this mechanism , the run time of the proposed solution can be shortened at the cost of a slight increase in the resulting circuit latency . qufd is implemented in java . trapped - ion parameters used for the simulations are as follows : move delay=10@xmath84 ; one - qubit primitive instruction delay=50@xmath85s@xmath86 and two - qubit primitive instruction delay=100@xmath84 . a workstation with intel core i7 - 2600 3.4ghz cpu and 8 gb ram was used to run the simulations . determining the concatenation level for the encoding , which is sufficient to get a correct result with high fidelity from the quantum circuit is a process based on the characteristics of the underlying quantum circuit technology and the implemented quantum circuit . this problem is outside the scope of this paper . in this experiment , we focus on one - level steane code [ [ 7,1,3 ] ] for quantum error correction and the steane syndrome extraction scheme . results for some other combinations of encoding and syndrome extraction schemes will also be presented . different ulb sizes are enumerated . for each ulb size , the latency of executing each of the ft operations ( cnot , t , s , x , y , and z ) encoded with steane code [ [ 7,1,3 ] ] are found by applying qufd . based on the results of the quantum physical mapper and target quantum circuit profiling results , the best ulb size is determined . the most important part of qufd is the quantum physical mapper for trapped - ion technology . there are different quantum mapping tools in the literature such as @xcite , quale @xcite , qpos @xcite , and qspr @xcite . in contrast to our work , all of these approaches rely on asap or alap scheduling and greedy placement of the qubits and instructions , which in turn results in long circuit latency . qspr improved on these approaches by repeating the scheduling and placement solution with random initial qubit placement solutions to reach a better initial qubit placement solution , which is critical in case of applying a greedy instruction placement solution . qspr algorithm provides about 41% improvement with respect to quale . in the following we provide a detailed comparison between qufd and qspr @xcite , which is the most advanced mapping tool in the literature . latencies of various ft operations implemented with different ulb sizes by using qufd and qspr are reported in fig . [ fig : fig12 ] . assuming simultaneous creation of ancillary qubits , there exist no feasible mapping solutions for the t instruction ( gate ) on 6@xmath876 and 8@xmath878 ulbs . this is because the number of the ancillary qubits for this instruction is more than the number of creation wells in the aforesaid ulbs . qspr fails to find any mapping solution for the 10@xmath8710 and 12@xmath8712 ulbs while qufd succeeds in these cases . the ft operation latencies obtained by qufd are from 35% ( for cnot ) to 41% ( for t instruction ) lower than those produced by qspr . although qufd uses larger computation time compared to qspr to produce its solutions , this is not a major concern , because we are only designing ulbs for seven ft operations and this procedure is done only once for the ft operation library design for each quantum circuit fabric . as mentioned in section [ sched ] , tightening the upper - bound on the number of instructions being scheduled in each scheduling level may decrease the total latency of the circuit by decreasing the routing time even though it may increase the number of scheduling levels at the end . this effect is examined for the cnot instruction mapped to a 10@xmath8710 ulb in fig . [ fig : fig13 ] . as can be seen in this figure , if the number of concurrent instructions is limited to 30% of the physical interaction wells on the fabric , the total latency is decreased by 18% . it is also seen that decreasing the maximum number of simultaneously active interaction wells does not change the circuit latency until this limit exceeds the maximum number of concurrent instruction in the qidg ( @xmath88 in fig . [ fig : fig13 ] ) . decreasing the maximum number of simultaneously active interaction wells after this value changes the scheduling , placement and routing solution by decreasing the resource contention in each scheduling level for placement and routing problems ( positive effect ) and possible qidg depth increase ( negative effect ) . the positive effect is dominant for a part of this process and after that the negative effect starts to dominate . as the coefficient of @xmath22,scaledwidth=70.0% ] to show the effect of enumerating different @xmath42 for the fds method , ( normalized ) total latency of the circuit with and without considering this factor for a complete set of ft operations ( cnot , h and t ) plus one of the pauli operations in a 10@xmath8710 ulb is reported in fig . [ fig : fig14 ] . as can be seen , for h and x operations , enumeration of different @xmath42 in the scheduling algorithm does not change the total latency but for cnot and t operations , this enumeration greatly reduces the total latency . in the scheduling algorithm on the circuit latency , scaledwidth=70.0% ] fig . [ fig : fig15 ] shows an exemplary case for the maximum scheduling level in different placement steps for the ft cnot instruction mapping on a 6@xmath876 ulb . clearly , as we go further in the placement iterations , the number of scheduling levels is increased when the deferred instruction(s ) is on the critical path of the circuit . the actual number of scheduling levels for this case is equal to 51 . the predicted number of scheduling levels after scheduling and placement solutions is increased from 26 ( without considering the routing latencies ) to 43 ( after iterative correction of the scheduling levels based on the placement solution ) . to show the effectiveness of the placement method that utilizes force - directed scheduling to modify the scheduling solution after each placement step , we implemented two other placement methods in the vlsi area . the first algorithm is based on the force - directed placement solution @xcite . in this approach , each instruction is considered to be a mass and data dependencies in the circuits are modeled by springs . the constant for each spring is set by the criticality factor of the source and the destination qubits . to avoid overlap between instructions with the same scheduling level , repulsive forces are used . the second algorithm is based on simpl algorithm @xcite with fixed scheduling levels for instructions . normalized latencies ( w.r.t . qufd results ) for a complete set of ft operations ( cnot , h and t ) plus one of the pauli operations in a 12@xmath8712 ulb for different placement mechanisms are presented in fig . [ fig : fig16 ] . as can be seen , the proposed placement method that modifies the scheduling level of instructions based on the placement solution generates on average 27% better results than the same placement method without using adaptive scheduling levels . moreover , the placement method based on the force - directed placement generates on average 72% worse results compared to the proposed placement solution . in this experiment , we focus on one - level steane encoding [ [ 7,1,3 ] ] and steane syndrome extraction scheme but to show the difference between encoding and syndrome extraction schemes , results of another concatenated code , that is , bacon - shor with the lowest concatenation level and different syndrome extraction schemes ( steane and knill ancilla ) are reported in fig . [ fig : fig17 ] . in this figure , latencies of cnot , h and x operations are measured in a 12@xmath8712 ulb . for the t operation , a 20@xmath8720 ulb is used because the number of required ancillary qubits for the t gate for some of the schemes is higher than the number of available creation wells on a 12@xmath8712 ulb . note , however , that these encoding / syndrome extraction schemes provide different levels of error detection / correction capabilities for the quantum circuit and can not be compared solely based on these latencies . based on the found latencies for ft operations , the total latency of the mapped toffoli circuit to a 2@xmath872 mesh tqa , which is calculated in section [ cadflow ] , is reported in fig . [ fig : fig18 ] . note that results for the 10@xmath8710 and 12@xmath8712 ulb sizes are not reported for qspr because this tool fails to find the total latency of the t instruction for these ulb sizes . as can be seen , the best ulb size for qufd is 10@xmath8710 and the best ulb size for qspr is 16@xmath8716 . moreover , the best total latency for the proposed qufd is 41% less than the best total latency of qspr . this paper presented quantum scheduling and placement algorithms for mapping fault - tolerant quantum operations to a universal logic block ( ulb ) . analytical formulations for the scheduling and placement problems were also presented . effective and state - of - the - art algorithms used in standard cad tools for classical circuits were modified to be applicable to the quantum mapping problem ( a quantum cad problem ) . in particular , modified list scheduling and force - directed scheduling were applied to the quantum instruction scheduling problem to capture the no - cloning dependencies among quantum instructions and determine the scheduling levels to minimize the number of instruction levels and reduce the resource contention . to consider the large impact of the qubit routing on the total latency of the quantum circuit , a net - weighting timing - driven placement solution based on simpl package was employed . this technique adaptively modifies the scheduling level of the instructions . moreover , a technique was described to determine the best ulb size for implementing ft quantum operations . all of the aforementioned algorithms were implemented in a cad tool called qufd . experimental results showed that qufd dominated the prior art quantum mappers by as much as 41% in terms of the average latency reduction . with some modifications , qufd can be applied to large qodg s to produce a mapping solution for any tile - based quantum architectures . moreover , the proposed flow is applicable to quantum mapping and ulb design in other quantum fabric technologies with small changes due to different qubit / information movement cost . * acknowledgment * : this research was supported by the intelligence advanced research projects activity ( iarpa ) via department of interior national business center contract number d11pc20165 . government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright annotation thereon . the views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements , either expressed or implied , of iarpa , doi / nbc , or the u.s . government . philipp schindler , julio t barreiro , thomas monz , volckmar nebendahl , daniel nigg , michael chwalla , markus hennrich , and rainer blatt . experimental repetitive quantum error correction . , 332(6033):10591061 , may 2011 . tzvetan s. metodi , darshan d. thaker , andrew w. cross , f. t. chong , and i. l. chuang . a quantum logic array microarchitecture : scalable quantum data movement and computation . in _ proceedings of the 38th annual ieee / acm international symposium on microarchitecture ( micro-38 ) _ , pages 305318 , washington , dc , usa , 2005 . ieee computer society . mark g. whitney , nemanja isailovic , yatish patel , and john kubiatowicz . a fault tolerant , area efficient architecture for shor s factoring algorithm . in _ proceedings of the 36th annual international symposium on computer architecture ( isca09 ) _ , pages 383394 , new york , ny , usa , 2009 . mark whitney , nemanja isailovic , yatish patel , and john kubiatowicz . automated generation of layout and control for quantum circuits . in _ proceedings of the 4th international conference on computing frontiers ( cf07 ) _ , pages 8394 , new york , ny , usa , 2007 . acm . m. j. dousti and massoud pedram . minimizing the latency of quantum circuits during mapping to the ion - trap circuit fabric . in _ design , automation and test in europe conference ( date12 ) _ , pages 840843 , march 2012 . larry mcmurchie and carl ebeling . pathfinder : a negotiation - based performance - driven router for fpgas . in _ proceedings of the 3rd acm international symposium on field - programmable gate arrays ( fpga95 ) _ , pages 111117 , new york , ny , usa , 1995 . m. a rowe , a. ben - kish , b. demarco , d. leibfried , v. meyer , j. beall , j. britton , j. hughes , w. m itano , b. jelenkovic , c. langer , t. rosenband , and d. j wineland . transport of quantum states and separation of ions in a dual rf ion trap . , may 2002 . w. k hensinger , s. olmschenk , d. stick , d. hucul , m. yeo , m. acton , l. deslauriers , c. monroe , and j. rabchuk . t - junction ion trap array for two - dimensional ion shuttling , storage , and manipulation . , 88(3):0341010341013 , january 2006 . d. r leibrandt , j. labaziewicz , r. j clark , i. l chuang , r. j epstein , c. ospelkaus , j. h wesenberg , j. j bollinger , d. leibfried , d. j wineland , d. stick , j. sterk , c. monroe , c. s pai , y. low , r. frahm , and r. e slusher . demonstration of a scalable , multiplexed ion trap for quantum information processing . , april 2009 . d. z. pan , halpin b. , and ren h. timing - driven placement . in alpert c. j. , mehta d. p. , and sapatnekar s. s. , editors , _ handbook of algorithms for vlsi physical automation _ , pages 223233 . crc press , 2007 .

the traffic of molecular motors through open tube - like compartments is studied using lattice models . these models exhibit boundary - induced phase transitions related to those of the asymmetric simple exclusion process ( asep ) in one dimension . the location of the transition lines depends on the boundary conditions at the two ends of the tubes . three types of boundary conditions are studied : ( a ) periodic boundary conditions which correspond to a closed torus like tube . ( b ) fixed motor densities at the two tube ends where radial equilibrium holds locally ; and ( c ) diffusive motor injection at one end and diffusive motor extraction at the other end . in addition to the phase diagrams , we also determine the profiles for the bound and unbound motor densities using mean field approximations and monte carlo simulations . our theoretical predictions are accessible to experiments .

cond-mat0304681

molecular motors are proteins that transform the free energy released from chemical reactions into mechanical work . in this article we consider a special class of motor proteins , namely cytoskeletal motors which perform directed walks along cytoskeletal filaments as reviewed in @xcite . in the cell , these motors have different functions related , e.g. , to vesicle transport and cell division . the best studied examples are kinesins , which walk along microtubules , and certain types of myosins , which walk along actin filaments . after a certain walking time , such a motor unbinds from its filament because its binding energy is finite and can be overcome by thermal activation . for kinesins , this typically happens after 100150 steps or after a walking time of about 1.21.8 seconds , see , e.g. , @xcite . in many motility assays , the filaments are immobilized on a substrate and are in contact with an aqueous solution . in such a situation , the unbound motors diffuse in the surrounding fluid until they eventually reattach to the same or another filament . recently , we have introduced lattice models to study the motors random walks , which consist of many diffusional encounters between motors and filaments in open and closed compartments @xcite . if many motors are placed in such a compartment , hard core exclusion between the motors has to be taken into account , since the motors are strongly attracted to the binding sites of the filaments , so that the filaments get overcrowded . these models are new variants of driven lattice gas models and exclusion processes , for which the active processes which drive the particles are localized to the filaments . lattice models of driven diffusive systems have been studied extensively in the last years , see e.g. @xcite . the simplest model is the asymmetric simple exclusion process ( asep ) in one dimension , where particles hop on a one - dimensional lattice with a strong bias towards one direction ( in the simplest case there are no backward steps at all ) and the only interaction of the particles is hard core exclusion , i.e. , steps to occupied lattice sites are forbidden . when coupled to open boundaries , this simple model already exhibits a complex phase diagram , see e.g. @xcite , which we will review below in some detail . the first model for the 1dimensional asep was introduced more than 30 years ago by macdonald _ @xcite in the context of protein synthesis by ribosomes on messenger rna ( mrna ) . at that time it was solved using a mean field approach and used to explain results of radioactive labeling experiments @xcite which showed that protein synthesis gets slower as the ribosome moves on the mrna template . the model of macdonald _ et al . _ explained this by the steric hindrance between successive ribosomes along the mrna track . two years later the same model was discussed by spitzer as a simple example for interacting particles in probability theory @xcite . since then , the asymmetric simple exclusion process ( and variants ) has been studied extensively as a generic model for non - equilibrium phase transitions @xcite and interacting stochastic systems @xcite as well as in other applications such as traffic flow @xcite . many properties of the 1dimensional asep are known exactly . as mentioned , the lattice models for random walks of molecular motors differs from the driven lattice gas models in an important way : walks of molecular motors are only driven as long as the motor is attached to a cytoskeletal filament , hence driving is localized to one or several lines . it can therefore be viewed as an asep which has the additional property , that particles ( molecular motors ) can unbind from the track with a small probability , diffuse in the surroundings and reattach to the same or another filament . in more mathematical terms , the asep is coupled to a _ symmetric _ exclusion process via adsorption and desorption of particles onto filaments . boundary conditions play an important role in driven systems . this becomes apparent , e.g. , if one compares a tube like system with periodic boundary conditions with one with closed boundaries . in the system with closed boundaries , a traffic jam of motors arises at one end of the system and the current of motors bound to the filament is balanced by diffusive currents of unbound motors as first shown in @xcite . with periodic boundary conditions , motors arriving at the right end of the system just restart their walk from the left end and a net current through the systems is obtained . in this article , we study the stationary states of tube - like compartments with open boundaries . these compartments have the shape of a cylinder and contain one filament which is placed along the cylinder axis in order to obtain the simplest possible geometry . the bound motors move along the filament and the unbound motors diffuse within the cylinder , see fig . [ fig : motor_tube_geom ] . at the ends of the tube , motors are inserted and extracted . such a system is accessible to _ in vitro _ experiments using standard motility assays , but it can also be viewed as a strongly simplified model for motor based transport in an axon @xcite , if these motors , which are synthesized in the cell body , are at least partly degraded at the axon terminal @xcite , a situation that can be mimicked by insertion and extraction of motors at the ends of a tube . the stationary states depend strongly on the way , in which the motors are inserted and extracted at the boundaries as we will explicitly demonstrate for three different types of boundary conditions , see fig . [ fig : motor_tube_randbed ] . our article is organized as follows . after introducing the model in section [ sec : model ] , we start in section [ sec : periodic_tube ] with periodic boundary conditions . this case can be solved exactly , since it satisfies local balance of currents in the radial direction , see appendix [ app_rdb ] . in section [ sec : open_tube ] , we discuss the situation in which the density of bound motors on the filament is fixed at the boundaries . finally , in section [ sec : bounddiff ] , we consider the case where the filament is shorter than the tube and the motors diffuse into and out of the tube . the main tool to study the open systems are monte carlo simulations . these are supplemented by dynamical considerations and self - consistent or mean field calculations . some details of the latter calculations are presented in appendix [ app_mf ] . we consider the motion of molecular motors in a cylindrical tube as shown in fig . [ fig : motor_tube_geom ] . the tube has length @xmath0 and radius @xmath1 . the total number of motors within the tube , denoted by @xmath2 , defines the overall motor concentration @xmath3 note that this concentration corresponds to the particle number density of the motors and , thus , is insensitive to the size of the motor particles . in general , the latter size depends on the type of motor and on the type of cargo attached to it . in the following , we will implicitly assume that the motor particle has a linear size which is comparable to the basic length scale @xmath4 as defined further below . if one wants to study the dependence on the motor particle size in a systematic way , one should measure the overall motor concentration in terms of the volume fraction of the motor particles . the cylindrical tube contains one filament located along its symmetry axis which is taken to be the x axis ; the two other cartesian coordinates are denoted by @xmath5 and @xmath6 . motors bound to the filament undergo directed motion , while unbound motors diffuse freely . since the motors are strongly attracted by the filament , a large fraction of these motors is in the bound state and mutual exclusion from the binding sites on the filament has to be taken into account even for relatively small overall motor concentrations . in order to include this mutual exclusion ( or hard core repulsion ) in the theoretical description , we map the system onto a lattice gas model on a simple cubic lattice . the lattice is oriented in such a way that its three primitive vectors point parallel to the @xmath7 , @xmath5 , and @xmath6axis , respectively . a rather natural choice for the lattice parameter @xmath4 is the repeat distance of the filament , which is @xmath8 nm in the case of kinesin motors moving on microtubule and @xmath9 nm for myosinv motors moving on actin filaments . the discretized tube consists of one line of binding sites , which represents the filament , and @xmath10 unbound channels , i.e. lines of lattice sites parallel to the filament . thus the cross section @xmath11 of the tube is equal to = ( 1+n_ch)^2 . for sufficiently large radii , one has @xmath12 , while for small radii , there are corrections due to the underlying lattice . in the following , we will measure all distances such as @xmath0 and @xmath1 in units of @xmath4 . thus , for the simulations , both @xmath0 and @xmath1 will be quoted as integers . the integer value of @xmath1 corresponds to a certain number of lattice sites along the cartesian coordinates @xmath5 and @xmath6 which are perpendicular to @xmath7 and run parallel to two basis vectors of the simple cubic lattice . thus , the interior of the tube contains all channels with @xmath13 . at each time step , a bound motor attempts to make a forward or backward step and to jump to the next lattice site to its right or to its left with probability @xmath14 or @xmath15 . in addition , the motor attempts to jump to each of the neighboring sites away from the filament with probability @xmath16 , and does not attempt to jump at all , i.e. , to rest at the filament site , with probability @xmath17 . since the sum of these probabilities is equal to one , we have @xmath18 . the velocity @xmath19 in the bound state is given by @xmath20 , where @xmath21 is the basic time scale of these random walks . in the following , we will measure all times and rates in units of @xmath21 and @xmath22 , respectively . since backward steps are rare for cytoskeletal motors , we will focus on the case @xmath23 , i.e. , we will ignore backward steps . if the particle unbinds from the filament , it attempts to jump to all nearest neighbor sites of the simple cubic lattice with equal probability @xmath24 . by choosing the time scale @xmath21 as @xmath25 , this hopping probability can be made to fit the diffusion coefficient of unbound motors , @xmath26 . when measured in units of @xmath4 and @xmath21 , the dimensionless diffusion coefficient @xmath27 . in principle , the resting probability @xmath17 can then be used to account for the ratio @xmath28 , which is quite large in the case of kinesin , and to adapt the velocity @xmath19 to the values obtained from experiments @xcite . for simplicity , we will often choose @xmath29 in order to eliminate one parameter from the problem . if an unbound motor attempts to hop to a filament site , the motor binds to it with sticking probability @xmath30 , while the step ( and hence binding ) is rejected with probability @xmath31 . both in the bound and in the unbound state , hopping attempts can only be successful , if the target site is not occupied by another motor ; otherwise the particle must stay where it is . in the following , we will mainly study overall motor concentrations @xmath32 in the range @xmath33 . for such small values of @xmath32 , mutual exclusion of the unbound motors can be safely ignored . however , because the motors are strongly attracted to the filament , the concentration of bound motors is much larger and it is crucial to take their mutual exclusion into account . first , we consider a cylindrical tube with periodic boundary conditions in the longitudinal direction . because of the translational invariance in the direction parallel to the filament , there are no net radial currents in the stationary state . indeed , a non - zero radial current in a state , which is translationally invariant in the longitudinal direction , would lead to net radial transport of motor particles , which is incompatible with the reflecting radial boundaries . this means that there is a bound current @xmath34 on the filament , but both currents of motors binding to and unbinding from the filament _ and radial currents of unbound motors _ are balanced locally . we call this situation _ radial detailed balance _ or _ radial equilibrium_. if there is only one unbound channel , @xmath35 , ( or @xmath10 equivalent channels ) radial equilibrium is equivalent to adsorption equilibrium . it is clear , that this will no longer be true , if translational invariance is broken by boundaries or blocked sites on the filament . another important property of systems with periodic boundary conditions is that , in this case , the number @xmath2 of motors in the system is conserved , which does not apply to open systems . because of translational invariance along the x axis , the bound and unbound motor densities , @xmath36 and @xmath37 , do not depend on @xmath7 . in addition , it follows from the absence of radial currents that @xmath37 is also independent of the radial coordinate @xmath38 . hence @xmath36 and @xmath37 are constant , and radial equilibrium implies the relation @xmath39 which leads to @xmath40 here and below , the densities @xmath36 and @xmath37 are local particle number densities which satisfy 0 _ b1 0 _ ub1 . in dimensionful units , this corresponds to @xmath41 and @xmath42 . since the total number @xmath2 of motors is conserved , we can use the normalization condition , @xmath43 to obtain a quadratic equation for the densities as a function of the system size and the total number of motors . since one root is always negative , the physically meaningful solution is : @xmath44.\end{aligned}\ ] ] from this expression for the unbound density , the bound density @xmath36 follows via eq . ( [ per_rb_bd ] ) and the stationary current is given by @xmath45 . the current calculated in this way is shown in fig . [ fig : strom_perrb ] as a function of @xmath46 . the data points ( circles ) are the results of monte carlo simulations for a system of length @xmath47 and radius @xmath48 and are in very good agreement with the analytical solution . it follows from the analytical solution that the current @xmath49 vanishes at @xmath50 , behaves as @xmath51 for small @xmath46 , and has the maximal value @xmath52 for @xmath53 , @xmath54 and @xmath55 . let us add two remarks : \(i ) eq . ( [ rdb_formel ] ) , as stated here , can be considered as a mean field equation . however , using the quantum hamiltonian representation of the stochastic process , it can be shown to hold exactly . the calculation is simple , but rather technical and is therefore presented in appendix [ app_rdb ] , which shows that stationary states can be constructed as product measures provided the bound and unbound densities satisfy eq . ( [ rdb_formel ] ) . \(ii ) note that in contrast to a homogeneously driven lattice gas such as the asymmetric simple exclusion process , there is no particle hole symmetry here . particles attempt to leave the filament with rate @xmath16 to a neighboring site while holes do so with rate @xmath56 , i.e. particles are strongly attracted by the filament , while holes are not . however , if one considers only the bound density , the current density relationship @xmath45 is invariant under the exchange of particles and holes . as we will see in the next section , this can lead to an apparent particle hole symmetry for systems with radial equilibrium , because the radial currents vanish and the state of the system can be determined by the bound density alone . now , let us consider the more interesting case , where the tube is open and the densities at the left and right boundary are fixed . to be precise , we consider two different sets of boundary conditions , ( b ) and ( c ) as shown schematically in fig . [ fig : motor_tube_randbed ] . in this section , we study case ( b ) while case ( c ) will be considered in the next section [ sec : bounddiff ] . for case ( b ) , we add two layers of boundary sites at @xmath57 and @xmath58 with @xmath13 . as before , the filament is located at @xmath59 . we then fix the density on the additional filament sites according to _ b(x = 0 ) _ b(x = l+ 1 ) _ b , ex , see fig . [ fig : motor_tube_randbed ] . furthermore , the densities on the nonfilament boundary sites are chosen in such a way that radial equilibrium as given by ( [ rdb_formel ] ) holds at both boundaries . these boundary conditions are implemented by the following choice of random walk probabilities . first , we eliminate two parameters from the problem , namely the jump probability @xmath15 to make backward steps on the filament and the resting probability @xmath17 to make no step at all on the filament . thus , we take @xmath60 throughout this section . next , when we choose a site within the left or right boundary layer during the monte carlo sweep , we first draw a random number @xmath61 which is uniformly distributed over the interval @xmath62 . the chosen left or right boundary site is taken to be occupied if @xmath63 or @xmath64 , respectively . if a boundary site with @xmath65 is occupied by a motor particle , this particle attempts to jump into the tube with probability @xmath24 . if the left boundary site with @xmath59 is occupied , the corresponding particle attempts to jump onto the left end of the filament with probability @xmath66 . if the right boundary site with @xmath59 is occupied , this particle can not enter since @xmath67 . in addition , all particles which jump from a site within the tube , i.e. , from a lattice site with @xmath68 , onto a boundary site , are extracted from the tube . in the limiting case with @xmath69 and @xmath70 , our system becomes equivalent to the 1dimensional asep , for which the density profiles and the phase diagram are known exactly @xcite . let us therefore summarize some of the known properties of this process , for which we use symbols without the subscript @xmath71. for the 1dimensional asep , there are three different phases . if the density @xmath72 at the left boundary is small and satisfies @xmath73 and if the density @xmath74 at the right boundary is not too large with @xmath75 , the system is in the low density ( ld ) phase , for which the bulk density @xmath76 is equal to the left boundary density , and the current is given by @xmath77 . because of the particle hole symmetry , an analogous situation holds for @xmath78 and @xmath79 [ high density ( hd ) phase ] . now , the bulk density is given by @xmath80 and the current is @xmath81 . at the line @xmath82 , a discontinuous phase transition takes place , and the bulk density jumps from the left boundary value to the right boundary value . finally , for @xmath83 and @xmath84 , the bulk density @xmath85 and the current attains its maximal value @xmath86 . therefore , this phase is called maximal current ( mc ) phase . the phase transition towards the maximal current phase is continuous with a diverging correlation length . the formation of three different phases can be understood in terms of the underlying dynamics of domain walls and density fluctuations @xcite . in the low density and high density phases , the selection of the stationary state is governed by domain wall motion . thus consider a domain wall , which forms between regions of different densities . if its velocity is positive , the domain wall travels to the right boundary and the bulk density is equal to the left boundary density ( low density phase ) . likewise , the bulk density is given by the right boundary density , if the domain wall velocity is negative ( high density phase ) . at the transition line with @xmath87 the domain wall velocity is zero , and domain walls diffuse through the system . the second mechanism is related to density fluctuations . if a small perturbation of the density , corresponding , e.g. , to some added particles , enters the system at the left boundary , it can move with positive or negative velocity , i.e. , it can spread into the bulk or is driven back towards the boundary . in the maximal current phase the velocity of such density perturbations , is negative and density fluctuations coming from the left boundary are driven back to the boundary . hence , increasing @xmath72 does not increase the bulk density , since additional particles can not enter the system ( overfeeding effect ) . both the velocities of domain walls and of density fluctuations are governed by the same density current relation which is @xmath88 for the 1dimensional asep @xcite . for the filament in a tube as considered here , the velocities of domain walls and density fluctuations are similar to those for the asep in one dimension . this can be understood from the fact that the density current relationship on the filament is the same as for the 1dimensional asep provided one rescales all currents by the factor @xmath66 which arises from the possibility to unbind from the filament . thus , let us first consider the case , where the behavior on the filament is determined by domain walls , i.e. the low density and high density phases . drift motion of domain walls is governed by the domain wall velocity on the filament , which is @xmath89 for the one - dimensional asep @xcite . in the tube system considere here , the domain wall velocity is slowed down compared to this value , because the domain wall of the unbound density must follow the bound density domain wall by binding and unbinding of motors to and from the filament . however , the sign of the domain wall velocity is the same for this tube system as for the one dimensional asep , and the domain wall velocity changes sign at the same values of the boundary densities . an explicit expression for @xmath90 can be obtained from the general expression for the domain wall velocity as given in ref . @xcite by integrating the density over the tube cross - section , which leads to @xmath91 for the geometry considered here . remember that the unbound densities are related to the bound densities by radial equilibrium at the boundaries . if a domain wall spreads from the left or right boundary into the system , radial equilibrium will hold approximately in the bulk because of translational invariance , but , in addition , all the way down to the dominating boundary , where we have imposed radial equilibrium via the boundary conditions . therefore , the current and the bulk density are determined by the bound density at the boundaries . the drift velocity of fluctuations of the bound density is @xmath92 @xcite . if the behavior on the filament is determined by this velocity , i.e. in the case @xmath93 and @xmath94 , radial equilibrium will not hold up to the boundary and unbound motors can enter the tube . however , after a short distance they will bind to the filament and act like an additional particle in the maximal current phase : the density fluctuation generated by the additional particle moves towards the boundary . therefore in the bulk , the bound density is 1/2 and the current is @xmath95 . summarizing these considerations , we predict to find exactly the same phase diagram as for the 1dimensional asep , see fig . [ fig : phasendiag_asep ] . in fact , we have chosen our boundary conditions ( b ) in such a way that we only have to replace the density @xmath96 of the 1dimensional asep by the bound density @xmath36 and the boundary densities @xmath72 and @xmath74 by the boundary densities on the filament , @xmath97 and @xmath98 . the unbound density in the bulk is obtained by radial equilibrium from the bound density . these expectations are confirmed ( i ) by a detailed analysis of the discrete mean field equations and ( ii ) by extended monte carlo simulations . a remarkable feature of the phase diagram is that it seems to exhibit particle hole symmetry , while the dynamics does not . the signature of particle hole symmetry in the phase diagram is the symmetry between the high density and low density phase . in both these phases , the bulk density is approximately constant and radial equilibrium holds approximately in the whole system except close to the left or right boundary . radial equilibrium holds exactly at that boundary which determines the bulk behavior . therefore radial currents vanish on average and the phase diagram is determined by the bound density alone , resulting in a phase diagram which gives the impression of particle hole symmetry , although this symmetry is broken . indeed , even though the substitution of the bound density @xmath36 by @xmath99 leads to the same bound current @xmath34 , it does not lead to the unbound density @xmath100 . two stationary states with bound densities that are related through particle hole symmetry are characterized by two unbound densities which are both smaller than the bound ones and thus break the apparent symmetry . to discuss the concentration profiles of the bound and unbound motors , we use continuum mean field equations and compare the mean field results with simulations . in addition , we make a two - state approximation , i.e. , we consider the case of a single unbound channel , so that a motor can be in only two states , bound to the filament or unbound . the two - state model is exact for an arbitrary number @xmath101 of _ equivalent _ unbound channels , but it can also serve as an approximation for the original tube systems : since the unbound density depends only weakly on the radial coordinate , we will consider the approximation in which @xmath37 is taken to be independent of @xmath102 and the unbound channels are taken to be equivalent . this approximation corresponds to a two state model in which the bound motors are described by the density @xmath103 and the unbound motors by the density @xmath104 . the effective diffusion coefficient for the unbound motors is given by @xmath105 . using again the mean field approximation , the total current @xmath106 parallel to the tube axis is now given by @xmath107 and the equality between incoming and outgoing currents at any lattice site leads to @xmath108= { \tilde\pi_{\rm ad}}\rho_{{\rm ub}}(1-\rho_{{\rm b}})-\tilde\epsilon\rho_{{\rm b}}(1-\rho_{{\rm ub } } ) , \label{e.twostate2}\ ] ] where @xmath109 and @xmath26 are the diffusion coefficients of motors in the bound and unbound state , respectively . in addition , we have introduced the rescaled binding and unbinding rates , @xmath110 and @xmath111 . in order to determine the density profiles far from the boundaries , we first calculate the homogeneous , @xmath7independent solutions , @xmath112 and @xmath113 , of the two mean field equations ( [ e.twostate1 ] ) and ( [ e.twostate2 ] ) . the first equation ( [ e.twostate1 ] ) for the total current @xmath106 then reduces to the current density relationship @xmath114 whereas the second equation ( [ e.twostate2 ] ) becomes @xmath115 which implies radial equilibrium for the @xmath7independent solutions . we then decompose the densities according to @xmath116 and expand the mean field equations ( [ e.twostate1 ] ) and ( [ e.twostate2 ] ) in powers of the density deviations @xmath117 and @xmath118 . the details of this expansion are described in appendix [ app_mf ] . for the high and low density phases with @xmath119 , the expansion of the mean field equations up to first order in @xmath117 and @xmath118 leads to an exponential approach @xmath120 of the density profiles towards the homogeneous solutions @xmath112 and @xmath113 , see appendix [ app_mf ] . the corresponding decay length @xmath121 satisfies the cubic equation @xmath122 which is solved numerically . for the asep in one dimension , one has @xmath123 and ( [ e.xi ] ) reduces to @xmath124 with @xmath125 thus , in this limit , mean field theory leads to the correlation length @xmath126 . for @xmath127 , we choose the unique solution of ( [ e.xi ] ) which approaches ( [ e.xi0 ] ) as @xmath10 vanishes . this solution behaves as @xmath128 for small @xmath10 . in addition to the mean field calculation , we again used monte carlo simulations in order to determine the density profiles as shown in figs.[fig : profile_ld ] and [ fig : profile_hd ] . as predicted by the mean field calculation , the constant bulk densities for the bound and unbound states are approached exponentially in the low and high density phases . the corresponding decay length @xmath121 is found to be the same for the bound and the unbound density and to diverge as one approaches the maximal current phase . some simulation results for the decay length @xmath121 are displayed in fig . [ fig : kurve_l_r ] . in this case , the radius @xmath1 of the tube was varied while the boundary densities were kept fixed . the latter densities were chosen to be @xmath129 and @xmath130 which lies within the low density phase but is close to the phase transition line which separates the low from the high density phase . one surprising feature of the monte carlo data for the decay length @xmath121 is that they exhibit a pronounced _ minimum _ as a function of tube radius @xmath1 . for comparison , we also display in fig . [ fig : kurve_l_r ] the @xmath121values as obtained from several mean field approximations corresponding to the dashed , dotted and solid lines . the dashed line is obtained from the solution of equation ( [ e.xi ] ) which we derived from the continuous mean field approximation for the two state model . we also determined this quantity using a lattice version of this approximation ( dotted line ) and a more elaborate mean field approximation ( solid line ) in which we solved the diffusion equation in the cylindrical compartment and matched this solution to the directed transport along the filament . inspection of fig . [ fig : kurve_l_r ] shows that all three mean field approximations are quite consistent with each other and lead to @xmath131 as follows from ( [ e.xi ] ) . such an increase of @xmath121 for large @xmath1 is in fair agreement with the monte carlo data displayed in fig . [ fig : kurve_l_r ] . however , in contrast to the monte carlo simulations , all three mean field approximations give a _ monotonic _ increase of @xmath121 with increasing @xmath1 . the largest discrepancy between the mean field results and the monte carlo data is found in the limit of small @xmath1 for which one recovers the 1dimensional asep . in this limit , the monte carlo data should be quite reliable as one concludes from the value obtained for @xmath132 which is in very good agreement with the exact solution for the 1dimensional asep as given by @xcite @xmath133}\ ] ] for @xmath134 in , ex . thus , we conclude that the decay length @xmath121 does indeed exhibit a nonmonotonic dependence on the tube radius @xmath1 and that this behavior is not reproduced by the mean field approximation . finally , we note that the different behavior of the decay length @xmath121 for large and for small @xmath1 is correlated with a qualitatively different behavior of the corresponding density profiles as observed in the monte carlo simulations . as an example , let us consider the density profiles within the low density phase as in fig . [ fig : profile_ld ] . in this case , the bound and unbound densities exhibit plateau regions which are determined by their values @xmath97 and @xmath135 at the left boundary . as one gets closer to the right boundary where the motors can leave the tube , the densities start to deviate from these constant values , and these deviations grow exponentially as @xmath120 . for small @xmath1 , the corresponding profiles are convex upwards for all values of @xmath7 . for large @xmath1 , on the other hand , the profile exhibits an inflection point close to the right boundary . this inflection point moves towards the interior of the tube as @xmath1 is further increased . in the maximal current phase , one has @xmath136 , and one has to consider terms up to second order in the density deviations @xmath117 and @xmath118 , see appendix [ app_mf ] . one then finds that @xmath117 satisfies the nonlinear differential equation @xmath137 with @xmath138 this differential equation can be solved by separation of variables . the solution behaves as @xmath139 i.e. , the deviation of the bound density from its asymptotic bulk value @xmath136 decays as @xmath140 within the mean field approximation . for @xmath123 , this becomes identical with the mean field solution for the asep in one dimension as discussed in @xcite . in this latter case , an exact solution is available @xcite , which shows that the density profile decays as @xmath141 , i.e. , with a different exponent . this is related to the fact that density fluctuations spread superdiffusively in the 1dimensional asep @xcite . the dispersion @xmath142 of an ensemble of particles in such a system behaves as @xmath143 for large times @xmath144 . this superdiffusive spreading of density fluctuations in one dimension can be taken into account within a mean field approximation if one considers a scale - dependent diffusion coefficient as shown in ref . @xcite . in the following , we will extend this approach to the tube geometry considered here . thus , we now replace @xmath109 in the mean field equation ( [ mf_rhobd_tube ] ) by a scale dependent diffusion coefficient @xmath145 and consider the modified mean field equation @xmath146{\frac{\partial}{\partial x}}\eta_{{\rm b}}\approx -v_{{\rm b}}\eta_{{\rm b}}^2 \quad . \label{e.meanfieldmodified}\ ] ] a convenient choice for @xmath147 which embodies the correct superdiffusive behavior for the asep in one dimension is @xmath148 the left boundary is located at @xmath149 , and @xmath150 and @xmath151 represent two scale parameters . with this choice , the modified mean field equation ( [ e.meanfieldmodified ] ) can again be solved by separation of variables . as a result , we obtain @xmath152^{-1}\ ] ] where we have introduced the abbreviations @xmath153 @xmath154 and @xmath155 the initial value @xmath156 denotes the density deviation at the left boundary . far from the left boundary , i.e. , for large values of @xmath157 , the expression ( [ e.bdensitydev ] ) leads to the asymptotic behavior @xmath158 \quad . \label{e.bdensitydev1}\ ] ] thus , the deviation of the bound density from its asymptotic value now decays as @xmath159 as for the asep in one dimension . in addition , we also obtain a correction term in ( [ e.bdensitydev1 ] ) which depends on @xmath160 . for large tube radius @xmath1 , one has @xmath161 and @xmath162 . therefore , the correction term becomes large for large @xmath1 . we can now define a crossover length @xmath163 at which the correction term has the same size as the leading term . this leads to the implicit equation @xmath164 and , thus , to @xmath165 ^ 2 \sim [ r^2 \ln(r^2 ) ] ^2 \qquad { \rm for } \quad { \rm large } \quad b \sim r^2 \quad . \label{e.crossoverlength}\ ] ] close to the left boundary , i.e. , for small values of @xmath166 , the expression ( [ e.bdensitydev ] ) for the deviation of the bound density from its asymptotic value @xmath136 leads to @xmath167 . \label{e.bdensitydev2}\ ] ] we may now define a second crossover length or extrapolation length @xmath168 at which the two terms in ( [ e.bdensitydev2 ] ) cancel . this extrapolation length is given by @xmath169 for large tube radius @xmath1 , the latter length scale grows as @xmath170 . in general , the extrapolation length can have both signs but , for the maximal current phase , one always has @xmath171 and , thus , @xmath172 . there is also an intermediate range of @xmath5values defined by @xmath173 . for these @xmath5values , the deviation @xmath117 of the bound density from its asymptotic value as given by ( [ e.bdensitydev ] ) simplifies and becomes _ b = . thus , for these intermediate @xmath5values the density deviation decays again as @xmath174 . in summary , the theory described here indicates ( i ) that the initial value at the left boundary is felt up to an extrapolation length @xmath175 , ( ii ) that the true asymptotic behavior of the density deviation is obtained for @xmath176 as follows from ( [ e.crossoverlength ] ) , and ( iii ) that the density deviation decays as @xmath177 on intermediate length scales with @xmath178 . these conclusions agree with the results of monte carlo simulations . in these simulations , the tube length @xmath0 is necessarily finite . this implies that the profiles observed in the simulations may not reach the asymptotic behavior present in a tube of infinite length . since both the crossover length @xmath179 and the extrapolation length @xmath180 increase quickly with increasing tube radius @xmath1 , we expect to find the true asymptotic behavior only for sufficiently small values of @xmath1 . this expectation is confirmed by the simulation results . for sufficiently small @xmath1 , the density deviation @xmath117 is found to decay as @xmath141 for large @xmath7 , as shown in fig . [ fig : deltarhob_asep_r357 ] for @xmath181 . for sufficiently large values of @xmath1 , on the other hand , the observed profiles decay as @xmath182 , see the monte carlo data in fig . [ fig : deltarhob_asep_r357 ] for @xmath183 and @xmath184 . in the latter cases , the true asymptotic behavior is not accessible and is cut off by the finite value of @xmath0 . in addition , a more detailed analysis of the monte carlo results for small values of @xmath1 but large values of @xmath0 explicitly shows that the decay of the bound density deviation behaves as @xmath185 for large @xmath7 but decays faster than @xmath186 for smaller values of @xmath7 . this crossover behavior is shown in fig . [ fig : decay_l6000 ] where the simulation data are compared with the thin dashed line corresponding to the decay law @xmath185 . furthermore , the data can be well fitted with a density profile as given by ( [ e.bdensitydev ] ) if one makes an appropriate choice for the scale parameters @xmath150 and @xmath151 . the monte carlo data shown in fig . [ fig : decay_l6000 ] correspond to a tube of radius @xmath187 and length @xmath188 with boundary densities @xmath189 and @xmath190 . in this case , a least squares fit of the data for @xmath191 leads to the parameter values @xmath192 and @xmath193 . the corresponding fitting curve corresponds to the thick dashed line in fig . [ fig : decay_l6000 ] . the fit becomes less reliable close to the left boundary , where the assumption that @xmath117 is small is no longer fulfilled . for fixed boundary densities , both fitting parameters @xmath150 and @xmath151 are found to depend on @xmath1 and to increase with increasing @xmath1 . in addition , for @xmath194 , it becomes rather difficult to determine these parameters since one would have to simulate rather long systems with @xmath195 in order to observe the true asymptotic behavior . for the asep in one dimension , it has been argued that the power law decay of the density profile as given by the asymptotic form @xmath196 for large @xmath7 is characterized by the universal amplitude @xmath197 . @xcite inspection of the relations ( [ e.ydefined ] ) , ( [ e.adefined ] ) , and ( [ e.bdensitydev1 ] ) shows that , in the present situation , @xmath198 where the parameters @xmath150 and @xmath151 depend on the tube radius @xmath1 . this implies that the amplitude @xmath199 will also depend on @xmath1 . using the numerically determined values for @xmath150 and @xmath151 and the bound state velocity @xmath200 , we obtain the estimates @xmath201 for @xmath132 , which corresponds to the 1dimensional asep and should be compared with the exact value @xmath197 , and @xmath202 for @xmath187 . thus , for small values of @xmath1 , the amplitude @xmath199 is found to decrease with increasing @xmath1 . finally , we consider the second type of open boundary conditions corresponding to case ( c ) in fig . [ fig : motor_tube_randbed ] . the length of the tube is again denoted by @xmath0 . this tube is now longer than the filament which has length @xmath203 . the left end of the tube is located at @xmath204 as before but the left end of the filament is at @xmath205 . likewise , the right end of the filament is at @xmath206 whereas the right end of the tube is at @xmath207 . thus , a motor particle which enters the tube on the left must diffuse over a distance @xmath208 before it can come into contact with the filament , and a motor particle which leaves the filament at its right end must also diffuse over a distance @xmath208 before it can leave the tube . at the left and right end of the tube , we now prescribe constant boundary densities as given by ( x=0 , y , z ) = _ ub , in ( x = l+1 , y , z ) = _ ub , ex for all values of @xmath5 and @xmath6 with @xmath13 . as before , the jump probability @xmath15 to make backward steps on the filament is taken to be zero . likewise , the resting probability @xmath17 to make no step at all on the filament is also zero with a possible exception at the last filament site with @xmath209 . indeed , in order to define the system in a unique way , we still have to specify the probability to make a forward step at this last filament site . two possible choices appear rather natural : ( i ) _ active _ unbinding from the last filament site , i.e. , the motor particle attempts to step forward with probability @xmath14 and makes a step if the adjacent nonfilament site is unoccupied . in this case , the forward step at the last filament site is governed by the same probabilities as all other forward steps along the filament ; and ( ii ) _ thermal _ unbinding in which the motor particle unbinds with probability @xmath16 both in the forward direction and in the four orthogonal directions . in the latter case , one has to choose the resting probability @xmath17 to be nonzero and to be given by @xmath210 . the choice ( i ) is suggested by the results of recent experiments on microtubules and kinesin motors @xcite which indicate that these motors unbind quickly at the filament ends . in the following subsections , we will first consider this choice ( i ) corresponding to active unbinding from the last filament site . in the last subsection [ s.activeversusthermal ] , we will show that the choice ( ii ) leads to rather similar behavior . in order to understand the behavior found for boundary condition ( c ) , it is instructive to partition the tube into three compartments which are defined as follows : ( i ) a left compartment with @xmath211 where transport is purely diffusive ; ( ii ) a middle compartment with @xmath212 where all directed ( or active ) transport occurs ; and ( iii ) a right compartment with @xmath213 where the transport is again purely diffusive . for a stationary state , the total current through the tube must be constant and , thus , must be the same in all three compartments . the current through the middle compartment is given by the bound current @xmath214 . thus , the diffusive currents @xmath215 and @xmath216 in the left and right tube segment must be equal and must satisfy the simple relation @xmath217 the relation as given by ( [ e.currentcasec ] ) is easily checked in the simulations since the density profile is found to be approximately linear in the left and in the right compartments provided @xmath218 is sufficiently large . for such a linear density profile in the right compartment , the diffusive current @xmath216 can be estimated as j_dif , r ( 1+n_ch ) d_ub . [ e.diffusivecurrent ] since the maximal density difference is one ( in the units used here ) , the maximal diffusive current behaves as ( j _ dif , r ) ~(1+n_ch ) d_ub ~r^2 d_ub . this shows that the maximal diffusive current depends on the tube radius @xmath1 , the lateral size @xmath218 of the diffusive segments , and the diffusion coefficient @xmath26 of the unbound motors . since these parameters can be chosen independently from the bound motor velocity @xmath19 , the diffusive current @xmath219 can be made smaller than the maximal bound current @xmath95 on the filament . in the latter case , the diffusive compartments act as diffusive bottlenecks and the maximal current phase characterized by the current @xmath95 is expected to be absent from the phase diagram . this expectation is indeed confirmed by the simulations as discussed next . one geometry , for which no maximal current phase has been observed in the simulations , is provided by a tube with radius @xmath220 , length @xmath221 , and filament length @xmath222 . the corresponding phase diagram as determined by the monte carlo simulations is shown in fig . [ fig : phasendiagramm_r5b5_ausschnitt ] . the largest part of the phase diagram is covered by the high density phase ; in addition , a low density phase is found for small values of the boundary densities @xmath135 and @xmath223 . the transition line displayed in fig . [ fig : phasendiagramm_r5b5_ausschnitt ] has been determined from the functional dependence of the bound density in the bulk , @xmath112 , on the left boundary density @xmath135 as shown in fig . [ fig : rho_b_ohnemc ] . inspection of this latter figure shows that the bound bulk density @xmath112 jumps at a certain value of the left boundary density @xmath135 . since @xmath0 is finite , this jump occurs over a small but finite interval of @xmath224 . thus , the jump can be characterized by two @xmath224values , say @xmath225 and @xmath226 , which represent the left and the right corner of the numerically determined jump . both @xmath135values have been included in the phase diagram of fig . [ fig : phasendiagramm_r5b5_ausschnitt ] . for @xmath227 , the simulations do not reach a stationary state within two days of computation . simulations also become very slow in the high density phase especially when the overall motor concentration gets so large that the bound density is close to one and the unbound density is no longer small compared to the bound density . in order to estimate the set of parameters , for which the phase diagram exhibits a maximal current phase , we return to the estimate ( [ e.diffusivecurrent ] ) for the diffusive current @xmath216 and make the simplifying assumption that the bound and unbound densities in the middle compartment of the tube are essentially constant . thus , we replace the true unbound density @xmath228 at the right end of the filament by its bulk value @xmath113 . the bulk density @xmath113 of the unbound motors is related to the bulk density @xmath112 of the bound motors via the radial equilibrium relation ( [ rdb_formel ] ) . for the maximal current phase , one has @xmath229 and ( [ rdb_formel ] ) leads to @xmath230 for small @xmath231 . in this way , we arrive at the estimate j_dif , r ( 1+n_ch ) d_ub _ ub , ex = 0 . [ e.estimate2 ] the maximal current phase should be present in the phase diagram if the diffusive current @xmath216 exceeds the maximal current @xmath95 on the filament . it then follows from ( [ e.estimate2 ] ) that the maximal current phase should be present for @xmath232 provided the tube radius @xmath1 satisfies r^2 > r_*^2 [ e.presencemcphase ] where @xmath233 has been used . using the same line of reasoning , a second , less restrictive condition can be obtained from an estimate for the diffusive current within the left compartment . the threshold value @xmath234 for the tube radius as given by ( [ e.presencemcphase ] ) has been confirmed by monte carlo simulations for the jump probabilities @xmath235 with @xmath236 , the sticking probability @xmath237 , the compartment size @xmath238 , and the boundary densities @xmath239 and @xmath240 . the jump probabilities imply the bound motor velocity @xmath241 ; the diffusion coefficient @xmath26 of the unbound motors has the value @xmath27 ( since the resting probability @xmath29 as mentioned above ) . when these parameter values are inserted into ( [ e.presencemcphase ] ) , one obtains the estimate @xmath242 . the corresponding monte carlo data are displayed in fig . [ fig : j_rhob_r_abh ] . inspection of this figure shows that the current does indeed attain its maximal value @xmath95 for @xmath243 with @xmath244 . the same fig . [ fig : j_rhob_r_abh ] also shows the transition from the low density to the high density phase which occurs for a tube radius @xmath245 which satisfies @xmath246 . a complete phase diagram for a tube with radius @xmath247 is shown in fig . [ fig : phasendiagramm_r17b5 ] . again , most of the phase diagram is covered by the high density phase ( hd ) , while a low density phase ( ld ) is found only for very small values of @xmath135 . the maximal current phase ( mc ) is present now but only for very small values of @xmath223 . it is interesting to note that similar effects also occur in the purely one - dimensional system if one considers a driven system which is bounded by two segments which exhibit only diffusive transport . in this 1dimensional case , quantitative predictions can be made using a mean field approximation as will be discussed elsewhere @xcite . the monte carlo data displayed so far have been obtained for active unbinding of a motor particle which is bound to the last filament site . as mentioned above , another possibility is that the motor particle gets stuck at the last filament site and unbinds only by thermal excitations , i.e. , with unbinding probability @xmath16 . for these two different unbinding mechanisms , one will , in general , obtain different density profiles . however , this difference is not dramatic as one can see from fig . [ fig : profile_abloesen_eps_v ] which exhibits density profiles for both cases . although the probability for a forward step at the last filament site differs by two orders of magnitude for the two cases , the bulk density exhibits a relatively small difference . the bound density , on the other hand increases and decreases close to the right end of the filament for thermal and active unbinding from the last site , respectively . let us summarize the main results and add a few comments . we have studied a lattice model for the motion of many molecular motors in an open tube which contains a single filament . when bound to the filament , the motor particles undergo an asymmetric simple exclusion process ( asep ) . in addition , motors can unbind from the filament and then diffuse freely in the tube . as for the asep in one dimension , the motor traffic in open tubes can exhibit three different phases : high density and low density phases which are characterized by an exponential decay of the density deviations from their bulk values and maximal current phases characterized by an algebraic decay . therefore , the molecular motor traffic in open tubes are promising candidates for the experimental observation of boundary induced non equilibrium phase transitions . in general , the location of the transition lines is found to depend on the precise choice of the boundary conditions . apart from periodic boundary conditions , case ( a ) , we studied two different boundary conditions ( b ) and ( c ) for an open tube . in case ( b ) , the bound and unbound densities are kept fixed at the boundaries and satisfy radial equilibrium . for this case , the location of the transition lines is independent of the model parameters , and the phase diagram of the asep in one dimension is recovered . in case ( c ) , the active compartment of the tube is bounded by two compartments where the transport is purely diffusive . in this latter case , the phase diagram depends on the geometry of the tube and on the transport properties in the bound and unbound motor states . in many cases , the maximal current phase is completely suppressed by the coupling to the diffusive compartments which act as bottle necks for the transport . the theoretical results described here should be accessible to experiments on cytoskeletal filaments and motors . in particular , the motor traffic through open tubes as discussed here provides new opportunities to study the transport properties of aseps by systematic experiments . we thank theo m. nieuwenhuizen for comments and discussions during the initial stage of this work . in the following , we show that eq . ( [ rdb_formel ] ) , the condition for radial equilibrium , holds exactly in the case of periodic boundary conditions using the quantum hamiltonian formalism @xcite . the exact stationary master equation can be written in the form @xmath248 where @xmath249 is the quantum hamiltonian of the stochastic process and @xmath250 is a vector in a product hilbert space ; each lattice site is represented by a two - state system with the orthogonal vectors @xmath251 for an occupied lattice site ( `` spin down '' ) and @xmath252 for a vacancy ( `` spin up '' ) . because of translational invariance in the direction parallel to the filament , there can not be any radial currents in our system , and the unbound density is independent of the radial coordinate ; thus we can restrict the analysis to the case of one unbound channel . we denote by @xmath253 and @xmath254 the state of site @xmath255 of the bound and unbound channel , respectively . using the general recipe given in chapter 2 of ref . @xcite we construct the quantum hamiltonian @xmath249 of our system : @xmath256 where @xmath257 represents the dynamics of the asymmetric exclusion process in the bound channel , @xmath258 the symmetric exclusion process in the unbound channel and @xmath259 the coupling of the two channels . each term can be written as a sum @xmath260 with @xmath261 @xmath262 is a creation operator for a particle at site @xmath255 of the bound channel , and @xmath263 is the corresponding annihilation operator . @xmath264 is the particle number operator at site @xmath255 of the bound channel and @xmath265 . operators for the unbound channel are defined in an analogous manner . the off - diagonal parts of the operators @xmath266 represent hopping of particles , while the diagonal parts are determined by conservation of probability , see chapter 2 of ref . @xcite . we now show that the product measure @xmath267 \otimes \left [ ( 1-\rho_{{\rm ub}})|0\rangle_{k,{{\rm ub } } } + \rho_{{\rm ub}}|1\rangle_{k,{{\rm ub}}}\right]\big),\ ] ] is a stationary state , if eq . ( [ rdb_formel ] ) holds , i.e. that the radial equilibrium condition implies @xmath268 . the product measure @xmath250 defines a state where the density is @xmath36 at each lattice site of the bound channel and @xmath37 at each site of the unbound channel and spatial correlations vanish . @xmath257 and @xmath258 do not couple the bound and unbound channel , therefore we can consider them separately and refer to the result , that for the symmetric as well as for the asymmetric exclusion process , the product measure @xmath250 is stationary in the case of periodic boundary conditions @xcite . for a proof using the quantum hamiltonian formalism , see chapter 7.1.2 of ref . @xcite : in the case of the asep in one dimension , for example , one can easily check , that @xmath269 , therefore the summation gives zero for periodic boundary conditions . hence @xmath270 and @xmath271 . concerning @xmath259 , it is sufficient to consider a single site @xmath255 in both channels . doing some steps , we find @xmath272 and @xmath273 for the product measure ( [ productmeasure ] ) . hence from @xmath274 we obtain eq . ( [ rdb_formel ] ) , which is the condition for radial equilibrium . in this appendix , we give the details of the continuum mean field approximation for the two state model introduced in section [ sec : profile ] . we insert the decomposition @xmath275 and @xmath276 as introduced in ( [ e.densitydecomposition ] ) into the mean field equations ( [ mf1 ] ) and ( [ mf2 ] ) and expand these equations up to second order in the density deviations @xmath117 and @xmath118 . as a result , we obtain @xmath277 and @xmath278 with @xmath279 note that radial equilibrium for the densities @xmath36 and @xmath37 would imply that the right hand side of ( [ mf2_entw ] ) vanishes . let us first consider the case @xmath119 which applies to the high and low density phases . in this case , we can neglect the second order terms and obtain two equations which are linear in the density deviations . the second equation ( [ mf2_entw ] ) can then be solved for @xmath118 which leads to @xmath280 with @xmath281 when this expression is inserted into ( [ mf1_entw ] ) , we obtain the first order relation @xmath282 for @xmath117 . we now make the ansatz @xmath283 which leads to the cubic equation ( [ e.xi ] ) for the decay length @xmath121 . now we consider the maximal current phase , i.e. the case @xmath136 . in this case the linear terms are zero . furthermore , we neglect terms of order @xmath284 and @xmath285 as can be justified a posteriori since @xmath117 is found to decay as an inverse power of @xmath7 . thus , up to leading order , we can ignore the left hand side of ( [ mf2_entw ] ) , and the right hand side of ( [ mf2_entw ] ) vanishes . this implies radial equilibrium for the asymptotic decay to the homogeneous solution . up to this order , we find @xmath286 for small @xmath117 and therefore @xmath287 if this latter expression is inserted into ( [ mf1_entw ] ) , we obtain @xmath288 which leads to @xmath289 for @xmath69 , this asymptotic behavior is identical to the one found from the mean field approximation for the asep in one dimension@xcite . 10 j. howard . . sinauer associates , sunderland ( mass . ) , 2001 . g. woehlke and m. schliwa . walking on two heads : the many talents of kinesin . , 1:5059 , 2000 . vale , t. funatsu , d. w. pierce , l. romberg , y. harada , and t. yanagida . direct observation of single kinesin molecules moving along microtubules . , 380:451453 , 1996 . thorn , j. a. ubersax , and r. d. vale . engineering the processive run length of the kinesin motor . , 151:10931100 , 2000 . r. lipowsky , s. klumpp , and th . m. nieuwenhuizen . random walks of cytoskeletal motors in open and closed compartments . , 87:108101.14 , 2001 . th . m. nieuwenhuizen , s. klumpp , and r. lipowsky . walks of motors in two and three dimensions . , 58:468474 , 2002 . s. katz , j. l. lebowitz , and h. spohn . nonequilibrium steady states of stochastic lattice gas models of fast ionic conductors . , 34:497537 , 1984 . kolomeisky , g. m. schtz , e. b. kolomeisky , and j. p. straley . phase diagram of one - dimensional driven lattice gases with open boundaries . , 31:69116919 , 1998 . macdonald , j. h. gibbs , and a. c. pipkin . kinetics of biopolymerization on nucleic acid templates . , 6:125 , 1968 . macdonald and j. h. gibbs . concerning the kinetics of polypeptide synthesis on polyribosomes . , 7:707725 , 1969 . assembly of the peptide chains of hemoglobin . , 47:247261 , 1961 . naughton and h. m. dintzis . sequential biosynthesis of the peptide chains of hemoglobin . , 48:18221830 , 1962 . winslow and v. m. ingram . peptide chain synthesis of human hemoglobin a and a@xmath290*. , 241:11441149 , 1966 . f. spitzer . interaction of markov processes . , 5:246290 , 1970 . j. krug . boundary - induced phase transitions in driven diffusive systems . , 67:18821885 , 1991 . g. schtz and e. domany . phase transitions in an exactly soluble one - dimensional exclusion process . , 72:277296 , 1993 . b. derrida , m. r. evans , v. hakim , and v. pasquier . exact solution of a 1d asymmetric exclusion model using a matrix formulation . , 26:14931517 , 1993 . t. m. liggett . . springer , berlin , 1999 . v. popkov , l. santen , a. schadschneider , and g. m. schtz . empirical evidence for a boundary - induced nonequilibrium phase transition . , 34:l45l52 , 2001 . goldstein and z. yang . microtubule - based transport systems in neurons : the roles of kinesins and dyneins . , 23:3971 , 2000 . dahlstrm , k. k. pfister , and s. t. brady . the axonal transport motor kinesin is bound to anterogradely transported organelles : quantitative cytofluorimetric studies of fast axonal transport in the rat . , 141:469476 , 1991 . h. van beijeren , r. kutner , and h. spohn . excess noise for driven diffusive systems . , 54:20262029 , 1985 . k. h. janssen and k. oerding . renormalized field theory and particle density profile in driven diffusive systems with open boundaries . , 53:4544 - 4554 , 1996 . j. s. hager , j. krug , v. popkov , and g. m. schtz . minimal current phase and universal boundary layers in driven diffusive systems . , 63:056110/1 - 056110/12 , 2001 . t. surrey , f. ndlec , s. leibler , and e. karsenti . physical properties determining self - organization of motors and microtubules . , 292:11671171 , 2001 . s. klumpp and r. lipowsky . to be published . schtz . exactly solvable models for many - body systems far from equilibrium . in c. domb and j. l. lebowitz , editors , _ phase transitions and critical phenomena _ , volume 19 , pages 1251 . academic press , san diego , 2001 . symbol= definition= @xmath291 abbreviation for @xmath292 used in subsection [ sec : profilemc ] + ( a ) periodic boundary conditions , see fig . [ fig : motor_tube_randbed ] + @xmath293 abbreviation used in appendix [ app_mf ] , defined in eq . ( [ def_ab ] ) + @xmath14 probability for a forward step of the bound motor + @xmath294 abbreviation for @xmath295 used in subsection [ sec : profilemc ] + ( b ) open tube , radial equilibrium at the boundaries , see fig . [ fig : motor_tube_randbed ] + @xmath296 abbreviation used in appendix [ app_mf ] , defined in eq . ( [ def_ab ] ) + @xmath15 probability for a backward step of the bound motor + ( c ) open tube , diffusive injection and extraction of motors , see fig . [ fig : motor_tube_randbed ] + @xmath297 diffusion coefficient for the 1dimensional asep + @xmath109 diffusion coefficient of the bound motor particle + @xmath218 distance between filament end and tube end for boundary conditions ( c ) + @xmath26 diffusion coefficient of the unbound motor particle + @xmath150 scale parameter for the scale - dependent diffusion coefficient + @xmath231 unbinding probability of a bound motor + @xmath298 rescaled unbinding probability @xmath299 + @xmath300 deviation of the bound density from the constant bulk value + @xmath301 deviation of the unbound density from the constant bulk value + @xmath302 parameter defined by eq . ( [ def_gamma ] ) + @xmath17 resting probability of a bound motor + @xmath106 ( global ) current through the tube + @xmath34 local bound current + @xmath303 local unbound current + @xmath215 diffusive current in the left diffusive compartment for boundary conditions ( c ) + @xmath216 diffusive current in the right diffusive compartment for boundary conditions ( c ) + @xmath0 length of the tube + @xmath4 lattice constant on filament + @xmath304 length of the filament + @xmath10 number of unbound channels + @xmath2 number of motors in the tube which is fixed for periodic boundary conditions ( a ) + @xmath11 cross section of the tube + @xmath30 sticking probability for a motor hopping to the filament + @xmath305 rescaled sticking probability @xmath306 + @xmath102 radial spatial coordinate + @xmath1 radius of the tube + @xmath234 minimal tube radius for which a maximal current phase is found in case ( c ) + @xmath307 density in the case of the 1dimensional asep + @xmath103 density of motors bound to the filament + @xmath308 right boundary density on the filament + @xmath97 left boundary density on the filament + @xmath112 constant bulk density on the filament + @xmath74 right boundary density for the 1dimensional asep + @xmath72 left boundary density for the 1dimensional asep + @xmath32 overall motor concentration + @xmath76 constant bulk density for the 1dimensional asep + @xmath309 density of unbound motors + @xmath223 density of unbound motors at the right boundary + @xmath135 density of unbound motors at the left boundary + @xmath113 unbound constant bulk density + @xmath21 basic time unit + @xmath310 velocity for the 1dimensional asep + @xmath19 velocity of bound motor + @xmath311 velocity of density fluctuations + @xmath90 domain wall velocity + @xmath7 spatial coordinate parallel to the filament + @xmath151 scale parameter for the scale - dependent diffusion coefficient , see eq . ( [ scale_dep_d_b ] ) + @xmath312 crossover lengths + @xmath121 decay length for the density deviations + @xmath313 localization length for the 1-dimensional asep + @xmath314 spatial location of the left tube end + @xmath5 rescaled and shifted spatial coordinate , @xmath315 + @xmath316 rescaled crossover lengths

several arguments suggest that stochastic condensation of cold gas and its accretion onto the central supermassive black hole ( smbh ) is essential for active galactic nuclei ( agn ) feedback to work in the most massive galaxies that lie at the centers of galaxy clusters . our 3-d agn jet - icm ( intracluster medium ) simulations show that the angular momentum of the cold gas crossing kpc is essentially isotropic . with almost equal mass in clockwise and counter - clockwise orientations , we expect a cancellation of angular momentum on roughly the dynamical time . this means that a compact accretion flow with a short viscous time ought to form , through which enough accretion power can be channeled into jet mechanical energy sufficiently quickly to prevent a cooling flow . the inherent stochasticity , expected in feedback cycles driven by cold gas condensation , gives rise to a large variation in the cold gas mass at the centers of galaxy clusters , for similar cluster and smbh masses , in agreement with the observations . such correlations are expected to be much tighter for the smoother hot / bondi accretion . the weak correlation between cavity power and bondi power obtained from our simulations also match observations . galaxies : clusters : intracluster medium galaxies : halos galaxies : jets

1611.02710

there is a general consensus that the intracluster medium ( icm ) in low - entropy galaxy cluster cores is able to remain in rough _ global _ thermal equilibrium because the powerful jets from the central agn inject sufficient energy to compensate for radiative losses ( for a review , see @xcite ) . there are , however , a number of key issues associated with this radio feedback " schema that have yet to be firmly pinned down , one of which is : how is the central supermassive black hole ( smbh ) able to transport sufficient energy to the cluster core ( @xmath1 10s of kpc ) at a timescale shorter than the core cooling time ( @xmath2 gyr ) . there are two scenarios proposed for agn feedback : ( 1 ) hot / bondi accretion which invokes spherical accretion with the mass flow rate set by the hot gas density and temperature on the cluster core scales ; and ( 2 ) the competing cold mode accretion model , in which a fraction of the gas in the cluster cores condenses into cold clouds and filaments ( when the ratio of the cooling time to the free - fall time [ @xmath3 falls below a threshold value in the range 5 - 20 ; e.g. , @xcite ) , which then decouple from surrounding hot intracluster medium ( icm ) , rain down upon the central smbh , and power agn feedback strong enough to prevent catastrophic cooling in the core . of the two , the bondi accretion model , though more frequently invoked in the literature , has a number of shortcomings . for one , bondi accretion is only applicable if the accretion flow is spherically symmetric and single phase , both conditions that are at odds with the multiphase gas with angular momentum observed in several cool cores ( e.g. , see @xcite ) . moreover , for a polytropic hot flow @xmath4 , the mass accretion rate is given by ( @xcite ) @xmath5 where @xmath6^{1/2}$ ] is the sound speed in the ambient medium far from the black hole ; the mean particle mass @xmath7 ; @xmath8 is the black hole mass in units of @xmath9 ; @xmath10 and @xmath11 are the ambient particle number density and temperature scaled to @xmath12 @xmath13 and 1 kev , respectively ; and since @xmath14 varies weakly with @xmath15 ( @xmath16 varies from 1/4 for @xmath17 to 1.1 for @xmath18 ) we adopt @xmath19 for convenience . in three nearby systems where both the ambient conditions governing the bondi flow rate and the mass accretion rate onto the black hole can be deduced , the former is at least two orders of magnitude too large . these nearby examples are sgr a * , the galactic center bh [ bondi rate of @xmath20 ( @xcite ) vs. the accretion rate at @xmath21 of @xmath22 ( @xcite and references therein ) ] ; ngc 3115 [ bondi rate of @xmath23 vs. the accretion rate onto the bh at least two orders of magnitude smaller ( @xcite ) ] ; and m87 [ bondi rate of @xmath24 @xmath25 vs. the accretion rate onto the bh of @xmath26 ( @xcite ) ] . @xcite find that the median efficiency ( defined as the ratio of the observed jet power and @xmath27 ; @xmath28 is the smbh accretion rate ) required to supply the cavity power of nearby radio galaxies is 300% , assuming @xmath29 typical of similar nearby hot accretion flows . this implies that hot accretion , with reasonable assumptions , is insufficient for powering most of the cavities observed in cluster cores . at the extreme end of the spectrum , the estimated jet / cavity power in some of the galaxy clusters ( e.g. , ms0735 , cen a ) is found to be much larger than @xmath30 ; i.e. , the bondi mass flow rate is insufficient to power the observed x - ray cavities _ even if all the hot gas in the bondi flow is accreted by the smbh _ ( @xcite ) . and finally , the accretion rate in the bondi regime takes a long time to adjust to the conditions ( @xmath31 ) changing in the cluster cores ( @xmath32 kpc ) at the core cooling timescale @xmath1 few 100 myr ( e.g. , @xcite ) . the cold mode feedback model sidesteps many of the challenges associated with the hot / bondi model . very briefly ( we refer the readers to @xcite , @xcite , and @xcite [ hereafter ] for a more expansive discussion ) , the model is based on the realization that even if the icm in the cluster core exists in rough global thermal balance , if the ratio of the cooling time and the gravitational free - fall time ( @xmath33 ) falls below a critical value ( @xmath34 or so ; @xcite ) , the gas will become susceptible to local thermal instabilities and fragment , leading to the formation of a multiphase medium consisting of cold dense clouds condensing from the hot diffuse icm itself . the clouds then fall towards central agn , resulting in increased smbh accretion and feedback that , in turn , quenches runaway cooling in the cluster core . the main question with this model is : will the infalling cold gas , which has non - zero angular momentum , end up forming a standard viscosity - mediated accretion flow ( saf ) in which the gas will flow inwards on a ( long ) viscous accretion timescale , or will the gas circularize sufficiently close to the smbh and accrete on a timescale shorter than the core cooling time . in this paper , we use 3-d hydrodynamic simulations of agn jets interacting with the icm to investigate this problem . section [ sec : sims ] briefly presents our numerical setup . in section [ sec : ang_mom ] , we calculate the circularization radius and viscous time of the cold gas based on the standard accretion physics . we show that the angular momentum distribution of cold gas crossing @xmath32 kpc is close to isotropic which will lead to angular momentum cancellation and the formation of a compact accretion flow . we use the mass distribution of angular momentum to estimate the accretion rate onto the smbh . in section [ sec : obs ] , we compare some of the results from our simulations against key observational . in section [ sec : astro ] we discuss the implications of our results , highlighting the need for better understanding of accretion . we summarize our paper in section [ sec : summary ] . [ tab : tab1 ] + the resolution of all runs , done in spherical @xmath35 coordinates , is @xmath36 . logarithmic grid is used in the @xmath37 direction , and a uniform one in others . @xmath38 is the average cold / hot mass flow rate across @xmath39 ; @xmath40 is the estimate of average bondi accretion rate using power - law ( pl ) and isothermal beta model ( @xmath41 ) extrapolations ; @xmath42 is the total cold ( @xmath43 kev ) gas mass in the simulation domain by the end . + @xmath44 is the smbh accretion rate calculated using eq . [ eq : mdot]/[eq : mdot_sca ] . + @xmath45 jet efficiency relative to @xmath46 ( see eq . 6 in ) . + @xmath47 different quantities ( e.g. , @xmath40 ) are averaged over this timescale . in this paper we discuss three 3-d agn jet - icm simulations carried out in spherical @xmath48 coordinates . the details of the numerical set up are given in . the three runs are : ( i ) an initially hydrostatic icm in a fixed nfw potential ( this is the fiducial run in with the halo mass of @xmath49 ) ; ( ii ) the same run but with the inner and outer radii reduced by half ; ( iii ) and a nfw+bcg ( the latter is modeled as a singular isothermal potential with circular velocity @xmath50 km s@xmath51 ) potential run . the nfw+bcg run is carried out with pluto ( @xcite ) code , whereas the nfw runs are done with zeus - mp ( @xcite ) . roughly similar results are obtained with the two codes . the initial entropy profile for both nfw and nfw+bcg runs are identical ( see eq . 7 in ) . the initial icm density for the nfw+bcg run is almost twice of the nfw runs ( which have identical initial density profiles ) . the different runs have different inner ( @xmath39 ) and outer ( @xmath52 ) radial boundaries ( see table [ tab : tab1 ] ) . the density and pressure at the outer radial boundary is fixed to their initial values . the total mass ( including both hot and cold phases ) crossing the inner boundary is calculated ( @xmath46 ) , and a fraction @xmath53 of @xmath54 is put back into the icm in the form of kinetic power ( @xmath55 ) of agn jets . all our runs behave in a very similar manner . there are cooling and heating cycles , and the feedback driven heating events are less frequent for a higher feedback efficiency . although the feedback efficiency , @xmath53 , is ten times higher for the nfw+bcg run , because of a deeper potential well at the center ( due to the bcg contribution ) , its behavior is similar to the fiducial nfw run . the nfw run with @xmath56 kpc uses the same efficiency parameter ( @xmath53 ) as the fiducial run with @xmath57 kpc . however , because of a smaller inner radius and non - zero angular momentum of the condensing gas , the mass accretion rate across the inner boundary is lower as some of the gas circularizes at @xmath58 kpc . consequently , the feedback energy input is smaller and the run shows a higher accumulation of cold gas in the central regions ( see @xmath42 in table [ tab : tab1 ] ) . density and temperature profiles , and other x - ray and jet properties are discussed in detail in , and will not be repeated in this paper . here we focus on the angular momentum distribution of cold gas in our simulations and the plausibility of stochastic accretion onto smbhs ( section [ sec : ang_mom ] ) , and comparison of simulation results with observations ( section [ sec : obs ] ) . our simulations , like @xcite , use the total ( dominated by cold gas ) accretion rate at @xmath32 kpc to estimate the feedback energy deposited by jets . one of the major open problems with cold mode feedback is : how does the cold gas originating at @xmath59 kpc lose angular momentum and get rapidly accreted onto the smbh . this is needed for feedback to prevent a cooling flow . although it is numerically formidable to resolve both the cluster core and accretion flow onto the smbh , we attempt to estimate the accretion rate onto the smbh based on the angular momentum distribution of cold gas within our simulation domain . while the large angular momentum gas forms a massive disk at @xmath59 kpc , the smaller angular momentum gas crossing the inner boundary has stochastic angular momentum changing over short timescales . rapid angular momentum cancellation in this stochastic cold gas can allow substantial cold gas to be channeled to the smbh sufficiently fast to prevent a cooling flow ( e.g. , see @xcite ) . in this section we review the standard accretion physics that we apply later to estimate the accretion rate onto the smbh ( section [ sec : saf ] ) . for a given specific angular momentum ( @xmath60 ) , the circularization radius of the gas is given by @xmath61 where @xmath62 is the mass enclosed ( including only smbh + dark matter contributions ; i.e. , ignoring gas mass ) within the circularization radius . the viscous accretion time at the circularization radius can be expressed in terms of the specific angular momentum , @xmath63 where @xmath64 is the angular velocity , @xmath65 is the viscosity parameter ( @xcite ) , and @xmath66 is the ratio of the disk height and radius . we scale the numerical values using @xmath67 . this is consistent with results from mhd simulations of magnetized accretion flows , which find that the effective value of @xmath65 is 0.1 over the bulk of the flow ( @xcite ) . although the realistic @xmath66 for standard agn disks is @xmath68 ( e.g. , see eq . 6 in @xcite ) , we scale our results to @xmath69 because the standard choice gives a very small smbh accretion rate . moreover , stochastic accretion simulations show a thick cold flow ( @xmath70 ; see section [ sec : sca ] ) supported by turbulent motions ( see fig . 16 in @xcite ) . incidentally , radiatively inefficient accretion flows ( riafs ) also have similar thick disks ( @xcite ) . ; bottom panel ) and viscous time ( eq . [ eq : tvisc ] ; top panel ) as a function of specific angular momentum ( @xmath60 ) for nfw+bcg and nfw+bcg+smbh potentials . as expected , the smbh dominates the enclosed mass within the bondi radius . magenta arrows represent the bondi radius , the corresponding @xmath60 ( eq . [ eq : rcirc ] ) , and @xmath71 ( eq . [ eq : tvisc ] ) . black arrows represent quantities corresponding to the inner simulation radius of 0.5 kpc ( dot - dashed line for nfw and dashed line for nfw+bcg ) and 1 kpc ( solid line ) . brown arrows show the viscous time of 0.2 gyr . ] figure [ fig : r_t_vs_l ] shows the viscous time ( top panel ) and circularization radius ( bottom panel ) , as a function of the specific angular momentum , for the two gravitational potentials ( nfw and nfw+bcg ) used in our simulations . note that the inner radius of our computational domain is well beyond the bondi radius , and the smbh contributionto gravity is negligible everywhere in the computational domain ( @xmath72 ) . the slope of @xmath71 and @xmath73 as a function of @xmath60 changes at the radius at which the enclosed mass is dominated by the extended potential rather than the smbh . the arrows in figure [ fig : r_t_vs_l ] mark important quantities such as bondi radius , inner radius of the computational domain , and a cooling time of 0.2 gyr . in this section we discuss the angular momentum distribution of cold gas in our numerical simulations . we show that the time - averaged angular momentum distribution of cold gas crossing our inner simulation boundary is stochastic , with almost equal mass going around in different directions . this implies that in the region of circularization , angular momentum cancellation will take place almost at the local dynamical time . moreover , the cold gas with @xmath74 @xmath75 s@xmath51 changes its mean angular momentum over a timescale shorter than the core cooling time ; i.e. , cold gas with small angular momentum is able to respond fast enough to close the feedback loop . to start with , we emphasize that when considering gas whose specific angular momentum ( @xmath76 ) distribution is stochastic , it is essential to explicitly account for the vector nature of the specific angular momentum , @xmath77 \bm{\hat{x } } + r[-v_\phi \cos \theta \sin \phi + v_\theta \cos \phi ] \bm{\hat{y } } $ ] ; where @xmath78 is the cylindrical / spherical radius ) . in standard accretion disk literature , @xmath60 is generally taken to be @xmath79 since the gas is considered to be coplanar . however , in our simulations the low angular momentum cold gas distribution shows almost equal importance of all the components of @xmath76 . ( mass inflow rate of cold gas across the inner simulation boundary ) distribution ( pdf ) of @xmath80 and @xmath81 for the fiducial nfw run with @xmath57 kpc . the slopes are similar to the mass distributions in fig . [ fig : l_pdf ] . the mass accretion rate is multiplied by @xmath82 , the free - fall timescale at the inner boundary ( 15 myr for the fiducial run ) . not only are the @xmath83 , @xmath84 and @xmath79 pdfs similar , the clockwise and counter - clockwise pdfs are also very similar . there is a slight bias in the @xmath79 distributiontoward larger values of @xmath79 in the clockwise sense , which is a signature of the massive rotating torus seen more prominently for the same run in fig . [ fig : l_pdf ] . ] to examine this , we consider the quantity @xmath85 , the infinitesimal mass in the computational domain with angular momentum between @xmath60 and @xmath86 ( we plot mass distribution in a logarithmic bin @xmath87dm / dl$ ] , as one can simple read off the mass from the figure ) . we can choose @xmath60 to be @xmath81 , or @xmath83 , @xmath84 or @xmath79 . the left panel of figure [ fig : l_pdf ] shows the time - averaged cold mass distribution as a function of @xmath88 ( both clockwise and counter - clockwise rotations are shown ) for our three runs . the right panel shows the time - averaged mass distribution as a function of the _ total _ specific angular momentum @xmath81 . comparing the @xmath81 and @xmath88 distributions , we find a dramatic lack of cold gas with small @xmath81 but a non - negligible mass of cold gas with very low @xmath88 . this gas with low @xmath88 has large @xmath89 and @xmath90 components , and therefore is not expected to circularize in the x - y plane . focussing on gas with @xmath91 ( corresponding to the circularization radius in eq . [ eq : rcirc ] equal to the inner radial boundary @xmath39 ) , where @xmath92 @xmath75s@xmath51 , we find that in all our simulations the low angular momentum gas ( @xmath91 ) has roughly similar distribution in @xmath79 and @xmath81 , with @xmath93 and @xmath94 . ( with respect to the velocity along a fixed axis ) is shallower compared to the distribution with respect to the absolute value of the total velocity @xmath95 . ] in fact , both the clockwise and counter - clockwise components of angular momentum distribution have very similar profiles for @xmath96 . since the mass probability density function ( pdf ) in fig . [ fig : l_pdf ] does not peak as abruptly for simulations with lower @xmath97 ( and smaller @xmath39 ) , we assert that the low-@xmath60 mass pdfs are robust ( in the sense that the low-@xmath60 pdf will be similar if @xmath39 is reduced further ) . also , as apparent in figure [ fig : l_pdf ] , there is an excess mass at specific angular momenta larger than @xmath97 . this matter with large angular momentum ( primarily @xmath79 ) circularizes within the computational domain ( see the cold torus in fig . 3 of ) and does not cross the inner boundary , unlike the low angular momentum cold gas . apart from the angular momentum pdf of cold gas mass _ within the computational domain _ , it is instructive to consider the pdf of the cold gas _ crossing the inner boundary _ , which is more relevant for smbh accretion estimates . figure [ fig : l_pdf_fid ] shows the time - averaged angular momentum pdf ( @xmath98 ) with respect to @xmath99 and @xmath81 of cold gas accreting through the inner boundary of our fiducial run . the pdfs relative to all three components ( @xmath83 , @xmath84 , @xmath79 ) are similar , including the clockwise and counter - clockwise components , suggesting that the cold gas crossing the inner boundary is roughly isotropic . the @xmath98 pdf with respect to @xmath81 is truncated at higher @xmath81 , cutting off sharply at @xmath100 @xmath75 s@xmath51 , which corresponds to the specific angular momentum at the inner boundary ( @xmath97 ; see fig . [ fig : l_pdf ] ) ; the gas with larger angular momentum can not overcome the centrifugal barrier and fall in through the inner boundary . the accretion rate pdfs in figure [ fig : l_pdf_fid ] and the mass pdfs in figure [ fig : l_pdf ] have similar slopes at low @xmath60 with respect to @xmath101 and @xmath81 . the left panel of figure [ fig : l_scatter ] represents the rotational direction ( @xmath102 ) of cold gas at all times in the fiducial nfw run . the right panel shows the corresponding rms @xmath103 ( @xmath104 stands for mass - weighted averaging ) . the red scatter circles are for the cold gas _ within _ the computational domain and the green triangles are for the cold gas _ crossing _ the inner boundary . the blue solid ( black dashed ) line is the time - averaged , mass - weighted orientation of cold gas within ( crossing ) the computational domain . the cold gas with @xmath105 @xmath75s@xmath51 has equal scatter around the value expected for the mean and rms angle between @xmath76 and the @xmath106 axis for isotropic distribution ( @xmath107 and @xmath108 ) , implying that the time - averaged angular momentum distribution of the cold gas with small angular momentum within the simulation domain and crossing the inner boundary is almost isotropic . the larger angular momentum gas , on the other hand , has a clockwise bias as also seen in figures [ fig : l_pdf ] & [ fig : l_pdf_fid ] . ( bottom panel ) and rms @xmath109 ( top panel ) of cold gas crossing the inner boundary as a function of time for the fiducial nfw run . the dashed red lines are for all the cold gas crossing the inner boundary . the purple dot - dashed lines with stars show the average orientation of cold gas with @xmath110 @xmath75s@xmath51 and solid blue lines are for @xmath81 @xmath111 @xmath75s@xmath51 . the orientation of cold gas with @xmath105 @xmath75s@xmath51 changes on a short timescale compared to the other two cases . ] now that we have shown that the _ time - averaged _ angular momentum distribution of cold gas with @xmath112 erg s@xmath51 is roughly isotropic , we want to study its time dependence . in particular , stochastic accretion of cold gas is a viable solution of the cooling flow problem only if the low angular momentum cold gas changes direction on a time scale shorter than the core cooling time ( taken to be 0.2 gyr ) . if the angular momentum variability time scale is shorter , the cold gas can undergo angular momentum cancellation ( and hence accretion ) faster than the core cooling time . figure [ fig : l_pdf_time ] shows the orientation of the average mass - weighted angular momentum distribution of cold gas crossing the inner radial boundary ( @xmath113 and rms @xmath114 , same quantities as in fig . [ fig : l_scatter ] ) as a function of time . the orientation of the total cold gas , and @xmath115 @xmath75s@xmath51 and @xmath105 @xmath75s@xmath51 cold gas are shown . the @xmath116 ratios of the total cold gas and with @xmath115 @xmath75s@xmath51 show that this gas only changes its orientation ( as measured by @xmath117 crossing zero ) till only 1.5 gyr , and attain a fixed sense of rotation thereafter . even though this cold gas ( dominated by large angular momentum gas ) shows small fluctuations on short time scales ( < 30 myr ) , it still maintains its orientation for the rest of the simulation time . time of 1.5 gyr coincides with the formation of the massive cold torus in the fiducial simulation ( see fig . 3 in ) . the rms @xmath118 value for this cold gas remains well above 0.58 , suggesting that most of it settles down in a disk - like structure . note that both @xmath113 and rms @xmath114 ( see the purple dashed lines with markers in fig . [ fig : l_pdf_time ] ) are much more variable for low angular momentum cold gas ( @xmath110 @xmath75s@xmath51 ) as compared to the total cold gas . the mean angle between @xmath76 and the @xmath106 axis ( @xmath119 ) fluctuates around zero and the rms @xmath120 fluctuates around the isotropic value ( 0.58 ) on a timescale shorter than 0.2 gyr . therefore , only the low angular momentum gas ( @xmath105 @xmath75s@xmath51 ) is expected to participate in feedback heating of the icm . results from our agn jet - icm simulations show that cold gas accretion at @xmath0 kpc is stochastic , with the angular momentum of cold gas with @xmath121 @xmath75s@xmath51 changing on time scales shorter than the core cooling time . the mixing of these randomly oriented cold gas clouds near the circularization radius will result in angular momentum cancellation . as these clouds lose angular momentum due to mixing , they will fall ballistically to the center , where they ought to establish a compact turbulent accretion flow with @xmath70 ( see @xcite ) and viscous time shorter than 0.2 gyr , the cooling time of the cluster core . from our simulation results , a specific angular momentum @xmath122 @xmath75s@xmath51 can be taken as the limit for stochastic cold accretion ( see figs . [ fig : l_scatter ] & [ fig : l_pdf_time ] ) . in this section we estimate the mass accretion rate onto the central smbh , first using the standard steady accretion flow model and then using the more plausible stochastic cold gas accretion . for the standard flow model estimate , we assume that cold gas forms a thick extended disk before accreting onto the central smbh on the local viscous time at the circularization radius . for stochastic cold accretion , the angular momentum of the randomly oriented gas clouds cancel each other and the clouds settle down in a compact accretion flow very close to central smbh . in this scenario accretion happens at the dynamical time ( roughly the timescale for angular momentum cancellation ) because instead of relying on viscosity to transport out angular momentum , accretion is powered by angular momentum cancellation among infalling clouds with random orientations . typical efficiency required with respect to the mass accretion rate @xmath46 at @xmath32 kpc to suppress the cooling rate to at least @xmath123% of the pure cooling flow value , @xmath124 , is @xmath125 . our cluster mass is @xmath126 ; this @xmath127 is expected to be smaller for lower mass halos ( see fig . 8 in ) , where @xmath55 is the jet power . this means that , assuming a smbh mechanical efficiency of @xmath128 ( @xmath129 ) , @xmath130 is required to sufficiently suppress a cooling flow ; i.e. , at least 0.1% of the cold gas infalling at 1 kpc must be accreted by the smbh on a timescale shorter than the core cooling time . thus , for our fiducial run ( for which @xmath131 ; see table [ tab : tab1 ] ) average @xmath132 is required to suppress a cooling flow . an estimate of the standard viscosity - mediated accretion flow ( saf ) rate onto the smbh , using the mass distribution of the cold gas crossing the inner boundary ( @xmath133 ; here @xmath134 ) , is given as @xmath135},\ ] ] where @xmath136 is the maximum angular momentum of the cold gas that can contribute to accretion and @xmath137 is the angular momentum corresponding to the innermost stable circular orbit ( isco ) . an appropriate upper angular momentum cut - off ( @xmath136 ) is the angular momentum for which the viscous accretion time equals 0.2 gyr , the core cooling time . from eq . [ eq : tvisc ] , the specific angular momentum for a viscous time of 0.2 gyr is @xmath138 @xmath75s@xmath51 . gas with a larger angular momentum accretes on a time scale longer than the core cooling time and hence can not stop a cooling flow . moreover , matter with angular momentum smaller than @xmath139 @xmath75 s@xmath51 ( assuming a non - rotating bh ) falls directly into the smbh with no way to extract its gravitational potential energy . note that we use the sum of the viscous time at the circularization radius ( which is a function of the specific angular momentum ; eq . [ eq : tvisc ] ) and the free - fall time at the inner boundary ( @xmath140^{1/2}=15 $ ] myr for the fiducial run ) as an estimate of the accretion time in the denominator of the integrand because accretion occurs at the local viscous time , only if mass from larger radii is supplied on a faster timescale ( i.e. , cold gas can not fall in faster than @xmath82 ) . the low angular momentum distribution of the gas crossing the inner simulation boundary ( with respect to @xmath81 ; see fig . [ fig : l_pdf_fid ] ) can be roughly approximated as @xmath141 ( @xmath142 ) . [ eq : mdot ] , with this approximation , can be written as @xmath143}.\ ] ] from eq . [ eq : tvisc ] , we can write @xmath144 taking @xmath145 , which corresponds to a viscous time equal to the core cooling time ( eq . [ eq : tvisc ] ) for the fiducial parameters , @xmath146 this value is uncomfortably close to the minimum value required to prevent a cooling flow ( @xmath147 ) . moreover , we have used a large @xmath69 instead of the standard thin agn disk value of @xmath148 . using @xmath149 give @xmath150 and @xmath151 , way too small compared to the required accretion rate . we list the actual value of the integral in eq . [ eq : mdot ] ( rather than making a power - law approximation ) in table [ tab : tab1 ] for @xmath69 . accretion for the stochastic cold gas ( figs . [ fig : l_scatter ] & [ fig : l_pdf_time ] show that @xmath110 @xmath75s@xmath51 cold gas is stochastic in space and time over 0.2 gyr ) is expected to proceed at almost the local free - fall time till the gas settles in a thick disk close to the isco where the viscous time is very short compared to the free fall at the inner boundary of our simulations , @xmath82 . from eq . [ eq : tvisc ] , the viscous accretion time for @xmath152 @xmath75 s@xmath51 is 14 myr . we assume that the stochastic cold gas with initial @xmath153 @xmath75 s@xmath51 cancels its angular momentum to @xmath154 @xmath75 s@xmath51 on a timescale shorter than 15 myr ( the free - fall time at the inner boundary ) . the stochastic cold accretion ( sca ) rate , therefore , can be approximated as @xmath155 where @xmath134 . this estimate assumes that most of the mass accreting onto the smbh originates as cold gas beyond @xmath32 kpc . not much gas condenses within 1 kpc because at most times @xmath156 at these radii . moreover , the available hot gas mass ( out of which cold gas can condense ) decreases inward . we take @xmath157 @xmath158 s@xmath51 as the upper cut - off of the integral in eq . [ eq : mdot_sca ] because cold gas is stochastic only for angular momentum lower than this . the integral in eq . [ eq : mdot_sca ] can be approximated as @xmath159 ( see table [ tab : tab1 ] for the exact value ) for our fiducial run for which @xmath160 . note that the sca estimate is independent of disk parameters ( @xmath65 & @xmath66 ) as we assume that accretion in the sca regime happens at the local free - fall time ( which is shorter than the free - fall time at the inner boundary ) . unlike saf integral ( eq . [ eq : mdot ] ) which is rather insensitive to @xmath136 , sca integral ( eq . [ eq : mdot_sca ] ) increases quadratically with @xmath136 . note that the smbh mass accretion rate in the sca regime is comfortably larger than the required @xmath161 to quench a cooling flow . in fact , we can also comfortably account for the reduction of @xmath28 in the riaf regime due to outflows . ) and stochastic cold accretion ( sca in green dashed line ; eq . [ eq : mdot_sca ] ) for our fiducial run . also shown in magenta solid lines is the expected mass accretion rate onto the smbh assuming that the smbh accretion rate is @xmath162 . red dot - dashed line is the jet power ( see section [ sec : bondi ] on how we calculate the jet / cavity power ) as a function of time . all quantities show large variations because of heating / cooling cycles driven by condensation / overheating . the accretion rate at @xmath39 is less time variable compared to @xmath163 and @xmath164 but overall trends are similar . ] figure [ fig : mdot_vs_time ] shows the jet power , and the mass accretion rates estimated from standard accretion flow ( saf ; eq . [ eq : mdot ] with @xmath69 ) and stochastic cold accretion ( sca ; eq [ eq : mdot_sca ] ) as a function of time . as expected , the sca mass accretion rate is higher on average as compared to saf . also @xmath163 , in comparisonto @xmath164 , is variable on a shorter time . notice the spikes in @xmath163 in the cooling phases of the core and the lack of stochastic accretion when jets overheat the core . similar trends are also seen for @xmath46 ; i.e. , all accretion rates higher in the cooling phases when jet power is small and suppressed in the overheating phases . we discuss the application of stochastic cold accretion estimate based on the angular momentum distribution of cold gas as the basis of a more robust feedback prescription in section [ sec : improv ] . table [ tab : tab1 ] lists important time - averaged quantities from our numerical simulations . the accretion rate estimates based on the standard accretion flow ( saf ; eq . [ eq : mdot ] with @xmath69 ) and stochastic cold accretion ( sca ; eq . [ eq : mdot_sca ] with @xmath165 @xmath75s@xmath51 ) are included , as is the average rate of mass crossing the inner boundary . the nfw run with @xmath56 kpc but the same @xmath53 as the fiducial nfw run ( with @xmath57 kpc ) shows that the average mass accretion ( hot+cold ) through the inner boundary is reduced by a factor of @xmath166 ( 11.6 @xmath25 as opposed to 21.6 @xmath25 for the fiducial run ) but the cold gas mass accumulating within the computational domain ( dominated by the high angular momentum gas ; e.g. , see fig . [ fig : l_pdf ] ) by the end of the simulation is about five times higher . in fact , the average cold gas mass accumulation rate for the @xmath56 kpc run is @xmath167 , only a factor of 2 lower than the pure cooling flow rate . a lot of cold gas crossing 1 kpc circularizes before crossing the inner boundary at 0.5 kpc ( compare the angular momentum pdfs and @xmath97 for different runs in the right panel of fig . [ fig : l_pdf ] ) , and the mass accretion rate at @xmath56 kpc ( @xmath46 ) is much smaller than the average mass cooling / deposition rate . importantly , a smaller @xmath46 for a fixed @xmath53 gives a much lower feedback heating , allowing a large amount of cold gas with sufficient angular momentum to accumulate within the computational domain . the nfw+bcg simulation also uses @xmath56 kpc but with a larger feedback efficiency ( @xmath168 ) , and therefore the mass of the accumulated cold gas is reasonable . these numbers illustrate the weakness of the current feedback models in which the suitable @xmath53 depends , rather sensitively , on the radius at which @xmath46 is measured . instead of measuring @xmath46 at an arbitrary inner radius and tuning @xmath53 to get a reasonable match to observations , our simulations suggest a more reliable way to estimated the smbh accretion rate based on the mass distribution of low angular momentum cold gas ( eq . [ eq : mdot_sca ] with @xmath112 @xmath75s@xmath51 ) . table [ tab : tab1 ] shows that @xmath163 for the nfw run with @xmath56 kpc is about 6.4 times larger than the nfw run with @xmath57 kpc but @xmath46 is about a factor of two smaller for the former . this implies that @xmath28 estimate based on the angular momentum distribution of cold gas is more robust because a larger mass accretion rate ( as obtained from sca ) will prevent excessive cooling and mass deposition seen in the nfw run with @xmath56 kpc . therefore , we anticipate similar outcomes ( similar cold gas mass , small scale accretion rate , etc . ) for feedback simulations with identical @xmath53s if the smbh accretion rate is estimated using the angular momentum distribution of the cold gas . in future we plan to carry out such simulations . while the results in this section strongly suggest the importance of stochastic cold feedback in maintaining thermal equilibrium in cluster cores , it is imperative for us to look at the limitations of our simulations . it is also important to compare our cold mode feedback results with bondi / hot model and with observations . section [ sec : obs ] deals with these issues . in this section we compare the results of cold feedback simulations with the key observations . ; this is the same as the cold gas mass as a function of time in fig . 9 of ) , cold gas secularly builds up with time . depletion of cold gas due to star formation leads to cycles in cold gas mass as a function of time . ] unlike @xcite , our numerical simulations do not allow for the depletion of cold gas due to star formation . this leads to a secular build up of cold gas to unrealistically large values . we can account for the influence of star formation on cold gas depletion using a simple post - processing model . the cold gas mass in presence of cold gas depletion due to star formation is given by @xmath169 where @xmath42 is the total cold mass in the computational domain , @xmath170 is the rate of condensation of cold gas from the hot icm ( i.e. , the derivative of the @xmath171 line in fig . [ fig : mcold_star ] ) , and @xmath172 is the cold gas depletion time scale ( due to star formation ) . therefore , the star formation rate @xmath173 . while in reality , star formation and associated feedback will impact the surrounding gas distribution , we do not take this into account as stellar feedback is subdominant on global scales in cluster cores ( although it may help drive local turbulence in the cold gas ; e.g. , @xcite ) . for a constant @xmath172 , eq . [ eq : cold_mass ] can be solved analytically to give @xmath174 figure [ fig : mcold_star ] shows the cold gas mass as a function of time for different cold gas mass depletion time scales ( @xmath172 ) . for a gas depletion time scale @xmath2 gyr the cold gas mass is always @xmath175 and most of the gas is channeled into stars . for @xmath176 gyr , there are times when there is no cold gas present in the cluster core . cold gas mass in our simulations is on the higher side and would lead to an average star formation rate of @xmath177 m@xmath178yr@xmath51 . this rate is comparable to that seen in some of the bcgs in cool core clusters ( e.g. , @xcite ) but is somewhat on the high side . this is understandable because in we deliberately chose the accretion efficiency of @xmath179 , the minimum value required to bring down the cooling flow rate by a factor of 0.1 . this suggests that the feedback efficiency ( with respect to the accretion rate at 1 kpc ) in typical clusters is between @xmath180 . accretion efficiency much above @xmath181 ( as in @xcite ) leads to strong feedback that wipes out cold gas at radii @xmath182 kpc at late times , and maintains the cluster in the hot state for most of the time . & fig . [ fig : mcold_star ] ) can also reduce the cold gas mass ( of course at the expense of creating more stars ) . ] the mass accretion rate onto the smbh in hot / bondi accretion is strongly dependent on the mass of the smbh ( eq . [ eq : mdotb ] ) . in the standard bondi scenario one does not normally consider the gas to cool to very low temperatures ( e.g. , @xcite ) . cold gas in cluster cores can be naturally understood as a result of local thermal instability in the core satisfying rough global thermal balance . under the assumption that the cold gas mass in the bondi scenario is proportional to the smbh accretion rate , the cold gas mass in cluster cores is expected to be @xmath183 ( i.e. , strongly dependent on smbh and halo properties ) . figure [ fig : mcold_vs_mbh ] , adapted from @xcite , shows the cold gas mass as a function of the smbh mass for different groups and clusters . the observations do not show a strong correlation ( anticorrelation ) between the smbh mass ( cluster temperature ; indicated by the colors of markers ) and the mass of molecular gas . in fact , the largest molecular gas mass occurs in the hottest cluster , unlike what is expected from @xmath184 scaling of @xmath40 ( eq . [ eq : mdotb ] ; this assumes that the temperature of the hot gas at the bondi radius scales with the core x - ray temperature ; x - ray spectra do not show x - ray emitting gas below half the cluster temperature ; e.g. , @xcite ; see also fig . 1a of @xcite ) . similarly , in figure [ fig : mcold_vs_mbh ] the data point with the smallest smbh mass has a large cold gas mass . there is a large scatter in cold gas mass , irrespective of the smbh mass and the cluster temperature , a hallmark of cold mode feedback that is stochastic and shows cooling and heating cycles ( @xcite ) . figure 9 in and figures 2 & 7 in @xcite show that the cold gas mass , for the same smbh and halo , are highly time variable and show a wide distribution . in fact , at some times one may not see any cold gas because star formation consumes all the cold gas before it is replenished by icm cooling ( this is seen in fig . 2 of @xcite who model star formation ; see also our fig . [ fig : mcold_star ] ) . the range of cold gas mass seen in numerical simulations with cold mode feedback is indicated by vertical intervals in figure [ fig : mcold_vs_mbh ] . the observed range of cold gas mass in cluster cores is not too different from that obtained in the simulations of cold mode feedback . kev ] density profiles to the bondi radius ) as a function of cavity power from our fiducial nfw run . simulation data is sampled every 10 myr and there is a large scatter for both extrapolations . also shown as green triangles is the instantaneous accretion power ( @xmath185 ) supplied by the smbh . at most times bondi power is within @xmath186 erg s@xmath51 , but cavity power has a much larger variation . a similar behavior is seen in our other runs listed in table [ tab : tab1 ] . data points from @xcite and the best - fit from @xcite are also shown . ] @xcite noted a tight correlation between the bondi power and the cavity power for nearby elliptical galaxies and argued for the validity of bondi accretion . @xcite recently refined the determination of bondi and cavities powers from allen et al.s sample and included more data points , and found that the correlation is much weaker . figure [ fig : bondi ] shows the bondi - power cavity - power relationship obtained for our fiducial nfw run where , as we have discussed , the jet is triggered by cold gas condensation . bondi power @xmath187 , where @xmath40 is given by eq . [ eq : mdotb ] , is calculated using the hot gas density extrapolated to the bondi radius , @xmath188 we use two different best - fit models for the hot gas density applied from @xmath39 to 10 kpc : a power - law [ pl ] model ( red squares in fig . [ fig : bondi ] ) ; and a @xmath41 model ( blue stars ) with the hot gas density going like @xmath189^{-3\beta/2}$ ] ( this is similar to what @xcite do ) . a power - law profile gives a larger bondi accretion rate ( see eq . [ eq : mdotb ] ) because the density extrapolated to the bondi radius is higher . the temperature at the bondi radius ( needed to calculate @xmath190 in eq . [ eq : rb ] ) is assumed to be the same as that at @xmath39 ; results are unchanged with other reasonable extrapolations of the temperature . also shown in figure [ fig : bondi ] is the instantaneous value of the feedback power , where @xmath185 ( green triangles ) , @xmath46 is the total [ hot+cold ] accretion rate measured at the inner radius @xmath39 . _ estimating cavity power : _ the cavity power in figure [ fig : bondi ] is calculated in the usual way . to estimate the jet / cavity energy we compute @xmath191 over the volume of low density regions ( density lower than 0.2 times the initial minimum density ; results are insensitive to the exact value of the threshold as long as it is smaller than the initial minimum density ) , corresponding to the enthalpy of a relativistic bubble . to convert this into power , we divide the energy by an estimate of the jet sound crossing time using the outermost radius of the bubble material ( @xmath192 $ ] ; 600 km s@xmath51 is the sound speed for the 1.5 kev cluster core plasma ; see the bottom - right panel in fig . 2 of ) . this definition is slightly different from what was used in , but the results are independent of the details of jet power and bondi estimates . as discussed in , there is a large scatter between cavity power and the instantaneous accretion power because of hysteresis in cold mode feedback . more importantly , the scatter and the lack of obvious correlation in the observational data from @xcite in figure [ fig : bondi ] is similar to that for the bondi power - cavity power scatter seen in our fiducial simulation . notice that the variation in bondi power is smaller than the variation in cavity power in the simulations . this is because the hot gas properties in the core ( on which the bondi estimate depends ) do not vary as much in time as the cold gas accretion rate . moreover , cooling and accretion events occur closely spaced in time and our feedback prescription can blow really powerful cavities ( see jet power in fig . [ fig : mdot_vs_time ] ) . the observational data point in figure [ fig : bondi ] with the highest bondi power is for m87 , for which @xcite use a smbh mass of @xmath193 , much larger than our fiducial choice of @xmath9 ; recalling that @xmath194 brings this point in agreement with the simulation results . the highest cavity power scales with the halo mass and our halo mass is on the higher side . also , unlike simulations , observations can miss out fainter parts of cavities , leading to an underestimate of the cavity power . to conclude , most observations of cool core clusters are explained by heating and cooling cycles driven by cold mode feedback . the problems with bondi accretion ( hot accretion in general ) are well recognized ( see section [ sec : intro ] ) . standard viscous mediated accretion flows onto smbhs , even with a large @xmath195 , leads to the angular momentum problem ; namely , the viscous accretion time scale for the accretion flow is much longer than the cooling time of the cluster core . therefore , enough gas can not accrete fast enough and provide sufficient feedback power to prevent a strong cooling flow . a number of published studies have invoked the stochastic nature of the condensing cold gas and its fundamentally different behavior compared to the standard accretion flow to overcome the angular momentum problem ( @xcite ) . through 3-d hydrodynamic simulations , we explicitly show that the angular momentum of cold gas condensing out of the hot phase indeed varies stochastically and has an isotropic distribution ( especially for the low specific angular momentum , @xmath153 @xmath75s@xmath51 , cold gas ) . cancellation of almost randomly directed angular momentum on short timescales leads to the formation of a compact accretion flow that accretes onto the smbh on the much shorter dynamical time scale ( compare eqs . [ eq : mdot ] & [ eq : mdot_sca ] ) . recent alma observations of infalling cold , clumpy co clouds in the elliptical galaxy at the center of abell 2597 ( @xcite ) , within @xmath196 pc of the smbh , provides an almost direct evidence for stochastic cold accretion ( sca ) . @xcite go onto argue that even in the higher @xmath197 regime , sca is responsible for the emergence of the smbh @xmath198 relation because accretion via a standard extended disk results in more massive smbhs than observed since a massive extended thin disk does not couple efficiently to an isotropic feedback - powered outflow . turbulence due to stochastic angular momentum and infall of cold gas ( viscous time is much shorter for a turbulent flow with @xmath199 ) also prevents the problem of gravitational fragmentation in stochastic accretion flow . most of the earlier studies of stochastic cold accretion were carried out in idealized numerical setups with turbulence in the cold gas driven artificially ( to realize the stochastic distribution of cold gas angular momentum ) . we analyze the angular momentum distribution of cold gas in our agn jet - icm simulations ( which appear consistent with most observations of cool cluster cores ; see section [ sec : obs ] ) and demonstrate that stochastic cold accretion is indeed realized in cool cluster cores stirred by time - dependent agn jets . in cluster simulations with limited resolution , the accretion physics input has to necessarily go in the form of a sub - grid model . one such model , which is much better than what is typically used ( including ours in this paper where we estimate @xmath197 at @xmath32 kpc and multiply it by an efficiency parameter @xmath53 to calculate the mechanical power injected by the smbh ) , is to use the mass accretion rate of the low angular momentum gas ( with say @xmath153 @xmath75s@xmath51 ; see eq . [ eq : mdot_sca ] & fig . [ fig : mdot_vs_time ] ) crossing the smallest resolved radius ( see section [ sec : improv ] for details ) . another improvement is the inclusion of star formation and realistic stellar feedback ( stellar feedback is increasingly important for lower mass halos such as clusters and elliptical galaxies ; e.g. , see fig . 4 in @xcite ) that can prevent artificial accumulation of massive cold gas in cluster cores ( e.g. , as done by @xcite ) . also , as mentioned earlier , our choice of @xmath53 ( feedback efficiency parameter ) is on the lower side , resulting in somewhat large mass deposition and star formation in the core . more 3-d simulations with larger @xmath53s are required to get a complete picture . much more needs to be done in understanding the details of stochastic cold accretion ( sca ) . in our simulations , which focus on larger ( @xmath59 kpc ) scales , we have to make assumptions about the underlying stochastic accretion . in particular , we need to understand how the stochastic cold gas condensing out of the icm is eventually accreted onto the smbh . whether a disk forms after the cancellation of stochastic angular momentum and on what timescale , and at what radii does the transition from sca to a viscous thin disk ( if it does at all ) and to a riaf ( radiatively inefficient accretion flow ) occurs . while the transition radius from a thin disk to an inner hot / thick riaf occurs at a smaller radius for a higher accretion rate , the details are not clear even for the standard ( non - stochastic ) accretion flows with a fixed angular momentum axis ( e.g. , see the vastly different estimates of disk to riaf transition radius in @xcite [ fig . 9b ] , @xcite [ fig . 1 ] , @xcite [ fig . 7b ] ) . strong agn jets and x - ray cavities ( which are powered by jets ) are expected only in the radiatively inefficient accretion regime for which @xmath200 ( @xcite ) . our smbh accretion rate at most times is lower than this threshold for jet formation ( see fig . [ fig : mdot_vs_time ] ) . infrequent , short duration high @xmath28 events , in fact , provide a basis for changing the direction of the jet , as observed in numerous clusters ( e.g. , @xcite ) . generally the mass accretion rate onto the smbh is expected to be even lower because of outflows in riafs ( larger suppression is expected for bigger riafs ) . therefore , advances in accretion physics are required to estimate the size of riafs and the mass accretion rate onto smbhs fed by sca . to sum up , majority of the cold gas condensing out of the icm due to thermal instability has large angular momentum such that the viscous accretion timescale is longer than the core cooling time for @xmath201 @xmath75s@xmath51 ( eq . [ eq : tvisc ] ) . most of this gas is expected to be toomre unstable and to form stars , instead of being accreted onto the smbh . the lower angular momentum cold gas ( @xmath153 @xmath75s@xmath51 ) has stochastic angular momentum and cloud collisions ( which lead to rapid angular momentum cancellation ) cause the cold , turbulent gas to be accreted rapidly , without much gravitational fragmentation . at innermost radii typically a riaf with powerful jets is expected , since @xmath202 . we finally summarize our major findings in the following points : * stochastic cold accretion ( sca ) , which was anticipated to play a key role in accretion onto smbhs from idealized simulations , is realized naturally in our realistic agn jet - icm simulations due to the turbulence induced by jets in a non - uniform icm . * we find that the low angular momentum cold gas ( @xmath112 @xmath75s@xmath51 ) condensing out of the icm has an isotropic distribution of angular momentum , which we expect will result in the cancellation of angular momentum on almost a dynamical time . this is unlike a standard viscosity - mediated accretion flow , in which an extended thin disk with a long accretion time is expected to form . another advantage of a stochastic accretion flow is that the net angular momentum of the low angular momentum cold gas varies on a timescale shorter than the core cooling time , implying that angular momentum cancellation and feedback driven by cold gas accretion can respond fast enough . * our work suggests an improved feedback prescription based on the angular momentum distribution of cold gas ( section [ sec : improv ] ) , which may be more robust than the usual models in which the mass accretion rate is estimated at @xmath1 kpc scales . in the latter approach one has to fine tune the feedback efficiency parameter ( @xmath53 ) for different radii at which the mass accretion rate is estimated ( compare our nfw runs with the inner radius of 1 and 0.5 kpc ) . * most of the features of cold accretion simulations , which show multiphase cooling and heating cycles driven by accretion of cold gas , match observations such as a large scatter of cold gas mass when compared to the smbh and halo masses ( this is not expected for hot / bondi accretion ) . a large scatter observed between bondi power and jet / cavity power is also consistent with our cold feedback simulations . this work is partly supported by the dst - india grant no . sr / s2/hep-048/2012 and an india - israel joint research grant ( 6 - 10/2014[ic ] ) . dp is supported by a csir grant ( 09/079[2599]/2013-emr - i ) . ab acknowledges support from nserc , through the discovery grant program , and pauli center for theoretical studies eth uzh . he also thanks the institute for computational sciences and university of zurich for hosting his visit . we acknowledge the support of the supercomputing education and research centre ( serc ) at iisc for facilitating our use of cray xc40-sahasrat cluster on which some of our runs were carried out . allen , s. w. , dunn , r. j. h. , fabian a. c. , taylor g. b. , & reynolds c. s. 2006 , , 372 , 21 babul , a. , sharma , p. & reynolds , c. s. 2013 , , 768 , 11 baganoff , f. k. , maeda , y. , morris , m. , et al . 2003 , , 591 , 891 bildfell , c. , hoekstra , h. , babul , a. , & mahdavi , a. 2008 , , 889 , 1637 cavagnolo , k. w. , donahue , m. , voit , g. m. , & sun , m. 2009 , , 182 , 12 churazov , e. , sazonov , s. , sunyaev , r. , forman , w. , jones , c. , & bhringer , f. 2005 , , 363 , l91 choudhury , p. p. & sharma , p. 2016 , , 457 , 2554 das , u. & sharma , p. 2013 , , 435 , 2431 fabian , a. c. 2012 , ara&a , 50 , 455 frank , j. , king , a. , & raine , d. 2002 , accretion power in astrophysics : third edition , " cambridge university press gaspari , m. , ruszkowski , m. , & sharma , p. , 2012 , apj , 746 , 94 gaspari m. , ruszkowski m. , & oh s. p. , 2013 , mnras , 432 , 3401 goodman , j. 2003 , , 339 , 937 hawley , j. f. & krolik , j. h. 2001 , , 548 , 348 hayes , j. c. et al . 2006 , , 165 , 188 hobbs , a. , nayakshin s. , power , c. , & king , a. 2011 , , 413 , 2633 hogan , m. t. et al . 2016 , preprint arxiv161004617 li , y. & bryan , g. l. 2014 , , 789 , 54 li , y. , bryan , g. , ruszkowski , m. , voit , g. m. , oshea , b. w. , & donahue , m. 2015 , , 811 , 73 liu , b. f. , yuan , w. , meyer , f. , meyer - hofmeister , e. , & xie , g. z. 1999 , , l17 loubser , s. i. et al . 2016 , , 456 , 1565 macchetto , f. et al . 1997 , , 489 , 579 marrone , d. p. , moran , j. m. , zhao , j. , & rao , r. 2007 , , 654 , l57 mccourt , m. , sharma , p. , quahaert , e. , & parrish , i. j. 2012 , 419 , 3319 mcdonald , m. , veilleux , s. , rupee , d. s. n. , & mushotzky , r. 2010 , , 721 , 1262 mcnamara , b. r. & nulsen , p. e. j. 2007 , ara&a , 45 , 117 mcnamara , b. r. , rohanizadegan , m. , & nulsen , p. e. j. 2011 , , 727 , 39 mcnamara , b. r. , nulsen , p. e. j. 2012 , new j. phys . , 14 , 055023 mignone , a. , bodo , g. , massaglia , s. , matsakos , t. , tesileanu , o. , zanni , c. , & ferrari , a. 2007 , , 170 , 228 mittal , r. , whelan , j. t. , & combes , f. 2016 , 450 , 2564 nayakshin , s. , power , p. & king , a. 2012 , , 753 , 15 nemmen , r. s. , bower , r. g. , babul , a. & storchi - bergmann , t. , 2007 , , 377 , 1652 nemmen , r. s. & tchekhovskoy , a. 2015 , , 449 , 316 peterson , j. r. , kahn , s. m. , faerels , f. b. s. et al . 2003 , , 590 , 207 pizzolato , f. & soker , n. 2010 , , 408 , 961 prasad , d. , sharma p. , & babul a. 2015 , , 811 , 108 ( ) quataert , e. & narayan , r. 2000 , , 528 , 236 rees , m. j. , begelman , m. c. , blandford , r. d. , & phinney , e. s. 1982 , nature , 295 , 17 russell , h. r. , mcnamara , b. r. , edge , a. c. , hogan , m. t. , main , r. a. , & vantyghem , a. n. 2013 , , 432 , 530 russell , h. r. , fabian , a. c. , mcnamara , b. r. , & broderick , a. e. 2015 , , 451 , 588 shakura , n. i. & sunyaev , r. a. 1973 , astron . & astrophys . , 24 , 337 sharma , p. , hammett , g. w. , quahaert , e. , & stone , j. m. 2006 , , 637 , 952 sharma p. , mccourt m. , quataert e. , & parrish i. j. , 2012a , mnras , 420 , 3174 sharma p. , mccourt m. , parrish i. j. , & quataert e. 2012b , mnras , 427 , 1219 soker , n. , sternberg , a. , & pizzolato , f. 2009 , in heinz s. , wilcots e. , eds , aip conf . ser 1201 , the moderate cooling flow model and feedback in galaxy formation . melville , ny , p. 321 soker , n. 2016 , new astronomy reviews , in press , eprint arxiv:1605.02672 shu , f. h. 1992 , the physics of astrophysics , volume ii , gas dynamics ( sausalito : university science books ) tremblay , g. r. et al . 2016 , nature , 534 , 218 voit , g. m. , donahue , m. , bryan , g. l. , & mcdonald , m. 2015 , nature , 519 , 203 wong , k .- w . , irwin , j. a. , yukita , m. , et al . 2011 , , 736 , l23 wong , k .- w . , irwin , j. a. , shcherbakov , r. v. , et al . 2014 , , 780 , 9 yang , h .- y . k. & reynolds , c. s. 2016 , , 829 , 90 yuan , f. & narayan , r. 2014 , , 52 , 529

we present a new series of supernova neutrino light curves and spectra calculated by numerical simulations for a variety of progenitor stellar masses ( 13- ) and metallicities ( and 0.004 ) , which would be useful for a broad range of supernova neutrino studies , e.g. , simulations of future neutrino burst detection by underground detectors , or theoretical predictions for the relic supernova neutrino background . to follow the evolution from the onset of collapse to 20 s after the core bounce , we combine the results of neutrino - radiation hydrodynamic simulations for the early phase and quasi - static evolutionary calculations of neutrino diffusion for the late phase , with different values of shock revival time as a parameter that should depend on the still unknown explosion mechanism . we here describe the calculation methods and basic results including the dependence on progenitor models and the shock revival time . the neutrino data are publicly available electronically .

1210.6841

supernova explosion is one of the most spectacular events in the universe . it is not only the death of massive stars but also an engine for the evolution of the galaxies . apart from the classification according to the spectroscopy , it is thought that there are two mechanisms of the explosion . while type ia supernova is caused by the thermonuclear explosion of white dwarfs , the other types of supernovae are driven by the core collapse of massive stars . unfortunately , however , details of the supernova explosion remain unsolved . as for the collapse - driven supernova , the physics which makes the explosion is not well understood . while many numerical simulations for the gravitational collapse of massive stars have been done so far ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * for recent reviews ) , there is no consensus on the explosion mechanism of collapse - driven supernovae . however , a rough sketch of the scenario shown below is widely accepted . the core of massive star becomes gravitationally unstable at the end of evolution and starts to collapse . the collapse is bounced by the nuclear repulsion force and the shock wave is launched . a supernova explosion is observed when the shock wave successfully propagates up to the stellar surface and expels the envelope material . finally the neutron star or black hole is formed as a remnant . however , the general difficulty in numerical simulations for successful explosions is that shock waves tend to stall before blowing out the stellar envelopes . a possible mechanism to revive the stalled shock waves is energy input by strong neutrino radiation from the newborn neutron stars ( the delayed explosion scenario ; e.g. * ? ? ? * ) , but it is still a matter of debate whether this single process is sufficient for successful explosions . an aspherical hydrodynamic turbulence , such as the convective instability and standing accretion shock instability , may help the shock revival ( e.g. * ? ? ? other proposed mechanism or physical processes to account for explosions include qcd phase transition @xcite , acoustic wave @xcite and magnetic field @xcite . the collapse - driven supernova is important also as a target of neutrino astronomy . as is well known , supernova neutrinos from sn1987a have been detected by the two water cerenkov detectors , kamiokande ii @xcite and imb @xcite . since the large magellanic cloud , where sn1987a appeared , is @xmath250 kpc away from the earth and detectors at the time were not as large as the present - day experiments , the event number was small ( eleven for kamiokande ii and eight for imb ) . nevertheless , many theoretical studies tried to extract various information about physics of supernovae and/or neutrinos from these data ( e.g. * ? ? ? * for a recent review ) . if a supernova occurs now near the galactic center ( @xmath210 kpc away from the earth ) , about 10,000 events will be detected by superkamiokande , which is the largest , currently operating neutrino detector around mev @xcite . therefore details of a collapse - driven supernova such as the explosion mechanism , as well as the neutrino physics such as mass and its hierarchy , mixing and oscillation , or any exotic physics , could be investigated by high - statistics light curve and spectrum of a future supernova neutrino burst event . another potential opportunity to observe neutrinos from supernovae is the relic background radiation , which gives valuable clues about e.g. , the cosmic star formation history and/or stellar initial mass function ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? the cosmic metallicity evolves with time and , as stated in section [ proge ] , the stellar evolution depends not only on its mass but also on its metallicity . since the metal - free and very metal - poor massive stars are formed in the high redshift universe and explode immediately , their contribution to the relic neutrino background would be small @xcite . however , galaxies whose metallicities are about one order of magnitude lower than the solar value reside within the redshift @xmath3 @xcite . to predict the flux and spectrum of the relic supernova neutrinos , one must model the integrated spectrum of supernova neutrinos for various ranges of progenitor stars . considering these situations , it would obviously be useful to provide the results of supernova neutrino emission calculated by state - of - art numerical simulations , for various types of progenitor stars . such a comprehensive datab ase of theoretical predictions can be used as templates to simulate supernova neutrino detection of a future neutrino burst event , or to construct a realistic prediction for the relic neutrino background . however , most of numerical simulations so far followed only within @xmath41 s ( e.g. * ? ? ? * ) after the core collapse because solving the neutrino transfer with hydrodynamics for a long term is a numerically tough problem . such simulations are not satisfactory for the above purpose , because a significant fraction of the total gravitational energy is emitted as neutrinos after this early phase , with a typical decay time scale of @xmath210 s. such a simulation result for a long term was shown in @xcite . recently , @xcite showed long term neutrino signals for two models , an ordinary collapse - driven supernova and an electron - capture supernova , which is a subcategory of collapse - driven supernovae . as for the electron - capture supernova , long term simulations were also performed by @xcite computing neutrino signals . on the other hand , neutrino diffusions in the late phase were separately investigated by evolutionary calculations ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? however , there is no comprehensive data set yet for numerical long - term supernova neutrino signals from various types of progenitor stars . in this study , we construct a publicly available database for numerical simulations of supernova neutrino emission , including eight simulations for normal single stars in the initial mass range of 13-@xmath0 for two different values of stellar metallicity ( @xmath5 and 0.004 , i.e. , the solar abundance and its @xmath6 , respectively ) . to follow the long - term evolution for many simulations , we use two different simulation methods : general relativistic neutrino - radiation hydrodynamic ( @xmath7rhd ) simulations for the early phase , and general relativistic quasi - static evolutionary calculations of neutrino diffusion in a nascent neutron star for the late phase . though the @xmath7rhd simulations do not lead to a natural supernova explosion , we introduce a parameter , shock revival time , which reflects the unknown explosion mechanism , and phenomenologically connect the early and late phases by this parameter based on physical considerations . although the connection between the two phases is not perfectly consistent as a physical simulation , this approach allows us to provide reasonably realistic supernova neutrino light curves and spectra from the onset of a core collapse to @xmath210 s after that , for various progenitor stellar models . the organization of this paper is as follows . in section [ proge ] , the progenitor models used in this study are described . the early phase of supernovae , from the onset of collapse to the shock propagation , is described in section [ colla ] . in section [ pnsc ] , the quasi - static evolutionary calculations of neutrino diffusion in the dense core for the late phase are described . the connection between the early and late phases are discussed in section [ neusg ] . section [ disc ] is devoted to a summary and discussion . massive stars with the initial mass @xmath8 are known to undergo gravitational collapse at the end of their lives . they synthesize heavy elements through some nuclear burning stages during the quasi - static evolutions . finally the iron core , where all nuclear reactions become equilibrium to inverse reactions of themselves and various elements called `` iron '' ( fe , ni , co , mn and so on ) are created , is formed at the center . the iron core has density of @xmath910@xmath10 g/@xmath11 and is supported by the degenerate pressure of relativistic electrons . on the other hand , the condition for the gravitational instability is written , using the adiabatic index , @xmath12 , as @xmath13 where @xmath14 , @xmath15 and @xmath16 are the pressure , density and entropy , respectively . since the adiabatic index of relativistic ideal gas is @xmath17 , the iron core is marginally stable . there are two processes which destabilize the iron core . the first one is an electron capture by protons belonging to nuclei : @xmath18 this reaction is caused because the sum of rest mass and kinetic energy of the relativistic electrons exceeds the mass difference between neutron and proton due to high density . diminution of electrons causes a deficit in pressure and makes the iron core unstable . the second is the photodisintegration reaction of nucleus such as @xmath19 this process works for high temperature as @xmath20 k. while the energy per nucleon , @xmath21 , reaches a minimum at @xmath22fe , the helmholtz free energy per nucleon , @xmath23 , is minimized in finite temperature . because of @xmath24 , matter becomes thermodynamically stable by the increase in entropy due to the photodisintegration of nuclei . nevertheless , this is an endothermic reaction so that the pressure does not increase so much as to suspend the core collapse . this is an onset of the collapse - driven supernova . the lower mass limit for the progenitors of collapse - driven supernovae is determined whether the iron core is formed or not . note that , for somewhat low mass cases , the core collapse may be triggered only by electron captures before ne ignition , which is called an electron - capture supernova . they can be regarded as a subcategory of collapse - driven supernovae because their scenarios after the collapse are thought to be similar . @xcite investigated stellar evolution sequences with initial masses between 6.5 and @xmath25 and the solar metallicity using three different codes and found that the lower mass limit for core collapse is 9-@xmath26 . observationally , @xcite found that it converges to @xmath27 investigating the collapse driven supernovae observed in a fixed 10.5-year period within a distance of 28 mpc . the upper mass limit is determined whether the supernova explosion succeeds or not . it is theoretically uncertain because the precise understanding on the explosion mechanism is still absent . nevertheless , generally to say , the density profile of progenitor is important because the shock wave should run through the star . what makes a situation more awkward is uncertainty in the theory of stellar evolution , the mass loss rate and convection . the effects of rotation and binary interaction may also affect the stellar evolution . @xcite showed numerically that nonrotating stars with initial masses @xmath28 fail to explode and form black holes directly while the mass loss was not considered in the progenitor models adopted by them . note that , the black holes could be also formed after the explosion , which is probably weak , via fallback accretion . recently , @xcite performed the core collapse simulations for many progenitor sets with the spherically symmetric ( 1d ) models involving simplified neutrino transfer . they implied that the upper mass limit for the explosion depends severely on the evolutionary calculations , while their estimation was based on the calculations with an artificially increased energy deposition . observational constraints on the upper mass limit are also obscure . @xcite suggested that the maximum mass for the progenitors of type ii - p supernovae is @xmath29 and the majority of more massive stars may collapse quietly to black holes where the explosions remain undetected . however , @xcite , who investigated the observed fractions of the collapse - driven supernovae , claimed that they will produce other types of supernovae which are not ii - p . on the other hand , massive stars with the mass of @xmath30 lose the outer layer due to their strong stellar winds and form wolf - rayet stars , where the helium cores are exposed @xcite . they are thought as progenitors of type ib and ic supernovae and , according to the light curve models by @xcite , the progenitors of some luminous type ic supernovae called hypernovae have the masses of @xmath31 . they implied that the fate of massive stars beyond @xmath32 depends on the stellar rotation and the hypernova progenitors are rotating . note that , @xcite suggested that a certain fraction of such massive stars produce type iin supernovae from luminous blue variables . they also pointed out the importance of binary interactions to account for type ib and ic supernovae . recently , extremely luminous supernovae such as sn2006gy ( type iin ) and sn2007bi ( type ic ) are observed and they are suggested to have a progenitor mass of @xmath33 , while they may be pair - instability supernovae , whose explosion mechanism is different from that of the collapse - driven supernovae . it should be emphasized that , as described above , the fate of massive stars is a hot subject under active discussion , but we deal progenitors provided from a single treatment for the stellar evolution in below . in this study , we prepare eight progenitor models with the initial masses @xmath34 , @xmath35 , @xmath36 and @xmath0 , and the metallicities of @xmath5 ( solar ) and 0.004 ( small magellanic cloud ) . they are computed by a henyey type stellar evolution code which is fully coupled to a nuclear reaction network . this code is also used in @xcite and its descriptions , such as the treatment of convection , are given in @xcite . as for the mass loss rate , we adopt the same model with `` case a '' of @xcite for the main sequence stage ( see also , * ? ? ? * ; * ? ? ? we set the empirical mass loss rate @xcite scaled with the metallicity as @xmath37 for the wolf - rayet stage @xcite . the effects of rotation and binary interaction are not taken into account . the density profiles of our progenitor models are shown in figure [ inidnsall ] and the composition profiles for two reference cases , @xmath38 and @xmath39 , are shown in figure [ inicom ] . note that , as shown in table [ keyprm ] , the total progenitor masses when the collapse begins , @xmath40 , are different from @xmath41 due to the mass loss . the progenitors have the onion - like structure : central core is surrounded by shells of lighter elements . we regard the region of oxygen depletion as a core hereafter . in figure [ coremas ] , the core mass is plotted as a function of the initial mass for all models . since the mass loss rate is larger for high metallicity , the models with @xmath42 have higher core mass than those with @xmath5 . the models with @xmath43 become the wolf - rayet stars losing large amount of their mass and final core masses are somewhat low , which may correspond to type ic supernovae . as a result , the core mass of the model with @xmath44 is the highest for each metallicity case . to evaluate the spectra of neutrinos emitted from the collapse of above progenitor models , which is the goal of this paper , we put the assumptions on the fate of our progenitors . for the equation of state by @xcite , which is adopted below in our simulations , the maximum mass of neutron stars is @xmath45 in the gravitational mass and @xmath46 in the baryonic mass . we assume that the model with @xmath47 forms a black hole , because its core mass is larger than the maximum baryon mass . while whether the other models can make explosions or not is unclear , we regard them as `` supernova '' progenitors . the correspondence between the initial mass and core mass is not well - established , however , the outer region far enough away from the core does not influence the dynamics and neutrino emission of the collapse . therefore , we believe that the systematics discussed in this study will be helpful for other progenitor sets if their core masses are known . in this section , we follow the early stage of the collapse - driven supernova showing the results of our numerical simulation . here , we utilize the general relativistic neutrino radiation hydrodynamics ( @xmath7rhd ) code , which solves the boltzmann equations for neutrinos together with the lagrangian hydrodynamics under spherical symmetry @xcite . we can compute the neutrino transport as well as the dynamics of collapse . we consider four species of neutrino , @xmath48 , @xmath49 , @xmath50 and @xmath51 , assuming that the distribution function of @xmath52 ( @xmath53 ) is equal to that of @xmath50 ( @xmath51 ) . neutrino oscillation is not taken into account , because it does not occur in the core region and also not affect the dynamics in this stage @xcite . a detailed description of the numerical simulations such as general relativistic hydrodynamics , transport and reaction rates of neutrinos can be found in @xcite . referring the resolution dependences shown in @xcite , we use 255 mesh points for the radial lagrangian coordinate and 20 and 4 mesh points for the energy spectrum and angular distribution , respectively . as the initial conditions , the eight progenitor models described in the previous section are adopted and the outer boundary is settled far from the core region . we utilize an equation of state by @xcite , which is based on a relativistic mean field model @xcite , and inhomogeneous matter distribution is described in the thomas - fermi approximation @xcite . the density profiles at the time with the central density of @xmath54 g @xmath55 are shown for all models in figure [ inidns ] . we can recognize that their profiles are very similar for the inner most @xmath56 and the progenitor dependence is seen in the envelope and outer region of the core . the difference will affect the emitted neutrino signal of the collapse - driven supernova . as the density increases due to the collapse , the mean free path of neutrinos gets shorter . in the collapse stage , the emitted neutrinos are mostly @xmath48 by the electron capture ( [ ecp ] ) . their main opacity source is coherent scattering off nuclei ; @xmath57 when the density exceeds @xmath58 g @xmath55 , neutrinos are trapped . the neutrino optical depth @xmath59 at the radius @xmath60 is defined as @xmath61 where @xmath62 is the stellar radius and @xmath63 is the mean free path of neutrino . roughly speaking , the neutrino trapping occurs inside the neutrino sphere , whose radius , @xmath64 , is defined as @xmath65 once the neutrinos are trapped , the inverse process of the electron capture ( [ ecp ] ) takes place and @xmath66-equilibrium for the weak interaction is achieved . as a result , the lepton fraction is kept nearly constant and the neutronization is moderated . these trends are shown in figure [ prebnc ] where the profiles of density , electron fraction and lepton fraction are plotted for the model with @xmath38 . here , the lepton fraction @xmath67 is related with the electron fraction @xmath68 and the electron - type neutrino fraction @xmath69 as @xmath70 . note that , the electron fraction of the inner region decreases even after the neutrino trapping because it is determined by the @xmath66-equilibrium for the given lepton fraction . the collapse of the core does not halt until the central density exceeds the nuclear density ( @xmath71 g @xmath55 ) . the infalling core is divided into two parts , to which we refer as the inner and outer cores . the inner core contracts subsonically and homologously ( @xmath72 ) while the outer core infalls supersonically like free - fall ( @xmath73 ) . roughly speaking , it is known that the inner core mass corresponds to the chandrasekhar mass @xmath74 , and does not depend on the progenitors . when the inner core density reaches the nuclear density , nuclei are closely packed and regarded as uniform matter . then , the adiabatic index increases suddenly due to the repulsive nuclear force and the core restores stability . the influence of nuclear repulsion propagates through the inner core and the collapse is decelerated . on the other hand , since the pressure wave transmits at the speed of sound , the outer core region is still falling supersonically . therefore , on the boundary of inner and outer cores , the shock wave is formed and launched outward . in the meanwhile , the falling outer core matter is swept by the shock wave and accretes onto the bounced inner core . this compact object is called a proto - neutron star . a successful explosion is achieved by the shock breakout through the stellar surface . the explosion without the shock stall is called a prompt explosion . however , this scenario is not promising according to recent studies and the shock wave is thought to be stalled . the shock wave propagation is prevented by the photodisintegration reaction of nucleus ( [ photodis ] ) and the neutrino emission . as already mentioned , the photodisintegration reaction is an endothermic reaction and consumes the kinetic energy of the shock wave . after the shock wave passes through the neutrino sphere , neutrinos produced by various sorts of reactions can escape carrying out the internal energy behind the shock front . moreover the shock wave should overcome the ram pressure of the supersonically infalling outer core . the dynamical features described above are seen in figure [ velo ] where the snapshots of velocity profile are shown for the model with @xmath38 . the stalled shock wave becomes accretion shock and would revive leading to a supernova explosion in some way . this is called a delayed explosion . unfortunately , however , the detailed scenario is still an open question . except the qcd mechanism in which a phase transition to deconfined quark matter makes the second collapse and shock formation and triggers the explosion @xcite , spherically symmetric ( 1d ) models are thought to fail . in this case , the ram pressure of accreting matter is higher and prevents the shock revival . on the other hand , if the sphericity is broken , the accretion rate is partially reduced and the explosion has an advantage . in fact , the morphology of collapse - driven supernovae is observationally indicated aspherical ( e.g. * ? ? ? therefore , most of the recently promising scenarios are based on multi - dimensionality . the neutrino heating mechanism has been discussed for years . while , as already mentioned , the neutrinos cool the inner region , they contribute the heating just behind the shock front , where free nucleons produced by the photodisintegration reaction ( [ photodis ] ) absorb a small part of neutrinos escaping from the inner region . with the aid of this heating , the shock wave is considered to revive on a time scale of the order of 100 ms . this scenario was initially proposed by @xcite according to their numerical simulation with 1d model whereas no other group was able to confirm . recently , employing multi - dimensional simulation , some groups have reported that the onset of the neutrino - driven explosion would be helped by aspherical hydrodynamic turbulence such as the convective instability ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and standing accretion shock instability ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? another candidate for the explosion scenario is the acoustic mechanism @xcite . the turbulence generated on the accretion shock travels inward and excites oscillation of the proto - neutron star . then , the acoustic wave originated in the oscillation propagates outward and deposits energy on the accretion shock , which causes the shock revival and explosion . in this hypothesis , the time scale for the shock revival is estimated as @xmath9500 ms , which is longer than that of the neutrino heating mechanism . magnetic fields may also be important for some supernovae ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? in fact , neutron stars with strong magnetic fields ( magnetars ) are observed and their progenitors might also have strong magnetic fields . in the mhd mechanism , the rotational energy converts to the explosion energy via magnetic fields amplified by the field wrapping and/or magnetorotational instability . while , as discussed above , neutrinos would play a key role for a successful explosion , it is a small part of them that are absorbed near the shock wave and contribute to the explosion . the amount of emitted neutrinos is mainly determined by the released gravitational potential of the accreted matter and thermal energy of the nascent proto - neutron star . since the neutrino energy deposition for the explosion has an insignificant effect , our @xmath7rhd simulation can be regarded to represent the neutrino signal before the shock revival . in reality , the shock wave would move outward again and the mass accretion stops while when it occurs may depend on the explosion mechanism . however , the shock may not be able to revive if the core is too massive . it is probable for our model with @xmath47 and we assume that it fails to explode . in this case , we follow the collapse and neutrino emission up to the black hole formation . for other models , our @xmath7rhd simulations are terminated at 550 ms after the bounce , within which the shock revival is assumed to occur . the evolution after the shock revival is dealt in the next section . in figure [ pstbnc ] , we show the evolutions of density , temperature and electron fraction after the bounce . they are the results of our @xmath7rhd simulation for the models with @xmath38 and @xmath39 . for both models , since the proto - neutron star mass gets larger by the accretion , the density and temperature rise due to gravitational compression . while the density profile is monotonic , the peak of temperature profile resides not at the center but at the medium region . this is because the shock wave does not run from the center and heats the outer matter , as seen in figure [ velo ] . since the protons created by the photodisintegration reaction ( [ photodis ] ) cause the electron capture ( [ ecp ] ) , the electron fraction decreases for the shocked region . as recognized in the comparison of the models with @xmath38 and @xmath39 , the profile at the bounce does not depend on the progenitor because , as already mentioned , the bounced inner core mass does not differ among progenitors . on the other hand , since the accretion rate is determined by the density profile of the outer core , the proto - neutron star mass differs among the models for several times 100 ms after the bounce . as a result , it is also reflected in the proto - neutron star structure . after the shock propagation , the proto - neutron star settles into a hydrostatic configuration . however , the proto - neutron star of the model with @xmath47 recollapses suddenly and forms a black hole at 842 ms after the bounce . the qualitative features of black hole formation are same as results in @xcite we evaluate the neutrino luminosities and spectra in the late phase ( after the shock revival ) by the proto - neutron star cooling ( pnsc ) simulation except for the black - hole - forming case . in this method , quasi - static evolutions of proto - neutron stars are solved with the neutrino transfer by multigroup flux limited diffusion scheme under spherical symmetry with general relativity @xcite . here , we follow the hydrostatic structure of the proto - neutron star at each time by the oppenheimer - volkoff equation while , in @xmath7rhd simulations , the equations for hydrodynamics are fully solved . we deal the boltzmann equations in the angle - integrated form for @xmath48 , @xmath49 and @xmath75 , where @xmath50 , @xmath51 , @xmath52 and @xmath53 are treated collectively as @xmath75 . in contrast , @xmath50 and @xmath76 are treated individually in our @xmath7rhd simulations . note that @xmath50 and @xmath76 have the same type of reactions and the difference in coupling constants is minor . in fact , the difference between the distribution functions of @xmath50 and @xmath76 is typically @xmath41% in our @xmath7rhd simulations . the equation of state , binning of neutrino energy and neutrino reactions taken into account are set to the same with the @xmath7rhd simulation shown in the previous section . in this section , we describe the late stage of collapse - driven supernova with our pnsc results . since the time scale of the neutrino diffusion is @xmath210 s , the evolution is followed till 20 s after the bounce . we use the results of our @xmath7rhd simulation as initial conditions of our pnsc simulation . the central parts up to just ahead of the shock wave are picked up . here , we take the profiles of electron fraction and entropy as functions of the baryon mass coordinate from the @xmath7rhd results and , using them , reconstruct hydrostatic configurations with almost steady flow of neutrinos , which are used as pnsc initial models . it is confirmed that , except for the close vicinity of surface , the obtained density profile is consistent with that of the original @xmath7rhd result as expected because the velocity of shocked region is negligible ( see figure [ velo ] ) . since , as already mentioned , detail of the explosion is not known , we set the shock revival time @xmath77 and @xmath7rhd profiles at @xmath77 are used as pnsc initial conditions . here , we investigate three cases as @xmath78 ms , 200 ms and 300 ms for each progenitor model . when an explosion mechanism is assumed , corresponding @xmath77 would be determined . therefore , we can regard that the explosion scenario described above are parameterized by the shock revival time . in the pnsc simulation , we follow the evolution of proto - neutron stars without accretion because amount of matter which falls back onto the proto - neutron star would be minor after the successful explosion . in figure [ plpnsc ] , we show the evolutions of density , temperature and electron fraction given by our pnsc simulation for the model with @xmath79 . one can see that the shocked outer mantle of proto - neutron star has relatively high temperature and is thermally expanded at initial moment ( solid lines ) . neutrinos can easily escape from there carrying out thermal energy and , therefore , the outer mantle shrinks in a short time scale to be denser and hotter while the entropy decreases there ( see figure [ etrp ] ) . note that diffusion fluxes of @xmath80 and @xmath75 initially transport heat inward from the mantle into the central region in parallel outward to the surface because number densities of thermal neutrinos have maxima in the hot mantle . this heat flux attributes to the initial entropy increase in the central region . once the temperature profile becomes monotonic , the entire proto - neutron star cools down gradually . in addition to the neutrino cooling , the net flux of electron - type neutrinos ( @xmath81 ) out of the proto - neutron star carry away the electron - type lepton number of the proto - neutron star . this deleptonization just corresponds to the neutronization of the proto - neutron star . the nascent proto - neutron star is lepton - rich and composed also of considerable protons initially . it becomes the ordinary neutron star with less protons achieving the neutrino - less @xmath66 equilibrium in which there remains small amount of protons and electrons . the net number flux of electron - type neutrinos is at most @xmath220% of the number flux of @xmath48 . the decrease of electron fraction shown in figure [ plpnsc ] represents this process well . the density evolution from the onset of collapse to 20 s after the core bounce is illustrated in figure [ denstot ] . the features of proto - neutron star evolution described above are qualitatively common among the models investigated in this study . after the revival , the shock wave propagates into the stellar envelope and finally blows it off . since , in the envelope , the energy losses owing to the photodisintegration and neutrino emission is quite tiny and the binding energy is subtle , the shock wave is not prevented from running outward . while the time scale of the neutrino diffusion is @xmath210 s , it takes from several hours to days for the shock wave to reach the stellar surface . since supernova explosion can be observed optically only after the shock breakout , the neutrino signal reaches to the earth in advance . finally , the shock wave blasts through the interstellar medium and forms a supernova remnant , such as the crab nebula . on the other hand , a neutron star remains at the center . in figure [ nsmass ] , the gravitational mass of neutron stars @xmath82 considered in our model is plotted . while , according to general relativity , the gravitational mass of a neutron star is different from its baryonic mass @xmath83 ( see table [ keyprm ] ) , there is a one - to - one correspondence between them . we evaluate @xmath82 from the baryonic mass accreted within the shock revival time @xmath77 assuming the equation of state by @xcite . since the accretion rate is higher , @xmath82 depends on @xmath77 especially for the progenitors with larger initial mass @xmath41 . while the neutron star mass could be increased by the accretion from a binary companion , the distribution of neutron star masses may give a hint of the explosion mechanism . neutrinos emitted from the collapse - driven supernova release the gravitational potential of the accreted matter and cool the nascent proto - neutron star . for convenience , we divide the neutrino flux after the shock stall , @xmath84 , into two terms as @xmath85 where subscript @xmath86 denotes the species of neutrinos and @xmath87 is a neutrino energy . the first term , @xmath88 , is an accretion term and related with the accretion luminosity , @xmath89 , which is approximated as @xcite @xmath90 where @xmath91 is the outer boundary radius of our @xmath7rhd simulation and @xmath92 is a radius of the neutrino sphere at the time @xmath93 defined by equations ( [ depth ] ) and ( [ sphere ] ) . @xmath94 , @xmath95 and @xmath96 are the gravitational constant , the mass accretion rate and the mass enclosed by @xmath92 , respectively . the second term , @xmath97 , is a cooling term which comes from the thermal energy loss of the proto - neutron star . while the accretion term is dominant for the early phase ( @xmath98 ms ) , only the cooling term remains after the shock revival leading the explosion . as already mentioned , supernova explosion is not successful in most of 1d simulations owing to the high mass accretion rate . this would be over estimate because some multi - dimensional effects such as the convective and standing - accretion - shock instabilities reduce the mass accretion in reality . therefore , we can regard that our 1d @xmath7rhd simulation gives the maximum case of the mass accretion rate . thus our @xmath7rhd results of the neutrino flux , @xmath99 , can be regarded as the upper limit : @xmath100 where @xmath101 is the maximum possible value of the accretion term . in figure [ nurhd ] , we show the time profiles of luminosities and average energies of emitted neutrinos evaluated from @xmath7rhd simulation for all models . the end point of the model with initial mass @xmath102 and metallicity @xmath42 is a moment of the black hole formation . the peak of @xmath48 just after the bounce ( @xmath103 s ) corresponds to the neutronization burst . when the shock wave propagates through the outer core , nuclei are dissociated into free nucleons and produce a large amount of @xmath48 by the electron capture ( [ ecp ] ) . a short burst of these neutrinos occurs after the shock breakout through the neutrino sphere . this is called a neutronization burst . while the peak luminosity exceeds @xmath104 erg s@xmath105 , the duration time is the order of 10 ms and the emitted energy is minor comparing with the whole emission of the supernova neutrino . the persistent emission after the neutronization burst originates from the mass accretion and proto - neutron star cooling , as described in equation ( [ devide ] ) . in this stage , @xmath48 and @xmath49 , which are abundantly emitted by the electron and positron captures , respectively , have higher luminosity than @xmath75 ( @xmath106 ) . nevertheless , the contribution of @xmath75 is not minor because the pair processes such as the electron - positron pair annihilation , plasmon decay and nucleon bremsstrahlung occur . thanks to the shock heating , the accreted matter is enough hot for electron - positron pair processes . on the other hand , the average energy of @xmath75 is higher than those of @xmath48 and @xmath80 . since @xmath75 does not have charged - current interactions with matter consisting of no @xmath107 , its mean free path is longer at the same position than those of @xmath48 and @xmath49 and its neutrino sphere is smaller as recognized by equations ( [ depth ] ) and ( [ sphere ] ) . therefore the temperature on the neutrino sphere is higher for @xmath75 , which makes their average energy also higher . as for the progenitor dependence , since the progenitors with higher density for the range 1.5-@xmath108 ( see also figure [ inidns ] ) have the higher mass accretion rate , their neutrino luminosities are higher as expected from equation ( [ acclumi ] ) . their average energies are also somewhat higher but the progenitor dependence is not clear especially for the first @xmath2100 ms . note that the neutrino emission stops when the proto - neutron star collapses to a black hole for the model with @xmath47 . the results of our pnsc simulation are just corresponding to the cooling term in equation ( [ devide ] ) . obviously , they give the lower limit of the neutrino flux because accretion induced neutrino flux is not included : @xmath109 in figure [ nupnsc ] , we show the time profiles of luminosities and average energies of emitted neutrinos evaluated from pnsc simulation for some models . they decrease in time as the proto - neutron star cools . the neutrino energy hierarchy ( @xmath110 ) is same as that in the accretion phase . the neutrino signals of the models with different initial mass @xmath41 and shock revival time @xmath77 are compared in figure [ nupnsc ] . the luminosity and average energy are higher for the models with larger neutron star mass @xmath82 ( see also figure [ nsmass ] ) while the difference is not so large . with inequalities ( [ frhd ] ) and ( [ fpnsc ] ) , we construct the light curve models of neutrino . for this , we introduce a fraction factor of the accretion term to its maximum , @xmath111 , as a function of time : @xmath112 while @xmath111 may also depend on the species and energy of neutrino , we ignore their dependences for simplicity . it is required for @xmath111 to satisfy @xmath113 for the early phase ( @xmath98 ms ) and @xmath114 for the phase after the explosion . using @xmath111 , the neutrino flux is expressed as @xmath115 the details of explosion dynamics would determine the function @xmath111 . for instance , a neutrino signal of the early explosion model corresponds to a rapidly decaying @xmath111 and a small neutron star mass @xmath82 . on the other hand , slowly decaying @xmath111 and large @xmath82 give a neutrino signal of the late explosion model . when supernova neutrinos are actually detected , this study would help us to probe the nature of progenitor and remnant . as discussed above , the neutrino luminosity in the accretion phase ( @xmath2100 ms after the neutronization burst ) is determined by the progenitor model especially for the density profile . the signals in the cooling phase ( @xmath210 s after the neutronization burst ) would provide hints for the mass of remnant neutron star . moreover , if the transition from the accretion phase to the cooling phase is observed , a restriction for the explosion mechanism may be possible . our results can be hopefully utilized as immediately comparable templates for a neutrino detection . one may be able to use our results for modeling of the supernova neutrino signals . in figure [ nucont ] , we demonstrate examples of the neutrino light curve and spectrum . they are drawn under the assumption , @xmath116 for the model with @xmath117 . the junctions of this interpolation are shown in figure [ nujct ] . since @xmath111 corresponds to the fraction of mass accretion rate to its maximum , the decay time scale @xmath118 would be a propagation time scale of the revived shock wave . when the shock revives at the radius @xmath119 with the escape velocity @xmath120 , it takes @xmath121 for the shock wave to pass the core with the size of @xmath122 . here , @xmath119 is the most ambiguous parameter but at least @xmath123 would be satisfied . therefore , we adopt 300 km as a typical value and get @xmath124 ms . note that , this value corresponds to also the free - fall time scale of inner core . the time shift @xmath125 is needed for the following two reasons . first , @xmath84 is evaluated as the neutrino flux on the outer boundary of our @xmath7rhd simulation at the time @xmath93 . the outer boundary of our pnsc simulation is the proto - neutron star surface while that of our @xmath7rhd simulation is located in the envelope . the correction of the light traveling time of the distance between two boundaries ( @xmath126 ms ) is needed . second , results of pnsc simulations for very early phase are not reliable because , as already mentioned , the static density profiles are not consistent with those of @xmath7rhd simulations in the close vicinity of surface . it is needed @xmath127 ms for the relaxation . therefore we set @xmath128 ms , as a typical value . incidentally , to evaluate the flux numerically , @xmath111 is settled to zero for @xmath129 ms , where @xmath130 in equation ( [ itplft ] ) . hereafter , for simplicity , we take @xmath124 ms and @xmath128 ms for all cases . note that , the neutrino number emitted during the period of this interpolation is about 5 - 15% of the total amount . in figure [ fdfit ] , the neutrino number spectra are compared with fermi - dirac distributions that have the same luminosity and average energy with zero chemical potential . while fermi - dirac spectra roughly fit our results , the deviation is not small . it is shown that high energy neutrinos are emitted in the accretion phase because the low density outer region has high temperature due to shock heating . on the other hand , after the shock revival , this region becomes dense and the proto - neutron star surface is not heated any more in the cooling phase . therefore , the deficit of high energy neutrinos can be seen in the spectra of late time . figure [ fdfitint ] compares , as in figure [ fdfit ] , numerical results and fermi - dirac distributions for some cases of time integrated neutrino number spectra over the duration of the simulations . we find that , for the supernova models , the spectrum given by our simulations is relatively well fitted by fermi - dirac distribution up to @xmath230 mev but has high energy tail originated in the accretion phase . even for the black - hole - forming model , the fermi - dirac distribution roughly fits our spectrum while the numerical uncertainty is larger for the high energy regime . in figure [ etot ] and table [ keyprm ] , we show the total neutrino energy emitted till 20 s after the bounce for all models with the interpolation ( [ itplft ] ) . note that , since the model with @xmath47 form a black hole , the total neutrino energy emitted till the black hole formation followed by @xmath7rhd simulation is plotted . we can see that the total emission energy is related to the core mass rather than the initial mass ( see also figure [ coremas ] ) , and it is larger for the explosion models with large @xmath77 because the accretion phase is longer and the neutron star mass is larger . these features are shown more explicitly in figure [ etotsp ] , where the total emission energy is plotted for the core mass . the shock - revival - time dependence of the total emission energy is larger for the models with higher core mass because the mass accretion rate is higher . since the neutrino luminosity is approximated as equation ( [ acclumi ] ) , the emission energy during the accretion phase is roughly proportional to the product of the mass accretion rate and shock revival time . moreover , the resultant proto - neutron star mass depends on this product . therefore , also in the late phase , the shock - revival - time dependence is larger for the models with higher core mass . on the other hand , the shock - revival - time dependence is larger for @xmath48 and @xmath131 than @xmath75 . this feature is also seen for the mean energy of emitted neutrinos , as shown in figure [ emean ] . as already stated , in the accretion phase , @xmath48 and @xmath131 are emitted more abundantly than @xmath75 . therefore the signals of @xmath48 and @xmath131 is sensitive to the duration of this phase , i.e. the shock revival time . the purpose of this study is to construct a comprehensive data set of long - term ( up to @xmath210 s from the onset of the collapse ) supernova neutrino signal for variety of progenitor stellar models with different initial masses and metallicities , which would be useful for a wide range of research related to supernova neutrinos . to achieve this goal avoiding the difficulty of long - term full numerical simulations , we combined two different schemes of numerical simulations . the early phase of the collapse - driven supernova , at which the collapsing core is bounced and the shock wave is stalled due to the matter accretion , has been followed by the general relativistic neutrino radiation hydrodynamics ( @xmath7rhd ) code . the late phase after the shock revival has been dealt by the general relativistic proto - neutron star cooling ( pnsc ) simulation which solves quasi - static evolutions with the neutrino diffusion . the two phases are combined phenomenologically , taking into account the uncertainty about the explosion mechanism , and the shock revival time is introduced as a parameter connecting the two phases . although this connection is not perfectly consistent as a single physical simulation , this is currently the best way to follow supernova neutrino signals up to @xmath210 s for many progenitor stellar models . there are still many uncertainties about collapse - driven supernova physics ( e.g. , equation of state ) , but we have chosen the standard or most popular parameters , to provide theoretical supernova neutrino emission models expected from the standard picture of collapse - driven supernovae . therefore , the database presented here would serve as a standard guideline or template for the supernova neutrino signals . it is interesting to compare our result with the past calculations of long - term supernova neutrino emission . @xcite presented the supernova neutrino model till 18 s after the bounce for a progenitor with @xmath132 and investigated the detectability in detail . in their model , the neutrino luminosity decreases by two orders of magnitude within @xmath210 s , which is similar to our results however , their average energy gets higher with time , which contradicts our results . recently , @xcite showed the supernova neutrino spectra till 20 s after the bounce for a progenitor with @xmath133 . while they mimicked explosion with an artificially enhanced neutrino reaction rate , their results are qualitatively consistent with ours . in particular , the drops of neutrino luminosity and average energy due to the onset of shock wave revival , where the matter accretion vanishes , are also seen in their models . roughly speaking , the light curve by @xcite is similar to our models with @xmath134 ms . incidentally , the neutrino average energy is important for the nucleosynthesis such as @xmath60-process and @xmath7-process ( e.g. * ? ? ? * ; * ? ? ? because we have calculated many models with different properties of progenitors in a consistent manner , we can examine the dependence of supernova neutrino emission properties on progenitors . it is quantitatively confirmed that the total emission energy of supernova neutrinos is related with the core mass of progenitors . thus it is larger for the progenitors with lower metallicity , but is not monotonically related to the initial mass of progenitors due to the mass loss during the pre - collapse stages . the total neutrino energy emitted also depends on the shock revival time that determines the explosion time ; it increases with @xmath135 because of more material accreting to the collapsed core . the increase is @xmath2 2050% by changing @xmath135 from 100 to 300 ms . in the following we discuss some potential applications of the theoretical supernova neutrino data set for various studies . first , our results could be used as immediately comparable templates for the future detection by neutrino detectors . as mentioned repeatedly , the explosion mechanism of collapse - driven supernovae is still unknown . the detection of neutrinos would give a clue to the diagnosis of the explosion mechanism , because they come from deep inside the supernova as shown in this paper . especially , the drops of luminosity and average energy of neutrinos are important as the observational signature of the explosion . our models would give a quantitative guideline to search such signatures in future detections . furthermore , a sudden stop of neutrino emission would be a signature of a black hole formation , like our model with @xmath102 and @xmath42 , providing an exciting opportunity to directly observe the birth of a black hole . the neutrino signal reaches to the earth earlier than the electromagnetic signal , and hence triggering the signal by working neutrino detectors and rapidly informing astronomical communities are of crucial importance . the various models presented here would be useful to construct such triggering systems , especially for optimizing the efficiency around the detection threshold level . it would be possible to predict the spectrum of supernova relic neutrinos using our results because we have evaluated the spectra for various progenitor models with different initial masses and metallicities . the supernova relic neutrinos are the integration of neutrino flux emitted by all collapse - driven supernovae in the causally - reachable universe . according to the search for supernova relic neutrinos at super - kamiokande @xcite , the signal was still not seen but the upper limit was close to the standard predictions @xcite . if the signal of supernova relic neutrinos is actually detected , it would give us a unique constraint on the cosmic star formation history and initial mass function . a significant excess of diffuse supernova neutrino flux compared with that expected from observed supernova rate may indicate a contribution from failed supernovae , most likely black hole forming events , like our model with @xmath102 and @xmath42 @xcite . supernova neutrinos are valuable not only for astrophysics but also for physics of neutrino itself . while we have not taken into account the neutrino oscillation in calculations of the models presented here , it can easily be dealt as a post - process to predict final neutrino signal reaching detectors on the earth . ( e.g. * ? ? ? recently , as indicated by the results from t2k @xcite and minos experiments @xcite , the mixing angle of neutrino oscillation @xmath136 is confirmed to be nonzero and evaluated as @xmath137 by the results from reactor neutrino experiments such as daya bay @xcite , reno @xcite and double chooz @xcite . thus , at present , the most undetermined parameter in the neutrino oscillation is the mass hierarchy . since , for @xmath137 , the survival probabilities of @xmath131 , @xmath76 and @xmath138 in the stellar envelope are different for the normal and inverted mass hierarchies , detections of the supernova neutrinos would give useful information of the mass hierarchy ( e.g. * ? ? ? the supernova neutrino signal depends on the equation of state , especially for the black - hole - forming case @xcite . the effect of hyperons , which is not taken into account in our equation of state @xcite , would be important not only for the black hole formation @xcite but also for the proto - neutron star cooling @xcite . however , hyperonic equation of state is an unsettled hot topic . in particular , recently , the mass of the binary millisecond pulsar j1614 - 2230 was evaluated as @xmath139 @xcite . unfortunately , this remarkable precision thanks to a strong shapiro delay signature excludes almost all models of hyperonic equation of state , because the maximum mass of neutron stars gets lower by the hyperon inclusion ( e.g. * ? ? ? * ; * ? ? ? neutrino interactions in matter are also affected by equation of state . in this study , we have adopted a single nuclear equation of state of @xcite without hyperons , which is based on mostly standard assumptions and can be regarded as a baseline model . while there are some issues beyond the scope of this study , we hope that the result of this paper will be useful for further progress of the related fields in astrophysics and neutrino physics . in this work , numerical computations were partially performed on the supercomputers at center for computational astrophysics ( cfca ) in the national astronomical observatory of japan ( naoj ) , research center for nuclear physics ( rcnp ) in osaka university , the university of tokyo , yukawa institute for theoretical physics ( yitp ) in kyoto university , japan atomic energy agency ( jaea ) and high energy accelerator research organization ( kek ) . this work was partially supported by grants - in - aid for research activity start - up from the japan society for promotion of science ( jsps ) through no . 23840038 , and for the scientific research on innovative areas from the ministry of education , culture , sports , science and technology ( mext ) in japan through no . the authors acknowledge supports by grants - in - aid for the scientific research from mext in japan through nos . 22540296 ( k.s . ) and 24105008 ( k.n . ) . the numerical data of supernova neutrino emission computed in this study are publicly available on the web at + http://asphwww.ph.noda.tus.ac.jp/snn/ + this data set is open for general use in any research for astronomy , astrophysics , and physics . not only the original data of @xmath7rhd and pnsc simulations but also combined data from the onset of collapse to 20 s after the core bounce with the interpolation ( [ itplft ] ) are provided . on the web , the differential neutrino number flux @xmath140 and differential neutrino number luminosity @xmath141 at the time @xmath142 are prepared for each neutrino species @xmath143 and energy bin @xmath144 . moreover , the spectral data integrated from the onset of collapse to 20 s after the core bounce with the interpolation ( [ itplft ] ) are also shown . abe , k. , et al . 2011 , phys . , 107 , 041801 abe , y. , et al . 2012 , phys . d , 86 , 052008 adamson , k. , et al . 2011 , phys . lett . , 107 , 181802 ahn , j. k. , et al . 2012 , phys . lett . , 108 , 191802 an , f. p. , et al . 2012 , phys . lett . , 108 , 171803 ando , s. , & sato , k. 2004 , new j. phys . , 6 , 170 bays , k. , et al . 2012 , phys . d , 85 , 052007 bethe , h. a. , & wilson j. r. , 1985 , , 295 , 14 beacom , j. f. 2010 , ann . , 60 , 439 bionta , r. m. , et al . 1987 , phys . lett . , 58 , 1494 blondin , j. m. , mezzacappa , a. & demarino , c. 2003 , , 584 , 971 burrows , a. , klein , d. & gandhi , r. 1992 , phys . d , 45 , 3361 burrows , a. & lattimer , j. m. 1986 , , 307 , 178 burrows , a. , livne e. , dessart , l. , ott , c. d. & murphy , j. 2006 , , 640 , 878 chakraborty s. , fischer , t. , mirizzi a. , saviano , n. , & toms , r. , 2011 , phys . , 107 , 151101 dasgupta , b. , oconnor , e. p. , & ott , c. d. 2012 , phys . rev . d , 85 , 065008 de jager , c. , nieuwenhuijzenm h. , & van der hucht , k. , 1988 , a&as , 72 , 259 demorest , p. b. , pennucci , t. , ransom , s. m. , roberts , m. s. e. & hesseles , j. w. t. 2010 , nature , 467 , 1081 fischer , t. , et al . , 2011 , , 194 , 39 fischer , t. , martinaz - piedo , g. , hempel , m. , & liebendrfer m. , 2012 , phys . d , 85 , 083003 fischer , t. , whitehouse , s. c. , mezzacappa , a. , thielemann , f .- k . , & liebendrfer m. , 2010 , a&a , 517 , a80 fryer , c. l. 1999 , , 522 , 413 fryer , c. l. , & warren , m. s. 2002 , , 601 , 391 herant , m. , benz , w. , hix , w. r. , fryer , c. l. , & colgate , s. a. 1994 , , 435 , 339 hirata , k. , et al . 1987 , phys . lett . , 58 , 1490 horiuchi , s. , beacom , j. f. , & dwek , e. 2009 , phys . rev . d , 79 , 083013 hdepohl l. , mller b. , janka h .- th . , marek , a. , & raffelt , g. g. , 2010 , phys . rev . lett . , 104 , 251101 ishizuka , c. , ohnishi , a. , tsubakihara , k. , sumiyoshi , k. , & yamada , s. , 2008 , j. of phys . g , 35 , 085201 janka h .- th . , 2012 , ann . nucl . part . , 62 , 407 keil , w. , & janka h .- th . , 1995 , a&a , 296 , 145 kotake , k. , sato , k. , & takahashi , k. 2006 , rep . phys . , 82 , 2644 kotake , k. , takiwaki , t. , suwa , y. , iwakami nakano , w. , kawagoe , s. , masada , y. , & fujimoto , s. 2012 , arxiv:1204.2330 [ astro-ph.he ] , accepted by advances in astronomy kudritzki , r. , pauldrach a. , puls j. , & abbott , d. , 1989 , a&a , 219 , 205 leblanc , j. m. , & wilson , j. r. 1970 , , 161 , 541 lien , a. , fields , b. d. , & beacom , j. f. 2010 , phys . rev . d , 81 , 083001 lunardini , c. , 2009 , phys . , 102 , 231101 malek , m. , et al . 2003 , phys . lett . , 90 , 061101 marek , a. , & janka h .- th . , 2009 , , 694 , 664 murphy , j. w. , dolence , j. c. , & burrows , a. 2012 , arxiv:1205.3491 [ astro-ph.sr ] , submitted to nakazato , k. , furusawa , s. , sumiyoshi , k. , ohnishi , a. , yamada , s. , & suzuki , h. 2012 , , 745 , 197 nakazato , k. , sumiyoshi , k. , & yamada , s. 2007 , , 666 , 1140 nomoto , k. , tominaga , n. , umeda , h. , kobayashi , c. , & maeda , k. 2006 , nucl . phys . , a777 , 424 obergaulinger , m. & janka h .- th . , 2011 , arxiv:1101.1198 [ astro-ph.sr ] , submitted to a&a oconnor , e. , & ott , c. d. 2011 , , 730 , 70 ott , c. d. 2009 , class . quantum grav . , 26 , 063001 oyamatsu , k. 1993 , nucl . phys . , a561 , 431 peeples , m. s. , & somerville , r. s. 2012 , arxiv:1208.4366 [ astro-ph.co ] , accepted by mnras . poelarends , a. j. t. , herwig , f. , langer , n. & heger , a. , 2008 , , 675 , 614 pons , j. a. , reddy , s. , prakash , m. , lattimer , j. m. , & miralles , j. a. , 1999 , , 513 , 780 qian , y .- z . , & woosley , s. e. 1996 , , 471 , 331 raffelt g.g . , 2012 , arxiv:1201.1637 [ astro-ph.sr ] roberts l.f . , 2012 , , 755 , 126 shen , h. , toki , h. , oyamatsu , k. , & sumiyoshi , k. 1998a , nucl . , a637 , 435 shen , h. , toki , h. , oyamatsu , k. , & sumiyoshi , k. 1998b , prog . phys . , 100 , 1013 shen , h. , toki , h. , oyamatsu , k. , & sumiyoshi , k. 2011 , , 197 , 20 smartt , s. j. , eldridge , j. j. , crockett , r. m. , & maund , j. r. 2009 , mnras , 395 , 1409 smith , n. , li , w. , filippenko , a. v. , & chornock , r. 2011 , mnras , 412 , 1522 sugahara , y. , & toki , h. 1994 , nucl . a579 , 557 sumiyoshi , k. , suzuki , h. , & toki , h. , 1995 , a&a , 303 , 475 sumiyoshi , k. , yamada , s. , & suzuki , h. , 2007 , , 667 , 382 sumiyoshi , k. , yamada , s. , suzuki , h. , & chiba , s. , 2006 , phys . 97 , 091101 sumiyoshi , k. , yamada , s. , suzuki , h. , shen , h. , chiba , s. , & toki , h. 2005 , , 629 , 922 suzuki , h. 1994 , in physics and astrophysics of neutrinos , ed . m. fukugita & a. suzuki ( tokyo : springer ) , 763 takiwaki , t. , kotake , k. & sato , k. , 2009 , , 691 , 1360 takiwaki , t. , kotake , k. & suwa , y. , 2012 , , 749 , 98 tanaka , m. , et al . , 2009 , , 699 , 1119 thielemann , f .- k . , hirschi , r. , liebendrfer m. , & diehl , r. , 2011 , in lecture notes in physics , vol . r. diehl , d.h . hartmann & n. prantzos ( berlin : springer verlag ) , 153 thompson , t. a. , burrows , a. , & pinto , p. a. , 2003 , , 592 , 434 totani , t. , sato , k. , dalhed , h. e. , & wilson , j. r. , 1998 , , 496 , 216 totani , t. , sato , k. , & yoshii , y. , 1996 , , 460 , 303 umeda , h. , & nomoto , k. 2008 , , 673 , 1014 umeda , h. , yoshida , t. , & takahashi , k. 2012 , prog . phys . , 2012 ( 1 ) , 01a302 vink , j. s. , de koter a. , & lamers , h. j. g. l. m. , 2001 , a&a , 369 , 574 woosley , s. e. , hartmann , d. h. , hoffman , r. d. , & haxton , w. c. 1990 , , 356 , 272 woosley , s. e. , heger , a. , & weaver , t. a. 2002 , rev . , 74 , 1015 yamada s. 1997 , , 475 , 720 yamada , s. , janka h .- th . , & suzuki , h. , 1999 , a&a , 344 , 533 yoshida , t. , & umeda , h. 2011 , mnras , 412 , l78

a new approach to the construction of index formulas for elliptic operators on singular manifolds is suggested on the basis of-theory of algebras and cyclic cohomology . the equivalence of toeplitz and pseudodifferential quantizations , well known in the case of smooth closed manifolds , is extended to the case of manifolds with conical singularities . we describe a general construction that permits one , for a given toeplitz quantization of a-algebra , to obtain a new equivalent toeplitz quantization provided that a resolution of the projection determining the original quantization is given .

math0503267

in recent years , the problem of finding index formulas for elliptic pseudodifferential operators on singular manifolds has been studied in numerous papers ( e.g. , see plamenevsky - rozenblum @xcite , melrose - nistor @xcite , rozenblum @xcite , schulze - sternin - shatalov @xcite , nazaikinskii - sternin @xcite ) . despite considerable progress in this direction , the situation is far from being clear yet . indeed , some of the formulas known in the literature fail to express the index via the principal symbol treated as an element of an appropriate calkin algebra ( e.g. , @xcite ) , in other formulas , separate terms are not homotopy invariant ( e.g. , see fedosov - schulze - tarkhanov @xcite ) , and the few formulas that combine both desirable properties ( e.g. , see @xcite ) are valid only for an important but rather narrow class of operators satisfying certain symmetry conditions . this situation is primarily caused by the complicated symbol structure for pseudodifferential operators on singular manifolds . these symbols consist of components ( in general , operator - valued ) corresponding to the strata of the manifold and satisfying certain matching conditions for the adjacent strata see , e.g. schulze @xcite . in a number of existing results , the index of the operator is expressed as a sum of contributions from these symbol components . these contributions are usually noninteger and lack homotopy invariance , which is not at all surprising in the presence of the matching conditions . it is also difficult to assign to them any straightforward topological or algebraic meaning . one can ensure their homotopy invariance only by severely restricting the class of operators to be considered ( say , by imposing some symmetry conditions ) . a detailed analysis of symmetry conditions and their role in obtaining invariant index formulas can be found in savin - schulze - sternin @xcite and in @xcite , and we do not dwell on the topic here . in the present paper , we propose another approach to the construction of index formulas on manifolds with singularities . this approach is based on @xmath0-theory of algebras and cyclic cohomology . we take a slightly different viewpoint as to what a `` good '' index formula must be . instead of trying to use topological invariants of separate components of the symbol ( which is hopeless due to results of @xcite ) , we consider the symbol as a whole , that is , as an element of an appropriate symbol algebra @xmath2 . moreover , instead of topological objects , one naturally deals with algebraic objects like the chern connes character viewed as an element of the cyclic cohomology group of the symbol algebra . in the abstract framework , the scheme is well known . if the algebra @xmath2 is separable , then every quantization of @xmath2 ( in particular , the pseudodifferential quantization , which is of interest to us ) is equivalent ( e.g. , see blackadar @xcite ) . ] to some ( generalized ) toeplitz quantization @xmath3 in the hilbert space @xmath4 defined as the range of an orthogonal projection @xmath5 in some hilbert @xmath2-module @xmath6 ; here @xmath5 is assumed to almost commute with the action of @xmath2 . under the additional condition that the commutators of elements of the algebra with the projection are not only compact but also belong to an appropriate von neumann schatten class , the toeplitz representation permits one to write out analytic index formulas ( see connes @xcite ) . namely , if @xmath7 is a dense local subalgebra of @xmath2 such that the restriction of the toeplitz quantization to @xmath8 is @xmath9-summable,$ ] belongs to the @xmath9th von neumann schatten class @xmath10 for every @xmath11 . ] then for an arbitrary elliptic element @xmath12 one has the index formula @xmath13\otimes\operatorname{tr}\bigr ) \bigl(\underbrace{\mathfrak{a}^{-1},\mathfrak{a},\mathfrak{a}^{-1 } , \mathfrak{a } , \ldots,\mathfrak{a}^{-1 } , \mathfrak{a}}_{\text{$n+1 $ arguments}}\bigr),\ ] ] where @xmath14 is odd , @xmath15 is the matrix trace , and the _ chern connes character _ @xmath16 $ ] of the quantization @xmath17 is given by the formula @xmath18(a_0,a_1,\ldots , a_n)\\=\sqrt{2i } ( -1)^{n(n-1)/2}\gamma{\biggl}(\frac n2 + 1{\biggr})^{-1}\operatorname{tr}\{a_0[p , a_1][p , a_2]\dotsm [ p , a_n]\}.\end{gathered}\ ] ] ( here @xmath19 is the operator trace in @xmath6 . ) to apply this formalism to manifolds with singularities , we do the following . for the case of a smooth compact closed manifold @xmath20 , there is a well - known toeplitz quantization @xmath21 of the symbol algebra @xmath22 with the help of the szeg - caldern projection @xmath23 in the space @xmath24 of square integrable functions on the cosphere bundle @xmath25 , see guillemin , @xcite , boutet de monvel @xcite , boutet de monvel - guillemin @xcite . this construction was explained in @xcite as showing that pseudodifferential operators are toeplitz operators in disguise . in the present paper , we generalize this quantization and construct a toeplitz quantization equivalent to the standard pseudodifferential quantization for manifolds with conical singularities . thus pseudodifferential operators on a manifold with conical singularities are toeplitz operators in disguise , too . ( this toeplitz quantization acts in a more complicated space , which , however , turns into @xmath24 if there are no singularities . ) in the smooth case , the equivalence ( and even the almost isomorphism ) of the toeplitz quantization to the pseudodifferential quantization as a homogeneous function of degree zero in the fibers is assumed . ] @xmath26 is given by the so - called guillemin transform @xmath27 ( see the cited papers ) , and we construct an analog of this transform for manifolds with singularities , thus proving the equivalence ( and even an almost isomorphism ) of quantizations in this case . this naturally results in an index formula of the form for elliptic operators on manifolds with conical singularities . this algebraic index formula has several advantages : * first , it expresses the index via the principal symbol alone and is homotopy invariant ; * second , it is valid for arbitrary elliptic symbols , not just for symbols satisfying some symmetry conditions ; * third , most importantly , it is expressed in terms of the cyclic cohomology class @xmath16 $ ] of the algebra @xmath8 . it is the cyclic cohomology of the symbol algebra that replaces the homology of a manifold as one passes from the algebra of functions on a smooth manifold to more general symbol algebras . a serious disadvantage of this formula ( and in general of the index formula for toeplitz quantizations given by projections of general form ) is that it fails to be local . at the same time , if the chosen projection is the positive spectral projection of an unbounded local operator , then a different representation of the index cocycle is possible , which results in more traditional , local index formulas . ( see connes - moscovici @xcite . ) in this connection , the construction of a local index formula involves the natural problem of finding an unbounded local operator such that the toeplitz quantization generated by its positive spectral projection is equivalent to the pseudodifferential quantization on a manifold with conical singularities . in the smooth case , the desired operator is the dirac operator . more precisely , the natural self - adjoint dirac operator @xmath28 acting on sections of the spinor bundle on the odd - dimensional contact manifold @xmath25 is associated with the almost complex structure on the distribution of contact hyperplanes . then the index of a pseudodifferential operator @xmath29 proves to be equal to the index of the toeplitz operator constructed from the symbol @xmath30 and the positive spectral projection of the dirac operator ( see baum - douglas @xcite ) : @xmath31 where @xmath32 is the natural projection . this equality of indices shows that ( at least , modulo torsion elements in the group @xmath33 ) the quantization given by the positive spectral projection @xmath34 of the dirac operator is equivalent to the pseudodifferential quantization and hence to the quantization with the help of the szeg caldern projection . there are @xmath0-theoretic proofs of this fact . ( one of then is based on the atiyah singer theorem , e.g. , see kaminker @xcite , and another , by baum - douglas - taylor @xcite , uses some facts concerning the @xmath35-neumann problem . ) both these approaches to the proof of encounter serious difficulties in the case of manifolds with singularities . the main difficulty is that so far one does not have even a hypothetical candidate for the `` dirac operator '' ( i.e. , an operator for which holds ) for the case of manifolds with conical singularities . a prototype for a possible generalization of formula to the conical case is the proof of ( as a special case of more general index formulas for toeplitz operators and quantized contact transformations on general contact manifolds ) due to epstein , melrose , and mendoza ( see @xcite and also @xcite ) . their proof _ relates the szeg caldern projection and the positive spectral projection of the dirac operator by a finite chain of transformations _ and hence is of interest to us as a model for the possible definition of the dirac operator in the conical case . indeed , we have already constructed a counterpart of the szeg caldern projection for manifolds with conical singularities and proved the equivalence of the toeplitz quantization and the pseudodifferential quantization . the epstein melrose mendoza construction comprises three steps ( we give more detail in the appendix ) : 1 . constructing a resolution of the szeg caldern projection ( in the case of the cosphere bundle of a smooth manifold , the resolution is given by the kohn rossi complex , e.g. , see folland - kohn @xcite ) ; 2 . proving that the toeplitz quantization associated with the positive spectral projection of the operator @xmath36 obtained as the standard roll - up of this resolution produces the operator with the same index as the toeplitz quantization associated with the szeg caldern projection ; 3 . deforming the self - adjoint operator @xmath36 to the dirac operator @xmath28 . in the present paper , we show that step 2 of this construction is a special case of a general assertion that permits one , starting from a toeplitz quantization of an arbitrary @xmath1-algebra @xmath2 , to obtain new equivalent ( i.e. , defining the same class in @xmath33 ) toeplitz quantizations associated with positive spectral projections of self - adjoint operators . the construction goes as follows : for a projection @xmath5 almost commuting with the action of @xmath2 , one takes a finite - length resolution in hilbert @xmath2-modules . rolling up the resolution in the standard way and adding the projection itself and the projection onto the cokernel in the last term of the resolution , one obtains a self - adjoint operator in the direct sum of all hilbert modules occurring in the resolution . the positive spectral projection of this self - adjoint operator is the desired new projection . in conclusion , we note that the so far open problem of defining the dirac operator @xmath28 in the conical case is undoubtedly of interest in that such an operator is a `` fundamental cycle '' of the algebra @xmath2 ( the noncommutative manifold ) and , in particular , defines the poincar duality on it : @xmath37 in the smooth case ( @xmath38 ) , this is reduced to the isomorphism @xmath39 the paper is organized as follows . in the first section , we construct a toeplitz representation of the pseudodifferential quantization on manifolds with conical singularities . section 2 contains the above - mentioned construction of equivalent quantizations from resolutions of projections . first we consider the case in which the resolution is formed by bounded operators and then the case in which the resolution is formed by unbounded operators . in the appendix , we briefly describe the construction of @xcite for the special case in which the contact manifold in question is the cosphere bundle of a smooth compact manifold without boundary . throughout the paper , we assume that the reader is acquainted with the notions and definitions of @xmath0-theory of operator algebras and related topics ( e.g. , see @xcite ) . on the other hand , the necessary definitions and facts of the theory of ( pseudo)differential operators on manifolds with conical singularities are given in subsection 1.1 together with appropriate bibliographic references . we consider pseudodifferential operators on a compact manifold @xmath20 with isolated conical singularities . we always assume that there is only one conical point @xmath41 ; extending the results to the general case encounters no difficulties . in this subsection we briefly list main definitions and facts . details can be found , e.g. , in schulze @xcite , egorov - schulze @xcite , schulze - sternin - shatalov @xcite and , as it concerns the cylindrical representation for cone - degenerate pseudodifferential operators , also in nazaikinskii - sternin @xcite . a _ manifold @xmath20 with conical singularity _ @xmath41 is a compact hausdorff topological space @xmath20 with a distinguished point @xmath41 such that the set @xmath42 is equipped with the structure of a @xmath43 manifold compatible with the topology and there is a given homeomorphism @xmath44 of a neighborhood @xmath45 of @xmath41 onto a cone @xmath46 with smooth compact base @xmath47 ; moreover , the mapping @xmath48 takes @xmath41 to the vertex @xmath49 of the cone , and the restriction of @xmath48 to the punctured neighborhood @xmath50 of the singular point is a diffeomorphism . the coordinate on the interval @xmath51 in the representation will be denoted by @xmath52 . points of @xmath47 ( and sometimes local coordinates on @xmath47 ) will be denoted by @xmath53 . the coordinates @xmath54 are called the _ conical coordinates_. however , we mainly use the _ cylindrical _ coordinates related to the conical coordinates by the change of variables @xmath55 . in the cylindrical representation , the smooth open manifold @xmath56 looks like a manifold with a cylindrical end , and the punctured neighborhood @xmath57 is represented as the direct product @xmath58 the function @xmath59 well defined on @xmath57 by the direct product structure can be continued to a smooth function @xmath60 everywhere nonpositive outside @xmath57 . we assume that the continuation is chosen and fixed . on @xmath20 we take some riemannian metric @xmath61 that has the form @xmath62 in the decomposition , where @xmath63 is a riemannian metric on @xmath47 . the _ weighted sobolev space _ @xmath64 , @xmath65 , is defined as the completion of the space @xmath66 of smooth compactly supported functions on @xmath56 with respect to the norm @xmath67 where @xmath68 and @xmath69 are , respectively , the beltrami laplace operator and the measure associated with the metric . the numbers @xmath70 and @xmath71 are called the _ smoothness exponent _ and the _ weight exponent _ , respectively . we deal only with pseudodifferential operators in the sobolev spaces @xmath72 with zero weight exponent ; the general case can be reduced to this with the help of the isomorphism and @xmath73 . ] @xmath74 moreover , we mainly deal with zero - order pseudodifferential operators , and so our main space is @xmath75 . occasionally , we shall use other spaces from the scale @xmath76 . the cosphere bundle @xmath25 of the manifold @xmath20 with a conical singular point is of interest to us as the space on which the interior principal symbols of pseudodifferential operators are defined . this space is defined in the standard manner as the quotient of the complement to the zero section in the _ compressed cotangent bundle _ @xmath77 by the action of the multiplicative group @xmath78 of positive numbers . an intrinsic geometric definition of the compressed cotangent bundle can be found , say , in melrose @xcite . we do not reproduce it here but just describe @xmath25 in cylindrical coordinates , which is sufficient for our aims . namely , we consider the cosphere bundle @xmath79 and compactify it by attaching the manifold @xmath80/\rr_+\ ] ] ( where @xmath81 is the zero section of the vector bundle @xmath82 ) at @xmath83 as follows . the direct product structure on @xmath57 induces a canonical isomorphism @xmath84/\rr_+\ ] ] for each point @xmath85 . now we say that a sequence @xmath86 , where @xmath87 and @xmath88/\rr_+$ ] , tends to @xmath89/\rr_+$ ] if @xmath90 , @xmath91 , and @xmath92 . the compactified space @xmath25 thus obtained is a smooth manifold with boundary @xmath93/\rr_+.\ ] ] ( the smooth structure on @xmath25 is defined via the radial coordinates . ) in the subsequent argument , it is now and then convenient to interpret operators on @xmath20 whose integral kernels are compactly supported in @xmath94 as operators on some closed manifold . the simplest way to obtain this closed manifold is to cut away the cylindrical end ( say , at @xmath95 ) and attach the second copy of the resulting manifold with boundary . this procedure gives a closed compact manifold , which is called the _ double _ of @xmath20 and denoted by @xmath96 . ( note that in contrast with the papers @xcite , dealing with index problems for operators with symmetry conditions , where the passage to the double has a topological meaning , in our case this passage is a purely technical tool of analysis . ) we can perform a similar cut - and - paste procedure with @xmath25 , which results in @xmath97 . ( note , however , that the fibres of the cosphere bundle must be glued together via the involution @xmath98 rather than identically , where @xmath9 is the dual variable of @xmath99 . this is because the directions of the @xmath99-axis on the first and second copies are opposite . ) we deal with scalar classical pseudodifferential operators of order zero . the algebra of such operators will be denoted by @xmath100 . every pseudodifferential operator @xmath101 is uniquely determined modulo compact operators by its _ principal symbol _ , which is the pair @xmath102 consisting of the _ interior symbol _ @xmath103 , which is a @xmath43 function on @xmath25 , and the _ conormal symbol _ @xmath104 , which is a family of classical zero - order pseudodifferential operators with parameter @xmath105 on the manifold @xmath47 in the sense of agranovich vishik @xcite . ( in particular , this means that the operator @xmath106 on the cylinder @xmath107 is a translation - invariant classical zero - order pseudodifferential operator . this operator will sometimes be also referred to as the conormal symbol . ) the elements of this pair must satisfy the matching condition @xmath108 where @xmath109 is the restriction to the spheres of the principal symbol of @xmath104 treated as an operator with parameter . the principal symbols @xmath110 thus defined are just elements of the calkin algebra @xmath111 of the algebra of zero - order classical pseudodifferential operators in @xmath112 . ( here @xmath113 is the ideal of compact operators in @xmath112 . ) the algebra @xmath8 is a dense local subalgebra of the symbol @xmath1-algebra @xmath2 obtained as the closure of @xmath8 in the direct sum @xmath114 , where @xmath115 is the algebra of continuous linear operators in @xmath116 . the multiplication law in @xmath2 is defined separately on each component ; it is given by pointwise multiplication of functions on the first component ( interior symbols ) and by multiplication of operator families on the second component . the _ pseudodifferential quantization _ @xmath117 on @xmath20 is a continuous linear mapping such that @xmath118 the construction of this mapping , which uses partitions of unity , can be found in the cited literature . this mapping is unique and multiplicatively homomorphic modulo compact operators . the invertibility of the principal symbol ( that is , the nonvanishing of @xmath103 and the invertibility of @xmath104 for each @xmath105 ) is a sufficient and necessary condition for the operator @xmath119 to be fredholm . invertible elements in @xmath2 ( and in @xmath120 are also said to be _ elliptic_. let us now present our results . to specify a toeplitz quantization of the algebra @xmath2 , one should describe a hilbert @xmath2-module @xmath121 and an orthogonal projection @xmath5 in @xmath121 almost commuting with the action of the algebra . let us construct both objects . to choose this module , we use the following motivation . irreducible representations of @xmath2 fall into two series see plamenevsky - senichkin @xcite . one series consists of one - dimensional representations parametrized by points @xmath123 and given by the formula @xmath124 the other series consists of the infinite - dimensional representations @xmath125 in the hilbert space @xmath126 . this suggests that the hilbert @xmath2-module @xmath121 ( see ) occurring in the definition of the toeplitz quantization can be obtained as a direct integral of these representations . more precisely , we take @xmath127 ( where the measure on @xmath79 is translation invariant on the cylindrical end ) and define the action of @xmath2 on @xmath121 componentwise by the formula @xmath128 the @xmath2-module thus defined is faithful . we construct a toeplitz quantization ( and prove its equivalence to the pseudodifferential quantization ) not for the entire algebra @xmath2 but for the subalgebra @xmath130 consisting of symbols @xmath131 stabilizing at infinity , i.e. , such that @xmath132 is independent of @xmath99 for @xmath133 . the passage to this subalgebra is sufficient at least in index theory , since @xmath129 has the following properties : * an element is invertible in @xmath134 if and only if it is invertible in @xmath120 ; * each elliptic element @xmath135 is homotopic in @xmath120 via elliptic elements with the same conormal symbol to an element of @xmath134 . the homotopy mentioned in the latter item `` sweeps '' all the variation of @xmath136 in @xmath99 on the cylindrical end into a small half - neighborhood ( @xmath137 ) of the section @xmath138 of @xmath20 . in this subsection we construct the projection @xmath5 by describing its range @xmath139 as the range of a continuous mapping @xmath140 which proves to be an almost isomorphism of @xmath112 onto @xmath141 . this mapping will be called the _ pseudo - guillemin transform _ ( or the guillemin transform for manifolds with singularities ) , since , as we shall see shortly , it is a counterpart of the ordinary guillemin transform @xcite for smooth compact manifolds and specifies an equivalence between the toeplitz quantization corresponding to @xmath5 and the pseudodifferential quantization . thus let us describe @xmath142 . on the real line @xmath143 , we consider a partition of unity @xmath144 such that the @xmath145 are smooth real - valued functions and @xmath146 next , let @xmath147 be a real - valued function such that @xmath148 and @xmath149 . all these functions can be viewed as functions on @xmath20 and @xmath150 as well as on the double @xmath96 if we extend them as continuous functions beyond the cut @xmath95 by constant values on the newly attached second copy of @xmath20 . using the natural projections , we also lift these functions to @xmath25 , @xmath151 , and @xmath97 . let @xmath152 be the guillemin transform @xcite for the compact closed @xmath43 manifold @xmath96 . we define a mapping @xmath153 by the formula @xmath154 here the right - hand side is well defined . indeed , the function @xmath155 is supported in @xmath156 and hence can be treated as a function on @xmath96 , whereby we can apply the guillemin transform @xmath157 . next , the multiplication by @xmath158 permits us to understand @xmath159 as a function on @xmath25 ( supported in @xmath160 ) rather than @xmath97 . likewise , @xmath161 can be viewed as a function on the cylinder @xmath150 . the continuous mapping , is called the _ pseudo - guillemin transform _ on a manifold @xmath20 with singularities . now we state our main theorem . [ main - theorem ] the following assertions hold . * the range @xmath162 of the mapping @xmath142 is closed . * the mapping @xmath163 is fredholm and almost unitary . ] where @xmath164 and @xmath165 are compact operators . ] * the orthogonal projection @xmath5 on @xmath141 in @xmath122 satisfies the condition @xmath166\in\mathfrak{s}_k\quad\text{for $ k>2n-1 $ and for arbitrary $ a\in\wt\ca$.}\ ] ] thus , @xmath5 determines a @xmath167-summable toeplitz quantization of @xmath129 . * one has @xmath168 modulo compact operators of the von neumann schatten class @xmath169 , so that @xmath142 determines an equivalence between the toeplitz quantization associated with @xmath5 and the pseudodifferential quantization @xmath170 . thus the theorem states that the pseudo - guillemin transform @xmath142 defines a representation of the pseudodifferential quantization on a manifold with isolated singularities as a toeplitz fredholm module of order @xmath171 . first we recall the main properties of the usual guillemin transform , which will be useful in our proof . the guillemin transform has the following properties : * @xmath157 is an operator of order zero in sobolev scales , that is , @xmath172 is bounded for all @xmath70 ; * the range @xmath173 of @xmath157 is closed ; * @xmath174 , and @xmath175 is the projection onto @xmath173 ; * [ 4 ] for every smooth function @xmath176 on @xmath97 , the commutator is interpreted as the operator of point - wise multiplication by @xmath176 . ] @xmath177 $ ] is an order @xmath178 operator in the sobolev scale on @xmath97 and belongs to the von neumann schatten class @xmath179 for @xmath171 ; * [ 5 ] for every smooth function @xmath176 on @xmath97 one has @xmath180 where @xmath181 is an arbitrary pseudodifferential operator of order zero on @xmath96 with principal symbol @xmath176 and @xmath0 is a pseudodifferential operator of order @xmath178 in the sobolev scale on @xmath96 thus belonging the von neumann schatten class @xmath182 for @xmath183 . now we proceed directly to the proof of the main theorem . one has @xmath184 where @xmath164 is a self - adjoint pseudodifferential operator of order @xmath178 on @xmath20 with integral kernel supported in the compact set @xmath185 . it follows that @xmath186 is fredholm and @xmath187 is finite - dimensional and consists of functions supported in @xmath188 . by standard argument , we find that @xmath189 is closed . let @xmath190 be the orthogonal projection onto @xmath191 . then the operator @xmath192 is an invertible pseudodifferential operator of order @xmath73 and its inverse @xmath193 has the form @xmath194 with @xmath165 a self - adjoint pseudodifferential operator of order @xmath178 on @xmath20 whose integral kernel is supported in the compact set @xmath185 . ( hence both @xmath164 and @xmath165 belong to @xmath169 . ) now the projection on @xmath189 has the form @xmath195 next , @xmath196 where @xmath197 is compact , and we have proved assertions 1 and 2 . since @xmath193 differs from the identity operator by an operator in @xmath169 , we can safely replace the operator @xmath5 by @xmath198 in the proof of assertion 3 and estimate the commutator @xmath199 $ ] . moreover , owing to the presence of the cutoff function @xmath158 we can assume that the interior symbol @xmath200 vanishes for @xmath201 and hence @xmath202 can be treated as an operator acting in @xmath203 when we proceed with the estimates . we have @xmath204=\begin{pmatrix } [ \psi { \mathrm{t}}\chhi2 ^ 2 { \mathrm{t}}^*\psi , a ] & \psi { \mathrm{t}}\chhi2\chhi1\bha -a\psi { \mathrm{t}}\chhi2\chhi1 \\ \chhi1\chhi2{\mathrm{t}}^*a-\bha \chhi1\chhi2{\mathrm{t}}^*\psi & [ \chhi1 ^ 2,\bha ] \end{pmatrix},\ ] ] where we for brevity write @xmath205 now , by the properties of @xmath157 , @xmath206\equiv[\pi , a]\equiv0 \,.\ ] ] ( here and in the following @xmath207 stands for equality modulo operators in @xmath169 . ) next , we shall estimate the commutator @xmath208 $ ] . to this end , we introduce a smooth partition of unity @xmath209 on the real line such that the following properties hold : ( a ) @xmath210 ; ( b ) @xmath211 ; ( c ) @xmath212 is contained in the interval @xmath213 $ ] and @xmath214 on @xmath215 ) . now we represent @xmath216 in the form @xmath217 and compute the commutator as follows : @xmath218&=\sum\limits_{j , k=0,1,2}[\chhi1 ^ 2,\ph_j\bha \ph_k]\\ & = [ \chhi1 ^ 2,\ph_0\bha \ph_0]-[\chhi2 ^ 2,\ph_0\bha \ph_1 ] + [ \chhi1 ^ 2,\ph_0\bha \ph_2]\\ & \quad{}-[\chhi2 ^ 2,\ph_1\bha \ph_0]+[\chhi1 ^ 2,\ph_2\bha \ph_0 ] + \sum\limits_{j , k=1,2}(-1)^{j}[\chhi{3-j}^2,\ph_j\bha \ph_k ] . \end{split}\ ] ] here we have used the fact that @xmath219 . now the first term can be interpreted as an order @xmath178 pseudodifferential operator on @xmath96 ( the support of its schwartz kernel with respect to the variable @xmath99 is contained in the square @xmath213\times[0,10]$ ] ) and hence belongs to the desired von neumann schatten class @xmath169 . all other terms have the form @xmath220 , where @xmath221 and @xmath222 are smooth functions with disjoint supports , constant at infinity . for example , @xmath223=\chhi2 ^ 2\ph_0\bha \ph_1 -\ph_0\bha \ph_1\chhi2 ^ 2=\chhi2 ^ 2\ph_0\bha \ph_1;\ ] ] here we have @xmath224 and @xmath225 . the integral kernel of such an operator has the form @xmath226 where @xmath227 is the integral kernel of @xmath216 . the kernel @xmath227 is smooth outside the diagonal @xmath228 and decays more rapidly than an arbitrary power of @xmath229 . with regard for the arrangement of supports of @xmath230 and @xmath231 , we have @xmath232 with some constant @xmath233 on @xmath234 it follows that @xmath235 is everywhere smooth and satisfies the estimates @xmath236 for all @xmath237 ; similar estimates are valid for the derivatives of @xmath235 . we conclude that the operator @xmath220 belongs to all von neumann schatten classes . thus , we have obtained the desired estimate for the commutator @xmath208 $ ] . consider the operator @xmath238 . we have @xmath239 where @xmath240 is some pseudodifferential operator with symbol @xmath200 , and the desired assertion follows ( since @xmath241 on @xmath242 ) by an argument similar to the preceding one . the estimate for the lower left entry of the commutator is similar . it remains to prove assertion 4 . we have @xmath243 as follows from our conditions on the symbols in @xmath129 . the proof is complete . we start with an important remark . in what follows , we sometimes use a slight generalization of the construction of toeplitz quantization . suppose that the space @xmath6 where the projection @xmath5 defining the quantization acts is itself a quantization space rather than a hilbert @xmath2-module , i.e. , is equipped with an _ almost _ representation is called an almost representation if it is an algebra homomorphism modulo compact operators : @xmath244 , where @xmath0 is a compact operator . ] ( quantization ) @xmath245 of the algebra @xmath2 . ( this quantization may well be toeplitz itself . ) let @xmath5 almost commute with the quantization , i.e. , satisfy @xmath246 where @xmath207 stands for equality modulo compact operators . ( this condition is independent of the specific choice of the representative @xmath247 . ) then a slight modification of formula permits one to define a toeplitz quantization @xmath17 in @xmath248 by setting @xmath249 the quantization @xmath17 is well defined ( since it is independent of the choice of a representative @xmath247 of @xmath250 ) and will be denoted by @xmath251 . if @xmath245 is a toeplitz quantization associated with a projection @xmath252 , then @xmath253 is associated with @xmath254 . now let us describe the class of operators that will be used in our resolutions . [ admi ] let @xmath255 and @xmath256 be hilbert @xmath2-modules . a bounded linear operator @xmath257 is said to be _ admissible _ if the following conditions hold . 1 . the range of @xmath258 is closed . the operator @xmath258 almost commutes with the action of the algebra , i.e. , @xmath259 in a similar way , we define admissible operators for the case in which @xmath255 and @xmath256 are not @xmath2-modules but quantizations of the algebra @xmath2 acting in these spaces . note that a projection is admissible if and only if it almost commutes with the action ( or quantization ) of the algebra @xmath2 , i.e. , specifies a toeplitz quantization . [ odin ] let @xmath257 be an admissible operator . then the following assertions hold . 1 . the operators @xmath260 , @xmath261 , and @xmath262 , as well as the projections on the kernel , cokernel ( the kernel of the adjoint operator ) , range , and corange of @xmath258 are admissible operators . if @xmath263 is a pair of admissible projections commuting with @xmath258 , i.e. , @xmath264 then the operator @xmath265 ( which is the restriction of @xmath258 to the range of @xmath266 ) is also an admissible operator . if , moreover , @xmath258 is self - adjoint , then its positive spectral projection @xmath267 is an admissible operator . \1 ) the fact that the ranges of @xmath258 and @xmath260 are or are not closed simultaneously is well known in functional analysis . under our assumptions , the ranges of @xmath261 and @xmath262 coincide with the ranges of @xmath260 and @xmath258 , respectively , and hence are also closed . the ranges of bounded projections are always closed . the assertion that these operators commute with the action of @xmath2 modulo compact operators is nontrivial only for the projections . it suffices to consider the projections on @xmath268 and @xmath269 . consider the first projection . ( the second projection can be treated in a similar way . ) the kernel of @xmath258 coincides with the kernel of the self - adjoint positive semidefinite operator @xmath261 . since the range of @xmath261 is closed , it follows that the restriction of @xmath261 to the orthogonal complement of @xmath268 is boundedly invertible by banach s theorem . hence the operator @xmath270 is invertible for nonzero @xmath271 sufficiently small in absolute value . thus , the projection @xmath5 on @xmath268 can be expressed by the cauchy integral formula @xmath272 where the integration is carried out over a circle of small radius centered at zero . accordingly , the commutator of @xmath5 with elements of @xmath2 is expressed by the formula @xmath273&=\frac1{2\pi i } \oint [ a,(z - d^*d)^{-1}]\,dz \\ & = \frac1{2\pi i } \oint ( z - d^*d)^{-1}[a , d^*d](z - d^*d)^{-1}\,dz , \end{aligned}\ ] ] which readily implies the desired compactness . \2 ) under our assumptions , the operator @xmath274 almost commutes with the action of @xmath2 , and the fact that its range is closed is obvious . \3 ) since the range of @xmath258 is closed , it follows that zero is either in the resolvent set of @xmath258 or an isolated point of spectrum of @xmath258 . consequently , the positive spectral projection can be expressed by a cauchy type integral and we can argue as in the proof of 1 ) . let @xmath275 be an admissible projection in a hilbert @xmath2-module @xmath276 . the corresponding toeplitz quantization will be denoted by @xmath17 . an _ admissible resolution _ of length @xmath277 of the projection @xmath5 is an exact sequence of hilbert spaces @xmath278 where all spaces are hilbert @xmath2-modules and all mappings are admissible operators . a trivial example is the resolution of length @xmath73 @xmath279 for a given resolution , we construct a self - adjoint operator @xmath258 in the direct sum @xmath280 of hilbert spaces as follows . first , we augment the exact sequence with the cokernel of @xmath281 : @xmath282 the projection in @xmath283 on the cokernel @xmath284 will be denoted by @xmath285 and called the projection _ dual _ to @xmath5 . by proposition [ odin ] , this is an admissible projection . next , let @xmath200 be the differential of degree @xmath286 defined in the graded sum as the direct sum of all operators @xmath287 , and let @xmath288 be the adjoint operator ( which is automatically of degree @xmath178 ) . we set @xmath289 where the operators @xmath5 and @xmath285 act in the respective components @xmath276 and @xmath283 of the direct sum . by the preceding reasoning , it is obvious that the operator @xmath258 is self - adjoint , invertible , and admissible . if @xmath17 is a quantization of the algebra @xmath2 , then by @xmath290\in k^1(\ca)$ ] we denote its equivalence class in the @xmath0-group of @xmath2 ( see @xcite ) . moreover , if @xmath17 is a toeplitz quantization determined by a projection @xmath5 , then we freely write @xmath291 $ ] instead of @xmath290 $ ] . let us now establish the main theorem of this section . [ 34 ] the positive spectral projection @xmath267 of the operator @xmath258 is admissible . the toeplitz quantization corresponding to the projection @xmath267 is equivalent to the toeplitz quantization defined by the projection @xmath5 : @xmath292=[p_+(d)]\in k^1(\ca)\,.\ ] ] the first assertion of the theorem follows from proposition [ odin],3 ) . let us prove the equivalence of quantizations . we can break the resolution at the @xmath293th space @xmath294 , thus obtaining an admissible resolution of length @xmath293 . we denote the projection @xmath285 obtained for the latter resolution by the same construction by @xmath295 , the direct sum with @xmath277 replaced by @xmath293 by @xmath296 , and the corresponding operator @xmath258 by @xmath297 . we prove the theorem by induction on @xmath293 . \(a ) the inductive base . let @xmath298 . then @xmath299 , the augmented resolution has the form @xmath300 the dual projection is @xmath301 , and @xmath302 the positive spectral projection of the operator @xmath303 is just @xmath5 , and so the assertion is trivially true . \(b ) the inductive step . suppose that the assertion has already been proved for @xmath304 , and let us prove it for @xmath305 . to this end , we write out the corresponding resolutions of lengths @xmath167 and @xmath306 next to each other : @xmath307 we expand the spaces @xmath308 and @xmath309 into the orthogonal sums @xmath310 where @xmath311 , @xmath312 . the terms in these expansions are not @xmath2-modules , but they are equipped with toeplitz quantizations determined in the ambient spaces @xmath294 by the projections @xmath295 ( in the case of @xmath313 ) and @xmath314 ( in the case of @xmath315 ) . one can readily see that the operators @xmath316 and @xmath317 can be represented as the direct sums @xmath318 where @xmath319 is an admissible self - adjoint operator acting in the direct sum @xmath320 , @xmath200 and @xmath288 are the restrictions of the operators @xmath321 and @xmath322 to the subspaces @xmath323 and @xmath324 , respectively ( these restrictions are still the adjoints of each other ) , the signs @xmath325 and @xmath326 depend on the parity of @xmath167 , and the projection @xmath295 are assumed to be restricted are equipped with the corresponding toeplitz quantizations . ] to the subspaces @xmath313 . thus , the positive spectral projections of these operators have the form @xmath327\ ] ] for the upper signs and @xmath328\oplus\wt p_{k+1}\ ] ] for the lower signs . we claim that the positive spectral projection of the self - adjoint operator @xmath329 defines a toeplitz quantization equivalent to the quantization associated with the projection @xmath330 . indeed , first we note that the operator is self - adjoint , admissible and invertible . next , consider the unitary transformation @xmath331 it is admissible ( by holomorphic functional calculus ) and reduces @xmath332 to the block diagonal form @xmath333 the positive spectral subspace of the operator coincides with @xmath334 which gives the desired equivalence . by @xmath335 we denote the quantization corresponding to the projection @xmath336 . in this notation , for the case of lower signs in , we have @xmath337=[\tau_k]+[1-\wt p_{k+1}]+[\wt p_{k+1}]=[\tau_k]\,,\ ] ] since the elements @xmath338 $ ] and @xmath339 $ ] are the inverses of each other in the @xmath0-group . for the case of upper signs , we have @xmath340=[\tau_k]+[1-\wt p_{k+1}]-[\wt p_{k}]\,.\ ] ] to complete the proof , let us show that @xmath341=[\wt p_{k}]\,.\ ] ] indeed , the operator @xmath200 is an admissible isomorphism @xmath342 whence follows . the proof is complete . now let us consider the case in which one deals with an unbounded resolution of the projection operator . resolutions of such form naturally arise in applications . consider the sequence @xmath343 where 1 . all @xmath294 , @xmath344 , are hilbert @xmath2-modules ; 2 . all @xmath287 , @xmath345 , are closed densely defined linear operators with closed range ; 3 . the operators @xmath287 boundedly commute with the action of the algebra @xmath2 , that is , for each @xmath346 and each @xmath345 one has @xmath347 and the commutator @xmath348 $ ] , well defined on @xmath349 by virtue of this inclusion , is bounded and hence can be extended by closure to an everywhere defined bounded operator from @xmath294 to @xmath350 , which will be denoted by the same symbol @xmath348 $ ] . further , we suppose that 1 . the sequence is a complex and , moreover , is exact in all but the extreme terms , that is , @xmath351 by @xmath352 and @xmath353 we denote the orthogonal projections on @xmath354 in @xmath276 and @xmath355 in @xmath356 , respectively ; these projections correspond to the only nonzero cohomology spaces of the complex . just as in the bounded case , @xmath353 will be referred to as the dual projection for @xmath352 . consider the laplacians @xmath357 of the complex . here , to unify the notation , we have set @xmath358 , so that actually @xmath359 the operators are densely defined and self - adjoint . indeed , it follows from condition 2 ) that @xmath360 is closed for all @xmath345 , and with regard to one has the orthogonal expansions @xmath361 ( see ( * ? ? ? * theorem iv.5.13 ) ) . one can readily see that for @xmath362 the laplacian @xmath363 has the direct sum decomposition @xmath364 corresponding to . it remains to note that for a closed densely defined operator @xmath365 , the operator @xmath366 is densely defined and self - adjoint ( see kato , ( * ? ? ? * theorem v.3.24 ) ) . we impose the following condition on the laplacians of the complex : 1 . for @xmath362 , the laplacian @xmath363 is an operator with compact resolvent . the laplacians @xmath367 and @xmath368 can not have a compact resolvent except for the trivial case in which the cohomology groups of the complex in dimensions @xmath73 and @xmath277 are finite - dimensional and hence the corresponding projections @xmath352 and @xmath353 define trivial quantizations . that is why we impose this condition only on the laplacians in intermediate dimensions . condition 5 ) is empty for @xmath369 . from now on , we assume that @xmath370 ; the subsequent results remain valid for @xmath369 provided that condition 5 ) is replaced by the following condition : 1 . the restrictions @xmath371 , @xmath372 , are operators with compact resolvent . now we can introduce the following definition . suppose that conditions 1)5 ) are satisfied . then one says that the complex is an _ admissible unbounded resolution _ of length @xmath277 of the projection @xmath352 . just as in the bounded case , let @xmath200 be the differential of degree @xmath286 defined in the graded sum @xmath373 as the direct sum of all operators @xmath287 , and let @xmath288 be the adjoint operator . we define the rolled - up operator @xmath258 by the same formula @xmath374 as in the bounded case , where it is assumed that the operators @xmath5 and @xmath285 act in the respective components @xmath276 and @xmath283 of the direct sum and @xmath375 since the sequence is a complex , one can readily establish the following assertion . the operator @xmath258 is self - adjoint . indeed , since @xmath352 and @xmath353 are self - adjoint and bounded , it suffices to prove the self - adjointness of @xmath376 on the same domain @xmath377 . the range @xmath378 is closed , and so we have the orthogonal decomposition @xmath379 now @xmath380 ( tautologically ) and @xmath381 ( since @xmath382 ) , and so we have the orthogonal decompositions @xmath383 where both embeddings are dense . it follows that @xmath384 now let us prove that @xmath385 ( and then automatically @xmath386 ) . the inclusion @xmath387 is obvious , and we have only to prove the opposite inclusion . let @xmath388 . this means by definition that there exists a constant @xmath233 such that @xmath389 for each @xmath390 . in particular , for arbitrary @xmath391 inequality becomes @xmath392 and now for arbitrary @xmath393 , @xmath394 , we have @xmath395 and @xmath396 it follows that @xmath397 . ( recall that @xmath200 is closed . ) next , for arbitrary @xmath398 inequality becomes @xmath399 and now for arbitrary @xmath400 , @xmath401 , we have @xmath402 it follows that @xmath403 . we conclude that @xmath404 , which completes the proof . now we can state the main theorem . [ main ] suppose that the sequence is an admissible unbounded resolution of length @xmath370 of the projection @xmath352 . then the following assertions hold . 1 . the projections @xmath352 and @xmath353 , as well as the positive spectral projection @xmath267 of the rolled - up operator @xmath258 given by , are admissible . the toeplitz quantizations defined by the projections @xmath267 and @xmath5 are equivalent . we split the proof into two lemmas . for @xmath345 , we define operators @xmath405 by the formula @xmath406 [ lemmaa ] the operators , are everywhere defined and bounded . moreover , they are admissible in the sense of definition , and @xmath407 the sequence @xmath408 is a ( bounded ) admissible resolution of length @xmath277 of the projection @xmath352 with the same dual projection @xmath353 . indeed , this readily follows from . in intermediate dimensions , we have the decomposition corresponding to the direct sum , and accordingly , @xmath409 ( this readily follows from the properties of the functional calculus of self - adjoint operators ; e.g. , see @xcite or ( * ? ? ? since @xmath287 is zero on the first component of the expansion , it follows that we can safely replace @xmath410 by @xmath411 in the formula for @xmath412 , so that we have @xmath413 ( for @xmath298 this is valid automatically , since @xmath414 . ) now we note that the following assertion is true . _ for a closed densely defined operator @xmath365 , the range of the operator @xmath415 coincides with the domain of @xmath365 . _ indeed , @xmath416 next , the domain @xmath417 is a core of @xmath418 as well as of @xmath365 ( * ? ? ? v , theorem 3.24 and lemma 3.38 ) . let us prove that @xmath419 if @xmath420 is a sequence such that @xmath421 then @xmath422 as @xmath423 , and hence the sequence @xmath424 is convergent . thus , @xmath425 reversing the argument , we see that @xmath426 now we see that , for a closed densely defined operator @xmath365 , the operator @xmath427 is bounded and one has @xmath428 moreover , @xmath429 since we have the following chain of equivalent relations : @xmath430 ( here we have used the fact that @xmath431 and hence , by the functional calculus , @xmath432 for any @xmath433 . ) applying this reasoning to the operator @xmath434 , we obtain all assertions of the lemma except for the fact that the commutators @xmath435 $ ] are compact operators for any @xmath346 . first , let us prove the compactness of the commutators for @xmath436 . since the domain @xmath349 is invariant under @xmath30 , we have @xmath437=[a , a_j](1+\delta_j)^{-1/2}+a_j[a,(1+\delta_j)^{-1/2}]\ ] ] ( both sides are defined everywhere on @xmath294 ) . the first term on the right - hand side in is compact as the product of the bounded operator @xmath348 $ ] by the compact operator @xmath438 . ( the compactness of the second operator follows , say , from ( * ? ? ? v , theorem 3.49 ) and the fact that @xmath439 is compact by assumption . ) it remains to prove that the operator @xmath440 $ ] is compact . let us temporarily omit the subscript @xmath293 for brevity . we have ( e.g. , see carey - phillips @xcite ) @xmath441 where the integral converges in operator norm . hence we have @xmath442=a\frac1\pi\int_0^\infty \la^{-1/2}(1+\delta+\la)^{-1}[\delta , a](1+\delta+\la)^{-1}\,d\la\\ = \frac1\pi\int_0^\infty \la^{-1/2}a(1+\delta+\la)^{-1}[a^*,a]a(1+\delta+\la)^{-1}\,d\la\\ + \frac1\pi\int_0^\infty \la^{-1/2}a(1+\delta+\la)^{-1}a^*[a , a](1+\delta+\la)^{-1}\,d\la\\ + \frac1\pi\int_0^\infty \la^{-1/2}a(1+\delta+\la)^{-1}[\wt a^*,a]\wt a(1+\delta+\la)^{-1}\,d\la\\ + \frac1\pi\int_0^\infty \la^{-1/2}a(1+\delta+\la)^{-1}\wt a^*[\wt a , a](1+\delta+\la)^{-1}\,d\la\,.\end{gathered}\ ] ] here we have denoted @xmath443 . the closed operator @xmath200 can be passed through the integrals , since all integrals in question converge in norm , as we shall see shortly . obviously , @xmath444 and so in each of the four integrals on the right - hand side in the integrand is compact and can be estimated in norm as @xmath445 at infinity and @xmath446 at zero . thus , the integrals converge absolutely and give compact operators . it remains to consider the case @xmath298 . the preceding argument would be of no use here , since the operator @xmath367 is not assumed to have a compact resolvent . however , @xmath447 for a closed densely defined operator @xmath365 , where the bar on the right - hand side stands for the closure . using this and the direct sum expansion , we readily obtain @xmath448 and then a computation similar to applies . the proof of lemma [ lemmaa ] is complete . now consider the rolled - up operator @xmath449 for the complex [ lemmab ] one has @xmath450 it suffices to prove that @xmath451 since the desired identity can be obtained from by adding @xmath352 ( and @xmath353 if @xmath277 is odd ) on both sides . next , @xmath452 since @xmath453 . ( here each laplacian acts in the respective direct summand in . ) it follows , with regard to lemma [ lemmaa ] , that @xmath454 it remains to note that , by the functional calculus of self - adjoint operators ( e.g. , see @xcite or ( * ? ? ? * chap . 9 ) ) , it is always true that @xmath455 for a self - adjoint operator @xmath365 , since the function @xmath456 vanishes at zero , is positive for positive @xmath457 , and is negative for negative @xmath457 . this proves the lemma . now the preceding results actually prove the theorem , since we have @xmath458=[p_+(\wt d)]$ ] by lemma [ lemmaa ] and theorem [ 34 ] , and it remains to apply the result of lemma [ lemmab ] . the proof of the theorem is complete . we note possible generalizations of the results obtained . first , there is no need , in fact , to assume that the sequence is exact in all intermediate terms ; it suffices to require that is has finite - dimensional cohomology in these terms . then the corresponding finite rank projections define trivial quantizations and hence the assertion of the main theorem remains valid . second , if we replace the condition that the resolvents of @xmath363 be compact by the condition that they belong to some von neumann schatten classes , then in the end we obtain , for some @xmath9 , an unbounded @xmath9-summable fredholm module determined by the operator @xmath258 . in this appendix , we describe the constructions of @xcite in slightly more detail , restricting ourselves to the case in which the contact manifold @xmath459 considered there is the cosphere bundle of a smooth closed compact manifold @xmath20 . grauert @xcite showed that the cosphere bundle @xmath460 can always be represented as the boundary of a pseudoconvex stein manifold diffeomorphic to the ball bundle @xmath461 ( the grauert tube ) . then the szeg caldern projection @xmath462 can be defined as the projection in @xmath463 onto the subspace of boundary values of holomorphic functions in the grauert tube . the manifold @xmath464 bears the natural kohn rossi complex ( the boundary @xmath465-complex ; e.g. , see @xcite ) @xmath466 where @xmath277 is the dimension of the manifold @xmath20 . here @xmath467 is the bundle of @xmath468-forms associated with the natural complex structure on the contact hyperplane distribution . the complex is acyclic in all dimensions except for @xmath73 and @xmath277 . moreover , its cohomology space in the zero term exactly coincides with the range of the szeg - caldern projection @xmath462 . note that if we `` roll up '' the complex in a standard way ( i.e. , pass to the operator @xmath469 in the direct sum of all spaces of the complex . ] ) , then the resulting self - adjoint operator is closely related to the dirac operator on @xmath459 . this relationship can be conveniently represented in the block @xmath470 expansion of this operator corresponding to the expansion of the space @xmath471 into the sum of even and odd components : @xmath472 where @xmath473 is the covariant derivative along a direction transversal to the contact distribution ( more precisely , along the reeb vector field ) . this suggests that the kohn rossi complex can be used to prove the equivalence of the quantizations given by the szeg caldern projection @xmath462 and the positive spectral projection of the dirac operator . the construction carried out in @xcite consists of two steps . \1 ) let @xmath474 be the orthogonal projection on the cokernel of the last operator @xmath465 in the kohn rossi complex . the operator @xmath475 ( where the projections @xmath462 and @xmath476 act in the first and the last component , respectively ) is self - adjoint and invertible . in accordance with the general results proved in sec . 2 , the quantization with the help of the positive spectral projection of @xmath36 is equivalent to the quantization with the help of the szeg caldern projection . ( in @xcite , only the equalities for the indices of operator obtained from the same symbol by these two quantizations were proved . ) \2 ) now it remains to deform the operator @xmath477 continuously in the class of fredholm operators to the dirac operator @xmath28 . this procedure is described in sufficient detail in the cited papers , and we only briefly mention the main points . first , the operator @xmath36 _ is not pseudodifferential _ ( owing to the presence of the projections @xmath462 and @xmath476 , which are not pseudodifferential operators ) . however , on contact manifolds there is a natural `` anisotropic '' version of pseudodifferential calculus , known as the heisenberg calculus . more precisely , in this calculus the differentiations along contact directions have the weight @xmath411 , and the differentiation along the transversal ( reeb ) direction has the weight @xmath478 . since the contact distribution is nonintegrable , the product in the algebra of principal symbols proves to be noncommutative . the principal symbol over each point @xmath479 is a pair of equivariant operators in the standard irreducible representation of the heisenberg group in @xmath480 . in particular , the projections @xmath462 and @xmath476 turn out to be heisenberg pseudodifferential operators whose symbols are the projections on the vacuum vector of the harmonic oscillator in one component of the pair and are equal to zero in the other component . the paper @xcite contains also an `` extended '' version of the heisenberg calculus , which includes heisenberg pseudodifferential operators as well as usual pseudodifferential operators . the homotopy of @xmath36 to the dirac operator @xmath28 is carried out in the framework of the extended calculus . agranovich and m.i . vishik , _ elliptic problems with parameter and parabolic problems of general type _ , uspekhi mat . nauk * 19 * ( 1964 ) , no . 3 , 53161 , english transl . in : russ . * 19 * ( 1964 ) , no . 3 , 53157 . j. kaminker , _ pseudodifferential operators and differentiable structures _ , operator algebras and @xmath0-theory ( san francisco , calif . , 1981 ) ( providence , r.i . ) , contemp . 10 , ams , 1982 , pp . 99128 . j. kaminker , _ operator algebraic invariants for elliptic operators _ , operator theory : operator algebras and applications , part 1 ( durham , nh , 1988 ) ( providence , ri ) , proc . pure math . 51 , ams , 1990 , pp . 307314 . b. a. plamenevsky and g. rozenblum , _ on the index of pseudodifferential operators with isolated singularities in symbols _ , algebra i analiz * 2 * ( 1990 ) , no . 5 , 165188 , english transl . in : leningrad math . j. , * 2 * ( 1991 , no . 5 , 10851110 . b. a. plamenevskij and v. n. senichkin , _ on a class of pseudodifferential operators on @xmath482 and on stratified manifolds _ , math . sb . * 191 * ( 2000 ) , no . 5 , 109142 , english transl . in : sb * 191 * ( 2000 ) , no . 5 - 6 , 725757 . schulze , b.yu . sternin , and v.e . shatalov , _ differential equations on singular manifolds . semiclassical theory and operator algebras _ , mathematical topics , vol . 15 , wiley vch verlag , berlin new york , 1998 . savin , b .- w . schulze , and b.yu . sternin , _ on invariant index formulas for spectral boundary value problems _ , differentsialnye uravnenija * 35 * ( 1999 ) , no . 5 , 705714 , english transl . in : differential equations * 35 * ( 1999 ) , no . 5 , 709718

we used _ hst _ to obtain surface brightness fluctuation ( sbf ) observations of four nearby brightest cluster galaxies ( bcg ) to calibrate the bcg hubble diagram of lauer & postman ( 1992 ) . this bcg hubble diagram contains 114 galaxies covering the full celestial sphere and is volume limited to 15,000 km s providing excellent sampling of the far - field hubble flow . the sbf zero point is based on the cepheid calibration of the ground method ( tonry et al . 1997 ) as extended to the wfpc2 f814w filter by ajhar et al . ( 1997 ) . the bcg globular cluster luminosity functions give distances essentially identical to the sbf results . using the velocities and sbf distances of the four bcg alone gives in the cmb frame , valid on,500 km s scales . use of bcg as photometric redshift estimators allows the bcg hubble diagram to be calibrated independently of recession velocities , yielding a far - field with an effective depth of,000 km s . the error in this case is dominated by the photometric cosmic scatter in using bcg as distance estimators . the concordance of the present results with other recent determinations , and a review of theoretical treatments on perturbations in the near - field hubble flow , argue that going to the far - field removes an important source of uncertainty , but that there is not a large systematic error to be corrected for to begin with . further improvements in depend more on understanding nearby calibrators than on improved sampling of the distant flow . 8bar_814 8bar_i_814 .+ .1em

astro-ph9708252

a key part of measuring the hubble constant , @xmath9 is to look out far enough so that the bulk velocities of galaxies are trivial compared to the hubble flow itself . due to the virgo cluster infall pattern , observation of the unbiased hubble flow can only be contemplated at distances in excess of @xmath10 km s@xmath4 . furthermore , bulk flows on even larger scales , such as those associated with the great attractor , may bias measurement of @xmath11 turner , cen , & ostriker ( 1992 ) and shi , widrow , & dursi ( 1996 ) , for example , show that under standard theories of structure formation , measurements of @xmath6 can depart significantly from its true `` global '' value due to the inhomogeneous distribution of matter in the universe , unless care is taken to sample deeply with large angular coverage . indeed , a common concern with many recent @xmath6 determinations is that they are not truly sampling the distant hubble flow ( bartlett et al . 1995 ) . characterizing the far - field requires observing large numbers of objects at large distances so that the hubble diagrams are insensitive to random peculiar velocities or bulk flows . hubble diagrams at present are largely based on the tully - fisher or @xmath12 relationships , the luminosities of supernovae ( sn ia or sn ii ) , and brightest cluster galaxies ( bcg ) . tully - fisher distances are available out to @xmath39,000 km s@xmath0 and have been recently used to measure a far - field @xmath6 ( giovanelli et al , 1997 ) , while @xmath12 have full - sky coverage out to only @xmath36,000 km s@xmath13 only a few sn ii have been observed in sufficient detail at large distances ( schmidt et al . 1994 ) , but the snia hubble diagram is becoming richer with time and provides some sampling of the hubble flow out to @xmath330,000 km s@xmath4 ( riess , press , & kirshner 1996 ; hamuy et al . 1996 ) . at present , however , calibration of the snia distance scale remains controversial ( see sandage et al . 1996 ) , and the sn ia diagrams remain relatively sparse at large distances . in this work we focus on calibrating the bcg hubble diagram , which is based on a recent characterization of bcg as relative distance estimators ( postman & lauer 1995 ) . in the classic work of sandage ( 1972 ) and sandage & hardy ( 1973 ) , bcg were used to show that the hubble flow was linear over a large range in redshift . lauer & postman ( 1994 ) observed bcg to define a frame for measuring the peculiar velocity of the local group , but as this work was in progress they realized that they could test for @xmath6 variations with distance with greater precision than was previously available in response to the concerns of turner , cen , & ostriker ( 1992 ) . lauer & postman ( 1992 ) presented a hubble diagram based on the 114 bcg that defined the volume - limited full - sky sample of abell clusters within 15,000 km s@xmath0 which is shown again here in figure [ bcg_hub ] . in brief , the absolute magnitudes of bcg , @xmath14 measured in apertures of fixed metric size , @xmath15 can be predicted from @xmath16 ( hoessel 1980 ) . figure [ bcg_hub ] shows the metric luminosities as apparent fluxes , corrected by the @xmath17 relationship to a standard value of @xmath18 the bcg hubble diagram slope is @xmath19 of the expected value , consistent with a uniform hubble flow over @xmath20 lauer & postman ( 1992 ) limit any variation in the _ apparent _ or local @xmath6 ( the hubble constant measured over a limited depth ) as compared to @xmath6 measured globally over the entire volume , to @xmath21 the snia hubble diagram also shows no evidence for @xmath6 variations with distance ; riess , press , & kirshner ( 1996 ) show its slope ( relative to euclidean ) to be @xmath22 the full - sky coverage of the abell cluster sample is crucial , as any dipole pattern caused by large bulk flows ( such at that advanced by lauer & postman 1994 ) will integrate out of the hubble diagram to first order . the linearity of the bcg hubble diagram shows that an excellent estimate of the far - field @xmath6 can be obtained once the zero point of the diagram is calibrated . we note that bcg presently provide the only volume - limited sample that explores the hubble flow at these distances . a hubble constant can be obtained from the bcg hubble diagram once an absolute distance is known to a subset of the galaxies . in essence , one transfers the full sample to a common distance , and finds the average absolute luminosity of the bcg on the assumption that the calibrating set is typical . random velocities and bulk flows of the bcg contribute to the `` cosmic scatter '' in their luminosity distribution , but cause no systematic offset ( with the caveats discussed in @xmath23[far_enough ] ) . we contrast this approach to others that use the apparent distance ratio between the virgo and coma clusters , or any other near and far aggregate of galaxies , to reach the far - field . instead , we are using the bcg as complete probes of the hubble flow over a large volume . we chose to calibrate the bcg hubble diagram with surface brightness fluctuation ( sbf ) distance estimates to four of the nearest bcg . the sbf method ( tonry & schneider 1988 ) uses the ratio of the second to first moments of the stellar luminosity function within early - type stellar systems as a distance estimator . the ratio of moments corresponds to an apparent magnitude , , that in the near - ir corresponds to the brightness of a typical red giant star . when the images are deep enough such that a star of apparent luminosity contributes more than a single photon to an observation , the random spatial point - to - point surface brightness fluctuations in a galaxy image are dominated by the finite number of stars it comprises , rather than photon shot noise . a power spectrum of the sbf pattern provides . use of the sbf method on galaxies with distances known from other methods ( see jacoby et al . 1992 for additional details ) provides the zero point , allowing absolute distances to be computed from . the most recent calibration of the sbf method is presented by tonry et al . major components of this work are : 1 ) understanding how varies with stellar population , 2 ) determining the zero point of the method , and 3 ) establishing the universality of the calibration . tonry et al . observe in the @xmath1 band , which minimizes variations in with stellar population _ ab initio . _ they also show that variations in are fully characterized by the ( @xmath24 ) colors of the stellar systems . based on 149 nearby galaxies they find @xmath25.\ ] ] this relationship has scatter of only 0.05 mag and agrees well with the theoretical calculations of worthey ( 1993a , 1993b ) both in slope _ and _ zero point . tammann ( 1992 ) was concerned that an earlier sbf calibration based on @xmath26 was incomplete and that additionally depended on the galaxies @xmath27 indices . in response , tonry et al . use their extensive sample to show that there is no correlation between the residuals of equation ( [ eqmibar ] ) and @xmath28 the zero point of equation ( [ eqmibar ] ) is based on cepheid distances to seven spiral galaxies with bulge sbf observations . tonry et al . present numerous comparisons of sbf to pnlf , tully - fisher , @xmath29 , snia , and sn ii distances , finding no evidence for any systematic offset between sbf bulge and elliptical galaxy measurements , nor any other systematic effect that challenges the calibration . although the nearest bcg are too far away for the sbf method to work from the ground , the high spatial resolution of _ hst _ allows sbf to be used beyond the 15,000 km s@xmath4 depth of the lauer & postman ( 1994 ) sample . an important caveat is that there is no direct match to the @xmath1 filter among the wfpc2 filter set . the f814w filter is a close analogue to @xmath1 ( see holtzman et al . 1995a ) , but requires additional calibration to tie it to the tonry et al . ( 1997 ) zero point . ajhar et al . ( 1997 ) accomplished this task in preparation for the present work , by comparing _ f814w sbf observations to the @xmath1 results for 16 galaxies in the tonry et al . sample . for the wfpc2 ccds and f814w filter , ajhar et al . find @xmath30,\ ] ] with scatter similar to that about equation ( [ eqmibar ] ) . a key difference between equation ( [ eqmibar ] ) and ( [ eqm8bar ] ) is the steeper relationship between 8bar and @xmath31 which ajhar et al . show is consistent with the differences between the f814w and @xmath1 filters . calibration of _ hst _ for sbf work is thus crucial for the present work . we selected the bcg in four abell clusters , a262 , a3560 , a3565 , and a3742 for observation . these are among the nearest of the lauer & postman ( 1994 ) sample so as to minimize _ hst _ exposure time . we also wanted to minimize the effects of bulk flows on placement of the calibrating bcg within the hubble diagram , so we selected bcg positioned such that their mean photometric offset about the hubble line in figure [ bcg_hub ] is largely insensitive to whether the hubble diagram is constructed from velocities referenced to the cosmic microwave background ( cmb ) , local group ( lg ) , or the abell cluster ( ac ) frame solution of lauer & postman ( 1994 ) . figure [ bcg_hub_frame ] shows how the positions of the bcg in the hubble diagram change with changing velocity system . complete frame invariance of the results is difficult to achieve with only four bcg , however ; we discuss this issue later where our results are affected by it . the bcg properties are given in table [ tabsamp ] . photometry , details of bcg identification , and so on are discussed by postman & lauer ( 1995 ) . velocities are given in the cmb , lg , and ac frames ( see lauer & postman 1994 ) . the velocities are a weighted average of all galaxies selected to be within the given abell cluster ; the estimated velocity error is 184 km s@xmath4 ( see postman & lauer 1995 ) . extinctions are from burstein & heiles ( 1984 ) . rrrcccccr 262 & 01 52 46.3 & + 36 09 05 & 4650 & 5130 & 5310 & 0.060 & 96/02/11 & 16,4003560 & 13 31 53.3 & @xmath3233 14 04 & 4020 & 3510 & 3360 & 0.038 & 96/01/16 & 9,2003565 & 13 36 39.1 & @xmath3233 57 57 & 4110 & 3630 & 3450 & 0.030 & 96/01/19 & 11,6003742 & 21 07 52.3 & @xmath3247 10 43 & 4680 & 4800 & 4740 & 0.018 & 96/04/22 & 16,500 [ tabsamp ] images were obtained in wfpc2 using the f814w filter . the galaxies were centered in the high - resolution pc1 chip . while suitable data were obtained in the flanking wfc ccds , we chose to analyze the pc data only , given its superior resolution , and the greater brightness of the central portions of the bcg with respect to the sky . at the bcg distances , the apparent luminosity of an f814w sbf `` star , '' 8bar , is extremely faint , thus long exposures are required ; the total exposures are given in table [ tabsamp ] . although our ideal criterion is to obtain at least five photons per 8bar star , our data contained only 2.33.8 photons per 8bar . the reasons for the shortfall were 1 ) _ hst _ was 5% less sensitive through f814w than prelaunch numbers suggested , 2 ) the galaxies were about 5% more distant than we had guessed from their redshifts , and 3 ) 8bar was significantly fainter ( 0.65 mag in the case of a262 ) because the galaxies were redder than anticipated and 8bar was more sensitive to color than our assumptions . nevertheless , all four galaxies yielded a strong sbf signal that could be accurately determined . because compact artifacts can strongly affect the sbf power spectrum , we built the total exposures from sets of `` dithered '' half - orbit images to eliminate hot - pixels , ccd defects , as well as cosmic - ray hits . each individual exposure was typically 1200s long , with the actual exposure time set to maximize the total exposure obtained with two roughly equal exposures per orbit . the dither pattern consisted of moving the telescope between exposures in a skewed - square - spiral pattern , designed to achieve integral - pixel shifts in the wfc ccds . the pattern optimized removal of fixed - pattern ccd defects , while preserving the exact shape of any sbf pattern from exposure to exposure . while one might argue for an approach that would instead optimize information recovered with sub - pixel stepping , we did not have the exposure time available to obtain an equal number of the required @xmath33 0.5-pixel steps and were concerned with the effects on the sbf pattern from any _ ad hoc _ interpolation scheme that might be required to assemble the completed image from a random set of offsets . ironically , we did not use the wfc images in the present analysis , given the excellent quality and superior resolution of the pc1 images ; however , this was a decision made after the data were in hand . in the dither pattern , each exposure would be shifted from the previous one , by @xmath34 wfc pixels ( @xmath35 ) in either the row or column direction , spiraling around the original pointing , with the exact shift adjusted by @xmath36 wfc pixel to avoid having any object land on the same row or column that it may have fallen on in a previous exposure . this latter criterion also meant that the exposure was simultaneously stepped @xmath36 pixel in a direction perpendicular to the major step , resulting in the spiral being skewed from perfect alignment with wfc row and column axes . the detailed shifts in all cases reflected the slight misalignment of the wfc ccds with respect to perfect quadrature with the _ hst _ sky axes ( see holtzman et al . 1995b ) . integral pixel offsets for the wfc produce non - integral steps in pc1 , but for f814w , the pc1 images are nearly nyquist - sampled , so shifting the images to a common center may be done with little error . as it is , however , we chose to stack the pc1 images with centering only to the nearest pixel to avoid the complexity of patching in the defects prior to interpolation . while this produces a slight blurring at the one - pixel scale , this has little effect on measurement of the sbf signal , because the final composite image remains photon shot - noise limited at the highest - spatial frequencies . this can be simply understood by considering the enclosed energy curve of the psf . holtzman et al . ( 1995b ) show that only 32% of the light within the f814w psf falls within its central core , corresponding to only a single photon for the typical exposure time in the present work . the final composite pc1 images were assembled with an algorithm that looks at the statistical properties of the data set at any pixel location , and rejects extreme values as might be caused by cosmic ray hits or hot pixels . the bcg images are shown in figures [ bcgpic1 ] to [ bcgpic4 ] , with the galaxies themselves largely subtracted to emphasize fine detail . globular clusters are readily visible into the centers of all images . this is the strength of using _ hst _ for sbf work globulars , background galaxies , and dust clouds are easily recognized and excluded when measuring the sbf power spectrum . dust clouds are also visible in three galaxies , a262 , a3560 , and a3565 ; indeed , their centers are completely or nearly obscured by dust . although sbf are measured from the pc1 images , we do need the wfc images to measure the sky levels , which are required as part of the analysis . sky levels were measured from a @xmath37 patch extracted from the far corner of wf3 . this is the corner of wf3 diagonally opposite the wfpc2 vertex , and is thus the portion of the wfpc2 field most distant from the bcg centers ; the typical displacement of the sky patch from the galaxy centers is @xmath38 the galaxies still contribute light to the sky measurements at this modest distance ; however , their contribution to the sky is readily estimated and corrected for using ground - based surface photometry extending to much larger radii . the sky values given in table [ tabsbf ] have been corrected for galaxy light contributions . rccccccccc 262 & @xmath39&2.3&21.40 & 1.401&1.317 & @xmath40 & 33.77&0.19&25.85 3560 & @xmath41&3.4&21.61 & 1.289&1.234 & @xmath42 & 33.35&0.14&25.12 3565 & @xmath43&3.8&21.70 & 1.286&1.239 & @xmath44 & 33.47&0.13&25.72 3742 & @xmath45&3.0&21.73 & 1.305&1.270 & @xmath46 & 33.88&0.12&25.77 [ tabsbf ] as equation ( [ eqm8bar ] ) shows , the 8bar value for a given bcg depends strongly on its @xmath26 color . we measure @xmath26 from ground - based photometry over an annulus between @xmath47 and @xmath48 in radius from the galaxy centers to match the area of the pc1 field used for the sbf measurements . the photometry for the three southern bcg was obtained under excellent conditions at the ctio 1.5 m telescope . for a262 , the @xmath1 image was obtained at the kpno 4 m , and john blakeslee kindly obtained the @xmath49 image at the mdm 2.4 m . for the small redshifts of the bcg , the k - corrections are @xmath50 and @xmath51 galactic absorption corrections are @xmath52 the observed @xmath31 and reduced @xmath53 colors are both given in table [ tabsbf ] . in passing , we note that ground @xmath54 and wfpc2 @xmath55 are essentially identical ( holtzman et al . while we did not obtain wfpc2 f555w data , we did compare the @xmath56 photometry to the ground @xmath1 data . for @xmath57 holtzman et al . ( 1995a ) find @xmath58 unfortunately , even with this transformation , the wfpc2 @xmath59 fluxes are still brighter than the ground values by 0.04 mag with a spread of 0.05 mag . ajhar et al . ( 1997 ) found excellent agreement between wfpc2 @xmath59 and ground @xmath1 transformed to @xmath59 for their sample , so the present mismatch is disappointing . since to first order , we might expect any deviations in ground @xmath49 to correlate with those in @xmath60 we chose to base @xmath53 entirely on ground photometry rather than ground @xmath49 and wfpc2 f814w . measurement of the fluctuation signal was straightforward , following the steps detailed in tonry et al . ( 1997 ) and ajhar et al . once the pc images were cleaned of cosmic rays and stacked , we subtracted sky ; fitted elliptical profiles to the galaxy light ; generated and subtracted a synthetic galaxy ; ran dophot ( schechter et al . 1993 ) to detect stars , globular clusters , and background galaxies ; determined a luminosity function of these objects and created a mask for removing objects brighter than a completeness cutoff ; and then performed the fluctuation analysis . the basic step in measuring fluctuations is to model the object s power spectrum as @xmath61 where @xmath62 is a constant giving the background level , and @xmath63 is another constant used to scale the power spectrum of the psf , @xmath64 ( properly speaking , @xmath65 is an expectation power spectrum that incorporates the galaxy profile and the mask that has been applied to the data . ) there are several sources of uncertainty in deriving the fluctuation power . the first is the normalization and match of the psf to the data . the second arises in fitting the power spectrum : the lowest wavenumbers are always corrupted , but there is no fluctuation signal at high wavenumbers , so the choice of precisely which wavenumbers to fit introduces uncertainty . finally , the raw @xmath63 variance is corrected for the variance from residual , undetected point sources , and this correction also carries an error . we used the same psf as ajhar et al . ( 1997 ) , which was constructed from several f814w star images obtained in 1995 . the wings of the psf were provided by archive exposures of stars centered in pc1 to provide routine monitoring of the f814w filter zero point calibration . the core of the psf was constructed from four pc1 images of the quadrupole gravitational lens 2237 + 0305 , which were dithered in a @xmath33 pattern of 0.5 pc pixel steps . this data set allowed the core to be recovered with `` perfect '' centering on a pc pixel . the composite psf has a total exposure of @xmath66 photons . it is an excellent match to the observed power spectra of the galaxies with the highest signal to noise , a3560 and a3565 , over the wavenumbers where sbf dominates , so we deemed it to be suitable for all our observations . we performed experiments with synthetic psf images ( created by software developed by jah ) with calibrated amounts of defocus and miscentering . we found that the results are completely insensitive to where the psf was centered , but that @xmath63 became larger as the psf became defocussed , as would be expected from its more compact power spectrum . using a psf from the extreme secondary displacement caused by `` breathing '' of the ota quoted by stsci , we found that brightened by about 0.3 mag . a synthetic psf with perfect focus gave nearly the same results as the observed psf , but was still about 0.1 mag brighter , revealing a bit _ less _ high wavenumber power than the observed psf . ajhar et al . also demonstrate that the observed psf will give very consistent results applied to a variety of observations , and indeed , we are working differentially with respect to those results anyway . we thus conclude that the observed psf is appropriate to use for the fluctuation analysis , with mismatch and normalization error amounting to about 0.05 mag . an observed power spectrum always has excess power at wavenumbers below 10 or so ( wavelengths longer than @xmath67 pixels ) , arising from poor flattening , poor galaxy subtraction , dust , etc . with the exception of a262 we found very good agreement with the psf beyond these contaminated wavenumbers . we normally use the rms variation in @xmath63 as a function of where we begin the power spectrum fit as an additional error component . in the case of a262 there is a range of @xmath63 values that are allowed by the power spectrum ; this range is reflected in the larger error budget for this galaxy . psf fits to the power spectra of the bcg are shown in figure [ sbfpow ] . the contribution from the residual point sources was small always less than a 0.20 mag correction to @xmath63 , and more typically 0.10 mag . since we think we know this variance contribution quite well , at least to 25% , the contributed error is small . as a test of this , we routinely analyze different annuli independently , and except for a262 we used 4 annuli at 1 , 2 , 4 , and 8 mean radius ( the central annulus was obscured by dust for a262 ) . if we have an error in the residual point source correction ( or an unknown source of variance which does not scale with galaxy brightness ) it will show up as a radial gradient in . a3560 had a gradient of about 0.1 mag in each of the outer annuli , but the other three had no gradient at the 0.05 mag level , giving us confidence that we have modeled the residual variance well and we have no unforeseen source of variance . the gradient in a3560 may very well be real ; there is a 0.04 mag color gradient in @xmath68 from the center to @xmath69 in radius , which would produce a gradient in consistent with what we see . the estimates and errors for are derived from the averages of the values determined for each annulus , and the formal uncertainty in the @xmath63 fitting procedure added in quadrature to 0.05 magnitude for the psf uncertainty . with the exception of a3560 , where we think we see a real gradient in , the scatter between the different annuli is consistent with the error estimates . we list the apparent sbf 8bar fluxes in table [ tabsbf ] uncorrected for extinction and prior to k - correction . the sbf signal strength in electrons is also given . a262 has the weakest sbf signal , which is reflected in its larger error bars . as discussed above , measuring requires characterization of the galaxy s globular cluster luminosity function ( gclf ) to estimate the residual variance contributed by the undetected ( faint ) portion of the gclf as well as by the undetected faint galaxies . figure [ fgclfs ] shows the luminosity functions of the objects found in the images of our sample along with the gclf , background galaxy , and combined luminosity function fits . the fitting procedure naturally produces an estimate of the gclf turnover magnitude in the @xmath70 band for an assumed gaussian @xmath71 width of 1.4 mag . the @xmath72 gclf turnover magnitudes are listed in table [ tabsbf ] . the estimated turnovers are all near the @xmath73 completeness limit and are uncertain by 0.20.3 mag . based on the sbf distance moduli , the mean absolute magnitude of the four bcg gclf turnovers in the @xmath70 band @xmath74 the error is consistent with the measurement errors and an estimated intrinsic scatter in the gclf technique of @xmath75 mag . few deep @xmath70-band gclfs of giant elliptical galaxies have been published for comparison ; however , @xmath76 is consistent with the m87 measurements of whitmore _ ( 1995 ) , who found @xmath77 . using the sbf distance modulus to the virgo cluster of @xmath78 ( tonry _ 1997 ) yields @xmath79 . for additional comparison , a crude estimate of @xmath80 can be based on the following assumptions : 1 ) combining the distance modulus to virgo with the average observed apparent magnitude for the gclf turnover in virgo in the @xmath49 band of @xmath81 ( blakeslee & tonry 1996 ) yields @xmath82 . 2 ) taking the mean color @xmath83 , based on the average of the mean found in coma s ic 4051 of @xmath84 ( baum _ et al . _ 1997 ) and that found in m87 of @xmath85 ( whitmore _ et al . _ 1995 ) and applying an uncertainty of 0.1 mag , yields an estimate of @xmath86 , also in good agreement with the bcg value . the present data set permits two separate approaches to measuring @xmath11 the first and most obvious approach is simply to form hubble ratios for the four bcg and average them in an optimal way . the bcg all have velocities in excess of 4,000 km s@xmath4 ( in the cmb frame ) , and may be far enough away so that their average hubble ratio might be close to the true value of @xmath11 at the same time , this approach does not make use of the bcg hubble diagram , nor does it transfer the results to the 15,000 km s@xmath4 far - field . the bcg instead are simply treated as test particles , without reference to their photometric properties ( although we do use cluster averages for the velocities ) . the second approach is to use the bcg as distance estimators to avoid any use of the velocities of the four sbf bcg themselves , in an effort to skip over the near velocity field this is the approach promised by the title of our paper . the frame independence of both the bcg hubble diagram and @xmath17 relationship argues that we can successfully reduce the bcg to a common distance . the sbf distances then permit calibration of the bcg as absolute rather than relative distance estimators . in either approach , the mathematical formalism used to derive a hubble ratio is the same . for each sbf observation , we compute a distance measurement in mpc , @xmath87 , from the expression @xmath88 where 8bar and 8bar are corrected for extinction and k - dimming . the hubble ratio is then just @xmath89 where @xmath90 is either the observed velocity of the bcg in the appropriate reference frame ( for the first approach ) or the estimated velocity of the bcg using the prescription in postman & lauer ( 1995 ) and summarized in 3.2 ( for the second approach ) . the bcg @xmath53 colors and the f814w sbf calibration of equation [ eqm8bar ] allow us to predict the absolute fluctuation magnitude 8bar . using a 8bar k - correction of @xmath91 ( tonry et al . 1997 ) and our measured values for the apparent 8bar , we derive the distance moduli and errors listed in table [ tabsbf ] . converting these to distances in mpc ( equation [ eqsbfdm ] ) and using the velocities in the cmb , lg , and ac frames , we compute hubble ratios in table [ tabho ] . the errors listed here include the estimated velocity error of 184 km s@xmath4 plus a nominal 100 km s@xmath4 allowance for peculiar velocity with respect to the local velocity field ( added in quadrature ) , although this term could plausibly be as large as 5% . there is no allowance for bulk flow since we are explicitly trying to remove this by examining different reference frames . the average hubble ratios are the weighted logarithmic averages , and the errors are those expected given the individual distance errors . while this sample is too small to solve for a preferred reference frame at 4,000 km s@xmath4 , we do see the insensitivity of the average hubble ratio among the three frames because of the sampling over the sky . it is also apparent that @xmath92 prefers the cmb frame to the local group or abell cluster frame , but again the numbers are small . because these sbf distances appear to be extremely consistent and accurate , a larger sample observed by _ hst _ throughout the sky could give a precise measure of peculiar velocities ; we discuss this issue further in @xmath93 our preferred value for @xmath6 is derived from the cmb frame since it has the lowest value of @xmath92 : @xmath94 km s@xmath4 mpc@xmath95 the error is discussed in detail below . the premise behind seeking the far - field is that the near - field may be strongly affected by random peculiar velocities or bulk flows . one measures distances in the near - field , but avoids using the corresponding near - field velocities by forming distance ratios between near and far objects , and then forming a hubble ratio based purely on the far velocities . the classic example of this approach is using the distance to the nearby virgo cluster , but adopting the velocity of the much more distant coma cluster by finding the ratio of the virgo to coma distance from some form of relative distance estimator , such as the @xmath12 method . in the present case , the transference to the far field is somewhat less obvious as we are not making an explicit comparison of the four sbf bcg to , say , the most distant bcg in the lauer & postman ( 1994 ) sample at 15,000 km s@xmath13 instead , we will implicitly compare the sbf bcg to the entire lauer & postman sample on the presumption that all bcg can be transferred to a common distance based on their observed velocities and a simple linear hubble flow model . one can add a bulk flow to this model , but in the end , this makes little difference , as we discuss below . the bcg @xmath17 distance estimator presented in lauer & postman ( 1992 , 1994 ) and postman & lauer ( 1995 ) , indeed is based on comparing the physical properties of bcg after adopting a linear hubble flow and an _ ad hoc _ @xmath11 the @xmath17 relation works by allowing @xmath96 to be predicted based on @xmath97 as measured at the metric radius @xmath98 the scatter in @xmath96 is 0.24 mag , which translates into a typical distance error of @xmath99 this is larger than the error expected for a pure inverse - square distance estimator , as an error in the metric radius also implies an error in the apparent luminosity within the metric radius ( gunn & oke 1975 ) . the @xmath17 relation is equivalent to assuming that bcg all have the same average enclosed surface brightness as a function of physical radius for a given value of @xmath100 postman & lauer ( 1995 ) showed that they could estimate bcg redshifts ( that is a relative distance expressed as a velocity ) to 17% accuracy from a surface brightness curve of growth by finding the angular aperture at which the enclosed surface brightness and @xmath97 were consistent with the @xmath17 relationship , then using the aperture size as a metric distance estimator . figure [ zest ] shows the plot of estimated versus observed redshift presented by postman & lauer ( 1995 ) . the estimated redshift results from calculating the velocity required to bring a given bcg onto the ridge line of the @xmath17 relation . the scatter in @xmath101 thus reflects the scatter of the bcg about the ridge line . lauer & postman ( 1994 ) showed that this scatter is strongly dominated by random photometric differences between the bcg rather than random peculiar velocities or velocity errors photometric scatter is constant with redshift , as are the residuals about the @xmath17 relationship , while the decreasing relative importance of random velocities with increasing hubble velocity would cause the scatter to decline with redshift . the estimated redshifts for the four sbf bcg ( given in table [ tabho ] ) imply the velocity scaling required to bring their photometric properties onto the @xmath17 ridge line ; hubble ratios can then be calculated , based on the sbf distances . the implied far - field hubble ratios are given in table [ tabho ] . again , the average value is best estimated with a error - weighted logarithmic average of the ratios . the far - field hubble constant implied is @xmath102 the error will be discussed in detail in the next section , but in this case it is dominated by the bcg photometric scatter about the @xmath17 relationship . in this context , it is worth noting that a262 is among the most deviant bcg with respect to its position in the @xmath17 relationship , a conclusion echoed in the highly deviant hubble ratio that it yields using its photometric redshift as the velocity ( see the last column in table [ tabho ] ) . the @xmath92 value for the far - field @xmath6 is also large . deleting a262 , yields a lower far - field @xmath103 this value is consistent with the former value , but does suggest that a more complete sampling of the nearby bcg may be an attractive way to improve the accuracy of the bcg far - field @xmath11 as the far - field @xmath6 does not explicitly depend on the velocities of the individual sbf bcg , it is explicitly independent of the velocity frame . there may be _ implicit _ dependences on frame , since different choices of hubble velocity will affect the particular placement of any given bcg within the @xmath17 relationship ; however , such effects are tiny . for example , at @xmath104 the ridge line value of @xmath96 varies by only 1% among the three frames . more to the point , the @xmath101 varies by only 2% for a262 , and less than 1% for the other bcg over the various frames . this result is not surprising , as any dipole pattern caused by a large bulk flow superimposed on the bcg recession velocities will integrate out of the @xmath17 relationship because it is defined from the full sky . a more relevant question is whether the nearby sbf bcg set may be offset from the more distant bcg by some global perturbation of the hubble flow that is limited to within , say , 5,000 km s@xmath4 or some other small fraction of the 15,000 km s@xmath4 limit of the abell cluster sample . this was the sort of problem posed by turner , cen , & ostriker ( 1992 ) and addressed by the lauer & postman ( 1992 ) bcg hubble diagram . the linearity of the bcg hubble diagram argued that @xmath105 between 3,000 and 15,000 km s@xmath13 further , lauer & postman ( 1992 ) showed that beyond 9,000 km s@xmath0 which includes the bulk of the bcg sample , @xmath106 on radial shells of 3,000 km s@xmath13 in other words , there is no evidence that scatter in the @xmath17 relationship or its ridge line is affected by any global variations in @xmath6 within 15,000 km s@xmath13 the @xmath17 relationship is defined by the full bcg sample , unweighted by distance ; the effective scale of the far - field bcg hubble constant is just the average bcg recession velocity , or @xmath311,000 km s@xmath13 both the `` sbf - alone '' and far - field @xmath6 values are affected by several random errors . the strong dependence of 8bar on @xmath26 puts a premium on accurate colors . we do not have multiple ground observations of the bcg , but previous experience plus the quality of the standard star solutions argues that the error in @xmath26 is likely to be close to 0.012 mag . we also include a 10% uncertainty in @xmath107 and @xmath108% errors in the @xmath26 and 8bar k - corrections . the final net color error gives a typical @xmath109 mag error for estimating 8bar from equation [ eqm8bar ] ; this is added in quadrature to the measurement error in 8bar given in table [ tabsbf ] to give the total random distance error . calculation of the hubble ratios also includes a 184 km s@xmath4 velocity error and 100 km s@xmath4 peculiar velocity term for the sbf - alone measures , and a 17% redshift error for the hubble diagram values , as noted above . the errors in the individual hubble ratios given in table [ tabho ] reflect the random error contributions only . important systematic errors include uncertainties in the f814w sbf calibration , the absolute zero point of the @xmath1 relationship on which it is based , and errors in the _ hst _ psf . scatter about the _ hst _ 8bar versus @xmath26 relationship argues that the uncertainty in predicting 8bar , given perfect @xmath31 is @xmath110 mag ( ajhar et al . 1997 ) . tonry et al . ( 1997 ) give @xmath111 mag as the error in the @xmath1 sbf zero point , which itself is a composite of the statistical errors in the ground @xmath1 versus @xmath26 relationship plus uncertainties in the cepheid calibration of the relationship . the systematic error associated with the psf is 0.05 mag , as discussed in @xmath112 if we add these errors in quadrature , we get a total systematic uncertainty of 0.13 mag in predicting distances . we include no systematic error in velocity for @xmath6 estimated from the hubble diagram . for the sbf - alone @xmath6 estimate , however , we do include an error for its relation to the true far - field value . since the four sbf bcg are among the nearest of the lauer & postman ( 1994 ) sample , we take the net uncertainty in @xmath6 measured at @xmath34,500 km s@xmath4 scales as 7% , given the variation of @xmath113 allowed out to 15,000 km s@xmath4 by the lauer & postman ( 1992 ) bcg hubble diagram . in terms of velocity , this error corresponds to @xmath114 km s@xmath4 at the sbf bcg distances , similar to plausible bulk flows on this scale , regardless of how the bcg sample the lauer & postman ( 1994 ) flow . the errors in the final average @xmath6 values given in table [ tabho ] reflect the distance error , and except for the hubble diagram solution , the 7% far - field error added in quadrature to the statistical error . there are a number of paths for improving the present results . the hubble diagram @xmath6 can best be refined by obtaining sbf distances to more nearby bcg . refining the @xmath6 estimated solely from sbf distances , however , requires observation of more distant galaxies , as the present use of near - field velocities is an important uncertainty . improving the f814w sbf calibration will also be useful . fortunately , _ hst _ sbf observations for the nearest galaxies can generally be done within an orbit , and in many cases may be made from images obtained for other purposes , as was the case for the ajhar et al . ( 1997 ) sample . because this calibration may improve , we have attempted to make the path from 8bar to distances readily visible . revised values of @xmath6 can thus be quickly derived when better data become available . the present @xmath6 values are somewhat larger than many recent @xmath6 measurements based on calibration of hubble diagrams of a variety of distance estimators , but they are consistent with the higher of the comparisons . giovanelli et al . ( 1997 ) find @xmath115 using cepheid distances to 12 galaxies , which were used to calibrate a composite tully - fisher relationship based on 24 clusters of galaxies within 9,000 km s@xmath13 as do we , giovanelli et al . emphasize their independence from the classic virgo / coma distance ratio approach . giovanelli et al . did not show a hubble diagram , nor list their clusters , but from their description of the cluster distribution , we estimate that the effective depth of their @xmath6 determination is @xmath36,000 km s@xmath13 calibration of the tully - fisher relationship has been a major goal of the _ hst _ cepheid key project as well ( freedman et al . mould et al . ( 1995 ) find @xmath116 based on a tully - fisher calibration that reaches beyond 4,000 km s@xmath0 but that also depends heavily on the virgo cluster sample . freedman ( 1997 ) gives @xmath117 as a provisional summary of the key project work to date . supernovae also are providing excellent probes of the hubble flow . riess , press , & kirshner ( 1996 ) present a hubble diagram based on their light curve - shape method applied to 20 sn ia s ; using three cepheid calibrators , they conclude @xmath118 @xmath119 on @xmath37,000 km s@xmath4 scales . hamuy et al . ( 1996 ) find @xmath120 an essentially identical result , using their @xmath121 decay rate estimator applied to 29 sn ia with an effective depth of @xmath312,000 km s@xmath13 proper use of sn ia as distance estimators remains controversial , however , with significant disagreement on how the light curve decay rate relates to the sn ia peak luminosity . sandage et al . ( 1996 ) , for example , emphasize a hubble diagram approach as well , using sn ia s , and conclude @xmath122 this value , while consistent with those of riess , press , & kirshner ( 1996 ) and hamuy et al . ( 1996 ) , argues for @xmath6 near the lower ends of their error bars , rather than the upper ends , as would be more consistent with the tully - fisher results cited above and the present bcg results . kennicutt , freedman , & mould ( 1995 ) present a figure showing large dispersion in @xmath6 values calculated from sn ia alone over the last few years ; agreement on how to calibrate the sn ia distance scale remains work for the future . there is an abundance of other recent measurements of @xmath6 that we could cite , but as many of them are based heavily or exclusively on the virgo or coma clusters , we find them less attractive than methods featuring rich sampling of the hubble flow . it is also worth noting that while we are comparing @xmath6 estimates that all depend on the _ hst _ cepheid calibration work , different methods make use of different calibrators , which may account for a portion of the variance in @xmath6 among authors . at this writing , feast & whitelock ( 1997 ) are arguing that the cepheid scale , itself , should be revised based on _ hipparcos _ parallax measurements . if so , then the tonry et al . ( 1997 ) and ajhar et al . ( 1997 ) sbf zero points will need revision . two promising alternative approaches for measuring @xmath9 which are completely independent of local calibration , are gravitational lens induced time delay observations and measurement of the sunyaev - zeldovich ( sz ) effect for @xmath123 clusters . as freedman ( 1997 ) summarizes , however , measurements of the sz effect are neither accurate nor consistent enough at this time to provide an @xmath6 that challenges the more local measures . gravitational lenses also have the potential to produce a far - field @xmath6 that steps over all the distance - ladder problems that bedevil more traditional methods ; however , detailed mass distributions of the lenses are required , which presently limits their accuracy . kundi ' c et al . ( 1997 ) find @xmath124 based on the observed time delay between the two qso images in the classic @xmath125 lens at @xmath126 . in contrast , schechter et al . ( 1997 ) present time delays and mass models for the lens pg @xmath127 that favor @xmath128 although they also present a model that gives @xmath129 it is thus difficult to make a strong case at present that gravitational lenses , which probe the hubble flow on extremely large scales , are yielding consistent @xmath6 values significantly smaller than the @xmath130 measurements . observational evidence suggests that the far - field has been reached . while the sampling of the hubble flow remains sparse beyond 15,000 km s@xmath0 the limiting radius of lauer & postman ( 1992 ) , the sn ia hubble diagrams of riess , press , & kirshner ( 1996 ) and hamuy et al . ( 1996 ) are consistent with linear flows out to @xmath330,000 km s@xmath13 going to cosmological distances , kim et al . ( 1997 ) use sn ia to constrain @xmath131 by comparing 28 sn with @xmath132 to 18 sn of the hamuy et al . ( 1996 ) sample . the motivation for developing such sn hubble diagrams is to measure @xmath133 and @xmath134 such cosmological tests presume an unbiased local hubble flow as a point of departure . on the other hand , with assumed @xmath133 and @xmath135 one can test for significant @xmath131 over extremely large scales . for @xmath136 kim et al . find @xmath137 ( @xmath138 or @xmath139 ( 95% confidence ) . for @xmath140 with @xmath141 or @xmath142 one can get @xmath143 corresponding to global @xmath6 actually _ larger _ than the local value . going out far enough to measure the unbiased @xmath6 means reaching the scale on which mass density fluctuations no longer generate significant velocity perturbations of the hubble flow . the most important bias to consider for a full - sky determination of @xmath6 is the global radial retardation or acceleration of the hubble flow that occurs within significant mass over- or under - densities . the possibility that we are within a large bubble of lower than cosmic density , for example has been proposed as a way of reconciling apparently high @xmath6 values measured nearby , with the estimated age of the universe , and the often - cited concern ( e.g. bartlett et al . 1995 ) that extremely far - field @xmath6 measures such as those from the sz effect or lenses are lower than the more local measures ( although , as noted above , the case for this is weak ) . berschinger ( 1985 ) and ryden ( 1994 ) present analytic treatments of how voids grow with time . the voids effectively expand faster than the cosmic scale factor ; observers well inside the voids would see linear , if spuriously rapid hubble flows . shi , widrow , & dursi ( 1996 ) in general find @xmath144 where @xmath145 is the relative mass - deficit of the void . when limited to popular initial power - spectra , however , one predicts only small @xmath131 over the large volumes sampled by bcg and sn ia . turner , cen , & ostriker ( 1992 ) , for example , confined their analysis to very modest scales compared to the hubble diagrams now available . the large variations in @xmath131 that they observed under cdm and pib power - spectra occurred for volumes limited to 3,000 km s@xmath13 for volumes limited to 6,000 km s@xmath0 they found @xmath146 depending on the power - spectrum . shi , widrow , & dursi ( 1996 ) likewise conclude that @xmath147 on 15,000 km s@xmath4 scales with `` reasonable '' models of galaxy and structure formation . the question of whether or not we have gone out far enough in measuring the hubble flow thus remains an issue only if there is significantly more power in mass - fluctuations on large scales than would be expected under standard theories . in this context , we can posit that the 689 km s@xmath4 bulk flow observed by lauer & postman ( 1994 ) in the volume limited to 15,000 km s@xmath4 is indeed evidence for such power on large scales . strauss et al . ( 1995 ) , and feldman & watkins ( 1994 ) both find the abell cluster bulk flow to be incompatible at @xmath148 confidence with all standard models of galaxy formation considered . tegmark , bunn , & hu ( 1994 ) further argue that the lauer - postman bulk flow is not compatible with degree - scale measurements of the cmb anisotropy power - spectrum . even so , however , shi , widrow , & dursi ( 1996 ) , taking the harrison - zeldovich power spectrum shape and normalization parameters that best fit the lauer - postman flow ( see jaffe & kaiser 1995 ) , still find that only modest @xmath149 would be expected . at the same time , shi , widrow , & dursi show that the range of power spectra considered by jaffe & kaiser to fit the lauer - postman bulk flow would admit @xmath131 as large as 0.12 in the limiting extreme case . of course , power - spectra give only a statistical expectation for the local distribution of matter . one remains free to argue that we are within a local density anomaly that exceeds the @xmath150 fluctuations at some level . in this case , shi , widrow , & dursi ( 1996 ) discuss the cmb dipole and quadrupole anisotropies that would be observed within the anomaly and conclude that they would put strong constraints on the allowed geometry of the local density fluctuation ( see also tomita 1996 ) . we conclude that while one can construct models of the universe for which the far - field remains at distances well in excess of those explored here , they are extremely unfavored by what we know of the power - spectrum of initial mass fluctuations in the universe . the accuracy of these sbf distances offers the means to test the validity of different reference frames , although we can not legitimately solve for an independent , best - fit reference frame with only four points . as described in section @xmath23[ho4000 ] , we consider three reference frames : cmb ( for obvious reasons ) , local group ( since for smooth flows this has zero dipole locally ) , and abell cluster ( ac ) frame following the lauer & postman ( 1994 ) bulk flow . although we tried to remove all systematic , common contributors to the errors in the sbf distances when calculating @xmath92 for our @xmath6 estimates , we find that @xmath151 in table [ tabho ] ( @xmath152 ) has a value of 0.3 for the cmb frame . this may indicate that we have overestimated our errors , since such a value or lower will occur only 18% of the time by chance , so we bear in mind that all the true @xmath92 values may be larger than what is listed in table [ tabho ] . at the same time , it seems unlikely that our random errors could be overestimated by a factor of almost 2 , necessary to raise @xmath151 in the cmb frame to unity , so at least some of the concordance of results in this frame surely is coincidence . taking the numbers at face value , the probability that @xmath151 is at least as large as 1.2 ( the lg value ) is 0.31 , and the probability that @xmath151 is at least as large as 2.4 ( the ac frame value ) is 0.07 . thus our observations offer the most support for these four clusters being at rest in the cmb frame . rrrrrrr 262 & @xmath153&0.0250&@xmath154&@xmath155&@xmath156&@xmath1573560 & @xmath158&0.0126&@xmath159&@xmath160&@xmath161&@xmath1623565 & @xmath163&0.0108&@xmath164&@xmath165&@xmath166&@xmath1673742 & @xmath168&0.0187&@xmath169&@xmath170&@xmath171&@xmath172 averages & & & @xmath154&@xmath173&@xmath174&@xmath175 @xmath151 & & & 0.3 & 1.2 & 2.4 & 2.6 [ tabho ] the very small number of sbf distances does not permit us say whether this rejects the ac frame or not , since the ac frame was chosen to minimize the scatter of the @xmath17 relationship for 119 clusters at distances much greater than our sample here . additionally , we are guessing at a random velocity component of 209 km s@xmath4 ( 100 and 184 in quadrature ) . if the random velocity is as large as 500 km s@xmath4 the ac frame will have unity @xmath151 ( 0.1 in the cmb frame ) . it is clear that a262 is problematic for @xmath17 as seen in figures [ bcg_hub ] , [ bcg_hub_frame ] , and [ zest ] ; it has a very low surface brightness and hence its distance is estimated to be large . likewise , the sbf and @xmath17 distances are consistent within the errors for a3560 , a3565 , and a3742 , but there is a @xmath176 inconsistency with a262 . the cosmic scatter in the @xmath17 is clearly related to intrinsic variations in metric surface brightness among the bcg . lauer & postman ( 1994 ) note that this scatter greatly dominates any variation in @xmath96 due to peculiar velocity , but rely the assumption that intrinsic variations in bcg surface brightness are not correlated over large angles and will average out of a large sample . in the case of a262 , however , the ac frame does partially compensate for the discrepancy between its @xmath17 distance and its redshift . the rms scatter in the @xmath17 relationship for just these four clusters is 0.41 mag in the cmb frame , but only 0.27 mag in the abell cluster frame . if we delete a262 the rms @xmath17 residuals drop to 0.29 mag ( cmb ) , and 0.20 mag ( ac ) . in their analysis lauer & postman tried deleting outliers ( such as a262 ) and found the ac frame to be quite stable , so this is not the whole story . nevertheless it seems clear that a larger sample of sbf distances could not only provide a very accurate reference frame and random velocity amplitude , but also allow us to understand why the lauer & postman sample and the @xmath17 relation point to the ac frame . lauer & postman ( 1992 ) originally attempted to calibrate the bcg hubble diagram on the presumption that ngc 4472 , the virgo cluster bcg , was typical . adopting a 14.4 mpc distance to ngc 4472 , they found @xmath177 however , if they adopted the sandage & tammann ( 1990 ) 21.9 mpc virgo distance for ngc 4472 , @xmath178 would be implied . our present results for @xmath6 are essentially identical to that obtained with the short distance to ngc 4472 , and are significantly different from that obtained from the with distance , arguing that the distance to the virgo core is indeed close to the shorter value . if ngc 4472 were at the distance of 21.9 mpc , then it would be among the brightest of the lauer & postman ( 1994 ) sample , deviating from the @xmath17 ridge line by more than @xmath179 of course , one could question this conclusion by challenging the sbf calibration , since it is already known that sbf distances imply a short distance to virgo ( tonry et al . 1997 ) ; however , this still remains as an important consistency check . the relative distances inferred from bcg are consistent with sbf distances . we have used _ hst _ to obtain sbf distances to four bcg beyond 4,000 km s@xmath4 to calibrate the lauer & postman ( 1992 ) bcg hubble diagram , producing an estimate of the global value of @xmath6 valid on @xmath311,000 km s@xmath4 scales . this method gives @xmath180 and is based on the full lauer & postman ( 1994 ) 15,000 km s@xmath4 volume limited bcg sample . as such , the result is independent of virgo or coma cluster distances and membership issues , as well as the recession velocities of the four bcg studied . the large error reflects the photometric scatter about the @xmath17 ridge line , which was used to transfer the bcg hubble diagram to the sbf distance scale . as more bcg are observed with _ hst , _ the formal errors in this far - field @xmath6 should decrease . our review of the present understanding of the formation of large scale structure argues that we are likely to have fairly sampled the far - field . even theories with enough power on large spatial scales to generate bulk flows as large as those observed by lauer & postman ( 1994 ) are unlikely to have deviations outside of @xmath181 for the volume sampled by the bcg hubble diagram . in contrast , the compatibility of our results with those based on more nearby objects argues that there is little effect on @xmath6 and the depth of the measurements . going to the far - field most likely removes a source of uncertainty , rather than correcting for a systematic error . indeed we find @xmath2 just from hubble ratios based on the sbf distances and observed recession velocities to the four sbf - calibrated bcg at @xmath34,000 km s@xmath4 alone , a result consistent with our far - field result . the present @xmath6 rests on calibration of the sbf method and an understanding of its systematic effects . at the fundamental level , we are tied to the nearby cepheid calibrators . changes in the cepheid scale will propagate to the present results through the tonry et al . and ajhar et al . calibrations . as noted in the introduction , tonry et al . ( 1997 ) sbf calibration is tied to seven spiral galaxies with cepheid distances . further , tonry et al . have observed enough galaxies to perform an exhaustive series of tests , finding no systematic offsets between sbf observations of bulges and elliptical galaxies . a weaker link is transferring the ground @xmath1 method to the wfpc2 f814w filter , a task accomplished by ajhar et al . ( 1997 ) ; we will attempt to refine this calibration as more nearby systems are observed with _ hst . _ we conclude that the major uncertainties in the distance scale are those close to home rather than far away . we thank guy worthey and barbara ryden for helpful discussions , and john blakeslee for the photometry of a262 . this research was supported in part by _ hst _ go analysis funds provided through stsci grant go-05910.03 - 94a . ajhar , e. a. , lauer , t. r. , tonry , j. l. , blakeslee , j. b. , dressler , a. , holtzman , j. a. , & postman , m. 1997 , , 114 , 626 bartlett , j. g. , blanchard , a. , silk , j. , & turner , m. s. 1995 , science , 267 , 980 baum , w. a , hammergren , m. , thomsen , b. , groth , e. j. , faber , s. m. , grillmair , c. j. , & ajhar , e. a. 1997 , , 113 , 1483 bertschinger , e. 1984 , , 58 , 1 blakeslee , j. p. , & tonry , j. l. 1996 , , 465 , l19 burstein , d. , & heiles , c. 1984 , , 54 , 33 feast , m. w. , & whitelock , p. a. 1997 , preprint , astro - ph/9706097 feldman , h. a. , & watkins , r. 1994 , , 430 , l17 freedman , w. l. et al . 1994 , , 371 , 757 freedman , w. l. 1997 , in @xmath182 texas symposium , eds . olinto , a. , frieman , j. , & schramm , d. ( world scientific ) giovanelli , r. , haynes , m. p. , da costa , l. n. , freudling , w. , salzer , j. j. , & wegner , g. 1997 , , 477 , l1 gunn , j. e. , oke , j. b. 1975 , , 195 , 255 hamuy , m. , phillips , m. m. , suntzeff , n. b. , schommer , r. a. , maza , j. , & avil ' es , r. 1996 , , 112 , 2398 holtzman , j. a. , burrows , c. j. , casertano , s. , hester , j. j. , trauger , j. t. , watson , a. m. , worthey , g. 1995a , , 107 , 1065 hoessel , j. g. 1980 , , 241 , 493 holtzman , j. a. , et al . 1995b , , 107 , 156 jacoby , g. h. , branch , d. , ciardullo , r. , davies , r. l. , harris , w. e. , pierce , m. j. , pritchet , c. j. , tonry , j. l. , & welch , d. l. 1992 , , 104 , 599 jaffe , a. h. , & kaiser , n. 1995 , 455 , 26 kennicutt , r. c. , freedman , w. l. , & mould , j. r. , 1995 , , 110 , 1476 kim , a. g. et al . 1997 , , 476 , l63 kundi ' c , t. et al . 1997 , , 482 , 75 lauer , t. r. , & postman , m. 1992 , , 400 , l47 lauer , t. r. , & postman , m. 1994 , , 425 , 418 mould , j. et al . 1995 , , 449 , 413 postman , m. , & lauer , t. r. 1995 , , 440 , 28 riess , a. g. , press , w. h. , kirshner , r. p. 1996 , , 473 , 88 ryden , b. s. 1995 , , 452 , 25 sandage , a. 1972 , , 178 , 1 sandage , a. & hardy , e. 1973 , , 183 , 743 sandage , a. , & tammann , g. a. 1990 , , 365 , 1 sandage , a. , saha , a. , tammann , g. a. , labhardt , l. , panagia , n. , & macchetto , f. d. 1996 , , 460 , l15 schechter , p. l. , mateo , m. , & saha , a. 1993 , , 105 , 1342 schechter , p. l. et al . 1997 , , 475 , l85 schmidt , b. p. , kirshner , r. p. , eastman , r. g. , phillips , m. m. , suntzeff , n. b. , hamuy , m. , maza , j. , & aviles , r. 1994 , , 432 , 42 shi , x. , widrow , l. m. , & dursi , l. j. 1996 , , 281 , 565 strauss , m. a. , cen , r. , ostriker , j. p. , lauer , t. r. , & postman , m. 1995 , , 444 , 507 tammann , g. a. 1992 , , t43 , 31 tegmark , m. , bunn , e. f. , & hu , w. 1994 , , 434 , 1 tomita , k. 1996 , , 461 , 507 tonry , j. l. , ajhar , e. a. , & luppino , g. a. 1990 , , 100 , 1416 tonry , j. l. , blakeslee , j. b. , ajhar , e. a. , & dressler , a. 1997 , , 475 , 399 tonry , j. l. , & schneider , d. p. 1988 , , 96 , 807 turner , e. l. , cen , r. , & ostriker , j. p. 1992 , , 103 , 1427 whitmore , b. c. , sparks , w. b. , lucas , r. a. , macchetto , f. d. , & biretta , j. a. 1995 , , 454 , l73 worthey , g. 1993a , , 409 , 530 worthey , g. 1993b , , 418 , 947

the complex of curves of a closed orientable surface of genus is the simplicial complex whose vertices , , are isotopy classes of essential simple closed curves in . two vertices co - bound an edge of the-skeleton , , if there are disjoint representatives in . a metric is obtained on by assigning unit length to each edge of . thus , the distance between two vertices , , corresponds to the length of a geodesic a shortest edge - path between and in . in , birman , margalit and the second author introduced the concept of _ efficient geodesics _ in and used them to give a new algorithm for computing the distance between vertices . in this note , we introduce the software package micc ( _ metric in the curve complex _ ) , a partial implementation of the efficient geodesic algorithm . we discuss the mathematics underlying micc and give applications . in particular , up to an action of an element of the mapping class group , we give a calculation which produces all distance vertex pairs for that intersect times , the minimal number of intersections needed for this distance and genus .

1408.4134

let @xmath13 or @xmath3 denote a compact , connected , orientable surface of genus @xmath14 , where @xmath1 . a simple closed curve on @xmath13 is _ essential _ if does not bound a disk in @xmath13 . the complex of curves , introduced by harvey @xcite , is the simplicial complex , @xmath15 , whose vertices ( or @xmath16-skeleton ) , @xmath17 , are isotopy classes of essential simple closed curves ; and , whose edges of the @xmath4-skeleton , @xmath18 , connect vertices that have disjoint representatives . for the remainder of this note , `` curve '' will mean `` simple closed curve '' . by declaring that each edge of @xmath18 has length @xmath4 , we endow @xmath17 with a metric . specifically , an _ edge path _ is a sequence of vertices @xmath19 such that @xmath20 . a _ geodesic path _ joining @xmath7 and @xmath8 is a shortest edge - path . the _ distance _ , @xmath6 , between arbitrary vertices is the length of a geodesic path . since it is known that the complex of curves is connected , which was stated by harvey @xcite and followed from a previous argument of lickorish @xcite , the value @xmath6 is well - defined for all vertex pairs . we note that if @xmath21 , there is a vertex @xmath22 and curve representatives in @xmath13 , @xmath23 , @xmath24 and @xmath25 , such that @xmath26 and @xmath27 . the generic situation ( when some component of @xmath28 has euler characteristic less than zero ) is that there are infinitely many isotopically distinct choices for @xmath29 and , thus , infinitely many possible geodesics for distance @xmath30 . in this case , the existence of infinitely many geodesics at distance @xmath30 forces infinitely many geodesics for all distances . it is this infinite local pathology which makes finding an effective distance computing algorithm challenging . the curve complex was first introduced by harvey @xcite . its coarse geometric properties were first studied extensively by masur minsky @xcite . the complex of curves has proved a useful tool for the study of hyperbolic @xmath31-manifolds , mapping class groups and teichmller theory . in particular , in @xcite , it was established that teichmller space is quasi - isometric to the complex of curves and is therefore @xmath32-hyperbolic . here , @xmath32-hyperbolic means that geodesic triangles in @xmath18 are @xmath32-thin : any edge is contained in the @xmath32-neighborhood of the union of the other two edges . in @xcite , aougab established uniform hyperbolicity@xmath32 can be chosen independent of genus ( for @xmath1 ) . in spite of this considerable advancement in understanding the coarse geometry of the complex of curves , the development of tools intended to explicitly compute distance has been difficult . in 2002 , jason leasure proved the existence of an algorithm to compute the distance between two vertices of @xmath17 ( @xcite , corollary 3.2.6 ) . later , other algorithms were discovered by shackleton @xcite and webb @xcite , but none of these algorithms were studied seriously from the viewpoint of doing explicit computations , and all seem unsuitable for that purpose . recently birman , margalit and the second author @xcite have given a new algorithm_the efficient geodesic algorithm_and we have developed an implementation of it called the _ metric in the curve complex _ applications of micc we will present in this note include : * establishing that the minimal geometric intersection number for vertices of @xmath33 with distance four is @xmath12 , * listing of all vertex pairs ( up to an action of an element of the mapping class group ) of @xmath33 with distance four and having minimally positioned representatives with intersection number at most @xmath34 , and * producing an explicit example of two vertices of @xmath35 that have distance four and intersection number @xmath36 the key idea in @xcite is the introduction of a new class of geodesics , _ efficient geodesics_. they are not the same as the ` tight geodesics ' that have dominated almost all published work on the curve complex following their introduction in @xcite , however they share with tight geodesics the nice property that there are finitely many efficient geodesics between any two fixed vertices in @xmath15 . for convenience , for a pair of curves , @xmath37 , we will refer to a component of @xmath38 as a _ segment_. we will use a slightly weaked definition for efficient geodesic than that given in @xcite . [ d : ie ] let @xmath39 with @xmath40 . an oriented path @xmath41 , in @xmath17 is _ initially efficient _ if there are representatives @xmath42 , @xmath43 and @xmath44 such that @xmath45 for any segment @xmath46 . we say @xmath47 is _ efficient _ if @xmath48 is initially efficient for each @xmath49 and the oriented path @xmath50 is also initially efficient . the efficient path algorithm is a consequence of the following . ( theorem 1.1 of @xcite ) [ t : simplify ] let @xmath51 , and let @xmath7 and @xmath8 be two vertices of @xmath52 with @xmath40 . there exists an efficient geodesic from @xmath7 to @xmath8 , and in fact there are finitely many . when @xmath53 , notice that an efficient geodesic @xmath54 yields an oppositely oriented efficient geodesic , @xmath55 . that is , distance @xmath31 vertices have non - oriented efficient geodesics . thus , for corresponding representatives @xmath56 , we have that @xmath57 ( respectively @xmath58 ) will intersect segments of @xmath59 ( respectively of @xmath60 ) at most once . from this observation , we will establish the following test for distance @xmath61 which micc implements . [ theorem : bmm test](*distance @xmath61 test * ) let @xmath62 be vertices with @xmath40 . let @xmath63 be the collection of all vertices such that the following hold : 1 . for @xmath64 , we have @xmath65 ; and 2 . for @xmath64 , there exist representatives @xmath66 of @xmath67 respectively , such that for each segment @xmath68 we have @xmath69 . then @xmath70 if and only if @xmath71 for all @xmath64 . moreover , the collection @xmath72 is finite . [ remark on z ] keeping with our previous observation regarding non - oriented efficient geodesics at distance @xmath31 , we can flip the roles of @xmath7 and @xmath8 . thus , the test can also be stated in terms of @xmath73 for @xmath74 where : * if @xmath74 then @xmath75 . * if @xmath74 , there exists representatives , @xmath23 , @xmath24 and @xmath76 such that for all segments , @xmath77 , we have @xmath78 . before giving the proof of the distance @xmath61 test , we recall a useful concept and its implications . let @xmath79 be a pair of curves such that @xmath80 is minimal with respect to isotopies of @xmath81 . that is , @xmath82 and @xmath81 are _ minimally positioned_. determining when @xmath83 and @xmath84 are minimally positioned is straightforward due to the bigon criterion ( propostion 1.7 , @xcite)no disc component of @xmath85 has exactly two segments of @xmath38 in its boundary . we say @xmath82 and @xmath81 ( or @xmath86 ) is a _ filling pair _ if @xmath87 is a collection of @xmath30-discs . it is readily seen that a pair is filling on @xmath13 if and only if their corresponding vertices in @xmath17 are at least distance @xmath31 apart . when a minimally positioned pair of curves is not filling but still intersects , some component of @xmath87 contains an essential curve . thus , the corresponding vertices are distance @xmath30 apart . algorithmically determining whether a minimally positioned pair is filling , or not , requires simple tools coming from classical topology . for @xmath26 , let @xmath88 be a regular neighborhood . the genus of this neighborhood , @xmath89 , can be algorithmically computed as discussed in [ sec : genus ] . ( for an oriented surface @xmath90 with boundary , recall @xmath91 , where @xmath92 is the euler characteristic and @xmath93 . ) if @xmath94 , then a component of @xmath95 contains an essential curve of @xmath3 and the vertices that @xmath82 and @xmath81 represent are distance @xmath30 apart . if @xmath96 , those vertices are distance at least @xmath31 apart . we will see in [ sec : genus ] that this _ filling calculation _ can be readily implemented . however , if one is handed a nice enough presentation of @xmath83 and @xmath84 in @xmath13 , determining whether they are a filling pair can be done by inspection . for example , we will do such filling determinations in example [ example : hempel ] . we now give the proof of distance @xmath61 test . * proof : * from the above discussion we see that the assumption , @xmath40 , translates into considering only minimally positioned filling pairs in @xmath13 . to determine whether the associated vertices in @xmath17 of a filling pair are at distance @xmath61 , we need only determine that they are not at distance 3 . thus , suppose @xmath82 and @xmath81 represent classes @xmath7 and @xmath8 such that @xmath40 . assume there exists a length @xmath31 path @xmath97 . from theorem [ t : simplify ] , we can further assume that this path is initially efficient . in particular , for representative @xmath98 of the vertices of this path , respectively , we can assume @xmath99 intersects segments of @xmath100 at most once . thus , @xmath101 is an element of the set @xmath72 . but , since @xmath102 , we need only establish that @xmath103 for all @xmath64 to contradict the assumption that there was a distance @xmath31 path . the fact that the set @xmath72 is finite is due to @xmath104 being a collection of @xmath30-discs and representatives of any vertex of @xmath72 having bounded intersection with any such @xmath30-disc component . the stated test for @xmath61 follows . [ example : hempel ] we consider an example of a pair of curves , @xmath82 and @xmath81 , on a genus @xmath30 surface which represent classes that are distance @xmath10 apart . ( see figure [ fig : hempel example ] . ) this is an example of j. hempel and appears in the lecture notes @xcite . these notes assert distance @xmath10 for the pair without proof . as an application of the distance @xmath105 test , we now give a proof establishing distance @xmath10 for hempel s example . @xmath106 at 140 100 @xmath81 at 165 119 @xmath81 at 363 244 @xmath107 at 97 280 @xmath108 at 380 276 in figure [ fig : hempel example ] , the surface @xmath109 is represented as a rectangular region minus two discs . the gray sides of the rectangle are identified , left - to - right and top - to - bottom , to form a torus minus two discs . the genus @xmath30 surface is obtained by identifying the two oriented red boundary curves , and the resulting single curve is @xmath81 ( the identification is initiated by lining up the six colored dots on the @xmath110 curves ) . these identifications induce identifications of the endpoints of the dark blue arcs , so as to form the curve @xmath82 . by inspection , one can see that @xmath111 is a minimally positioned filling pair . we now apply the distance @xmath61 test . we wish to find curves , @xmath112 , that represent vertices , @xmath64 . such a @xmath112 will be in the complement of @xmath82 , intersecting any segment of @xmath113 at most once . three such @xmath114 can be immediately identified . this is because the complement of @xmath115 in @xmath109 is a collection of some number of @xmath10-gon regions and one single @xmath12-gon region . the boundary of any one of these regions is an alternating joining of segments in @xmath82 and @xmath81 . thus , any @xmath10-gon boundary has two segments in @xmath81 ; and , the boundary of the single @xmath12-gon has @xmath116 segments in @xmath81 . requiring that any @xmath112 intersect segments of @xmath81 at most once forces it to either not intersect a @xmath10-gon , or intersect each of the two @xmath81 segments of a @xmath10-gon once . however , there six different ways a @xmath112 can exit / enter the @xmath12-gon , giving us three possible @xmath114 that intersect the @xmath12-gon region once . in figure [ fig : hempel example ] , the dashed green and purple curves @xmath112 and @xmath117 illustrate two of the three curves generated by the exit / enter possibilities . it is readily apparent that both @xmath118 and @xmath119 is a collection of @xmath30-discs , none of which are bigons . thus , the corresponding vertex pairs are at least distance @xmath31 apart . the remaining possibilities for a @xmath107 can be dealt with in a similar straightforward manner ( theoretically , there are also @xmath120 that intersect the @xmath12-gon region @xmath30 and @xmath31 times ) . @xmath121 the @xmath72-calculation above illustrates the primary computing capabilities of the micc software package @xcite . micc is a computational tool that can determine whether the distance between two vertices in @xmath122 is @xmath30 , @xmath31 , or @xmath61 . its input is readily produced from any representation of two curves on a closed surface . it has functionality that can be used to search for new curve pairs or manipulate existing examples . its output can be used to construct geodesic paths between curves of short distances . as such , micc is an additional tool scholars can utilize in answering a number of basic questions about the local pathology of the complex of curves . as an illustration , we consider the relationship between distance and minimal intersection number . it is known that the theoretical minimal intersection number for a filling pair on a @xmath3 is @xmath123 due to the euler characteristic of the surface . for @xmath11 , this theoretical minimum is not realizable and the realizable minimum is in fact @xmath10 . recent work of aougab and huang @xcite has given a construction for realizing the theoretical minimum for @xmath124 . additionally , they show that all such minimum filling pairs are distance @xmath31 . for fixed @xmath1 , using his uniform hyperbolicity result , aougab proved that the theoretical minimum intersection number grows exponentially as a function of distance ( theorem 1.2 , @xcite ) . also , aougab and taylor @xcite give a recipe for producing filling pairs at a given distance whose intersection numbers are close to the minimum in an asymptotic sense ; see their paper for the precise statement . ido , jang and kobayashi @xcite also have a construction for producing filling pairs of a prescribed distance . the arguments in these last three citations employs the high power machinery of masur and minsky , including the _ bounded geodesic image theorem _ @xcite . thus , the growth bounds and constructed examples inherit a `` coarse geometry '' quality , which so far in the literature has not been used to produce the exact minimal intersection number with accompanying filling pairs for a specified distance and genus . in contrast , micc can be used to find explicitly all minimum intersecting filling pairs of distance @xmath10 . using the distance @xmath61 test , we give a `` proof of concept '' calculation that constructs all minimum intersecting distance @xmath10 filling pairs in @xmath125 up to homeomorphism . we can next use micc to calculate distance for curve pairs of increasing intersection number starting at this minimum . the result of this calculation is the following theorem . [ theorem : dwh - g2-d4-i12 ] the minimal intersection for a filling pair , @xmath126 , representing vertices @xmath127 , respectively , with @xmath128 is @xmath12 . combining this theorem with the distance @xmath61 test we obtain a partial test for distance four . [ corollary : dwh - g2-d4-i12 ] let @xmath70 for two vertices in @xmath129 . let @xmath130 be curves in minimal position representing @xmath7 and @xmath8 , respectively . let @xmath131 be a curve . if 1 . @xmath132 and @xmath133 , or 2 . @xmath134 and @xmath135 , then @xmath128 . the proof of concept calculation for theorem [ theorem : dwh - g2-d4-i12 ] involves finding all solutions to an integer linear programing problem so as to identify all potential candidates for minimally intersecting distance four curve pairs . however , such a comprehensive search is not necessarily needed to find examples of distance @xmath61 pairs . utilizing all of the functionality of micc , one can `` experiment '' with different curve pairs in a search distance four pairs . remark [ remark : experiment ] discusses how such experimentation led to the discovery of the first known filling pair representing distance @xmath10 vertices in @xmath136 . in particular , we have the following result : [ theorem : dwh - g3-d4-i29 ] the minimal intersection number for a pair of filling curves in @xmath137 that represent distance @xmath10 vertices in @xmath136 is less than or equal to @xmath36 . the outline for our paper is as follows . in [ sec : dwh ] , we discuss a method of representing a filling pair on @xmath138 . in [ sec : dwh - g2-d4-i12 ] , we give the proof of theorem [ theorem : dwh - g2-d4-i12 ] . in particular , the proof can be viewed as giving a general strategy for calculating theoretical minimal intersections of distance @xmath10 filling pairs for any higher genus . this strategy employed in a limited manner allowed us to verify that the explicit example given in @xcite ( cf . 2 ) of a distance @xmath10 vertex pair in @xmath129 establishes theorem [ theorem : dwh - g2-d4-i12 ] . we finish [ sec : dwh - g2-d4-i12 ] with an analysis of a genus @xmath31 distance @xmath10 pair establishing theorem [ theorem : dwh - g3-d4-i29 ] . finally , in [ sec : micc commands ] , we discuss the complete functionality of micc . to make this discussion concrete , we illustrate the range of micc commands with a running example . at the end of this manuscript , we attach the current known _ spectrum _ of pairs of distance @xmath10 or greater in @xmath129 with up to @xmath139 intersections . the full distance @xmath10 or greater spectrum in @xmath129 with up to @xmath34 intersections is available at @xcite . _ an expository remark_throughout we will continue to use @xmath83 and @xmath84 as representatives of @xmath7 and @xmath8 , respectively . similarly , indexed @xmath140 curves will be used as representatives of indexed @xmath141 vertices . this is meant to be consistent with the notation used in @xcite . for all other curves in the surface , we will use various `` flavors '' of @xmath107 , and @xmath142 will denote the corresponding vertex . we will always assume that any pairing of curves are in minimal position . in deciding how to represent curves on surfaces , we must first choose how we will represent closed oriented surfaces . representing surfaces with boundary as _ disc with handles _ ( or @xmath143 , example in figure [ fig : dwh - g2 - 2 ] ) is a well known method among working geometers and topologists , and it can be readily adapted to closed surfaces in our situation . for an essential curve @xmath144 , we can split @xmath13 along @xmath82 to produce a surface @xmath145 having two boundary curves , @xmath146 and @xmath147 . if @xmath82 is separating , @xmath145 will have two connected components , i.e. @xmath148 . the genus of each component will be less than @xmath14 and their sum would be @xmath149 . as such , a @xmath143 representation will have @xmath150 ( respectively , @xmath151 ) being a single @xmath30-disc with @xmath152 ( respectively , @xmath153 ) @xmath4-handles attached . we recover @xmath13 by gluing the boundary of these two components together . when @xmath82 is non - separating @xmath145 will be a @xmath154 connected surface with two boundary curves @xmath146 and @xmath147 . a @xmath143 representation of @xmath145 would be a single @xmath30-disc with @xmath155 @xmath4-handles attached . as before , @xmath13 is recovered by giving a gluing of its two boundary curves . now suppose @xmath86 is a filling pair in minimal position on @xmath13 . we obtain a @xmath143 representation of @xmath13 by splitting along @xmath82 . since @xmath82 and @xmath81 are in minimal position , we know that @xmath156 will be a collection of properly embedded essential arcs , @xmath157 . we require that each @xmath4-handle in our @xmath143 representation contain at least one @xmath158-arc . for example , skipping ahead to figure [ fig : dwh - g2 - 2 ] , the arcs with labels @xmath159 are the needed arcs . we refer to such an arc as a _ co - cores _ of a @xmath4-handle . skipping ahead to figure [ fig : dwh - g2-d4-i12 ] , the reader will find an example of a genus @xmath4 @xmath143 surface with two boundary components . the properly embedded essential black arcs are examples of @xmath158-arcs and each @xmath4-handle contains at least one such arc . since we must be able to recover both @xmath81 and @xmath13 by a gluing of @xmath146 and @xmath147 , we must have @xmath160 . more precisely , for @xmath161 there are three possible configurations : @xmath158 is a @xmath162 arc ( respectively , @xmath163 arc ) when @xmath164 ( respectively , @xmath165 ) ; and , @xmath158 is a @xmath166 arc when @xmath167 and @xmath168 . thus we can have any number of @xmath166 @xmath158-arcs but the number of @xmath162 arcs must equal the number of @xmath163 arcs . finally , observe that the @xmath158-arcs divide both @xmath146 and @xmath147 into @xmath169 intervals . in order to specify a gluing of @xmath146 and @xmath147 , we orient and cyclically label these @xmath169 intervals , by convention , @xmath16 through @xmath170 . again , referring to figure [ fig : dwh - g2-d4-i12 ] , we illustrate a gluing of @xmath146 and @xmath147 by the @xmath16 through @xmath171 labels , shown in red and blue respectively . we can reverse engineer this construction with an eye towards finding filling pairs of distance greater than @xmath31 . suppose we are interested in finding such a filling pair @xmath86 with @xmath82 non - separating . any associated @xmath145 will be a connected @xmath143 with two boundary curves . initially , let us fill @xmath145 with a maximal collection , @xmath172 , of properly embedded essential arcs that are pairwise non - parallel . we specify @xmath173 arcs to be arcs associated to the @xmath4-handles . thus , when we split the @xmath143 along these @xmath174 arcs , we obtain the underlying @xmath30-disc . figure [ fig : dwh - g2 - 2]-left illustrates such a configuration for @xmath11 . @xmath175 at 116 100 @xmath176 at 107 245 @xmath177 at 340 123 @xmath178 at 73 305 @xmath179 at 380 305 @xmath180 at 225 90 @xmath175 at 625 93 @xmath176 at 619 245 @xmath177 at 840 123 @xmath178 at 585 305 @xmath179 at 885 305 @xmath180 at 735 94 is a maximal collection of @xmath116 weighted arcs . the weights , @xmath181 , are non - negative integers . the green graph in the right illustration is @xmath182 , the dual graph . each edge of @xmath182 intersects exactly one arc of @xmath183 once.,title="fig:",scaledwidth=90.0% ] to obtain the collection of @xmath158-arcs that will be used to produce a curve @xmath81 , we will assign weights to each arc of @xmath172 such that there is a reasonable expectation that the distance @xmath61 test is satisfied . once we have determined the weight of an arc of @xmath172 , we will place that number of parallel copies of the arc in the @xmath143 . to this end , we consider the dual graph , @xmath184 , to @xmath172 in @xmath145 . ( figure [ fig : dwh - g2 - 2]-right illustrates such a dual graph for the left configuration . ) in graph - theoretic terms , we consider _ elementary circuits_edge paths that form simple loops in @xmath184 . each elementary circuit , @xmath185 , represents a possible vertex in @xmath72 of the distance @xmath61 test . let @xmath186 be the sum of the weights of all the arcs in @xmath172 that a circuit @xmath107 intersects . since our speculative @xmath81 curve will be the union of arcs parallel to those in @xmath172 and @xmath187 should be a filling pair for any circuit @xmath107 , we require that @xmath188 be greater than or equal to the minimal intersection number for a filling pair , i.e. @xmath10 when @xmath11 and @xmath189 when @xmath190 @xcite . thus , if @xmath191 is the complete set of elementary circuits , for each circuit we get an inequality of the form @xmath192 ( when @xmath190 ) or @xmath61 ( for @xmath11 ) . this gives us @xmath193 inequalities that make up an _ integer linear program _ ( ilp ) . we add to this the equality that states the sum of the weights of @xmath162 arcs equals the sum of the weights of @xmath163 arcs . these combined equations are the constraints for the object equation , @xmath194 , the sum of all the weights which we wish to minimize . figure [ fig : dwh - g2 - 2]-right illustrates this correspondence between a @xmath184 and a ilp when @xmath195 . in particular , for this dual graph , there are @xmath116 elementary circuits which yield @xmath116 weight equations . we get a seventh equation coming from having the weights of the @xmath162 and @xmath163 arcs being equal . all this yields the following constraints for minimizing @xmath196 : @xmath197 using any popular computing software ( e.g. maple ) one can readily check that the optimal value of this @xmath194 is @xmath198 . thus , we have a theoretical minimal intersection number for a filling pair of distance @xmath10 on @xmath109 . this optimal value happens to be uniquely realized by the _ weight solution _ @xmath199 = [ 2,2,2,0,2,0]$ ] . placing the corresponding set of @xmath198 @xmath158-arcs into the @xmath143 of figure [ fig : dwh - g2 - 2]-left , we then have @xmath198 possible ways to identify @xmath146 to @xmath147 . as discussed in [ subsec : perm ] , micc s permutation functionality can be used to check which of these boundary identifications will result in a single @xmath8 curve . for @xmath158-arcs corresponding to the weight solution @xmath200 $ ] there happen to be four identifications that yield a @xmath86 filling pair . finally , by employing micc s distance functionality ( described in [ subsec : distance ] ) to search for all elementary circuits , one can determine that all of these intersection @xmath198 filling pairs whose corresponding vertices have distance @xmath31 in @xmath129 . although the case counting is extensive , the discussion in [ sec : dwh ] gives us a straight forward strategy for proving theorem [ theorem : dwh - g2-d4-i12 ] . first , we need to generate all possible genus @xmath30 @xmath143 diagrams with weighted @xmath158-arcs so that we may then generate the corresponding ilp s . initially we divide this generating process into two cases corresponding to whether @xmath82 is a separating or non - separating curve . if @xmath82 is separating , we have exactly one possible @xmath143 diagram with weighted @xmath158-arcs . @xmath82 splits @xmath109 into two genus one surfaces with boundary , and each genus one surface has three weighted @xmath158-arcs . ( see figure [ fig : dwh - g2 - 4 ] . ) @xmath179 at 163 100 @xmath176 at 245 345 @xmath178 at 120 330 @xmath179 at 666 100 @xmath176 at 748 347 @xmath178 at 623 330 we leave it to the reader to generate the unique ilp in this situation . for the ilp corresponding to figure [ fig : dwh - g2 - 4 ] , we determined that the theoretical minimal intersection number for a filling pair @xmath201 of distance @xmath10 on @xmath109 is @xmath12 ; and , intersection @xmath12 is uniquely realized when all the weights equal @xmath30 . ibm s software package _ cplex optimization studio _ @xcite was utilized in solving this and all other ilp s in this paper . next , employing micc s permutation functionality , we determined that there are six possible identifications of @xmath146 and @xmath147 which yield a single @xmath84 curve . however , micc s distance functionality determined that all of these filling pairs of intersection @xmath12 were distance @xmath31 . so we turn to the non - separating case for @xmath82 . @xmath175 at 106 110 @xmath176 at 245 349 @xmath177 at 315 135 @xmath178 at 120 332 @xmath179 at 297 280 @xmath180 at 255 85 @xmath175 at 606 110 @xmath176 at 747 349 @xmath177 at 855 124 @xmath178 at 620 332 @xmath179 at 797 280 @xmath180 at 755 85 when @xmath82 splits @xmath109 into a connected genus one surface with two boundary components , we first need to generate all possible @xmath143 diagrams along with all possible complete collections of weighted @xmath158-arcs . such a @xmath143 will have three handles with each handle having the feature that it intersects one or both boundary curves . since we must have at least one handle intersecting both boundary curves , the possibilities are : all three handles intersect both boundaries ( as in figure [ fig : dwh - g2 - 2 ] ) ; two handles intersect both boundaries ( as in figure [ fig : dwh - g2 - 3 ] ) ) ; or , only one handle intersects both boundary curves . in the latter case , it is straight forward to see that , due to the requirement that the number of @xmath162 and @xmath163 arcs are equal , there will be boundary parallel @xmath158-arcs . since this would mean that @xmath202 is not minimal , we conclude that only the first two possibilities occur . having settled on the @xmath143 diagram of either figure [ fig : dwh - g2 - 2 ] or [ fig : dwh - g2 - 3 ] , we consider other choices for a maximal collection of weighted @xmath158-arcs . for example , figure [ fig : dwh - g2 - 3]-left and [ fig : dwh - g2 - 3]-right illustrates two different choices ( w.r.t . the @xmath143 structure ) in the case where we have just two handles intersecting both boundaries . note that the difference between the two collections is a different choice for the arcs associated with the weights @xmath203 . however , by interchanging the roles of the @xmath179 and @xmath177in figure [ fig : dwh - g2 - 3]-right , we view @xmath177 as the co - core of a handle instead of @xmath179we obtain figure [ fig : dwh - g2 - 3]-left ( after a relabeling weights ) . moreover , there is a similar re - choosing of the co - cores of handles in figure [ fig : dwh - g2 - 3]-left that will yield the weighted arc collection of figure [ fig : dwh - g2 - 2 ] . finally , for the alternate choice of the @xmath203 @xmath158-arcs in figure [ fig : dwh - g2 - 2 ] , one can again re - choose the handle co - cores to produce the collection of figure [ fig : dwh - g2 - 2 ] . ( we leave the details to the reader . ) thus , we need only consider the collection of weighted arcs in figure [ fig : dwh - g2 - 2 ] and its associated ilp ( [ llp ] ) . @xmath204 at 226 140 as mentioned at the end of [ sec : dwh ] , the optimal value of ilp([llp ] ) is @xmath198 and @xmath200 $ ] is the unique weight solution . again , this was determined by utilizing the software package cplex @xcite . next , we employed micc to determine that any of the four associated filling pairs are only distance @xmath31 . continuing , we utilized cplex to find all weight solutions for values @xmath205 . cplex found five weight solutions for value @xmath206 : @xmath207 $ ] , @xmath208 $ ] , @xmath209 $ ] , @xmath210$],and @xmath209 $ ] . using micc we determined all associated intersection @xmath211 filling pairs are distance @xmath31 . similarly , for values @xmath212 and @xmath213 there are @xmath214 and @xmath215 , respectively , weight solutions . again , micc found only distance @xmath31 filling pairs . however , using cplex to find all weight solutions to ilp ( [ llp ] ) for value @xmath216 we found @xmath217 solutions , @xmath206 of which had associated filling pairs that are of distance @xmath10 . to list these solutions : @xmath199 \in \ { [ 2 , 2 , 2 , 2 , 2 , 2]^4,$ ] @xmath218 ^ 2,$ ] @xmath219 ^ 2,$ ] @xmath220 ^ 2,$ ] @xmath221,$ ] @xmath222,$ ] @xmath223 ^ 2,$ ] @xmath224 ^ 2,$ ] @xmath225 ^ 2\}$ ] . ( the power notation is to dedicate a multiplicity of distance @xmath10 filling pairs . ) figure [ fig : dwh - g2-d4-i12 ] is a realization of @xmath221 $ ] . we credit its original discovery to j. birman , d. margalit and the second author @xcite . finally , we remark that there are undoubtedly duplications up to homeomorphism which we do not determine . to summarize , utilizing the distance @xmath61 test capability of micc , we have shown that the filling pair of figure [ fig : dwh - g2-d4-i12 ] is distance @xmath61 . to establish distance @xmath10 , we must produce a length @xmath10 path between @xmath82 and @xmath81 . we refer the reader to the green arc in figure [ fig : dwh - g2-d4-i12 ] that crosses the @xmath158-arc labeled @xmath30 and has endpoints in the @xmath226 segments of @xmath146 and @xmath147 . thus , this green arc can be understood as a closed curve that intersects @xmath82 once and @xmath81 once . this closed curve , @xmath58 , will represent a vertex @xmath227 of a length @xmath10 path @xmath228 . we can obtain a curve , @xmath57 , representing @xmath101 by taking a regular neighborhood of @xmath229 and letting @xmath57 correspond to its unique boundary curve . ( here , @xmath230 and @xmath231 . ) similarly , @xmath232 is represented by a curve coming from the boundary curve of a regular neighborhood of @xmath233 . the fact that these neighborhoods are each topologically a torus - minus - disc makes all of these curves essential . thus , we have a length @xmath10 path . @xmath234 at 80 100 @xmath234 at 470 90 @xmath234 at 630 90 @xmath58 at 265 90 @xmath235 at 170 90 @xmath235 at 522 90 we will prove the @xmath143 diagram of figure [ fig : genus 3 ] is a genus @xmath31 distance @xmath10 example of a filling pair have intersection @xmath36 . as described in [ subsec : input ] , the input for micc can be obtained by traversing @xmath146 & @xmath147 and reading off the labels of the @xmath158-arcs as they are crossed . the labels of the segments of the boundary curves indicate the identification map that forms the curve @xmath82 . as before , the black @xmath158-arcs join together to form @xmath84 . for convenience , we give the micc input below : * input top identifications : 1,11,3,27,8,15,7,24,0,10,2,12,4,21,19,17 , + 24,14,6,23,28,9,16,25,13,5,20,18,16 * input bottom identifications : 0,10,2,26,7,14,6,23,28,9,1,11,3,22,20 , + 18,25,13,5,22,27,8,15,26,12,4,21,19,17 in our genus @xmath30 example , we saw that there were @xmath116 curves representing vertices in the elementary circuit set @xmath72 . in fact , micc utilizes the set @xmath236 in its calculation . for the filling pair of figure [ fig : genus 3 ] , the set @xmath236 has @xmath237 vertices . we can specific a curve representing a vertex of @xmath236 by a sequence of boundary segment labels . for example , the sequence corresponds to a curve in this figure that intersects in cyclic order the boundary segments in this list . the set of green arcs in the figure correspond to this label sequence and should be to understood as a curve , @xmath234 , representing @xmath238 . by inspection observe that @xmath239 is a filling pair . similarly , the brown arcs of the figure correspond to the label sequence . the reader can also check by inspection that this curve , @xmath235 , representing @xmath240 , is also filling when paired with @xmath82 . having had micc determine that all of the curves representing vertices in @xmath236 fill when paired with @xmath82 , it remains for us to find a length @xmath10 path so as to establish @xmath128 . the magenta arc of figure [ fig : genus 3 ] corresponds to a curve , @xmath58 , representing a @xmath227 vertex of such a path . notice @xmath58 intersects @xmath84 only twice . thus , as a pair they can not be filling . taking an appropriate boundary curve of a regular neighborhood of @xmath241 will yield a @xmath242 . similarly , since @xmath243 we know @xmath244 and we can construct an @xmath57 as we did for our genus @xmath30 example to give us a representative of @xmath101 . [ remark : experiment ] as previously mentioned , the argument establishing ( the genus @xmath30 ) theorem [ theorem : dwh - g2-d4-i12 ] can be thought of as a proof of concept calculation . for higher genus the calculation is exactly the same , although more extensive due to the fact that there are more @xmath143 representations and , thus , a @xmath245-calculation with accompanying ilp equations for each @xmath143 . however , if we willing to settle for an estimate the calculation can be limited to just one @xmath143 representation and we can use micc to `` discover '' some rough bounds . this is essentially how the example of figure [ fig : genus 3 ] was found . we choose a somewhat symmetric genus @xmath31 @xmath143 and placed an initial set of disjoint proper essential arcs so that any simple closed curve drawn in the @xmath143 intersected these arcs at least @xmath246 = [ 2 \cdot g - 1 ] ) $ ] times . ( for example , observe that a curve that goes over only the left most @xmath4-handle of figure [ fig : genus 3 ] can be made to intersect the black arcs exactly @xmath226 times . ) using the perm command discussed in [ subsec : perm ] we were able to find identifications of the two boundary curves that connected these arcs into a single curve . from there we could use micc to compute the distance of the resulting filling pair . when such an attempt yields a distance @xmath31 pair , we can use the curves command discussed in [ subsec : curves ] to list all possible curves in @xmath245 . by adding in more proper arcs and/or altering their placement and iterating the procedure above , we can reduce the size of @xmath245 until we find a distance @xmath10 filling pair . such experimentation may be useful in understanding the relationship between the distribution of the proper arcs in the @xmath143 and the possible curves in @xmath245 . in this section we discuss the input format for micc and the commands for analyzing and manipulating that input . the input is two sequences of numbers which corresponds to a `` ladder '' representation of curves in the surface . the commands are _ genus _ , _ distance _ , _ curves _ , _ matrix _ , _ faces _ and _ perm _ ( for permutation ) . we will illustrate each of these features by further developing the example in figure [ fig : dwh - g2-d4-i12 ] . given a @xmath143 presentation of filling pair of curves @xmath201 on a surface @xmath13 , as previously discussed in [ sec : dwh ] , we have specified the gluing of @xmath146 and @xmath147 by cyclically labeling their segments . assuming we used @xmath169 labels for the segments of @xmath247 , we now label the @xmath158-arcs of @xmath84 , @xmath16 through @xmath170 . looking at figure [ fig : dwh - g2-d4-i12 ] , we have a labeling using @xmath16 through @xmath171 . for the purpose of extracting the micc input from this example , the labeling assignment could have been done in a random fashion . however , for aesthetic reasons we have taken care to label the @xmath158-arcs in the cyclic order they occur in @xmath84 . with the @xmath158-arcs of @xmath84 in our @xmath143 representation labeled as described above , we can now extract the micc input for our filling pair . starting at segment @xmath16 in @xmath147 , we traverse this boundary component in the positive direction and record the labels of the @xmath158-arcs as we pass over their endpoints . this sequence of labels will be the input for micc s `` input top identifications '' . repeating this process on @xmath146 we get the sequence of labels that will be the input for micc s `` input bottom identifications '' . together these two sequences are the basis of a _ ladder representation _ , @xmath248 , of a filling pair @xmath201 . a ladder representation is readily understood by again considering our example in figure [ fig : dwh - g2-d4-i12 ] . starting at the @xmath16 segments of @xmath247 and reading off the two label sequences , our input for micc would be : * input top identifications : 1,6,11,4,3,2,7,0,5,9,8,7 * input bottom identifications : 0,5,10,3,2,1,6,11,4,10,9,8 the reader should readily grasp the ladder metaphor by considering the representation in figure [ fig : d4-l - m]-left of a regular neighborhood of @xmath249 coming from figure [ fig : dwh - g2-d4-i12 ] . in this illustration , our two boundary components which have been glued together to form @xmath82 is represented by the horizontal segment that has its ends identified . ( our convention forces this identification to always occur in the middle of the @xmath16-segment of @xmath247 . ) each vertical segment above or below this horizontal @xmath82 is half of an @xmath158-arc . from left - to - right there are @xmath12 labels above these vertical @xmath158-halves correspond to micc s input top identifications . similarly , the @xmath12 labels below these vertical @xmath158-halves correspond to micc s input bottom identifications . @xmath83 at 1 115 @xmath84 at 25 155 it will be useful to the reader to have an understanding of how micc utilizes a ladder representation . specifically , for a ladder @xmath248 we have , from left - to - right , @xmath16 through @xmath170 intersection points of @xmath82 with @xmath81 . for the @xmath250 intersection we will associate to it a @xmath251 row vector @xmath252}_i = [ v^-(i),w^+(i),v^+(i ) , w^-(i)]$ ] . figure [ fig : d4-l - to - m ] illustrates the scheme for determining the values of the @xmath253 and @xmath254 . in particular , @xmath255 and @xmath256 ; and @xmath257 and @xmath258 correspond to the adjacent ( ladder ) vertices along @xmath84 with the parity determined by whether @xmath84 is pointing down ( @xmath259 and @xmath260 ) or up ( @xmath261 and @xmath262 ) at the @xmath250 intersection . @xmath83 at 7 75 @xmath84 at 10 110 @xmath263 at 253 95 @xmath264 at 283 95 @xmath265 at 313 95 @xmath266 at 343 95 } _ 0 = [ v^-(0 ) , w^+(0 ) , v^+(0 ) , w^-(0 ) ] = [ -11 , -5 , 1 , 7]$].,title="fig:",scaledwidth=50.0% ] stacking the row vectors @xmath252}_i$ ] in order , we produce a @xmath267 matrix . again , @xmath169 is the intersection number of @xmath82 and @xmath81 . figure [ fig : d4-l - m]-right is the associated _ characteristic matrix _ , @xmath268 $ ] , for the ladder in figure [ fig : d4-l - m]-left . it should be readily apparent to the reader that the information in @xmath268 $ ] is sufficient to reproduce the ladder @xmath248 , and @xmath248 is sufficient to reproduce a @xmath143 representation . thus , up to any permutation of labels and changes of orientation of our curves , @xmath248 is dependent only on the classes @xmath7 and @xmath8 . given the prompt , what would you like to calculate ? , a reply of matrix will produce @xmath268 $ ] . we now make concrete the filling calculation . we will use @xmath248 and its characteristic matrix @xmath268 $ ] to compute the minimal genus of the surface that @xmath82 and @xmath81 fill . for a surface of genus @xmath14 , with a minimally intersecting filling @xmath10-valent graph , @xmath269 , we have @xmath270 , where @xmath271 is the set of vertices and @xmath272 is the set of edges of the graph @xmath273 . also , @xmath274 is a set of @xmath275-gon disc regions . then @xmath276 . thus , determining @xmath14 requires a count of the number of @xmath275-gon regions of @xmath277 . listing the components of @xmath277 using @xmath268 $ ] is achieved by specifying their edge - path boundaries in @xmath273 . when traversed such a edge - path alternating_an edge @xmath278 followed by an edge @xmath279 followed by @xmath280 and so on , will be a cyclically ordered set @xmath281 . thus , @xmath282 ( respectively , @xmath283 ) starts at vertex having integer label @xmath284 ( respectively , @xmath285 ) and terminates at vertex having integer label @xmath286 ( respectively , @xmath287 ) ; and , @xmath288 and @xmath289 , all modulo @xmath290 . the entries of the @xmath252_i}{\rm ' s}$ ] gives us that @xmath291 additionally , we have @xmath292 now fixing a @xmath275-gon region and , by convention , traversing its edge path boundary so as to always keep the region to our left , we have the following scheme for finding the terminus endpoint for the _ next _ @xmath293 or @xmath294 edge . * if @xmath295 then @xmath296 ( with @xmath297 ) . * if @xmath298 then @xmath299 ( with @xmath297 ) . * if @xmath300 and @xmath301 then @xmath302 ( with @xmath303 ) . * if @xmath304 and @xmath305 then @xmath306 ( with @xmath303 ) . * if @xmath300 and @xmath305 then @xmath307 ( with @xmath303 ) . * if @xmath304 and @xmath301 then @xmath308 ( with @xmath303 ) . we illustrate this scheme using our @xmath268 $ ] in figure [ fig : d4-l - m]-right . starting at vertex @xmath30 we can traverse the edge between @xmath309 and @xmath310 . thus , we have the assumption of t2 for @xmath311 . this gives us @xmath312 with @xmath313 . since @xmath314 , we have the t3 assumption and @xmath315 . but this gives us the t1 assumption for @xmath316 . thus , we have that @xmath317 which is @xmath318 . since @xmath319 we again have the t3 assumption which implies @xmath320 . now , we again have the t2 assumption for @xmath321 which implies @xmath322 , back where we started . ( refer to the region in figure [ fig : dwh - g2-d4-i12 ] contain @xmath204 . ) working out all such boundary edge paths we get a count for @xmath323 and , thus , are able to compute genus . having traversed all the boundaries of the @xmath275-gon regions of @xmath85 , micc records this calculation as a vector . specifically , let @xmath324 be the number of @xmath275-gon regions for @xmath325 . then associated with the graph @xmath326 we have the vector @xmath327 $ ] . the output of micc is actually formatted as @xmath328 . this vector solution to euler characteristic equation of lemma 4.1 of @xcite . the prompting inquiry and command appears as : what would you like to calculate ? faces . additionally , it lists each boundary edge in a truncated matter it lists only the @xmath329 . for our extended example of figure [ fig : d4-l - to - m ] , we would get : @xmath330 an alternate reply to the prompt , what would you like to calculate ? , is curves . micc applies the theorem [ theorem : bmm test ] test by listing all of the curves representing vertices of @xmath236 and computing the genus of their graphs when paired with @xmath83 . micc finds all curves representing elements of @xmath236 by applying a classical depth - first search @xcite for elementary circuits of the graph @xmath331 , the dual graph to the proper arcs of @xmath83 in @xmath332 . the micc output is a cyclic sequence of @xmath333 edges . that is , a curve,@xmath112 , representing a vertex @xmath334 and a region @xmath335 , @xmath336 will be a collection of proper arcs having their endpoints on @xmath333 edges of the boundary of @xmath337 . as we traverse @xmath112 , we will travel between regions by passing through @xmath338 . thus , @xmath142 can be characterized by giving a cyclic listing of these @xmath338 . continuing with our extended example of figure [ fig : d4-l - to - m ] , the output response to curves would yield : @xmath339}$ \\ $ { \tt curve genus : } \ 2 $ \\ \\ $ { \tt path \ [ 2 , 10 , 11 , 7]}$ \\ $ { \tt curve genus : } \ 2 $ \\ \\ $ { \tt path \ [ 1 , 6 , 5 , 4 , 3 , 8]}$ \\ $ { \tt curve genus : } \ 2 $ \\ \\ $ { \tt path \ [ 0 , 5 , 4 , 9 , 2 , 7]}$ \\ $ { \tt curve genus : } \ 2 $ \\ \\ $ { \tt path \ [ 0 , 5 , 4 , 9 , 10 , 11]}$ \\ $ { \tt curve genus : } \ 2 $ \\ \\ $ { \tt path \ [ 0 , 11 , 10 , 9 , 3 , 8 , 1 , 6]}$ \\ $ { \tt curve genus : } \ 2 $ \\ \end{tabular}\end{aligned}\ ] ] to determine @xmath6 one replies to the prompt , what would you like to calculate ? , by typing distance . if the genus of any of the pairs @xmath340 is less than @xmath14 , then micc will respond with distance : 3 . if the genus of all such pairs is @xmath14 , then micc will respond with distance : 4 + . all possible @xmath112 representing elements of @xmath236 are determined as described above . finally , micc has an experimental functionality . as most topologist who have attempted to construct filling pairs on surfaces know , it is difficult to do so while avoiding the production of multi - curves . for example , starting with a @xmath143 representation of the surface , after placing down some collection of @xmath158-arcs finding an identification of @xmath341 and @xmath342 so as to have @xmath84 be a single curve is tedious at best . fortunately , micc automates this process . given any ladder top / bottom identification , it will first determine whether @xmath84 is a single curve or a multi - curve . if it is a multi - curve , then it will produce the inquiring prompt would you like to shear this multi - curve ? . with the reply yes , it will search through all possible @xmath341/@xmath342 identifications for those that yield a single @xmath84 curve and print them out along with their distance . if micc has been given a ladder identification that corresponds to a single @xmath84 curve , one can still find all other @xmath341/@xmath342 identifications that yield a single curve . when given the prompt what would you like to calculate ? just reply with perm ( for permutation ) . for our extended example , the output would be : @xmath343 $ \\ $ [ 12 , 5 , 10 , 3 , 2 , 1 , 6 , 11 , 4 , 10 , 9 , 8]$ \\ \\ $ { \tt curve \ 2 \ distance : \ 3}$ \\ $ [ 5 , 9 , 8 , 7 , 1 , 6 , 11 , 4 , 3 , 2 , 7 , 12]$ \\ $ [ 12 , 5 , 10 , 3 , 2 , 1 , 6 , 11 , 4 , 10 , 9 , 8]$ \\ \\ $ { \tt curve \ 3 \ distance:\ 4+}$ \\ $ [ 1 , 6 , 11 , 4 , 3 , 2 , 7 , 12 , 5 , 9 , 8 , 7]$ \\ $ [ 12 , 5 , 10 , 3 , 2 , 1 , 6 , 11 , 4 , 10 , 9 , 8]$ \end{tabular}\end{aligned}\ ] ] micc was originally created as a tool that would help in the search for distance @xmath10 filling pairs on surface of genus greater than @xmath30 . we have since realized that it can be set to other uses . for example , it was recently used to find geodesic triangles where any pair of vertices correspond to filling pairs in minimal position . as mentioned previously , we hope researchers will find additional uses . remarking on the complexity of the algorithms employed in micc , the most computationally expensive task involves finding all cycles in the graph @xmath344 of [ subsec : curves ] . we know that the runtime of the process of counting all cycles in a graph is at best exponential . this gives a lower bound on the complexity of finding all cycles in @xmath344 , and thus on the complexity of the program itself . there are many ways in which we would like to improve the current version of micc . currently , micc is only a partial implementation of the algorithm presented in @xcite . we plan to extend the functionality of micc to encompass the full scope of the efficient geodesic algorithm of @xcite . yet this partial implementation is also manifestation of the complexity barrier involved with the current exponential running time of the graph search . an improved algorithm will allow for more intricate curves to be studied , and parallelization of the micc would help future users to fully utilize their multicore computers in their research . the first , third and fourth authors are grateful to joan birman , dan margalit and the second author for sharing results of their joint work as it developed . our thanks goes to john ringland , joaquin carbonara and the urge to compute program at the university at buffalo and buffalo state college for supplying a nurturing environment for our work . this work was supported in part by nsf csums grants 0802994 and 0802964 to the university at buffalo and buffalo state college . finally , we thank the referees for alerting us to the recent results in the literature , for suggesting a strengthening of the statement of corollary [ corollary : dwh - g2-d4-i12 ] and numerous other expository improvements . below we list with multiplicity all solutions to ilp ( [ llp ] ) for @xmath194-values @xmath12 , @xmath346 and @xmath139 . * @xmath350 $ ] , @xmath351 ^ 2 $ ] , @xmath352 $ ] , @xmath353 $ ] , @xmath354 $ ] , @xmath355 $ ] , @xmath356 $ ] , @xmath357 $ ] , @xmath358 $ ] , @xmath359 $ ] , @xmath360 $ ] , @xmath361 $ ] , @xmath362 $ ] , @xmath363 $ ] , @xmath364 $ ] , @xmath365 ^ 2 $ ] , @xmath366 ^ 2 $ ] , @xmath367 ^ 2 $ ] , @xmath368 $ ] , @xmath369 $ ] , @xmath370 $ ] , @xmath371 ^ 2 $ ] , @xmath372 $ ] , @xmath373 $ ] , @xmath374 $ ] , @xmath375 ^ 2 $ ] , @xmath376 $ ] , @xmath377 $ ] , @xmath378 $ ] , @xmath379 $ ] @xmath380 $ ] , @xmath381 ^ 2 $ ] , @xmath382 $ ] , @xmath383 $ ] , @xmath384 $ ] , @xmath385 $ ] , @xmath386 $ ] , @xmath387 ^ 2 $ ] , @xmath388 $ ] , @xmath389 $ ] , @xmath390 $ ] , @xmath391 $ ] , @xmath392 ^ 2 $ ] , @xmath393 ^ 2 $ ] * @xmath395 ^ 2 $ ] , @xmath396 $ ] , @xmath397 ^ 2 $ ] , @xmath398 $ ] , @xmath399 ^ 2 $ ] , @xmath400 $ ] , @xmath401 ^ 4 $ ] , @xmath402 $ ] , @xmath403 $ ] , @xmath404 $ ] @xmath405 $ ] , @xmath406 $ ] , @xmath407 ^ 3 $ ] , @xmath408 $ ] , @xmath409 ^ 2 $ ] , @xmath410 ^ 2 $ ] , @xmath411 ^ 3 $ ] , @xmath412 $ ] , @xmath413 ^ 2 $ ] , @xmath414 $ ] , @xmath415 $ ] , @xmath416 ^ 2 $ ] , @xmath417 $ ] , @xmath418 ^ 2 $ ] , @xmath419 ^ 2 $ ] , @xmath420 $ ] , @xmath421 $ ] , @xmath422 $ ] , @xmath423 $ ] , @xmath424 $ ] , @xmath425 $ ] , @xmath426 ^ 2 $ ] , @xmath427 $ ] , @xmath428 ^ 2 $ ] , @xmath429 $ ] , @xmath430 ^ 2 $ ] , @xmath431 $ ] , @xmath432 $ ] , @xmath433 $ ] , @xmath434 $ ] , @xmath435 $ ] , @xmath436 ^ 2 $ ] , @xmath437 ^ 2 $ ] , @xmath438 $ ] , @xmath439 $ ] , @xmath440 $ ] , @xmath441 $ ] , @xmath442 ^ 2 $ ] , @xmath443 ^ 4 $ ] , @xmath444 $ ] , @xmath445 ^ 2 $ ] , @xmath446 ^ 2 $ ] , @xmath447 $ ] , @xmath448 $ ] , @xmath449 $ ] , @xmath450 $ ] , @xmath451 $ ] , @xmath452 $ ] , @xmath453 $ ] , @xmath454 ^ 2 $ ] , @xmath455 ^ 3 $ ] , @xmath456 $ ] , @xmath457 ^ 2 $ ] , @xmath458 $ ] @xmath459 $ ] , @xmath460 $ ] , @xmath461 ^ 2 $ ] , @xmath462 $ ] , @xmath463 ^ 2 $ ] , @xmath464 $ ] , @xmath465 $ ] , @xmath466 $ ] , @xmath467 $ ] , @xmath468 ^ 2 $ ] , @xmath469 ^ 2 $ ] , @xmath470 $ ] @xmath471 $ ] , @xmath472 $ ] , @xmath473 $ ] , @xmath474 ^ 3 $ ] , @xmath475 ^ 2 $ ] , @xmath476 $ ] , @xmath477 ^ 2 $ ] , @xmath478 ^ 2 $ ] , @xmath479 $ ] , @xmath480 ^ 2 $ ] . the weight solutions @xmath481 $ ] for @xmath216 is particularly intriguing since it suggests there might be a high level of symmetry . figure [ fig : 2,2,2,2,2,2 ] is a @xmath31-dimensional rendering of the one the four associated distance @xmath10 filling pair . the aesthetic of this rendering is so appealing that it was placed at the begin of @xcite . we thank those authors for its use here . micc software package , software tutorial , and all known weight solutions yielding filling pairs having distance @xmath61 for @xmath482 with @xmath11 ( approximately 72,000 weight solutions ) are posted for download at micc.github.io . w. j. harvey , _ boundary structure of the modular group _ , riemann surfaces and related topics : proceedings of the 1978 stony brook conference ( i. kra and b. maskit eds . ) , ann . math . stud . 97 , princeton , 1981 .

we generate inherent structures , local potential - energy minima , of the ``-space overlap potential '' in two - dimensional many - particle systems using a cooling and quenching simulation technique . the ground states associated with the-space overlap potential are stealthy ( _ i.e. , _ completely suppress single scattering of radiation for a range of wavelengths ) and hyperuniform ( _ i.e. , _ infinite wavelength density fluctuations vanish ) . however , we show via quantitative metrics that the inherent structures exhibit a range of stealthiness and hyperuniformity depending on the fraction of degrees of freedom that are constrained . inherent structures in two dimensions typically contain five - particle rings , wavy grain boundaries , and vacancy - interstitial defects . the structural and thermodynamic properties of inherent structures are relatively insensitive to the temperature from which they are sampled , signifying that the energy landscape is relatively flat and devoid of deep wells . using the nudged - elastic - band algorithm , we construct paths from ground - state configurations to inherent structures and identify the transition points between them . in addition , we use point patterns generated from a random sequential addition ( rsa ) of hard disks , which are nearly stealthy , and examine the particle rearrangements necessary to make the configurations absolutely stealthy . we introduce a configurational proximity metric to show that only small local , but collective , particle rearrangements are needed to drive initial rsa configurations to stealthy disordered ground states . these results lead to a more complete understanding of the unusual behaviors exhibited by the family of `` collective - coordinate '' potentials to which the-space overlap potential belongs .

1104.5622

recently , we have been interested in a family of soft , long - range pair interactions that give rise to novel physical behaviors of many - particle systems including classical disordered ground states for a range of densities , negative thermal expansion , and vanishing normal - mode frequencies.@xcite this family of soft pair interactions includes those pair potentials @xmath1 with fourier transforms @xmath2 that are positive , bounded , and vanish to zero at some finite wavenumber @xmath3 and beyond . the potential energy @xmath4 of a system of @xmath5 particles in a fundamental cell with volume @xmath6 under periodic boundary conditions can be written as @xmath7 , \end{aligned}\ ] ] where @xmath8 is the distance between particles @xmath9 and @xmath10 , @xmath11 are the fourier coefficients of the density field , and @xmath12 are the wave vectors appropriate for the system size and shape . these soft pair interactions , called `` collective - coordinate '' potentials , are bounded and possess long - range oscillations in real space.@xcite because of the finite cutoff in @xmath0-space , an analytic lower bound for the potential energy per particle @xmath13 is easily obtainable , and numerical methods allow for the construction and investigation of ground states with extremely high precision . a series of studies has examined the structural characteristics of ground states in one,@xcite two,@xcite and three dimensions.@xcite three structural ground - state regimes were found to characterize these two - dimensional systems . increasing the fraction of degrees of freedom that are constrained @xmath14 , equivalently decreasing density , spans regimes that are disordered , wavy - crystalline , and crystalline . recently , an analytical connection between the fraction of constrained degrees of freedom within the system and the disorder - order phase transition for a class of target structure factors has been provided@xcite by examining the realizability of the constrained contribution to the pair correlation function . more recently , these potentials were used to construct `` stealthy '' and `` hyperuniform '' materials in two dimensions.@xcite `` stealthy '' materials refer to point patterns that completely suppress single scattering of radiation for certain wavelength ranges . while nonstealthy materials may allow for some small amount of scattering , stealthy materials are absolutely transparent at those wavelengths . `` hyperuniform '' refers to point patterns in which infinite wavelength density fluctuations vanish.@xcite ( these terms are defined more precisely in sec . [ sec : background ] . ) the hyperuniformity notion enables the rank ordering of crystals , quasicrystals and special disorderedmany - particle systems.@xcite interestingly , disordered hyperuniform many - particle ground states , and therefore also point distributions , with substantial clustering can be constructed.@xcite constructed ground states are now the basis for novel disordered materials with tunable , photonic band gaps.@xcite using a model potential from the family of collective - coordinate potentials in two dimensions , the `` @xmath0-space overlap potential , '' described in sec.[sec : background ] , we observed negative thermal expansion and vanishing normal - mode frequencies for ground - state configurations . we attributed these phenomena to the nature of the underlying energy landscape , which we described as having ground - state `` valleys '' that weave between the higher - energy portions of the landscape.@xcite despite the discovery of the existence of such unique properties , the fundamental mechanisms allowing for disordered ground states and stealthy point patterns still need to be fully elucidated . here , we analyze the energy landscape further by examining the associated inherent structures , or local potential - energy minima , for the aforementioned @xmath0-space overlap potential in two dimensions to better understand these properties . we have two primary motivations for this work . first , through the course of our previous research , we have observed several paradoxical phenomena related to local potential - energy minima and glassy behavior . while an `` inherent structure '' is in general any local potential - energy minimum , the term `` glass '' refers to an amorphous solid that is kinetically trapped in a potential - energy well . in ref . [ 2 ] , we observed an unusual equation of state when cooling a system compared to a more familiar ( _ e.g. _ , lennard - jones ) glass - forming system as fig . [ fig : glasses ] demonstrates this schematically . in fig.[fig : normglass ] , the dark and dotted lines represent the equilibrium equation of state and nonequilibrium isobaric cooling path for a lennard - jones system . figure [ fig : ccglass ] shows the corresponding cooling paths for a system interacting with the @xmath0-space overlap potential . in the more familiar glass - forming system , there is a well - defined glass transition below the freezing point that is dependent on the rate at which the system is cooled.@xcite it is well known that the glass - forming behavior of a system is a function of the underlying energy landscape.@xcite the depths of the potential - energy wells in the energy landscape dictate the dynamics of structural rearrangements of the system.@xcite for the @xmath0-space overlap potential , we observed that at high pressure the nonequilibrium curve deviated from the equilibrium curve at a temperature above the melting temperature , as demonstrated in fig . [ fig : ccglass ] . at lower pressures where there is no well - defined melting temperature , as in the wavy - crystalline regime , the nonequilibrium curve deviates similarly from the equilibrium curve . in addition , the @xmath0-space overlap potential gives rise to negative thermal expansion , generally for nondimensional @xmath15.@xcite other paradoxical behaviors have led us toward a broad analysis of the energy landscape . we have observed that some ground - state configurations , particularly crystalline structures , had many zero - energy normal modes . these ground states are therefore not mechanically rigid . however , while using numerical minimization algorithms to construct ground - state configurations , we encountered many mechanically stable local potential - energy minima . lastly , there exists a large density range in which the energy landscape was evidently devoid of all local minima . an increment in @xmath14 just outside this range introduced local potential - energy minima to the landscape . while we have previously observed these phenomena,@xcite the fundamental mechanisms that underlie them are not fully understood . the inherent - structure analysis has proven to be a fruitful method for relating the energy landscape to low - temperature phenomena . many studies have examined inherent - structure characteristics in glass - forming liquids with strong repulsive cores such as the stillinger - weber potential,@xcite water - like pair potentials,@xcite binary lennard - jones - like systems,@xcite and general repulsive potentials.@xcite the inherent structures for lennard - jones and steeply repulsive potentials are in general not hyperuniform or stealthy due to the dominance of grain boundaries and vacancy defects . in contrast to the above list of strongly repelling potentials , the @xmath0-space overlap potential is a soft interaction . soft interactions are often useful models for soft - matter systems such as colloids , polymers , and microemulsions.@xcite in addition , these @xmath0-space overlap interactions are also qualitatively similar to friedel oscillations in molten metals.@xcite the @xmath0-space overlap potential is localized in @xmath0-space and delocalized in real space as a result of the fourier transform . certain duality relations link the ground - state energies of the @xmath0-space overlap potential to the ground - state energies of real space analog.@xcite there have been several investigations of the inherent structures of various soft interactions . the gaussian - core model in two dimensions has polycrystalline inherent structure with a large correlation length even when sampled from the liquid state.@xcite energy landscape analyses revealed that the range of the morse potential affects the relation between temperature and the potential - energy distribution of sampled inherent structures.@xcite while for the yukawa potential , inherent structures varied depending on whether they were obtained from the liquid , crystal or hexatic phase.@xcite novel , stealthy dielectric materials are currently being fabricated,@xcite and nearly stealthy ceramic materials are of interest for optical applications.@xcite because of these recent experimental applications and the unusual physical behaviors discovered,@xcite we are also motivated to understand the fundamental differences between point patterns that absolutely suppress scattering for certain wavelengths and those that nearly suppress scattering . our ground - state construction procedure @xcite automatically distinguishes stealthy and hyperuniform configurations from those that do not possess such properties . in particular , we want to understand whether the particle rearrangements required to transform nearly hyperuniform and stealthy materials to configurations that are perfectly hyperuniform and stealthy are global or local in nature . while a general method for understanding these particle rearrangements would require a new algorithm to search for collective motions that increase stealthiness , the collective - coordinate approach provides an excellent framework from which to address this question . here , we provide two methods for identifying rearrangements in particle systems to achieve stealthy and hyperuniform materials . for large @xmath14 , we use the nudged - elastic - band algorithm to connect inherent structures , which are nearly hyperuniform and nearly stealthy , to ground states along a minimum - energy path . for small @xmath14 values , we have identified that there are no inherent structures higher up in the energy landscape . therefore , we study the rearrangements from a saturated random sequential addition ( rsa ) of hard disks@xcite to ground states of the @xmath0-space overlap potential . we use saturated rsa systems as initial conditions because upon saturation they suppress scattering for small wavenumbers and are nearly hyperuniform.@xcite we also introduce a stealthiness metric and a configurational proximity metric to quantify characteristics of these transitions . in this paper , we use a collective - coordinate potential and a simulation methodology to find stealthy point patterns and local potential - energy minima above the ground state . we probe the following fundamental questions : * how are inherent structures and their thermodynamic properties in collective coordinate systems different from those found in other soft - potential systems ? * to what extent are inherent structures stealthy and/or hyperuniform ? * how are the features of inherent structures related to the pair potential function ? * what collective particle rearrangements are necessary to construct a path from an inherent structure to a ground state ? * what global and/or local particle rearrangements are necessary to convert a nearly stealthy and nearly hyperuniform system in to one that is absolutely stealthy and hyperuniform ? the remainder of this paper is organized as follows . in sec.[sec : background ] , we define the @xmath0-space overlap potential , introduce several definitions , and briefly review the ground - state structural regimes . in sec . [ sec : methods ] , we detail the methods we use to obtain and characterize inherent structures . in addition , we discuss the methods we use to find transition states between inherent structures and ground states . in sec.[sec : cooling ] , we examine the thermodynamic properties of inherent structures as a function of system size and cooling rate . the inherent structures are characterized in sec . [ sec : structure ] . paths connecting inherent structures to ground states are provided in sec . [ sec : mapping ] while the rearrangements from rsa systems to ground states are explored in sec.[sec : rsa ] . concluding remarks are provided in sec . [ sec : disc ] . while a fully detailed summary of collective coordinates is provided in ref . , we provide a brief summary of the relevant mathematical relations and definitions here . recall that the potential energy of the system of interacting particles in a periodic simulation box is defined in eq . ( [ eq : phi ] ) . the collective density variables @xmath11 are given by @xmath16 where @xmath17 is the position of the @xmath18 particle . the wave vectors for a periodic box correspond to linear combinations of the reciprocal lattice vectors associated with the periodic box . for a rectangular box of dimensions @xmath19 and @xmath20 , those wave vectors are @xmath21 $ ] , where @xmath22 and @xmath23 are integers . the structure factor @xmath24 , proportional to the intensity of scattering of radiation , is related to the collective density variables via @xmath25 where @xmath26 is the appropriate ensemble average . when angularly averaged , the structure factor @xmath27 is only dependent on the wavenumber @xmath28 and zero for @xmath29 , any configuration for which @xmath27 is constrained to be zero for all @xmath30 is a ground state . for such potentials , it is useful to refer to the dimensionless parameter @xmath31 as the fraction of degrees of freedom that are constrained , where @xmath32 is the number of independent wave vectors @xmath33 and @xmath34 is the total degrees of freedom.@xcite in two dimensions @xmath14 is limited to the range @xmath35 . this limit was discussed in ref . [ 2 ] and arises from the inability to suppress bragg scattering of the triangular lattice.@xcite here , we fix @xmath36 , and therefore @xmath14 is inversely related to the number density @xmath37 . some relations between @xmath14 and @xmath38 are given in ref . [ 2 ] . three structural regimes exist for ground - state configurations in two dimensions . for @xmath39 , ground states are disordered . for @xmath40 , they are wavy crystalline , while for @xmath41 , crystalline structures are the most viable ground - states configurations identified via numerical search techniques.@xcite we have observed that quenching systems with @xmath42 always resulted in a ground state . the energy landscape is evidently devoid of local potential - energy minima above the ground state when @xmath42 and thus inherent - structure analysis is limited to those systems with @xmath43 . in the infinite - volume limit and its corresponding fourier transform @xmath2 , the @xmath0-space overlap potential , as adapted from ref . [ 2].,scaledwidth=80.0% ] we continue to use the @xmath0-space overlap potential , introduced by torquato and stillinger , @xcite since it is exactly representable in real and reciprocal space , and is sufficiently short - ranged in real space for computational purposes ; see fig . [ fig : vkvr ] . the @xmath0-space overlap potential is proportional to the intersection area between two disks of diameter @xmath3 with centers separated by @xmath0 , and hence for @xmath44 is @xmath45,\ ] ] and zero for all @xmath46 . in the infinite - volume limit , the associated real space pair potential is @xmath47 ^ 2,\ ] ] where @xmath48 is the bessel function . henceforth , we refer to this @xmath1 as the `` overlap potential . '' the potential @xmath1 is bounded at @xmath49 and behaves as @xmath50 for large @xmath51 . in this paper , we choose to report a scaled potential energy @xmath52 , omitting the structure - independent contribution at @xmath53 and removing the additive constants , i.e. , @xmath54 so that ground state energies are equal to zero regardless of @xmath14 for @xmath55 . this new scaled energy provides a sense as to how much a configuration differs energetically from a ground state , and is related to the actual potential energy , choosing @xmath56 , via @xmath57 \\ & = & \frac{\rho}{2 } - \frac{1}{2\omega}\sum_{{{\bf k}}}v({{\bf k } } ) + \varepsilon\end{aligned}\ ] ] note that the structure - independent terms sum to be nonnegative because @xmath1 is a nonnegative function . we choose the cutoff for ground states to be those configurations in which @xmath58 . we also compute a stealthiness metric @xmath59 as the average of the values of the structure factor for all @xmath60 , defined to be @xmath61 this metric is a measure of the stealthiness because it vanishes to zero for completely transparent systems but will remain nonzero for nonstealthy systems . while there is a numerical limitation of @xmath62 for double precision computing , there are certain practical limits on stealthiness . because of the weights assigned by @xmath2 for @xmath0 near @xmath3 , the metric @xmath59 may not necessarily be suppressed to zero for ground states due to numerical precision of the quantity @xmath63 . therefore , the stealthiness metric is best used as a relative comparison of stealthiness between configurations and not as a determination of a configuration as a ground state . our method to generate inherent structures follows that of sastry and coworkers,@xcite which is based on the initial algorithms of stillinger and weber.@xcite initially , we simulate the atomic motions of the system interacting via the @xmath0-space overlap potential in eq . ( [ eq : phi ] ) by integrating the newtonian equations of motions , assuming unit mass , @xmath64 we use the velocity verlet algorithm@xcite with a time step @xmath65 of 0.4 , which is chosen so as to accurately conserve total energy in the constant @xmath66 ensemble . throughout the dynamical simulation , configurations were sampled and subjected to a quenching of the potential energy . here , the sampled configuration is used as the initial condition for a conjugate gradient minimization . upon termination of the conjugate gradient quenching , we subject the system to an additional quenching via the minop algorithm.@xcite we previously reported the efficiency of the minop algorithm compared to the conjugate gradient method.@xcite we add this final quench for reasons of increased efficiency and precision compared to simply using the conjugate gradient method to high precision . many of the inherent structures that we have found have very slight differences in the scaled potential energy @xmath52 that can be on the order of magnitude @xmath67 . our implementation of the conjugate gradient method becomes inefficient at these tolerances . while there is a possibility that the minop algorithm can traverse to other capture basins , we suspect that the system is quenched sufficiently deep into a potential - energy well after the conjugate gradient step that the minop algorithm will not push the system into another capture basin . we initialize the dynamical trajectories at a temperature associated with the liquid state . here , we report the dimensionless temperature @xmath68 , where @xmath69 is the boltzmann constant and @xmath70 is the simulation temperature . we previously reported that the transitions from ground states to highly disordered liquids occur in the range of @xmath71-@xmath72 , depending on @xmath14.@xcite we initialize our simulations at @xmath73 . configurations are sampled every @xmath74 time steps until at least fifty are obtained . then the velocities of the system are rescaled to a lower temperature by @xmath75 , typically a value of 0.00015 , and the configurations are sampled again at the reduced temperature . the rate at which the system is effectively cooled becomes @xmath76 . we have explored a range of coolings rates from @xmath77 with @xmath74 varying from 3 to 750 . surprisingly , as discussed in sec.[sec : cooling ] , while the properties of the thermal structures are sensitive to the cooling rate , the properties of the inherent structures are not . to construct pathways between ground - state configurations and inherent structures for @xmath78 , we use the `` nudged - elastic - band algorithm '' used for finding minimum - energy transition paths.@xcite this method requires as an input two local minima that are nearby in the energy landscape . we generate this pair of structures by initializing a known ground - state configuration ( obtained by the search methods in ref . [ 3 ] ) , assigning a small temperature to the system , and integrating the equations of motion . the configurations are quenched by the conjugate gradient / minop minimization every five time steps . a simulation run is terminated when the quenching of a configuration produces a local minimum in a new capture basin . this new inherent structure is presumed to be near the original ground state in configuration space . we then apply the nudged - elastic - band ( neb ) algorithm@xcite to the ground state / inherent structure pair . the algorithm discretizes the path between two local minima in a potential - energy landscape into `` beads , '' or image configurations . these beads , representing configurations , are connected via harmonic springs of zero natural length . then a minimization algorithm is applied to minimize simultaneously the force parallel to the string of beads and the force perpendicular to the true force . the algorithm bends the string of beads around the `` hills '' in the energy landscape . the end result is a discrete path that provides the minimum - energy path from one potential - energy minimum , through a low - order saddle point , and to the other potential - energy minimum . while a large number of `` beads '' allows for a more refined minimum - energy pathway , we find that using fifty beads serves our purposes well by yielding a smooth transition pathway and a reasonable approximation to the saddle point . occasionally , there are other inherent structures in the vicinity of the straight - line interpolation between the inherent structure and ground state used as an initial condition . in these cases , the neb algorithm will not produce a smooth transition from the ground state to the original inherent structure , and therefore we refine our initial ground state / inherent structure pair . for example , when using the triangular lattice as an initial condition for say @xmath79 , there are number of zero - energy modes that the system can traverse during the molecular dynamics trajectories . when applying the neb algorithm to a triangular lattice / inherent structure pair , the pathway has a preference to bend toward the zero - energy path before moving uphill toward the saddle point . this caused the energy of the transition - state pathway to appear rugged since the string of beads was pushed near a different inherent structure . in these cases , we searched the string of beads for a new ground - state / inherent structure pair . we minimized the potential energy of each bead until we found a new ground state / inherent structure pair closer together than the original pair and applied the neb algorithm again . this method does not guarantee that we find the closest inherent structure to a ground state nor does it guarantee that the saddle point is a first - order ( single negative eigenvalue ) saddle . for @xmath42 , we examine particle rearrangements from rsa configurations to ground states . to generate rsa packings , disks of diameter @xmath80 are randomly , sequentially , and irreversibly placed inside a simulation cell so that they do not overlap with any other particles . the process is saturated when additional particles can not be placed without overlapping other particles . in practice , this occurs after a very large number of attempted particle insertions are rejected . here , we terminate the process after 1.5 million attempted particle placements . since @xmath3 is fixed to be unity , we must select a target number density @xmath81 that corresponds to our target @xmath14 . using a desired number of particles in a saturated system of 750 particles , we then identify the appropriate system area . the simulation cell is chosen to be a square box . we can then assign the particles a diameter @xmath80 so that upon saturation of the rsa process , we achieve close to the desired number density . for rsa processes in two dimensions , the saturation packing fraction @xmath82 is 0.547.@xcite the assigned diameter becomes @xmath83 . in our rsa patters , the number of particles did not match the target number of particles due to finite system effects . given the dimensions of the box and the number of particles , we can then assign the appropriate wave vectors and @xmath14 value to the system . the potential energy of the system is then quenched using the conjugate gradient method followed by a quenching using the minop algorithm as was detailed above . to characterize inherent structures , we use several quantitative metrics . the structure factor @xmath27 , _ cf . _ eq . ( [ eq : sk ] ) , provides structural information on the long - ranged ordering of the system . defined above , @xmath52 is a scaled potential energy and @xmath59 is a metric for the extent of stealthiness . in addition , we use the small-@xmath0 features of @xmath27 to compare the extent of hyperuniformity between structures . hyperuniform systems have the feature that @xmath27 vanishes as @xmath0 approaches zero . we fit the small-@xmath0 regime ( _ i.e. _ @xmath84 ) to a log - polynomial equation @xmath85 the parameter @xmath86 can be considered to be a metric for the extent of hyperuniformity , where a value that diverges to @xmath87 indicates that a system is hyperuniform . this metric is best used for relative comparisons among systems since numerical precision is limited . for example , one could say system a is more hyperuniform than system b because it has a lower value of @xmath86 . however , @xmath86 is not be a sufficient test to determine if a configuration is absolutely hyperuniform ( _ i.e. , _ one should refrain from saying system a is absolutely hyperuniform ) . we also employ the bond - order parameter @xmath88 defined as @xmath89 where @xmath90 is the angle between two particles with respect to a fixed , but arbitrary , coordinate axis , to quantify the local orientation order . particles are considered `` bonded '' if @xmath91 , chosen so as to include nearest - neighbors . for the perfect triangular lattice , @xmath88 has a value of unity while for the ideal gas @xmath88 vanishes . wavy crystals have value of @xmath88 that can range from about 0.4 to 0.8 . we introduce a configurational proximity parameter @xmath92 defined as @xmath93 where @xmath94 and @xmath95 are the positions of the rsa configuration and ground state configuration respectively and @xmath96 is the mean - nearest - neighbor in the rsa configuration . this metric @xmath92 quantifies the closeness of two different particle configurations . in what follows , we will apply it to characterize the relative particle displacements required for a configuration to relax to the ground - state energy . before discussing the results of the quenching and cooling simulations , we briefly discuss some aspects of system - size effects related to the inherent structures of this system . while we previously showed that system - size effects were negligible when characterizing the ground states@xcite and the equilibrium properties,@xcite we find that there are appreciable system - size effects when considering the inherent structures and energy landscape . we have compared inherent structures consisting 418 particles to 2340 particles sampled from the same temperature . in general , larger systems provide a slightly deeper quenching of the potential energy . in addition , the bond - order parameters for larger systems tend to be less than that of smaller systems . this is expected since larger systems often have various polycrystalline - like domains that impart destructive interference on @xmath88 . the fraction of quenchings from a fixed temperature that achieve the ground state varied considerably across system sizes . while the quenching of the energy of small systems occasionally resulted in achieving a ground - state structure , quenches of the large systems never resulted in a ground - state structure . the measure of the ground - state manifold becomes increasingly small as the system size increases . given that the number of inherent structures increases exponentially with respect to system size,@xcite it is unsurprising that larger systems would not sample the capture basins for ground states as frequently as smaller systems . system - size effects have been also been identified in the binary lennard - jones system.@xcite due to the system - size effects , we should not extrapolate true dynamical information from our results . here , we present the analysis for 418-particle systems , noting that the trends across @xmath14 values are similar to those in the 2340-particle system . the properties of the excited - state ( or thermalized ) structures behave as expected . the thermodynamic properties lagged the equilibrium values as the system was cooled , and for slower cooling rates , the lag in thermodynamic properties was smaller than that for fast cooling rates . however , we find that the rate at which the system was cooled provided no distinguishable effects on the properties of the inherent structures when compared across various cooling rates . this should be contrasted with conventional glass - forming systems such as binary lennard - jones where systems cooled under slow rates found deeper wells in the energy landscape than systems cooled under faster rates.@xcite in our previous research , we have identified the approximate range for transitions from crystalline or wavy - crystalline to liquids to be in the temperature range of @xmath97.@xcite we find that the temperature at which the inherent structures were sampled from had no distinguishable effect on the thermodynamics properties or structures . as the system was cooled , the systems hopped across various capture basins without regard to temperature . figure [ fig : pottime2 ] exhibits the dynamics of systems as they sample inherent structures for @xmath14=0.6004 and 0.8086 for 418 particles cooled at a rate of @xmath98=@xmath99 . the figure shows the potential energy of inherent structures sampled every 30 time steps . in the figure , there are two classes of behaviors . for @xmath79 , the system samples higher - energy inherent structures and also frequently samples ground states , those systems with @xmath100 . for @xmath101 , the system samples mostly higher - energy structures and occasionally , though far less frequently than for smaller @xmath14 , samples ground states . for both values of @xmath14 , the system hops to a new inherent structure between samplings . there were no consecutive samples that were identical except for very low temperatures far below the melting point ( _ e.g. , _ @xmath102 ) . the rate at which the system samples capture basins does not appear to slow down significantly when the temperature is reduced as one might expect . this phenomenon occurred for all cooling rates we tested . in addition , even when sampling inherent structures every three time steps , the system never remained in the same basin for consecutive samplings . note that the depths of the quenches for ground - state structures were deeper for @xmath101 than for @xmath79 . this demonstrates the issue that arises with finite - precision computing because these scaled energies should vanish to zero in a mathematical sense . in a practical sense , systems with energies below this threshold and orders of magnitude below the clusters of systems higher up are ground states . we find that the structural features of the inherent structures did not vary with temperature either . figure [ fig : skavgtemp ] shows the structure factor for inherent structures with @xmath79 ensemble - averaged for several temperatures . however , when observing the structure factors for individual configurations , there is some small variability in the shape of @xmath27 . however , there are no clear correlations with temperature . in the figure , the structure factors associated with each temperature overlay nearly perfectly . even at very low temperatures ( @xmath102 ) , the structure factor for the inherent structures does not differ from those obtained from a liquid state . for familiar systems , as the temperature is reduced , the potential energy of the inherent structures typically becomes lower as the system continues to reduce the potential energy through structural rearrangements , sampling deeper and deeper basins . however , we find as the temperature is reduced , the energy of the inherent structures does not appear to decrease . this reveals that the energy landscape is relatively flat and insensitive to temperature . the heights of the barriers , which we explore in sec . [ sec : mapping ] , are therefore expected to be relatively small . = 0.6004 and @xmath5=418 and at various dimensionless temperatures @xmath103 - 0.00195 ( circles ) , 0.00135 ( squares ) , 0.0075 ( diamonds ) , and 0.00015 ( triangles ) . each temperature is ensemble averaged over 50 configurations sampled from a molecular dynamics trajectory . despite the range of temperatures , temperature has little effect on the inherent structures.,scaledwidth=60.0% ] in fig . [ fig : gsfrac ] , we plot the frequency in which the cooling and quenching procedure yielded a ground state for 418 particles . the plot shows the fraction of inherent structures that were ground states as a function of @xmath14 for 3000 inherent structures for each @xmath14 . we have found that the temperature from which the inherent structures are sampled does not have a significant effect on the rate at which the ground state is achieved . however , the value of @xmath14 affects the frequency at which the ground state is found . in general , as @xmath14 is increased from 0.6 to 0.9 , the rate at which the ground states are found is reduced . however , for the `` disordered - ground - state '' region , @xmath104 , ground states were found infrequently . for example , in fig . [ fig : gsfrac ] , the frequency in which ground states are found for @xmath105 is much smaller than that of @xmath79 . this suggests that the relative availability of the ground - state manifold is small for @xmath106 , becomes larger in the wavy - crystalline region ( @xmath107 ) , and then diminishes as the terminal @xmath14 of 0.9 is reached . this shows that changes in @xmath14 ( and equivalently , density for fixed @xmath3 ) , fundamentally affect the shape and size of the ground state manifolds . an expansion or compression of the system does not simply rescale the energy landscape while maintaining its overall shape . systems containing 2340 particles never achieved ground state energies , presumably because of the exponential growth in the number of inherent structures with system size.@xcite and @xmath108 that are ground states ( _ i.e. , _ scaled potential energy @xmath58 ) . data represent the average of four independent cooling runs with 750 samples for each run . @xmath14 values of 0.6004 and 0.7033 achieved the highest fraction of ground states . , scaledwidth=60.0% ] figures [ fig : inhsmall ] and [ fig : inhlarge ] show inherent structures obtained via the cooling and quenching procedure . these structures display features that are common among inherent structures . upon examining the inherent structures , we found several structural motifs - five - particle rings , wavy grain boundaries , and vacancy - interstitial defects . the frequency of these features varies depending on @xmath14 . for @xmath109 , the five - particle ring , as demonstrated in fig . [ fig : inhsmall ] , is the most prevalent feature . the five - particle ring arises due to the local interactions of constituent particles . the nearest- and second - nearest neighbor distances lie within the first and second wells of @xmath1 respectively . the ratio of the relative distances creates a frustration that favors the local five - particle ring . evidently , these local forces are more dominant than the long - range forces that promote stealthiness . for @xmath110 , a common feature of inherent structures is a wavy grain boundary as evident in figs . [ fig : inhsmall ] and [ fig : inhlarge ] . in these images , the grain boundaries manifest themselves in a more wavy form than one might see in a lennard - jones inherent structure , in which particles prefer to align in straight lines . this @xmath14 range is associated with wavy - crystalline ground states , which appear as a uniformly sheared triangular or square lattice . this behavior evidently manifests itself in the inherent structures where the grain boundaries appear to have some waviness . for these @xmath14 values , the long - range nature of the potential is sufficiently strong to inhibit five - particle rings in favor of the wavy grains . lastly , for @xmath111 , the last image in fig . [ fig : inhlarge ] , the inherent structures either have a large grain boundary of perfectly aligned particles , vacancy - interstitial pairs , or both . these inherent structures also exhibit polycrystallinity . in the cases of @xmath107 , the multiple wavy domains appear to be in structural conflict with each other . this is best demonstrated by fig.[fig : inhsmall ] for @xmath79 where several domains appear to meet in the middle of the system . this conflict evidently `` jams '' the system and prevents it from reaching the ground state . ) for @xmath112 ( left ) , 0.5622 ( middle ) , and 0.6004 ( right ) for 418 particles . these structures demonstrate the five - particle clusters that are common in inherent structures for these @xmath14 values . the five - particle rings are highlighted by a surrounding circle . the buildup of wavy grains are visible for @xmath79 . , title="fig:",scaledwidth=30.0% ] ) for @xmath112 ( left ) , 0.5622 ( middle ) , and 0.6004 ( right ) for 418 particles . these structures demonstrate the five - particle clusters that are common in inherent structures for these @xmath14 values . the five - particle rings are highlighted by a surrounding circle . the buildup of wavy grains are visible for @xmath79 . , title="fig:",scaledwidth=30.0% ] ) for @xmath112 ( left ) , 0.5622 ( middle ) , and 0.6004 ( right ) for 418 particles . these structures demonstrate the five - particle clusters that are common in inherent structures for these @xmath14 values . the five - particle rings are highlighted by a surrounding circle . the buildup of wavy grains are visible for @xmath79 . , title="fig:",scaledwidth=30.5% ] ) for @xmath113 ( left ) , 0.8086 ( middle ) , and 0.8995 ( right ) for 418 particles . these structures demonstrate the wavy grains and vacancy - interstitial defects common to inherent structures for @xmath114 . as @xmath14 increases , the inherent structures are more ordered , though grain boundaries prevent the systems from minimizing the energy.,title="fig:",scaledwidth=30.0% ] ) for @xmath113 ( left ) , 0.8086 ( middle ) , and 0.8995 ( right ) for 418 particles . these structures demonstrate the wavy grains and vacancy - interstitial defects common to inherent structures for @xmath114 . as @xmath14 increases , the inherent structures are more ordered , though grain boundaries prevent the systems from minimizing the energy.,title="fig:",scaledwidth=30.0% ] ) for @xmath113 ( left ) , 0.8086 ( middle ) , and 0.8995 ( right ) for 418 particles . these structures demonstrate the wavy grains and vacancy - interstitial defects common to inherent structures for @xmath114 . as @xmath14 increases , the inherent structures are more ordered , though grain boundaries prevent the systems from minimizing the energy.,title="fig:",scaledwidth=30.0% ] observing the structure factor for inherent structures shows that inherent structures with smaller @xmath14 are more stealthy and more hyperuniform than those at higher @xmath14 . however , inherent structures that are not ground states were neither completely stealthy nor hyperuniform . in fig . [ fig : skavginh ] , we show the ensemble - averaged @xmath27 for 50 inherent structures sampled at @xmath115 for various @xmath14 . for the smallest @xmath14 value , the small-@xmath0 tail of the structure factor initially decreases and then rapidly increases as @xmath0 is increased toward unity . the shape of the small-@xmath0 tail of @xmath27 remains constant but is shifted to higher values ( less hyperuniform ) as @xmath14 increases to @xmath116 . however , as @xmath14 leaves the disordered - ground - state regime , the shape of the curve changes , as exemplified by the curves for @xmath14=0.6004 and 0.7033 . in fig . [ fig : skavginh ] , the curve for @xmath79 demonstrates the new shape . for the small-@xmath0 region , @xmath27 initially has a negative slope which quickly turns positive , and nearly flattens before diverging near @xmath3 . the shape of this ensemble - averaged @xmath27 is due to the relatively large number of ground states that arise in the inherent structure analysis for this @xmath14 . for @xmath79 , nearly 40% of inherent structures are ground states , and this @xmath27 represents the average of the ground - state @xmath27 and the inherent structure @xmath27 . a similarly large fraction of inherent structures are ground states for @xmath14=0.7033 . if the ground states are removed from the ensemble - averaging , the structure factor appears more like those associated with @xmath14=0.8086 and 0.8995 . for these @xmath14 , @xmath27 has a initial negative slope that turns positive near @xmath0=0.5 , which then diverges near @xmath3 . it is interesting to note that the way in which systems get `` stuck '' in an inherent structure for @xmath14 in the disordered regime is different than that of the wavy - crystalline regime . for @xmath106 , the systems struggle most to minimize @xmath27 for @xmath117 . the length scale associated with @xmath117 corresponds to local interactions in real space ( _ i.e. , _ on the order of a few particle diameters ) . this manifests itself with the five - particle rings . for @xmath107 , the systems have difficulty minimizing @xmath27 near zero and unity . the @xmath0-values near zero represent longer - range interactions than the @xmath0-values near unity . the inherent structures for @xmath107 are therefore `` stuck '' on a longer length scale . it is no surprise then that the inherent structures appear locally ordered but are frustrated with a grain boundary . ensemble averaged over 50 configurations sampled from a molecular dynamics trajectory at @xmath115 for a system of 418 particles . as @xmath14 increases , the inherent structures become less stealthy . also , the extent of hyperuniformity decreases when @xmath14 is increased . , scaledwidth=60.0% ] it is obvious that as @xmath14 increases the extent of hyperuniformity of the inherent structures diminishes . this is shown in an ensemble sense in fig.[fig : skavginh ] . however , we have also shown this general relationship for individual structures . in fig . [ fig : c0stealth ] , we display the hyperuniformity coefficient @xmath86 for the log - polynomial fit to @xmath27 in eq.([eq : logpoly ] ) . in the figure , each data point represents one inherent structure . there are two clusters of points , where the cluster of points on the lower left are typically ground states where @xmath27 vanishes as @xmath0 approaches zero , but due to numerical precision the hyperuniformity coefficient @xmath86 remains finite . in the cluster in the upper right of fig . [ fig : c0stealth ] , there is a linear relation between @xmath86 and @xmath118 . also , it is clear that as @xmath14 increases , the hyperuniformity parameter @xmath86 also increases . a few outlying structures are present for @xmath14=0.7003 , 0.8086 and 0.8995 . these structures appear as triangular lattices with very slight waves or shear applied to them , a sufficient amount that they are not ground states , but still have very small limiting values of @xmath27 as @xmath0 approaches zero . versus stealthiness metric @xmath59 for inherent structures generated from the md and quench procedure . there is a clear relation between the hyperuniformity coefficient @xmath86 and the stealthiness metric @xmath59 , although there exist outlying configurations . there is also a positive relation between @xmath86 and @xmath14 . those configurations in the lower left with @xmath119 are typically ground states . 800 configurations are shown for each @xmath14 value.,scaledwidth=60.0% ] the stealthiness of the inherent structures , as characterized by the metric @xmath59 , in general does not have a relation to the six - fold bond - order orientiational order of a system . figure [ fig : potbond2 ] shows the bond - order parameter @xmath88 as a function of stealthiness @xmath59 for various inherent structures . for each @xmath14 value , there is a range of available @xmath59 and @xmath88 . initially , for @xmath14 near 0.5 , these ranges are rather narrow , and as @xmath14 increases , these ranges become much broader . for large @xmath14=0.8086 there is a large diversity of @xmath88 and @xmath59 available for inherent structures . however , for @xmath14=0.8995 , there is a narrow band of available @xmath59 but a large range of @xmath88 depending on the number and orientation of opposing crystalline domains . the group of structures for @xmath14=0.8995 and @xmath120 are those structures with only vacancy - interstitial defects . versus stealthiness metric @xmath59 for inherent structures generated from the md and quench procedure . as @xmath14 increases , there is an increasing degree of six - fold ordering . however , for large @xmath14 , the range of available @xmath88 extends larger due to the polycrystalline nature of the configurations . 800 configurations are shown for each @xmath14 value . for clarity , we omit those structures with @xmath121 so that the plot can provide more detail about the inherent structures . , scaledwidth=60.0% ] the bond - order parameter also shows considerable variation when plotted against the hyperuniformity parameter @xmath86 . figure [ fig : c0bond ] shows @xmath86 as a function of @xmath88 for various @xmath14 values . as @xmath14 increases , the variation in @xmath88 grows while @xmath86 generally falls into a narrow range of values . this figure demonstrates that hyperuniformity does not necessarily require local six - fold ordering . systems with @xmath14=0.7033 have the largest range of available local six - fold ordering for more hyperuniform structures because @xmath88 ranges from as low as 0.1 to just below unity . versus bond - order parameter @xmath88 for inherent structures generated from the md and quench procedure . despite the broad range of six - fold ordering at larger @xmath14 , the level of hyperuniformity remains consistent across @xmath14 values . 800 configurations are shown for each @xmath14 value . for clarity , we omit those structures with @xmath122 so that the plot can provide more detail on the inherent structures . , scaledwidth=60.0% ] for ( a ) @xmath123 , ( b ) @xmath79 and 0.7033 , and ( c ) @xmath101 and 0.8995 . the local maximum corresponds to a saddle point in the energy landscape . , title="fig:",scaledwidth=31.0% ] for ( a ) @xmath123 , ( b ) @xmath79 and 0.7033 , and ( c ) @xmath101 and 0.8995 . the local maximum corresponds to a saddle point in the energy landscape . , title="fig:",scaledwidth=32.0% ] for ( a ) @xmath123 , ( b ) @xmath79 and 0.7033 , and ( c ) @xmath101 and 0.8995 . the local maximum corresponds to a saddle point in the energy landscape . , title="fig:",scaledwidth=31.0% ] the nudged - elastic - band algorithm was used to determine the minimum - energy path from a ground - state structure to an inherent structure nearby in configurational space . we find that the height of the barrier between the ground state and the inherent structure generally increases with @xmath14 . however , there are instances that deviate from this general situation . figure [ fig : barriers ] illustrates the minimum - energy paths found for various @xmath14 from the ground state ( point zero on the reaction pathway ) to the inherent structure ( last point on the reaction pathway ) . the pathway is discretized over fifty images . in all the cases , the landscape is locally quadratic near the ground state and inherent structure . the maximum for each pathway represents the saddle point in the energy landscape separating the ground state and inherent structures . connections from the triangular lattice to inherent structures were challenging to construct because of the zero - energy modes present in triangular - lattice systems . the nudged - elastic - band algorithm encountered difficulty in traversing through zero - energy valley and then uphill toward a saddle point . in all cases except @xmath14=0.8995 , we report transitions from non - lattice ground states to inherent structures . for @xmath124 ( left in fig . [ fig : barriers ] ) , the barrier height , @xmath125 is significantly large compared to the difference in the energies of the ground - state and inherent structures ( _ i.e. , _ @xmath126 ) . when @xmath14 becomes larger , as in the case for @xmath14=0.7033 and higher , middle and right in fig . [ fig : barriers ] , @xmath125 is nearly equivalent in height . for the case of @xmath14=0.8995 , the minimum - energy path appears to flatten out as it approaches the inherent structure . in reality , there is a maximum energy along the path , although it is difficult to discern due to the scaling in the plot . in the cases where the energy of the inherent structure is nearly equal to the barrier height , small perturbations from the inherent structure can allow the system to relax to the ground state . on the other hand , for smaller @xmath14 , relatively higher - energy perturbations are necessary for the inherent structure to climb the energy barrier and fall to the ground state . while there are exceptions to this phenomenon , they usually occur when the inherent structure and ground state are not nearby in the energy landscape ( _ i.e. , _ there is a another inherent structure closer to the ground state ) . along these paths , the configurational proximity metric @xmath92 monotonically decreases along the path from the ground state to the inherent structure . the metric @xmath127 for the proximity per particle between the inherent structure and the ground state is of the order 0.1 . on average , particles are collectively displaced by 10% of the nearest neighbor spacing to achieve a ground state . . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] + . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] . the transitions are from the set described in fig . [ fig : barriers ] . , title="fig:",scaledwidth=30.0% ] in figs . [ fig : path ] and [ fig : path2 ] , we display the ground state , saddle point , and inherent structure and the associated voronoi diagrams for @xmath14=0.5383 and 0.7033 . these are the same paths shown in fig.[fig : barriers ] . these systems are representative of the types of local rearrangements found in the transition from ground states to inherent structures . in fig . [ fig : path ] , the ground state has wavy - crystalline characteristics where there is alignment in one direction . the voronoi diagram appears to have highly ordered arrangement of polygons . in the saddle point images , one can see the appearance of the five - particle rings . in the inherent structure , the defect appears to be localized , however , the defects are more easily discerned in the voronoi diagram . for @xmath14=0.7033 , the buildup of the grain boundary is clear in fig.[fig : path2 ] . the ground state is wavy - crystalline and the saddle point shows the appearance of a vertical and horizontal grain . the inherent structure shows the apparent mismatch between lines of particles that is responsible for the inability to reach the ground state . the voronoi diagrams in this case visually display the source of the frustration in the system . while far from the grain , the polygons in the voronoi digram for the inherent structure are quite regular and ordered , those near the grain boundary have much more variation in shape . for @xmath14=0.8995 , the rearrangements found in the transition from a perfect lattice to one with a vacancy - interstitial pair are very local ( albeit collective ) , only involving the few particles immediately surround the interstitial - vacancy pair . in all cases observed , the rearrangements from ground states to inherent structures involve only local , but collective , rearrangements ( within a few particle diameters ) . as highlighted here , they typically involve the formation of a grain boundary or a five - particle ring . configurations generated by random sequential addition near the saturation densities are known to suppress long - wavelength scattering , described as `` ultratransparent,''@xcite though images of the structure factor for the rsa configuration demonstrates that it is not hyperuniform and not absolutely stealthy.@xcite from ref . , as @xmath0 approaches zero , the structure factor approaches a value of 0.059 , corresponding to a value of the hyperuniformity parameter @xmath86 of -1.23 . in figs . [ fig : rsax0991 ] and [ fig : rsax3657 ] , we display the initial rsa configurations and the final ground - state configurations as obtained from the collective coordinate approach , for @xmath14 values of 0.0991 and 0.3657 . each figure also displays the corresponding voronoi tessellation derived from the point patterns . in both cases when observing the dynamics of the minimization algorithm , one observes a general spreading of particles due to the repulsion for small @xmath51 . these collective rearrangements are very local and there are no large global rearrangements involved . while every particle is displaced collectively from their initial location , each particle is displaced by a small fraction of the mean - nearest neighbor distance , as measured by the proximity metric per particle , @xmath127 . in fig . [ fig : rsax0991 ] , one can observe similar arrangements of particles in both the rsa configuration and the ground - state configuration . these images would be difficult to distinguish visually . several structural features remain in tact aside from the general spreading of particles . in the voronoi tessellations of the rsa configuration , there are areas where the local density appears higher than in other areas . however , in the tessellation of the ground state , these areas of higher local density have been relaxed away . the density appears much more uniform in the ground state than the rsa configuration . however , the visual difference between the two images are subtle . in fig.[fig : rsax3657 ] , the spreading remains local , although it is clear that the repulsion is stronger . the particles , on average , travel a larger distance and fewer structural features of the rsa pattern are maintained . the voronoi tessellation of the ground state appears to have more uniform cell sizes than the rsa configuration , though the differences are subtle . and @xmath128 and ( bottom ) the associated voronoi diagrams . the left images are the initial rsa configurations and the right images are the ground states . the trajectory from the rsa configuration to the ground state appears to arise from a gentle repulsion . the diameters of the particles correspond roughly to the assigned rsa diameter . the dark lines represent the system box.,title="fig:",scaledwidth=40.0% ] and @xmath128 and ( bottom ) the associated voronoi diagrams . the left images are the initial rsa configurations and the right images are the ground states . the trajectory from the rsa configuration to the ground state appears to arise from a gentle repulsion . the diameters of the particles correspond roughly to the assigned rsa diameter . the dark lines represent the system box.,title="fig:",scaledwidth=40.0% ] and @xmath128 and ( bottom ) the associated voronoi diagrams . the left images are the initial rsa configurations and the right images are the ground states . the trajectory from the rsa configuration to the ground state appears to arise from a gentle repulsion . the diameters of the particles correspond roughly to the assigned rsa diameter . the dark lines represent the system box.,title="fig:",scaledwidth=40.0% ] and @xmath128 and ( bottom ) the associated voronoi diagrams . the left images are the initial rsa configurations and the right images are the ground states . the trajectory from the rsa configuration to the ground state appears to arise from a gentle repulsion . the diameters of the particles correspond roughly to the assigned rsa diameter . the dark lines represent the system box.,title="fig:",scaledwidth=40.0% ] and @xmath129 and ( bottom ) the associated voronoi diagrams . the left images are the initial rsa configurations and the right images are the ground states . the trajectory from the rsa configuration to the ground state appears to arise from a gentle repulsion . the diameters of the particles correspond roughly to the assigned rsa diameter . the dark lines represent the system box.,title="fig:",scaledwidth=40.0% ] and @xmath129 and ( bottom ) the associated voronoi diagrams . the left images are the initial rsa configurations and the right images are the ground states . the trajectory from the rsa configuration to the ground state appears to arise from a gentle repulsion . the diameters of the particles correspond roughly to the assigned rsa diameter . the dark lines represent the system box.,title="fig:",scaledwidth=40.0% ] and @xmath129 and ( bottom ) the associated voronoi diagrams . the left images are the initial rsa configurations and the right images are the ground states . the trajectory from the rsa configuration to the ground state appears to arise from a gentle repulsion . the diameters of the particles correspond roughly to the assigned rsa diameter . the dark lines represent the system box.,title="fig:",scaledwidth=40.0% ] and @xmath129 and ( bottom ) the associated voronoi diagrams . the left images are the initial rsa configurations and the right images are the ground states . the trajectory from the rsa configuration to the ground state appears to arise from a gentle repulsion . the diameters of the particles correspond roughly to the assigned rsa diameter . the dark lines represent the system box.,title="fig:",scaledwidth=40.0% ] we have quantified the differences between the rsa structures and the ground states for several rsa configurations at various @xmath14 values . table [ tab : rsa ] displays the parameters of these systems as well as the differences in potential energy @xmath13 between the rsa system and ground state , the stealthiness metric for rsa configurations @xmath130 , and the configurational proximity metric @xmath92 . the stealthiness metric @xmath59 for ground states vanishes identically to zero . there are some variations in energies and metrics that are attributed to finite system size . structural features of the rsa patterns can vary across samples , and these variations show up in table [ tab : rsa ] particularly for small @xmath14 . in general , the difference between the potential energy of the rsa system and the ground state increases uniformly with @xmath14 , but is on the order of @xmath131 . the stealthiness metric also increases uniformly with @xmath14 as is expected , since decreasing the density of rsa configurations will rescale the structure factor to be less stealthy . the configurational proximity metric @xmath92 increases uniformly with @xmath14 , ranging on a per particle basis of 0.00621 to 0.01478 for these @xmath14 . these metrics show that particles in the rsa configuration need to be collectively displaced locally by a small fraction of the mean - nearest - neighbor distance in order to produce stealthy , hyperuniform systems . .[tab : rsa ] comparison of rsa configurations to their associated ground states . [ cols="<,<,<,^,^,^,^,^",options="header " , ] in this paper , we have shown that the energy landscape associated with the @xmath0-space overlap potential is relatively flat and devoid of deep energy wells . this is to be contrasted with lennard - jones systems , which possess rugged energy landscapes . the sampling of inherent structures from the energy landscape is independent of the temperature from which the system was sampled and the rate at which the system was cooled . five - particle rings , which are related to the local interactions between particles , disrupt the ability of a system to become a stealthy ground state , while grain boundaries , which arise for @xmath14 values in the wavy - crystalline regime , are attributed to the longer - range interactions in a system . while the hyperuniformity parameter @xmath86 and the stealthiness parameter @xmath59 are positively correlated with each other , the bond - order parameter @xmath88 can display a broad range of characteristics for various @xmath14 values . we have used the nudged - elastic - band algorithm to show that local particle rearrangements can allow systems to be perturbed over a relatively small energy barrier from an inherent structure to a ground state . these rearrangements correspond to a low configurational proximity metric @xmath92 . lastly , we have shown that highly local collective particle displacements , as quantified via the configurational proximity metric @xmath92 , are sufficient to convert a rsa configuration into a stealthy , hyperuniform ground state . while these studies represent another contribution to understanding the nature of these unusual long - ranged potentials , there are some natural extensions to this work that can further illuminate the relationship between particle interactions , stealthiness , and hyperuniformity . we have used simple metrics @xmath86 and @xmath59 as measures of stealthiness and hyperuniformity , there are a multitude of other metrics that can be constructed . measuring hyperuniformity for a single finite system is challenging due to the fluctuations and noise that can arise in the small-@xmath0 behavior of @xmath27 . ideally , one would prefer to have a large system where the number variance in an observation window can be measured accurately . however , we are limited in system size because of numerical methods . while our measurement of stealthiness is directly related to the structure factor , we assumed equal weights for all the @xmath27 , while one could assign weights differently to favor small-@xmath0 scattering or near-@xmath3 scattering . currently , there are no obvious metrics that stand out as superior to those used here . however , the development of new , improved metrics remains a potential direction of future research . while our paths connect ground states to inherent structures , a more generic algorithm taking inherent structures to ground states would be valuable . for instance , our methods relied on knowledge of the ground state and inherent structure _ a priori_. however , a more useful method would be to find the nearest ground - state structure given only an inherent structure . this algorithm would be far more powerful than the current method . one could further quantify the possible rearrangements in the inherent structures in order to help produce such an algorithm that relies on trial displacements that search for small energy barriers nearby . this could lead to a statistical analysis of the types of rearrangements made in a system to determine which are more likely to lead to a ground state . in addition , a more in - depth study of the rearrangements from rsa configurations to hyperuniform and stealthy configurations would be meaningful . perhaps one can distinguish the relative contributions of the features of the interaction potential that contribute to the hyperuniformity and stealthiness of a system . the precise role of the pair interactions in the formation of disordered ground states and stealthiness is unknown . here , we have determined that five - particle rings are a consequence of the first two minima in the overlap potential @xmath1 . the ground - state behavior holds for a class of potentials@xcite and we strongly suspect that similar behavior in the inherent structure analysis may generally hold as well . a systematic study involving the determination of ground states for the truncated overlap potential in eq . ( [ eq : overlapvr ] ) might help understand how many local minima in @xmath1 are necessary for disordered ground states to arise . the potential could be truncated at the first , second , third , _ etc_. minimum and numerical procedures could be used to find ground - state structures . this would encounter issues of rigor in that the structures found by numerical methods may not touch the lower bound on @xmath13 . the trivial lower bound for @xmath13 is zero ( due to the nonnegativity of the overlap potential ) , but for the densities associated with the disordered and wavy - crystalline regimes , the improved lower bound , that being @xmath27 must vanish for all @xmath132 , is not applicable to the truncated @xmath0-space overlap potential . in addition , we observed that the shape of @xmath2 , which assigns weights to @xmath27 in the potential - energy function , apparently influences the shape of the function @xmath27 for the inherent structures for @xmath106 . above this @xmath14 value , there was no obvious connection between @xmath2 and @xmath27 for the inherent structures . this raises the question as to whether the shape of @xmath2 actually influences the shape of the function @xmath27 for inherent structures . a simple study varying the shape of @xmath2 , but maintaining the compact support at @xmath3 and the positivity of @xmath2 would help answer this question . lastly , while this collective coordinate setup was used understand and design new materials in two - dimensions , such as photonic bandgap materials,@xcite the extension to three dimensions should be particularly fruitful . while we have some understanding of stealthy ground states in three dimensions,@xcite the relations between inherent structures and ground states may be different in three dimensions than two dimensions . connecting these structural rearrangements and physical properties can help to contribute to the next class of three - dimensional photonic bandgap materials . this research was supported by the u.s . department of energy , office of basic energy sciences , division of materials sciences and engineering under award de - fg02 - 04-er46108 .

we describe models of nonthermal photon emission from a homogeneous distribution of relativistic electrons and protons . contributions from the synchrotron , inverse compton , nonthermal bremsstrahlung and neutral - pion decay processes are computed separately using a common parameterization of the underlying distribution of nonthermal particles . the models are intended for use in fitting spectra from multi - wavelength observations and are designed to be accurate and efficient . although our applications have focused on galactic supernova remnants , the software is modular , making it straightforward to customize for different applications . in particular , the shapes of the particle distribution functions and the shape of the seed photon spectrum used by the inverse compton model are defined in separate modules and may be customized for specific applications . we assess the accuracy of these models by using a recurrence relation and by comparing them with analytic results and with previous numerical work by other authors .

astro-ph0607574

cosmic - rays with energies up to about 1000tev are thought to be accelerated in supernova remnant shocks @xcite . in the simplest , idealized model of the acceleration process , test particles interact with a shock discontinuity to produce a power - law momentum distribution of cosmic - rays . photon spectra produced by power - law momentum spectra , as in the xspecmodels srcut and sresc @xcite , have been widely used to fit the x - ray synchrotron spectra of several snrs . however , in the diffusive shock acceleration picture @xcite , nonlinear processes are expected to cause some deviation , or curvature , away from a power - law momentum distribution . recent observations of cas a @xcite and sn1006 @xcite have shown evidence for a curved synchrotron spectrum . such spectra can be studied using simulations of diffusive shock acceleration in supernova remnants ( see _ e.g. _ * ? ? ? * ; * ? ? ? * ; * ? ? ? these simulations predict the momentum distribution of nonthermal particles and the resulting photon emission spectrum . unfortunately , such detailed calculations are still too time - consuming for widespread use in iterative fitting of models to observed spectra . therefore , we have developed models of the synchrotron ( [ sec : sync ] ) , inverse compton ( [ sec : invc ] ) , nonthermal bremsstrahlung ( [ sec : ntbrem ] ) and neutral - pion decay ( [ sec : pizero ] ) spectra produced by homogeneous emitting regions having nonthermal particle momentum distributions with arbitrary shape . our primary goals in developing these models were to make them as general and accurate as possible , and to make them computationally efficient enough to be practical for use in iterative fitting of observational data . an additional goal was to adhere to a modular design to simplify customizing the models for specific applications . by providing alternate implementations of the appropriate modules , one can customize the shapes of the particle distributions and the shape of the photon spectrum used to compute the inverse compton model . for example , @xcite used the synchrotron spectrum model with the curved particle momentum spectrum described in [ sec : pdf ] to detect curvature in the cosmic - ray electron spectrum of sn1006 . we assess the accuracy of the computed photon spectra by applying a recurrence relation and by comparing with analytic results and with published numerical models ( [ sec : accuracy ] ) . when using these models to fit ( [ sec : observ ] ) simultaneously radio , x - ray and gamma - ray observations of supernova remnants , we have occasionally found it useful to reduce the number of degrees of freedom by introducing additional physical constraints on the fit parameters . gamma - ray spectra are particularly troublesome because they may contain significant contributions from inverse compton emission , neutral - pion decay and nonthermal bremsstrahlung . by introducing additional physical constraints , one can usefully reduce the set of linear combinations of these models that fit the gamma - ray data . in [ sec : constraints ] , we discuss some of these constraints and describe how they may be imposed . the algorithms used to compute the photon emission impose relatively few limitations on distribution functions suitable for use in fitting . the spectral models described below are derived assuming a particle momentum distribution function that depends only on the magnitude of the particle momentum and not its direction . for most practical applications , the momentum distribution function should depend on a reasonably small number of parameters and should be integrable by adaptive quadrature rules . in applications to date , we have used a nonthermal particle distribution function of the form @xmath0 where @xmath1 and where @xmath2 , and @xmath3 . the same functional form is used for both protons and electrons . the normalization parameter , @xmath4 , represents the density of particles with momentum @xmath5 and has units @xmath6 . when @xmath7 , equation ( [ eq : pdf ] ) describes a power - law distribution in momentum with an exponential cutoff at @xmath8 . when @xmath9 , the power - law exponent changes with momentum for @xmath10 ( see figure [ fig : density ] ) . the effect is such that , for each factor of ten increase in the momentum above @xmath11 , the spectral index changes by an amount equal to the curvature parameter , @xmath12 . positive values of @xmath12 cause the particle spectrum slope to flatten toward higher momenta , as shown in figure [ fig : density ] . positive values are expected due to nonlinear behavior of the diffusive acceleration mechanism ( see _ e.g. _ * ? ? ? * ; * ? ? ? * ; * ? ? ? the qualitative effect of a small positive curvature on the shape of the nonthermal photon spectrum is shown in figure [ fig : curvature ] . each of the photon emission models described below depends on parameters associated with the underlying momentum distribution of nonthermal particles . from equation ( [ eq : pdf ] ) , these parameters are @xmath13 , @xmath12 and @xmath14 . the normalization parameter for each photon emission model includes the normalization parameter , @xmath4 , for the relevant nonthermal particle distribution function . the low end of the particle momentum distribution is dominated by a thermal maxwellian as shown in figure [ fig : density ] . as described in [ sec : constraints ] , one can impose charge conservation in the injection mechanism to set the value of the cosmic - ray proton normalization , @xmath15 relative to the cosmic - ray electron normalization , @xmath16 . in computing nonthermal photon spectra , integrals over particle momenta generally include only relativistic particles with @xmath17 ; in practice , we use @xmath18 . from @xcite , the total synchrotron power emitted per unit frequency from an energetic electron ( @xmath17 ) spiraling in a magnetic field is @xmath19 where @xmath20 is the electron charge , @xmath21 is the electron mass , @xmath22 is the magnetic field strength , @xmath23 is the pitch angle between the electron s velocity vector @xmath24 and the magnetic field vector @xmath25 , @xmath26 is the critical frequency , defined as @xmath27 and @xmath28 , the first synchrotron function , is defined as @xmath29 where @xmath30 is an irregular modified bessel function . note that equation ( [ eq : p - emitted ] ) applies to frequencies well above the gyro - frequency , where the synchrotron spectrum may be regarded as continuous . assuming a steady state electron distribution , we define @xmath31 as the density of nonthermal electrons with pitch angles @xmath23 within a solid angle @xmath32 , and momenta @xmath33 within @xmath34 . from @xcite , @xmath35 is the total synchrotron power received per unit volume and per unit frequency , integrated over pitch - angles . combining equations ( [ eq : p - emitted ] ) , ( [ eq : nu - c ] ) and ( [ eq : dwdnudt ] ) , and assuming an isotropic distribution of pitch - angles so that @xmath36 , we obtain @xmath37 expressing this result in terms of a photon emission rate per unit energy , equation ( [ eq : sync - emis1 ] ) yields the differential emissivity spectrum in the form @xmath38 where @xmath39 is the photon energy and the angular integral is @xmath40 @xcite showed that equation ( [ eq : sync - angular - integral ] ) may be expressed analytically in terms of whittaker functions , @xmath41 , \label{eq : whittaker } \ ] ] where @xmath42 and @xmath43 is a confluent hypergeometric function of the second kind . the synchrotron flux is obtained by evaluating equation ( [ eq : sync - emis ] ) . to speed numerical computations , we compute @xmath44 using a cubic spline interpolation on a pre - computed table ; we use the gnu scientific library @xcite to perform numerical integration and one - dimensional spline interpolation and to evaluate selected special functions . this table is constructed by evaluating equation ( [ eq : whittaker ] ) on an adaptive grid of @xmath45 values chosen to accurately sample the behavior of the function over the range @xmath46 . a log - spaced grid of @xmath45 values was refined by adding interpolation points until the interpolated value of @xmath47 had a fractional error smaller than @xmath48 at the midpoint of each @xmath45 interval ; we linearly interpolated the logarithms of @xmath45 and @xmath44 only while refining the grid for the lookup table . in the final table , the associated cubic spline interpolation errors are @xmath49 over the entire range of @xmath45 . the parameters of the synchrotron model are @xmath13 , @xmath12 , @xmath14 , the total magnetic field strength , @xmath50 and the normalization , @xmath51 where @xmath16 is the normalization of the nonthermal electron momentum distribution , @xmath52 is the distance to the source and @xmath53 is the synchrotron emitting volume . a homogeneous emitting volume is assumed . note that the parameters of the model describe the physical properties of the synchrotron - emitting plasma rather than the properties of the observed synchrotron emission . in some situations , two or more of the fit parameters are degenerate . for example , when fitting x - ray observations alone , the data constrain the critical frequency , @xmath26 , but , because @xmath26 depends on the product @xmath54 , the model parameters @xmath50 and @xmath14 are degenerate . similarly , when fitting radio observations alone , the normalization , @xmath55 , and @xmath50 are degenerate . in both cases , freezing @xmath50 solves the problem . although one could group the physical parameters to remove degeneracies for special cases , we have chosen to keep the physical parameters separate . the advantage of this choice is that , by introducing additional observational constraints from other energy bands , it may be possible to constrain the physical parameters separately . for example , by fitting radio , x - ray and gamma - ray data simultaneously , one might individually constrain @xmath55 , @xmath13 , @xmath12 , @xmath14 and @xmath50 . given a distribution of relativistic ( @xmath17 ) electrons @xmath56 , immersed in an isotropic radiation field with photon number density @xmath57 , the differential emissivity spectrum of compton scattered photons for single - scattering is @xmath58 where @xmath59 , and @xmath60 \ ] ] is the klein - nishina scattering cross - section @xcite , where @xmath61 @xmath62 and @xmath63 is the classical electron radius . the allowed range for @xmath64 follows from the kinematics of compton scattering . in the frame in which the electron is initially at rest , the incident photon has energy @xmath65 and the scattered photon energy may span the range @xmath66 . in the lab frame , the scattered photon energy lies in the range @xmath67 . therefore , the allowed range for @xmath64 is @xmath68 the lower limit of the integral over electron momenta in equation ( [ eq : ic - rate ] ) corresponds to the minimum electron momentum , @xmath69 , that can compton scatter a photon from initial energy @xmath70 to final energy @xmath39 . from the kinematics , one can show that the threshold electron lorentz factor is @xmath71 , \label{eq : ic - threshold}\ ] ] which corresponds to @xmath72 . the inverse compton model is obtained by evaluating equation ( [ eq : ic - rate ] ) for the cosmic background radiation field so that @xmath73 where @xmath74 is the electron compton wavelength , @xmath75 , and @xmath76 = 2.725k @xcite . for computational convenience , it is useful to change the order of integration . performing the integral over photon energies first , we let the electron momentum integral extend over the full range of electron momenta and use the cross - section @xmath77 we can then speed up numerical computations by tabulating the photon energy integral , @xmath78 for a given spectrum of seed photons , @xmath57 . in practice , we use @xmath64 as the variable of integration to evaluate the integral in equation ( [ radiation - field - integral ] ) . because the range of @xmath79 spans many orders of magnitude over the domain of interest , we find it useful to interpolate on the logarithm of the function value . in constructing an interpolation table for @xmath80 , we explicitly incorporate the existence of the threshold lorentz factor @xmath81 . because @xmath82 asymptotically goes to zero as @xmath83 , we set @xmath84 for @xmath85 where @xmath86 is the maximum incident photon energy of interest . introducing a change of variables , @xmath87 and @xmath88 , we tabulate @xmath89 on a @xmath90 uniform rectangular grid in @xmath91 and @xmath92 covering the range @xmath93 and @xmath94ev . the smoothness of the resulting table allows accurate two - dimensional interpolation with a @xmath95-order spline @xcite in each coordinate . the parameters of the inverse compton model are @xmath13 , @xmath12 , @xmath14 , the blackbody temperature , @xmath76 and the normalization , @xmath96 where @xmath97 is the homogeneous emitting volume . by default , the temperature parameter , @xmath76 , is not used and the model uses a lookup table for @xmath80 appropriate for seed photons from the cosmic background radiation field . the error associated with interpolation in this table is typically @xmath98 . a switch is provided to force the model to compute the radiation field integral by direct integration instead of table interpolation . the computational expense of direct integration currently makes this mode impractical for use in spectral fitting . alternatively , one can use a linear combination of lookup tables to describe a more complicated radiation field . for example , one can construct lookup tables corresponding to a dilute stellar radiation field or to emission from molecular clouds . during the fit , one can vary the relative proportions of these components . note that the current implementation restricts the input radiation field shape to one that is integrable by adaptive quadrature rules . the inverse compton emitting volume , @xmath97 , need not be the same as the synchrotron emitting volume , @xmath53 . because the cosmic background radiation photons will fill the entire synchrotron emitting volume , @xmath97 will usually be at least as large as @xmath53 . however , if nonthermal electrons are found in a volume with a relatively weak magnetic field , that volume will produce inverse compton emission , but little synchrotron emission , and @xmath99 . for this reason , it is useful to allow the inverse compton and synchrotron norms to be different . in this section , we consider nonthermal bremsstrahlung emission from a population of nonthermal electrons incident on a stationary target containing free electrons and ions . the total bremsstrahlung emissivity is computed as a sum of contributions from electron - electron and electron - ion bremsstrahlung . given two populations of relativistic particles with momentum distributions , @xmath100 and @xmath101 , and with interaction cross - section , @xmath102 , the general expression for the collision rate per unit volume is @xmath103 where the integrals extend over the momenta of the interacting distributions @xcite . the kronecker delta , @xmath104 corrects for double - counting when the particles are identical . when @xmath105 , equation ( [ eq : general - rate ] ) reduces to the familiar non - relativistic form , and one of the populations may be treated as a stationary target . for example , when the target particles may be characterized by a maxwellian thermal distribution , the thermal motions of the target particles may be neglected as long as @xmath106 . in the limit that @xmath105 , equation ( [ eq : general - rate ] ) yields a differential emissivity spectrum for electron - electron bremsstrahlung of the form @xmath107 where @xmath108 is the target electron density , and @xmath109 is the lab - frame differential cross - section for this interaction between identical particles and includes the factor @xmath110 . the lab - frame cross - section , @xmath111 , differential in both photon energy , @xmath39 , and photon emission angle , @xmath112 was taken from @xcite and was computed using software kindly provided by e. haug . the lab - frame cross - section , differential in photon energy , @xmath109 , is obtained by integrating over photon emission angles . the angular integration limits are given by @xcite and follow from energy - momentum conservation . in the center of momentum ( cm ) frame , the photon emission is symmetric along the direction of motion and all values of @xmath113 are accessible . for an electron incident with momentum @xmath114 in the lab frame , beaming restricts photon emission angles to a narrow cone in the forward direction . from @xcite , photons with energies @xmath115 are emitted into a cone with the maximum emission angle , @xmath116 , given by @xmath117 lower energy photons may span @xmath118 . in carrying out the angular integration for ultra - relativistic electrons , some care must be taken to minimize numerical problems due to round - off errors . in particular , the expression for the differential cross - section , @xmath119 , includes terms that divide by the quantity @xmath120 where @xmath121 , @xmath122 , and @xmath76 is the incident electron kinetic energy in units of @xmath123 . for @xmath124 , @xmath45 is identically zero in double - precision for @xmath125 , causing a division by zero error . in the ultra - relativistic regime , we computed @xmath45 using the first few terms of its series expansion in powers of @xmath126 . substituting @xmath127 , and using @xmath128 as the variable of integration , the relevant cancellations can be handled analytically . to check the result , we used the lorentz invariant , @xmath129 , to transform haug s cm - frame cross - section into the lab frame ( see _ e.g. _ * ? ? ? * ) . in the cm frame , the cross - section computation is less affected by round - off errors because the relevant electron lorentz factor is @xmath130 . applying the lorentz transformation and treating round - off errors by handling the relevant cancellations analytically , we verified that the transformed result agreed with the lab - frame cross - section in the ultra - relativistic regime . to speed up the numerical integrations , the cross - section is evaluated using a two - dimensional cubic - spline interpolation on pre - computed tables . in constructing an interpolation table for @xmath131 , we explicitly incorporate the fact that @xmath132 for kinetic energies @xmath133 , where @xmath134 is determined by the kinematics . we define @xmath134 numerically as the locus of points at which @xmath135 for @xmath136 . introducing a change of variables , @xmath137 and @xmath138 , we tabulate @xmath139 on a @xmath90 rectangular grid in @xmath140 and @xmath141 covering @xmath142 and @xmath143 . the smoothness of the resulting table allows accurate two - dimensional cubic spline interpolation . the error associated with interpolation in the cross - section table is @xmath144 . from equation ( [ eq : general - rate ] ) the differential emissivity spectrum for electron - ion bremsstrahlung can be written @xmath145 where @xmath146 is the density of target ions with charge @xmath147 , and @xmath148 is the lab - frame differential cross - section . the bethe - heitler cross - section @xcite determines the probability that deflection of a relativistic electron in the unscreened field of an ion of charge @xmath147 will yield a photon of energy @xmath149 . the differential cross - section may be written in the form @xmath150 , \label{eq : heitler - xsec } \ ] ] where @xmath151 , \ ] ] @xmath152 @xmath153 , \quad a \equiv 2\ln\left[\gamma \left(1+\beta \right)\right],\ ] ] @xmath154.\ ] ] in these expressions , @xmath23 is the fine structure constant and @xmath155 and @xmath156 are the lorentz factors of the scattered and incident electron , respectively . the bethe - heitler cross - section is derived using the born approximation which is appropriate in the limit of high kinetic energies , such that @xmath157 for both the incident and scattered electron . the recoil of the nucleus is neglected so @xmath158 . the accuracy at low energies is improved by including the elwert correction factor , which we apply at all energies @xcite , so that @xmath159 where @xmath160 and where @xmath161 when the target ions retain bound atomic electrons , it is necessary to modify the cross - section to account for screening of the nuclear charge . we assume that the target material is completely ionized so that screening corrections are not necessary . the relative simplicity of the cross - section makes it practical to perform the integration over electron momenta by evaluating ( [ eq : heitler - xsec ] ) directly rather than interpolating values from a pre - computed table . the parameters of the nonthermal bremsstrahlung model are @xmath13 , @xmath12 , @xmath14 and the normalization , @xmath162 where @xmath163 is the total ion number density of the target , @xmath146 is the number density of ions with charge @xmath147 and @xmath164 is the emitting volume . the user interface includes parameters to control the relative contributions of electron - electron and electron - proton bremsstrahlung . the electron - electron contribution has weight @xmath165 and the electron - proton contribution has weight @xmath166 . by default , the target is assumed to consist of hydrogen and helium with @xmath167 so that the default weights are @xmath168 and @xmath169 . a common alternative is to view @xmath170 as the target proton density and to compute the weights using that assumption . as long as the weights are consistent with the definition of @xmath170 , the result is the same . nonthermal protons produce gamma - ray emission primarily through collisions with thermal protons . these collisions yield neutral pions via @xmath171 , and the neutral pions decay via @xmath172 . we include only the contribution from proton - proton collisions and ignore contributions from processes other than the decay of neutral pions . in its rest - frame , the neutral pion decays within @xmath173 sec , producing two gamma - rays , each with an energy of @xmath174 . because the pion has zero spin , the gamma - rays are emitted isotropically . following @xcite , the number of gamma - rays emitted into a rest frame angle between @xmath175 and @xmath176 is @xmath177 since a gamma - ray that is emitted at an angle @xmath175 in the pion rest frame has a lab - frame energy @xmath178 the number of gamma - rays with lab - frame energy between @xmath39 and @xmath179 is @xmath180 where @xmath181 is the pion momentum . for a population of pions with momentum distribution , @xmath182 , the gamma ray spectrum is then @xmath183 the lower integration limit is set by the minimum pion momentum required to yield a lab - frame photon of energy @xmath39 . from equation ( [ eq : labframe - gamma - energy ] ) , the low energy gamma - ray has energy @xmath184 , corresponding to a pion velocity of @xmath185 a lab frame photon of energy @xmath39 therefore requires a pion lorentz factor of at least @xmath186 the upper integration limit corresponds to the maximum pion momentum that can be produced by a proton of kinetic energy @xmath187 . from the collision kinematics , one can show that @xmath188 the production spectrum of secondary pions in proton - proton collisions is @xmath189 where @xmath190 is the density of target protons , @xmath191 is the nonthermal proton velocity , @xmath192 is the nonthermal proton momentum distribution , and @xmath193 is the differential cross - section for production of a neutral pion with lab - frame momentum @xmath194 from a proton with lab - frame momentum @xmath33 . the lower integration limit is the threshold proton momentum for producing a pion with momentum @xmath194 . from the collision kinematics , one can show that the threshold proton kinetic energy is @xmath195 for the purpose of interactive spectral fitting , computing the photon spectrum from neutral - pion decay by direct integration over the proton momentum distribution and over the resulting pion spectrum is quite computationally demanding . the cost of the numerical integrations is compounded by the fact that the differential cross - sections are computationally expensive for certain energies ( see _ e.g. _ * ? ? ? * ; * ? ? ? to reduce the cost of these computations , we evaluate the integral over the proton distribution using the delta - function approximation described by @xcite . the parameters of the neutral - pion decay model are @xmath13 , @xmath12 , @xmath14 and the normalization , @xmath196 where @xmath15 is the normalization for the nonthermal proton momentum distribution , @xmath190 is the density of target protons , and @xmath197 is the emitting volume . note that for this model , the parameters @xmath13 , @xmath12 and @xmath14 refer to the proton momentum distribution . consider spectral models of the form @xmath198 where @xmath39 is the photon energy , @xmath33 is the particle momentum , @xmath199 is a scaling constant , and @xmath200 depends on the physical process . taking the derivative with respect to @xmath201 , we find that @xmath202 this identity specifies a recurrence relation between spectra of the form shown in equation ( [ eq : recur - direct ] ) . numerical evaluation of equation ( [ eq : recur - direct ] ) yields model spectra of the form @xmath203 , where @xmath204 is the exact result and @xmath205 represents the error in the computation . because accurately computed spectra must satisfy equation ( [ eq : recur ] ) , it follows that @xmath206 must also satisfy equation ( [ eq : recur ] ) . although an error in the computation of @xmath200 might generate such @xmath207 dependent errors in the computed spectrum , such an error should be detectable on comparison with an analytic solution or with an independent numerical calculation . eliminating that possibility , the error term in computed spectra that satisfy equation ( [ eq : recur ] ) must be independent of @xmath23 and @xmath13 so that @xmath208 . the magnitude of the energy - dependent error , @xmath209 must be estimated by other means , such as by comparing with analytic solutions and with independent numerical results . although satisfying the recurrence relation does not prove that the spectral computations are correct , it does provide strong constraints on the magnitude and parameter dependence of the error term , @xmath205 . to verify equation ( [ eq : recur ] ) numerically , we introduce a finite - difference approximation for the derivative . it follows that computed spectra of the form shown in equation ( [ eq : recur - direct ] ) should obey a relationship of the form @xmath210 where @xmath211 for clarity in equation ( [ eq : recur - finite ] ) , we have avoided writing out the explicit dependence of the right - hand side upon photon energy , @xmath39 . the centered difference used to approximate the derivative in equation ( [ eq : recur - finite ] ) should yield quadratic convergence in the limit that @xmath212 . for each of our spectral models , we verified that , in the limit @xmath213 , @xmath214 as @xmath215 ; observing smooth convergence at the expected rate confirms that the models behave as expected . in subsequent evaluations of equation ( [ eq : recur - finite ] ) , we adopt a value of @xmath216 . because individual terms in equation ( [ eq : recur - finite ] ) can be @xmath217 , it is useful to compute the fractional error , @xmath218 , where the denominator is the @xmath219-norm of the three terms in equation ( [ eq : recur - finite ] ) . to test the @xmath13 and @xmath23 dependence of @xmath205 , we computed values of @xmath220 at many points along each of several spectra . one spectrum was computed for each pair of ( @xmath13 , @xmath23 ) values on a grid sampling the range @xmath221 and @xmath222tev . for the synchrotron model , @xmath220 was evaluated at photon energies in the range @xmath223 , where @xmath224 is the critical energy . for the other models , @xmath220 was evaluated at photon energies in the range @xmath225 . each photon energy grid was logarithmically spaced and used 80 points per decade . figure [ fig : recur - distrib ] shows that , throughout this range , @xmath220 was @xmath226 . we conclude that , throughout the parameter range of interest , the @xmath227 dependence of @xmath205 is no larger than about a part in @xmath228 . this result suggests that the errors are essentially independent of the parameters of the particle distribution function . to test the dependence of the errors on the other parameters , including the photon energy , we compare our results with other analytic and numerical solutions . @xcite derive the well - known analytic result that the synchrotron spectrum from a power - law distribution of electrons , @xmath229 , is itself a power - law of the form @xmath230 . to compare with this result , we used a power - law electron distribution function to compute numerically @xmath231 , as described above in equation ( [ eq : sync - emis ] ) , and we computed @xmath232 from equation ( 4.59 ) of @xcite . in deriving the analytic solution , the lower limit of the integral over @xmath233 is extended to zero @xcite , effectively neglecting the electron mass . consistent with this assumption , the value of @xmath234 in the numerical integrand must be computed using the approximation @xmath235 otherwise , the computed spectrum departs from the analytic solution at low frequencies . we find that the absolute value of the fractional error is @xmath236 for frequencies in the range @xmath237hz and for values of @xmath13 and @xmath50 in the ranges @xmath238 and @xmath239 g , respectively . the primary source of error in this comparison is associated with interpolation of @xmath44 in our precomputed table ; computing @xmath44 directly in terms of hypergeometric functions , we reproduce the analytic solution to within @xmath240 over the specified range of @xmath241 , @xmath13 and @xmath50 . @xcite also discuss the inverse compton spectrum produced by a power - law distribution of electrons scattering photons from a blackbody radiation field . they give asymptotic analytic solutions valid in the thomson limit and in the extreme klein - nishina limit . to compare our inverse compton model with these analytic solutions , we used a power - law electron distribution function , @xmath229 , to compute numerically @xmath242 , as described above in equation ( [ eq : ic - rate ] ) . for the purpose of these tests , we evaluated the integral over incident photon energies ( equation [ radiation - field - integral ] ) by direct numerical integration rather than by spline interpolation in precomputed tables . in the thomson limit , the energy of the incident photon in the electron rest frame is small compared to the electron rest energy and the momentum transferred by the electron is small compared to its initial momentum . in this regime , corresponding to @xmath243 , the scattering cross - section is essentially independent of the energy of the incoming photon . the thomson limit approximation is most accurate in the limit of low radiation temperature , @xmath76 , and for scattered photon energies near the middle of the applicable energy range . such a case provides the best test of the accuracy of our inverse compton model in the thomson limit . for power - law slopes in the range @xmath244 and radiation field temperatures @xmath245k , we used our model to compute inverse compton spectra , @xmath246 , for scattered photon energies in the range @xmath247 , using @xmath248 . for comparison , we computed @xmath249 using equation ( 2.65 ) of @xcite . we found that our inverse compton model converged smoothly to the analytic result in the appropriate limit . the smallest fractional difference is @xmath250 broader comparisons are less useful as a test of computational accuracy because the asymptotic solution itself becomes less accurate with increasing @xmath76 and for energies approaching the endpoints of the applicable range . for @xmath251k , the fractional error is smallest ( @xmath252 ) at @xmath253ev , then increases toward lower and higher energies with @xmath254 for the range @xmath255ev . increasing @xmath76 narrows the applicable energy range so that for @xmath256k , @xmath257 only for the range @xmath258kev . in the extreme klein - nishina limit , the energy of the incident photon in the electron rest frame is large compared to the electron rest energy . in this regime , corresponding to @xmath259 , the scattering cross - section is strongly peaked near the maximum scattered photon energy so that individual compton scatterings tend to involve a large energy transfer . the characteristic scattered photon energy is then @xmath260 . it follows that this regime may be characterized by the requirement that @xmath261 so that , for a blackbody radiation field with @xmath262 , the extreme klein - nishina limit corresponds to scattered photon energies @xmath263 . we computed @xmath264 using equation ( 2.88 ) of @xcite , that is defined in terms of an infinite series that converges extremely slowly : @xmath265 . from its series definition , it follows that this constant is equivalent to @xmath266 , where @xmath267 is the first derivative of the riemann zeta function . to enable more precise quantative comparison with the analytic solution , we used the numerical value @xmath268 , obtained using the symbolic algebra package maple . ] . in the extreme klein - nishina limit , we computed inverse compton spectra , @xmath246 , for scattered photon energies in the range @xmath269 for power - law slopes in the range @xmath238 and radiation field temperatures in the range @xmath270k . we found that the inverse compton model spectra converged smoothly to the analytic result , @xmath271 . the fractional error decreased smoothly with increasing photon energy from @xmath272 at @xmath273 to @xmath274 for @xmath275 . numerical round - off errors become important at higher energies . to examine the absolute accuracy of our computed photon spectra for more realistic particle spectra , we compared our results with those of @xcite . to facilitate this comparison , sturner kindly computed photon spectra for a particle distribution function of the form @xmath276 where @xmath277 has units of @xmath278 and @xmath76 is the kinetic energy . sturner provided synchrotron , inverse compton and nonthermal bremsstrahlung spectra for @xmath279 1.8 , 2.0 and 2.3 and @xmath14=10tev . the synchrotron spectrum was computed using @xmath50=1@xmath280 g . the inverse compton spectrum was computed for the cosmic background radiation field , using a temperature of 2.7k . the nonthermal bremsstrahlung spectrum was computed for a fully ionized target with ion density 0.11@xmath281 consisting of protons ( 0.1@xmath281 ) , alpha - particles ( 0.01@xmath281 ) , and free electrons ( 0.12@xmath281 ) , corresponding to relative weights of @xmath282 and @xmath283 , for electron - electron and electron - ionbremsstrahlung , respectively . we used our models to compute photon spectra for the same parameters and particle distribution function . figures [ fig : sturner - sync - diff ] and [ fig : sturner - gamma - diff ] show our spectra divided by those obtained from sturner . over most of the energy range , our spectra agree quite well with sturner s ; the inverse compton and synchrotron spectra agree to within @xmath284% , while the nonthermal bremsstrahlung spectra typically differ by @xmath285% . aside from the weak @xmath13 dependence in the nonthermal bremsstrahlung differences below about 100mev , the differences between sturner s spectra and ours are largely independent of the spectral index , @xmath13 . because the @xmath13 dependent errors in our spectra are constrained by the recurrence relation , the weak residual @xmath13 dependence seen in the nonthermal bremsstrahlung differences appears to be associated with sturner s spectra . the reason for the overall @xmath2866% discrepancy below 0.25mev is unclear ; features in the nonthermal bremsstrahlung ratio near 30tev and near 10mev correspond to points in sturner s spectra at which the slope changes discontinuously . because sturner s synchrotron and inverse compton fluxes become negative at the extreme end of the high - energy photon spectra , rather than asymptotically approaching zero as ours do , we attribute the larger differences near the cutoff in these spectra to numerical errors in sturner s spectra . however , since the largest differences occur at extremely low flux levels , they are probably not important for observational comparisons . we also compared our synchrotron model with the srcut model @xcite in xspec version 11.3.2d . srcut is designed to compute the synchrotron spectrum from an exponentially cut off power - law distribution of electrons in a homogeneous magnetic field . it depends on a break - frequency parameter , @xmath287 , defined as the critical frequency for @xmath288 and @xmath289 . for @xmath290 , we verified that srcut is consistent with the analytic result for a power - law electron distribution . for frequencies comparable to @xmath287 or larger , we find that srcut spectra differ from our spectra by as much as a factor of two or more . figure [ fig : srcut ] shows that the ratio of the two models depends on photon energy and on the power - law index . the ratio also depends on the break frequency in the sense that the models agree for @xmath290 but disagree for frequencies near the break and above . we are confident that our computed synchrotron spectrum is accurate , first , because the validity of equation ( [ eq : recur ] ) indicates that any errors are independent of the power - law index , @xmath13 , and cutoff - energy , @xmath14 and , second , because our computed spectrum is consistent with sturner s model . we conclude that , for frequencies near and above @xmath287 , the srcut model does not accurately represent the synchrotron spectrum from an exponentially cut - off power - law distribution of electrons in a homogeneous magnetic field . we tested our neutral - pion decay model by comparing our results with the results of @xcite . for the purpose of this comparison , we used his proton momentum distribution function of the form @xmath291 where @xmath292 1.67 \left(\frac{p}{\rm gev / c}\right)^{-2.7 } \left[1 + \left(\frac{\rm 2.5\,gev / c}{p}\right)^2 \right]^{-1/2 } , \\[10pt ] \hspace*{4 cm } e \le 100\,{{\rm gev } } , \end{cases}\ ] ] where @xmath293 and @xmath294 . figure [ fig : mori ] shows our computed neutral - pion decay gamma - ray spectrum divided by the fluxes given in table 1 of @xcite . the largest difference occurs near 100mev , where our flux is about a factor of two larger than that of mori . because @xcite used a much more detailed model of pion production , we do not expect exact agreement . yet , for photon energies above 1gev , the two gamma - ray fluxes agree to within about 10% . because we are primarily interested in fitting data in the 1 - 10tev band , our simplified pion - decay model is adequate for our needs . fitting a model to an observed spectrum involves minimizing a goodness - of - fit statistic that compares the observed spectral data to the spectral model . this comparison often involves binned data . for example , consider x - ray observations which provide the observed number of counts in each energy bin . neglecting nonlinear effects in the x - ray detector , the expected number of counts is usually computed using an expression of the form @xcite @xmath295 where @xmath296 is the total number of counts in bin @xmath297 , @xmath298 is the number of counts due to the instrumental background , @xmath299 is the exposure time and @xmath300 is the spectral model describing the incident flux of photons with energy @xmath301 . in equation ( [ eq : forward - folding ] ) , @xmath302 is the redistribution function , describing the probability that incident photons with energy @xmath301 contribute counts to bin @xmath297 , and @xmath303 is the effective area , accounting for the telescope collecting area , the transmission efficiency of the optical system and the quantum efficiency of the detector . in spectral fitting software , equation ( [ eq : forward - folding ] ) is usually implemented as a discrete sum of the form @xmath304 the redistribution function , @xmath302 , is represented as a matrix , @xmath305 , and the effective area and source models are represented as vectors , @xmath306 , and @xmath307 , respectively . to accurately represent the integral in equation ( [ eq : forward - folding ] ) , the software must compute the model , @xmath307 , as an integral @xmath308 over the width @xmath309 of each spectral bin , @xmath310 . for binned data , we evaluate these integrals using the well - known simpson s rule . although this approach requires three function evaluations per bin , the bin - edge function values may be shared between neighboring bins , reducing the total cost of each spectrum computation by about 30% . this approach is accurate to the extent that a quadratic polynomial is a good approximation to the underlying function , @xmath300 , within each spectral bin . when simultaneously fitting radio , x - ray and gamma - ray observations of supernova remnants , the degeneracy of certain fit parameters ( [ sec : sync ] ) and the variety of emission mechanisms in the gamma - ray band may make it difficult to determine a unique set of fit parameters unless additional constraints are available . based on reasonably general principles , a number of constraints can sometimes be imposed . for example : 1 . when the magnetic field associated with the synchrotron emission is generated primarily by cosmic - ray streaming @xcite , the energy density in cosmic - rays should set an upper limit on the magnetic energy density . the corresponding upper limit on the magnetic field strength is @xmath311 where @xmath312 and @xmath313 are the energy density in cosmic - ray electrons and protons respectively . because @xmath50 and @xmath14 are degenerate , introducing an upper limit on @xmath50 effectively sets a lower limit on @xmath14 . if electrons and protons are injected into the accelerator at the same rate , the normalization of the proton momentum distribution @xmath15 can be fixed by requiring equal densities in nonthermal electrons and protons at some characteristic injection kinetic energy @xmath314 : @xmath315 when both distribution functions are of the form @xmath316 and share the same power - law exponent , @xmath13 , this constraint implies that @xmath317 if @xmath318 @xcite . by fixing the ratio @xmath319 , this constraint reduces the variation in the ratio @xmath320 to the variation associated with the mass ratio of the associated targets , @xmath321 ) . 3 . in many cases it should be reasonable to assume that the emitting volume that produces synchrotron emission will also produce inverse compton emission due to up - scattering of cosmic background photons . in such cases , one can impose a lower limit on the normalization associated with this inverse compton process so that @xmath322 . this constraint ensures that fits to the gamma - ray spectrum will include an appropriate contribution of inverse compton emission . a similar argument can be used to constrain the minimum value of the nonthermal bremsstrahlung normalization , @xmath323 for an assumed minimum target density . to support imposing constraints based on charge conservation , our software provides a function to compute the proton norm , @xmath15 , that , for a given electron norm , @xmath16 , will yield equal nonthermal electron and proton densities at a given injection kinetic energy @xmath324 . for constraints that depend on the energy density in nonthermal particles , our software provides functions that compute the energy density of each particle population . the energy density is defined to be @xmath325 where the lower integration limit , @xmath326 , is the momentum at which the thermal and nonthermal particle densities are equal . note that @xmath326 depends on the density and temperature of the thermal particles and on the nonthermal particle momentum distribution . in practice , complicated fit constraints such as the upper limit on @xmath50 may be imposed using a technique analogous to the method of lagrange multipliers @xcite . rather than minimizing @xmath327 , the idea is to construct a constraint function @xmath328 which may depend on a vector of parameters , @xmath329 , and then to minimize the sum , @xmath330 , where @xmath331 is a parameter that determines the importance of the constraint . the constraint function should be constructed to ensure that @xmath332 when the constraint is satisfied . for example , to impose the constraint that @xmath333 , one might choose @xmath334^{-1 } , & { b_\mathrm{tot}}^2 \ge 8\pi(\varepsilon_e + \varepsilon_p ) ; \\ 0 , & { \rm otherwise}. \end{array } \right.\ ] ] similar terms may be added to impose additional constraints . we have described models that can be used to compute the synchrotron , inverse compton , nonthermal bremsstrahlung and neutral - pion decay spectra of homogeneous sources containing nonthermal electrons and protons . we have implemented these models as an importable module for use in the spectral fitting package isis @xcite ; the software is available from the isis web page[multiblock footnote omitted ] . the models are designed to be accurate and fast enough for use in interactive data analysis on a typical workstation . to the best of our knowledge , this is the first implementation of some of these models in a form suitable for interactive fitting with publicly available data analysis software . results derived using the synchrotron and inverse compton models have been presented elsewhere by @xcite and by @xcite . we assessed the accuracy of these models by using a recurrence relation , by comparing with analytic solutions for synchrotron and inverse compton scattering and by comparing with published work by other authors . the accuracy with which our solutions obey the recurrence relation in equation ( [ eq : recur - finite ] ) demonstrates their correct dependence upon the power - law index and cutoff - energy parameters . we also showed that , over most of the energy range of interest , our models agree with those of @xcite . the largest differences between our models and those of @xcite , of order @xmath28610% , occur near the cutoff of each model spectrum and are essentially independent of the spectral index . we found much larger differences between our synchrotron spectrum and the srcut model from xspec . although consistent with the analytic solution for @xmath290 , srcut differs from our model and from sturner s by as much as a factor of two or more in the x - ray band near @xmath287 ( see figure [ fig : srcut ] ) . the most important difference is that the normalization derived using srcut overestimates the radio flux at 1ghz by an amount dependent upon the spectral index as shown in figure [ fig : srcut ] . the spectral index from srcut tends to be a few percent too steep and the break frequency tends to be a few percent too low but , in practice , such differences may be detectable only with very high quality data . note that , by overestimating the radio flux , srcut may suggest the existence of positive curvature in the underlying particle momentum distribution . in future work , we hope to improve the neutral - pion decay model by explicitly computing the integral over proton momenta using improved pion - production cross - sections . this refinement will extend the applicable range to sub - gev photon energies and will improve the overall accuracy of the model . these improvements will be more important as better observations of the gamma - ray spectrum become available from the _ gamma - ray large area space telescope _ ( glast ) and from future advances in instrumentation . we thank e. haug for providing software to compute values for the electron - electron bremsstrahlung cross - sections and s. sturner for providing numerical tables used to verify our results for selected electron momentum distributions . we also acknowledge useful conversations with john e. davis and a number of comments from the referee which helped us to improve the paper . for reference , tables [ table : sync ] , [ table : gamma - powerlaw ] and [ table : gamma - curved ] contain sample spectra for each emission process described in this paper . the nonthermal electron and proton momentum spectra have the same shape , with slope @xmath335 , and a cutoff energy @xmath336tev . table [ table : sync ] gives synchrotron spectra for curvatures @xmath7 and @xmath337 . tables [ table : gamma - powerlaw ] and [ table : gamma - curved ] give the inverse compton , nonthermal bremsstrahlung and neutral - pion decay spectra for curvatures @xmath7 and @xmath337 , respectively . all normalization coefficients were set to unity . the synchrotron spectrum was computed for @xmath338 g . the inverse compton spectrum was computed using a 2.725k blackbody distribution of seed photons @xcite . contributions to nonthermal bremsstrahlung due to the electron - electron and electron - proton processes are listed separately , with their respective weights set to unity . the spectral models described in this paper were designed to fit multi - wavelength spectral data interactively on a typical workstation and to achieve a high degree of accuracy as efficiently as possible . in practice , to be sure that the fit has fully converged and to derive confidence limits , it is necessary to thoroughly examine the parameter space in the neighborhood of the best - fit parameters . to conduct this search more quickly , we have found it useful to distribute the task of computing single - parameter confidence limits over a number processors running in parallel @xcite . one master process manages the computations being performed by a number of slave processes that run on different computers . all of the computers are connected by a local network and all processes have access to the relevant data and spectral models . the master process assigns each slave process the task of computing confidence limits for a single parameter . if a slave process finds an improved fit , that new parameter set is sent to the master process . if that fit is the best yet found by any slave , the master commands all the slave processes to re - start the search from the beginning , using the new parameter set . we have found that the time required to obtain a set of converged single parameter confidence limits using this approach is often reduced by more than a factor of @xmath339 , where @xmath339 is the number of slave processes . we have implemented this algorithm using the parallel virtual machine ( pvm ) @xcite to handle message passing between the master and slave processes . we used the spectral fitting package , isis @xcite to perform model fits and confidence limit searches using a set of s - langscripts . an s - lang module provides a scriptable interface to the pvm library , making it possible for these scripts to communicate with the pvm . cccccc 1.0000e-07 & 6.8294e+15 & 6.8613e+15 & 1.7783e-01 & 2.5564e+06 & 7.0374e+06 + 1.7783e-07 & 2.8798e+15 & 2.9011e+15 & 3.1623e-01 & 1.0403e+06 & 3.1004e+06 + 3.1623e-07 & 1.2144e+15 & 1.2284e+15 & 5.6234e-01 & 4.1918e+05 & 1.3546e+06 + 5.6234e-07 & 5.1206e+14 & 5.2114e+14 & 1.0000e+00 & 1.6680e+05 & 5.8530e+05 + 1.0000e-06 & 2.1591e+14 & 2.2168e+14 & 1.7783e+00 & 6.5346e+04 & 2.4928e+05 + 1.7783e-06 & 9.1040e+13 & 9.4594e+13 & 3.1623e+00 & 2.5107e+04 & 1.0423e+05 + 3.1623e-06 & 3.8386e+13 & 4.0507e+13 & 5.6234e+00 & 9.4173e+03 & 4.2578e+04 + 5.6234e-06 & 1.6184e+13 & 1.7409e+13 & 1.0000e+01 & 3.4291e+03 & 1.6897e+04 + 1.0000e-05 & 6.8229e+12 & 7.5091e+12 & 1.7783e+01 & 1.2040e+03 & 6.4693e+03 + 1.7783e-05 & 2.8762e+12 & 3.2506e+12 & 3.1623e+01 & 4.0433e+02 & 2.3701e+03 + 3.1623e-05 & 1.2123e+12 & 1.4121e+12 & 5.6234e+01 & 1.2859e+02 & 8.2272e+02 + 5.6234e-05 & 5.1092e+11 & 6.1560e+11 & 1.0000e+02 & 3.8275e+01 & 2.6741e+02 + 1.0000e-04 & 2.1528e+11 & 2.6929e+11 & 1.7783e+02 & 1.0511e+01 & 8.0233e+01 + 1.7783e-04 & 9.0683e+10 & 1.1819e+11 & 3.1623e+02 & 2.6180e+00 & 2.1844e+01 + 3.1623e-04 & 3.8186e+10 & 5.2042e+10 & 5.6234e+02 & 5.7921e-01 & 5.2861e+00 + 5.6234e-04 & 1.6072e+10 & 2.2985e+10 & 1.0000e+03 & 1.1101e-01 & 1.1090e+00 + 1.0000e-03 & 6.7607e+09 & 1.0181e+10 & 1.7783e+03 & 1.7883e-02 & 1.9570e-01 + 1.7783e-03 & 2.8416e+09 & 4.5213e+09 & 3.1623e+03 & 2.3343e-03 & 2.8009e-02 + 3.1623e-03 & 1.1931e+09 & 2.0124e+09 & 5.6234e+03 & 2.3621e-04 & 3.1107e-03 + 5.6234e-03 & 5.0025e+08 & 8.9740e+08 & 1.0000e+04 & 1.7564e-05 & 2.5414e-04 + 1.0000e-02 & 2.0938e+08 & 4.0070e+08 & 1.7783e+04 & 8.9943e-07 & 1.4316e-05 + 1.7783e-02 & 8.7436e+07 & 1.7903e+08 & 3.1623e+04 & 2.9327e-08 & 5.1409e-07 + 3.1623e-02 & 3.6404e+07 & 7.9973e+07 & 5.6234e+04 & 5.5363e-10 & 1.0702e-08 + 5.6234e-02 & 1.5099e+07 & 3.5680e+07 & 1.0000e+05 & 5.3930e-12 & 1.1512e-10 + 1.0000e-01 & 6.2322e+06 & 1.5878e+07 & 1.7783e+05 & 2.3578e-14 & 5.5658e-13 + lrrrr 1.0000e+06 & 2.0062e-10 & 1.3845e-10 & 1.6562e-10 & 8.2454e-18 + 1.7783e+06 & 8.4541e-11 & 5.3800e-11 & 6.2697e-11 & 2.4513e-17 + 3.1623e+06 & 3.5616e-11 & 2.0632e-11 & 2.3393e-11 & 7.0967e-17 + 5.6234e+06 & 1.5000e-11 & 7.7773e-12 & 8.5860e-12 & 1.9707e-16 + 1.0000e+07 & 6.3147e-12 & 2.8778e-12 & 3.1021e-12 & 5.1353e-16 + 1.7783e+07 & 2.6568e-12 & 1.0460e-12 & 1.1053e-12 & 1.2201e-15 + 3.1623e+07 & 1.1170e-12 & 3.7410e-13 & 3.8916e-13 & 2.5548e-15 + 5.6234e+07 & 4.6913e-13 & 1.3195e-13 & 1.3565e-13 & 2.7122e-15 + 1.0000e+08 & 1.9677e-13 & 4.6000e-14 & 4.6881e-14 & 2.7122e-15 + 1.7783e+08 & 8.2389e-14 & 1.5880e-14 & 1.6085e-14 & 1.9824e-15 + 3.1623e+08 & 3.4416e-14 & 5.4375e-15 & 5.4842e-15 & 9.0019e-16 + 5.6234e+08 & 1.4332e-14 & 1.8492e-15 & 1.8596e-15 & 3.6492e-16 + 1.0000e+09 & 5.9441e-15 & 6.2516e-16 & 6.2745e-16 & 1.3641e-16 + 1.7783e+09 & 2.4521e-15 & 2.1026e-16 & 2.1075e-16 & 4.8273e-17 + 3.1623e+09 & 1.0045e-15 & 7.0378e-17 & 7.0481e-17 & 1.6490e-17 + 5.6234e+09 & 4.0771e-16 & 2.3445e-17 & 2.3468e-17 & 5.5070e-18 + 1.0000e+10 & 1.6351e-16 & 7.7725e-18 & 7.7775e-18 & 1.8116e-18 + 1.7783e+10 & 6.4567e-17 & 2.5626e-18 & 2.5637e-18 & 5.8906e-19 + 3.1623e+10 & 2.4988e-17 & 8.3928e-19 & 8.3953e-19 & 1.8934e-19 + 5.6234e+10 & 9.4223e-18 & 2.7251e-19 & 2.7257e-19 & 5.9973e-20 + 1.0000e+11 & 3.4360e-18 & 8.7461e-20 & 8.7476e-20 & 1.8582e-20 + 1.7783e+11 & 1.2001e-18 & 2.7620e-20 & 2.7624e-20 & 5.5578e-21 + 3.1623e+11 & 3.9651e-19 & 8.5250e-21 & 8.5262e-21 & 1.5693e-21 + 5.6234e+11 & 1.2191e-19 & 2.5465e-21 & 2.5469e-21 & 4.0316e-22 + 1.0000e+12 & 3.4130e-20 & 7.2548e-22 & 7.2560e-22 & 8.8745e-23 + 1.7783e+12 & 8.4456e-21 & 1.9286e-22 & 1.9290e-22 & 1.5161e-23 + 3.1623e+12 & 1.7719e-21 & 4.6256e-23 & 4.6267e-23 & 1.7037e-24 + 5.6234e+12 & 2.9674e-22 & 9.4868e-24 & 9.4896e-24 & 9.4934e-26 + 1.0000e+13 & 3.6207e-23 & 1.5226e-24 & 1.5232e-24 & 1.6037e-27 + 1.7783e+13 & 2.7845e-24 & 1.6433e-25 & 1.6443e-25 & 3.4422e-30 + 3.1623e+13 & 1.0582e-25 & 9.1509e-27 & 9.1588e-27 & 1.9973e-34 + 5.6234e+13 & 1.2918e-27 & 1.6454e-28 & 1.6477e-28 & 1.9824e-41 + 1.0000e+14 & 2.3039e-30 & 4.1544e-31 & 4.1640e-31 & 2.4571e-53 + lrrrrrr 1.0000e+06 & 2.4116e-10 & 1.3856e-10 & 1.6572e-10 & 1.1455e-17 + 1.7783e+06 & 1.0553e-10 & 5.3856e-11 & 6.2750e-11 & 3.1243e-17 + 3.1623e+06 & 4.6340e-11 & 2.0662e-11 & 2.3422e-11 & 8.4245e-17 + 5.6234e+06 & 2.0415e-11 & 7.7932e-12 & 8.6015e-12 & 2.2119e-16 + 1.0000e+07 & 9.0225e-12 & 2.8864e-12 & 3.1106e-12 & 5.5285e-16 + 1.7783e+07 & 3.9996e-12 & 1.0506e-12 & 1.1099e-12 & 1.2763e-15 + 3.1623e+07 & 1.7779e-12 & 3.7663e-13 & 3.9168e-13 & 2.6248e-15 + 5.6234e+07 & 7.9236e-13 & 1.3332e-13 & 1.3701e-13 & 2.7832e-15 + 1.0000e+08 & 3.5389e-13 & 4.6739e-14 & 4.7619e-14 & 2.7832e-15 + 1.7783e+08 & 1.5832e-13 & 1.6278e-14 & 1.6483e-14 & 2.0479e-15 + 3.1623e+08 & 7.0899e-14 & 5.6505e-15 & 5.6971e-15 & 9.5028e-16 + 5.6234e+08 & 3.1756e-14 & 1.9621e-15 & 1.9725e-15 & 3.9827e-16 + 1.0000e+09 & 1.4211e-14 & 6.8424e-16 & 6.8653e-16 & 1.5601e-16 + 1.7783e+09 & 6.3444e-15 & 2.4051e-16 & 2.4101e-16 & 5.8712e-17 + 3.1623e+09 & 2.8206e-15 & 8.5278e-17 & 8.5385e-17 & 2.1653e-17 + 5.6234e+09 & 1.2459e-15 & 3.0488e-17 & 3.0513e-17 & 7.9260e-18 + 1.0000e+10 & 5.4508e-16 & 1.0983e-17 & 1.0988e-17 & 2.9006e-18 + 1.7783e+10 & 2.3532e-16 & 3.9805e-18 & 3.9818e-18 & 1.0644e-18 + 3.1623e+10 & 9.9771e-17 & 1.4483e-18 & 1.4486e-18 & 3.9144e-19 + 5.6234e+10 & 4.1292e-17 & 5.2743e-19 & 5.2752e-19 & 1.4370e-19 + 1.0000e+11 & 1.6555e-17 & 1.9143e-19 & 1.9145e-19 & 5.2220e-20 + 1.7783e+11 & 6.3677e-18 & 6.8852e-20 & 6.8859e-20 & 1.8516e-20 + 3.1623e+11 & 2.3209e-18 & 2.4353e-20 & 2.4355e-20 & 6.2578e-21 + 5.6234e+11 & 7.8871e-19 & 8.3813e-21 & 8.3822e-21 & 1.9414e-21 + 1.0000e+12 & 2.4462e-19 & 2.7658e-21 & 2.7661e-21 & 5.2052e-22 + 1.7783e+12 & 6.7266e-20 & 8.5667e-22 & 8.5679e-22 & 1.0934e-22 + 3.1623e+12 & 1.5746e-20 & 2.4116e-22 & 2.4121e-22 & 1.5278e-23 + 5.6234e+12 & 2.9588e-21 & 5.8625e-23 & 5.8639e-23 & 1.0730e-24 + 1.0000e+13 & 4.0826e-22 & 1.1297e-23 & 1.1301e-23 & 2.3213e-26 + 1.7783e+13 & 3.5900e-23 & 1.4881e-24 & 1.4888e-24 & 6.4948e-29 + 3.1623e+13 & 1.5844e-24 & 1.0307e-25 & 1.0315e-25 & 5.0034e-33 + 5.6234e+13 & 2.2946e-26 & 2.3534e-27 & 2.3565e-27 & 6.7148e-40 + 1.0000e+14 & 4.9900e-29 & 7.7046e-30 & 7.7217e-30 & 1.1451e-51 + _ note added in proof._after this manuscript was submitted , papers by kamae _ ( astro - ph/0605581 ) and by kelner , aharonian & bugayov ( astro - ph/0606058 ) appeared , presenting parameterized cross - sections for various particles produced of @xmath340 collisions . we plan to revise our pion - decay model to incorporate these new cross - sections . details will be discussed in a subsequent paper .

we present imaging and spectro- polarimetric observations of the ultraluminous infrared galaxy iras p09104 + 4109 using the keck 10-m telescope . we detect the clear presence of broad , , and mg ii 2800 emission lines in the polarized flux spectra of the nucleus and of an extranuclear emission region 4 away , confirming the presence of a hidden central quasar . the polarization of the broad mg ii emission line is high ( 29% ) , consistent with the remarkably high polarization ( 30%40% ) observed in the extended continuum emission . this indicates that the off - nuclear continuum is dominated by light scattered from the hidden quasar , most probably by dust mixed with the line emitting gas . the high polarizations , combined with the `` foreshortened '' morphology of the polarized brightness distribution allow us to constrain the scattering biconical structure to be at inclination with a half - opening cone angle 40 . the narrow emission lines are polarized in a stratified fashion , with the high ionization lines ( , , ) being polarized 0.7%1.7% and essentially unpolarized . the line polarizations are positively correlated with critical density , ionization potential , and velocity width of the emission lines . this indicates that , as is the case with the narrow - line radio galaxies , which also often contain powerful quasars , the narrow - emission line region may be partially shadowed by the putative torus , with the higher ionization lines originating closer to the nucleus . one notable characteristic of the extranuclear knot is that all species of fe are markedly absent in its spectrum , while they appear prominently in the nucleus . in addition , narrow mg ii is observed to be much weaker than predicted by ionization models . our favored interpretation is that there is a large amount of dust in the extranuclear regions , allowing gaseous refractory metals to deposit . near the nucleus , dust is destroyed in the strong radiation field of the quasar , inhibiting metal depletion onto grains . the extended emission regions are most likely material shredded from nearby cluster members and not gas condensed from the cooling flow or expelled from the obscured quasar . the higher temperature inferred from [ o iii ] lines compared to that from [ n ii ] and the general better agreement with models of line ratios , especially 5007/4363 and he ii/ , provide strong evidence for matter - bounded clouds in addition to ionization - bounded clouds in the nlr . ionization by pure velocity shocks can be ruled out . shocks with photoionizing precursors may be present , but are probably not a dominant contributor to the energy input . # 1#1 # 1#1

astro-ph0004383

discovered by the @xmath6 all - sky survey , ultraluminous infrared galaxies ( ulirgs , see review by @xcite ) emit most of their energy in the infrared ( @xmath7 ) , and could harbor infant quasars enshrouded in a large amount of dust ( @xcite ) . on the other hand , they may also represent energetic , compact starbursts ( @xcite ; @xcite ) . much recent research effort has been devoted to understanding the dominant energy source in these ulirgs whether it is obscured quasars or intense bursts of star formation . iras p09104 + 4109 ( @xmath8 = 0.44 ; @xcite ) is one of the few exceptionally luminous ulirgs sometimes referred to as the `` hyperluminous '' ( @xmath9 ) infrared galaxies . the others include f15307 + 3252 ( @xmath8 = 0.926 ; @xcite ) and f10214 + 4724 ( @xmath8 = 2.286 ; @xcite ) . all of these galaxies have been shown to harbor a quasar nucleus obscured from direct view ( p09104 + 4109 , @xcite ; f15307 + 3252 , @xcite ; f10214 + 4724 , @xcite ) . this has led to the suggestion that perhaps all warm , @xcite ; @xcite ) ; @xmath10 and @xmath11 are the @xmath6 flux densities in units of jy at 25 @xmath12 m and 60 @xmath12 m , respectively . ] ulirgs contain buried qsos and that they may be the misdirected type 2 qsos ( agn with seyfert 2-like emission - line characteristics but qso - like luminosities ) . a diagnostic diagram ( @xcite ) involving emission line and infrared color @xmath13 does indeed show that essentially all warm ulirgs with sufficiently high ionization harbor energetic quasars in their centers . iras p09104 + 4109 has been identified with a central cd galaxy in a rich cluster ( @xcite ; hall 1997 ) . it has a high - ionization spectrum characteristic of a seyfert 2 galaxy . ground - based and @xmath14 ( @xmath15 ) images show that has an off - nuclear northern extension or `` plume '' of ionized gas ( @xcite ; @xcite ; @xcite ) that may serve as a scattering mirror of the light originating from the nucleus . we wish to probe the origin of this extension and constrain the scattering geometry for by obtaining polarimetric observations of this knot . in this paper , we confirm the high polarization and the broad emission lines of , and mg ii in polarized flux of the nucleus . in addition to the @xmath16-band imaging polarimetry , we also present new spectropolarimetric data for the nucleus and off - nuclear emission regions of . we study in detail the emission - line spectra of the nuclear and extra - nuclear regions in order to derive an understanding of the ionization mechanism of the narrow - line region ( nlr ) of this galaxy . throughout this paper , we assume @xmath17 = 75 , @xmath18 = 0 and @xmath19 = 0 . at the redshift @xmath20 of p09104 + 4109 , 1 corresponds to a projected size of 5 kpc . spectropolarimetric observations were made with the polarimeter ( @xcite ) installed in the low resolution imaging spectrometer ( lris , @xcite ) on the 10-m keck i telescope on the night of 1994 december 31 ( ut ) . a 1 wide , long slit , was centered on the nucleus of , and oriented at @xmath21 = 14 to include the extended structure . the emission - line regions extend @xmath0 5 north - northeast of the nucleus ( @xcite ) , as well as a few arcsecs south . we used a 300 grooves mm@xmath22 grating which provided a dispersion of 2.46 pixel@xmath22 and a resolution of @xmath0 10 ( fwhm ) . the observations were made by following standard procedures of rotating the half waveplate to four position angles ( 0 , 22.5 , 45 , and 67.5 ) , and dividing the exposure times equally among them . two sets of observations , lasting 60 and 52 min . , were made under photometric conditions , giving a total exposure time of 112 min . the results presented here are the average of the two sets . spectra of the nucleus were extracted from an aperture 34 wide , and those of the ne off - nuclear region were extracted from a region 36 wide centered 34 from the nucleus . this includes the entire ne plume seen in the hst images by armus ( 1999 ) . the data were reduced using vista , following standard polarimetric reduction techniques as described by cohen ( 1997 ) . hereafter , we shall refer to the nucleus spectrum as nuc and that of the ne extension as ext . imaging polarimetry was obtained on the night of 1995 december 16 with the keck i telescope . the images were taken through the @xmath16 filter , which contains mainly continuum emission . exposure times were 10 min . at each of the four waveplate positions . sky was clear but seeing was variable between 1 18 . reductions were done analogously to spectroscopic observations as described by cohen ( 1997 ) . table 1 presents the integrated emission - line flux ratios with respect to and their rest - wavelength equivalent widths for the nuc and ext spectra , shown in figure [ nucextsp ] . the emission line fluxes of nuc were measured from the starlight - subtracted spectrum . the contribution from an old stellar population represented by the elliptical galaxy ngc 821 was used to remove the starlight . the starlight fraction ranges from 5% in the blue end to a maximum of 40% in the red . we did not correct for starlight in the ext spectrum as the galaxy contribution in the outer regions is negligible . the largest source of uncertainty in the flux measurement is the placement of the continuum . for strong lines ( f / f ( ) 20% ) the uncertainty is 10% ; for weaker lines ( f / f ( ) 20% ) the uncertainty is 10%25% . the line ratios in nuc have been corrected for a small amount of extinction = 0.24 determined from the / , and / ratios , assuming the intrinsic case b values of 0.47 and 0.26 , respectively ( osterbrock 1989 ) , and the extinction curve of cardelli , clayton , & mathis ( 1989 ) with @xmath23 . this correction is small and does not significantly alter the conclusions of this paper . no corrections have been applied to ext as its observed balmer line ratios are close to case b values , indicating little extinction due to foreground dust . the small extinction derived might seem surprising , given the high infrared luminosity of this object . however , this is consistent with the result reported by veilleux , kim & sanders ( 1999 ) in their spectroscopic study of a large sample of ulirgs , that there is no obvious correlation between extinction and ir luminosity among the ir - bright galaxies . the mean for the seyfert 2 galaxies in their sample is 1.2 with a range of 0.3 to 2.7 . our reddening estimate is intermediate between those found by kleinmann ( 1988 , @xmath24 0.12 ) , who used the same optical line ratios , and soifer ( 1996 , = 0.380.79 ) , who used infrared [ s ii ] and hydrogen lines in combination with optical lines . soifer et al . also suggested that the low reddening may result from the dust being confined in a disk - like structure , which effectively obscures the central nucleus but does not block our line of sight to the nlr . as veilleux ( 1999 ) noted , the color excess derived from the optical line ratio method tends to underestimate the amount of dust in these objects . in the nucleus , the narrow lines all display a strong blueward asymmetry , as reported by kleinmann ( 1988 ) , crawford & vanderriest ( 1996 ) and hines ( 1999 ) , and can not be adequately fit with a single gaussian profile . two gaussian profiles including a main , central component with ( intrinsic ) fwhm @xmath25 600 and another component blueshifted 720980 with fwhm @xmath25 700 are able to fit the observed profiles well . our measured range of velocity shifts of the blue - shifted component is somewhat smaller than the 1250 reported by crawford & vanderriest ( 1996 ) , and more consistent with 950 reported by kleinmann ( 1988 ) . aside from the smaller line widths in ext , the most remarkable difference between the nuc and ext spectra is the complete absence of any fe species in ext , as figure [ nucextsp ] shows . the imaging polarization of is shown in figure [ polmap ] . the polarization is very high ( @xmath0 25%30% ) throughout the nucleus and extensions up to 5 away . the polarization increases strongly across the source . in the south , a number of pixels shows @xmath26 30% , and the highest is 40.3% @xmath27 12.8% . we note that the distribution of surviving vectors is biased because those which have positive errors are preferentially saved ; however , there is no doubt that the polarization is remarkably high . unlike most of the narrow - line radio galaxies like 3c 195 and 3c 33 ( @xcite ) , there is no well - defined v - shape bi - cone of scattered light , although the entire region is highly polarized . the polarization field displays a strong similarity to the patterns in the blrg fsc 2217 + 259 and the highly - polarized nlrg 3c 234 ( @xcite ; @xcite ) . to the ne and sw , the vectors are perpendicular to the radius and this suggests scattering from a central illuminating source . however , the curvature of the polarization vectors is small , as in 3c 234 and fsc 2217 + 259 . this may suggest that the viewing angle is small , indicating that the line of sight is just outside the ionization cone ( see 4.3 ) , a geometry that has also been suspected in 3c 234 ( @xcite ; @xcite ) . the southern extension appears to display somewhat higher polarization than its northern counterpart . this may be a result of forward scattering and would imply that the cone axis is pointed toward us in the south , analogous to cygnus a ( @xcite ) . dilution by line emission is not a major factor since there are no strong emission lines through the @xmath16 band used . however , dilution by hot stars or nebular continuum could play an important role in the north as there is more ionized gas there . our imaging polarimetry confirms and extends the polarization map obtained with @xmath15 by hines ( 1999 ) , which shows a reflection `` cone '' in the central 1 . the right panel of figure [ polmap ] shows the polarized flux @xmath28 which is commonly called the `` polarized light '' . as can be seen , the polarized light distribution extends more to the north than to the south , reaching up to @xmath0 4 ( 20 kpc ) to the northeast along @xmath21 = 19 . this coincides well with the extended [ o iii ] emission seen in the ground - based ( @xcite ) and @xmath15 images ( @xcite ) . the polarized extension is thus better aligned along this emission - line structure ( @xmath29 11 ) than the radio axis ( @xmath21 = 333 ) , a condition also exhibited by many high and low - redshift radio galaxies ( @xcite ; @xcite ) . this indicates that ext serves as a reflection region of nuclear light , and that it lies exposed to the ionizing radiation escaping through the ionization cones from the hidden quasar . the polarized flux distribution peaks at the total flux peak , suggesting that scattered light dominates the observed radiation . the spectropolarimetry for nuc is displayed in figures [ nucspol ] . high polarizations are seen , with magnitudes consistent with those seen in the imaging polarimetry ( fig . [ polmap ] ) . the keck data of the nucleus confirm the general results of hines ( 1993 , 1999 ) . the observed continuum polarization uncorrected for starlight ranges from @xmath0 8% at 6000 ( rest ) to about 18% at 3000 (rest ) . the polarization @xmath21 is constant with wavelength at 97 . in addition to broad , , and probable fe ii multiplets near 3200 reported by hines , we also detect broad mg ii in polarized flux . we measured a fwhm of 7540 @xmath27 700 for the broad in polarized flux , somewhat smaller than that reported by hines ( 1999 , 12,000 @xmath27 2500 ) . this is comparable to the fwhm of 6000 @xmath27 700 measured for . the resemblance of the nuclear to a quasar spectrum also extends to its shape . the galaxy - corrected total flux spectrum ( @xmath30 , @xmath31 ) is redder than ( @xmath32 ) , which is similar in slope to the more core - dominated composite qso spectrum of baker & hunstead ( 1995 , @xmath33 ) , or that of francis ( 1991 , @xmath34 ) . our measured spectral indices agree well with those of hines ( 1999 ) . this suggests that there is a second `` featureless continuum '' fc2 in the total flux spectrum that makes it redder . this component is likely due to a combination of hot stars and nebular continuum plus reddened transmitted light from the hidden agn ( e.g. , @xcite ) . quasars viewed at high inclination angle have been shown to exhibit redder continuum due to dust extinction ( @xcite ) . the presence of hot stars , perhaps arising in a starburst , is suggested by the broad feature underlying he ii 4686 . this feature can not be a broad scattered he ii component since it appears very strongly in the total flux spectrum , suggesting that the light is viewed directly , and it is not detected in the spectrum ( see fig . [ nucspol ] ) . it can be attributed to wolf - rayet ( w - r ) stars , as has been identified by heckman ( 1997 ) in mrk 477 and storchi - bergmann , cid fernandes , & schmitt ( 1998 ) in mrk 1210 , and tran ( 1999 ) in tf j1736 + 1122 . in addition , there is nebular continuum giving rise to the balmer jump near 3650 . if this is the case we would expect the balmer jump to be unpolarized . is it ? to answer this , we have constructed an `` unpolarized spectrum '' = @xmath35 @xmath36 , assuming an intrinsic continuum polarization of 20% . we then measured the balmer jump index defined as @xmath37 , where @xmath38 is the continuum flux just blueward of 3727 , and @xmath39 that just redward of it . we obtain an index of 1.74 @xmath27 0.1 for the total flux spectrum , and 2.0 @xmath27 1 for the unpolarized flux . naturally , there is a larger uncertainty in the value obtained from the unpolarized flux spectrum . however , a balmer discontinuity appears to still be present , meaning that it is unpolarized , and supporting the notion that fc2 contains nebular continuum plus starlight from a starburst . we find that @xmath40 , the intrinsic polarization produced by scattering , is @xmath40 ( ) @xmath25 16% while @xmath40(mg ii ) @xmath25 29% ( see below ) . this strengthens the reality of the nebular continuum discussed above , since a rising @xmath40 toward the blue would leave most of the balmer discontinuity in the unpolarized flux spectrum . this blue @xmath40 is possible with dust but not electron scattering . thus , dust scattering is important , while electron scattering could contribute but not dominantly . the polarization of ext is noisier than nuc , and we have binned the results ( except for ) , presented in figure [ extspol ] , by 5 pixels to improve s / n . as can be seen , the polarization shows similar behavior and magnitude to those in nuc . in addition , a broad mg ii is clearly seen in the total flux spectrum . this indicates that the continuum in ext is dominated by scattered light from the obscured quasar . the polarization @xmath21 is roughly constant with wavelength at 106 @xmath27 2 , slightly higher than that measured in the nucleus . the @xmath21 is accurately perpendicular to the slit , as expected for an off - nuclear measurement of a reflection nebula ( see @xcite for similar cases ) . we show the ext smoothed by 7 pixels in figure [ extpf ] , which shows clearly that broad and mg ii are present . in , the spectral index is @xmath41 , significantly bluer than the average quasar spectrum . assuming that the incident spectrum is typical for a quasar , this suggests that , as for the nucleus component , dust is the dominant scatterers which could bluen the light . we model the spectropolarimetric observations using the procedure described by tran ( 1997 ) , to separate the various components of the emission lines and continuum . in a region surrounding the + complex , we perform a fit of the continuum and emission - line components in the total flux , , and spectra . the broad is assumed to have the same profile as in the polarized flux spectrum . the narrow lines of and were fitted assuming that they are similarly polarized , and consist of one main core component and another blue - shifted component , both with gaussian profiles . our keck data do not have sufficient spectral resolution to warrant the addition of more than two components in the profile , and we do not attempt to fit the line polarizations with the three narrow components identified by hines ( 1999 ) . the results indicate that the continuum and broad are similarly polarized at 16% @xmath27 3% and @xmath21 = 98 . similar modeling at the mg ii line shows that the polarization of the broad mg ii line , at 29% , is substantially higher than broad , but is consistent with values seen in the imaging polarimetry . if real , it suggests that the intrinsic polarization rises to the blue , which is only possible if dust scattering is present . our fitting also shows that the core component of the narrow lines has a small amount of polarization , 0.70% @xmath27 0.03% at @xmath21 = 90 , which is different from that of the broad line and continuum . on the other hand , the 3727 line is virtually unpolarized . we measure a polarization of 0.22% @xmath27 0.12% . this indicates radial stratification of the nlr gas ( see below ) . the high s / n data allowed us to measure the polarization of various emission lines in the nucleus . the results are in table 2 , where we list the ( biased ) , , their associated errors , the ionization potential ( ip ) of the lower and upper stages of ionization for the ions of interest , and the critical density ( @xmath42 ) for collisional de - excitation of each line ( calculated at 10@xmath43 k ) . the polarizations are flux - weighted means in the core component only , and the uncertainties on and are 1@xmath44 statistical errors alone and do not include systematics such as those due to continuum placement and data reduction . in figure [ linep ] , we plot the measured line polarizations versus @xmath42 , the mean of the lower and upper ip , and fwhm . a clear trend is apparent in all three plots : lines of higher @xmath42 , ip and fwhm exhibit higher polarization . such a correlation has also been observed in the liner ngc 4258 ( @xcite ) . to our knowledge , ngc 4258 and are the only two agns where this phenomenon has been shown to occur systematically over a number of forbidden lines with a wide range of ionizations . the tendency for to be more polarized than , especially in radio galaxies ( @xcite ) , has been known for some time , and an anisotropic emission structure has been suspected as the cause ( @xcite ) . together with the well - known correlation of linewidth with @xmath42 and ip ( e.g. , @xcite ) in seyfert galaxies , this result provides strong indication that the narrow emission - line structure in is radially stratified , with lines of higher ionization and critical density being emitted closer to the nucleus , and therefore more likely to be partially occulted . the higher polarization results from less dilution by direct light . the temperature determined from the 5007/ 4363 ratio ( using table 1 ) is @xmath25 1.3@xmath45 k for both nuc and ext . we also used the 6583 line from the near - ir spectrum of evans ( 1998 ) to measure the temperature from the 5755/6583 ratio . this is only available for nuc , giving @xmath46 k. this is about 4000 k cooler than the region , and may suggest that matter - bounded clouds are present ( e.g. , @xcite ) . the electron density estimated from the 4712/4740 ratio is @xmath47 and 5,000 for nuc and ext , respectively . this can be compared to a value of 6,000 for nuc determined from [ cl iii ] 5518/5538 line ratio . the [ ar iv ] 4712/4740 ratio is clearly better determined than the [ cl iii ] 5518/5538 ratio since the stronger ar lines are less affected by starlight subtraction and yield more certain flux measurements . to help us understand the rich emission - line spectrum of , we compare the line ratios with model calculations . in particular , we would like to understand the source of the ionization and the excitation mechanism of a large number of emission lines encompassing a wide range of ionization , from the low - excitation [ n i ] 5200 , and [ s ii ] to the coronal lines such as and . theoretical models from three main mechanisms are considered : \1 ) simple photoionization of essentially radiation - bounded nlr clouds by the central source having a powerlaw radiation spectrum . the emergent emission - line spectrum is characterized by the slope @xmath48 of the ionizing continuum , and the ionization parameter @xmath49 , where @xmath50 is the number of ionizing photons , @xmath51 is the radius and @xmath52 is the electron density . we build the agn models with the photoionization code mapping ic ( @xcite ) , using a power - law with index @xmath53 . the models are isobaric , so the density varies across the nebula . recently , a new set of models has been computed for the nlr in an attempt to account for more realistic physical conditions . in these models the line emission originates from a wide range of temperatures and densities , although the observed emission - line spectrum results only from clouds best able to emit it . these are called `` locally optimally emitting clouds '' ( loc ) models ( ferguson 1997a , 1997b ) . we shall discuss these models but compare our observations only with the simple , isobaric calculations that we made . \2 ) two - component photoionization model of binette ( 1996 , 1997 ) , consisting of both optically thin gas ( matter - bounded , mb ) clouds and optically thick gas ( ionization - bounded , ib ) clouds . the sequence is parameterized by the parameter which is the ratio of the solid angle subtended by mb clouds relative to that of ib clouds , as viewed by the observer . most of the higher ionization lines ( i.e. , coronal lines and [ o iii ] ) are emitted by mb clouds , which lie closer to the nucleus , filtering the radiation field that reaches the ib clouds further away , where essentially all of the lower ionization lines ( e.g , [ n ii ] , [ o i ] ) originate ( @xcite ; @xcite ) . a large value of indicates a larger weight to the mb component , and therefore a higher excitation spectrum . \3 ) shocks and shock + precursors models of dopita & sutherland ( 1995 , 1996 ) . mechanical energy deposited by velocity shocks due to turbulent cloud motions or the outflowing radio jets can produce powerful local uv radiation which can ionize the gas . two types of shock models have been considered : pure shocks , in which mechanical energy from shocks is solely responsible for collisionally heating the gas , and the hybrid shocks + precursors , also known as `` photoionizing shocks '' , in which the shock itself generates an abundant supply of ionizing photons in the hot post - shock gas and produce an extensive precursor h ii region . in figures [ shock][ibmb ] , we present several diagnostic diagrams involving a number of emission lines with a wide range of ionization as an attempt to identify the main underlying mechanism for ionizing the nlr gas in nuc and ext . model calculations based on the above three mechanisms are shown for comparison with the observations . simple power - law @xmath5 sequences are plotted for density @xmath54 ( of the front layer ) as determined from the nuclear gas . sequences with @xmath55 ( as determined from the extended gas ) produce very similar results and we do not plot them on the diagrams . two different sequences are shown for the ib+mb photoionization models of binette ( 1997 ) , each characterized by the value of the ionization parameter _ at the irradiated face of the mb component _ , @xmath56 : high @xmath57 and low @xmath58 . the ionization parameter of the ib component is @xmath59 for both . on all figures , the parameter @xmath60 goes from 0.01 to 100 , increasing from left to right . pure shocks and shocks + precursors models are plotted as grids parameterized by the shock velocity @xmath61 and the magnetic parameter @xmath62 . @xmath61 varies from 150 to 500 , and @xmath62 varies from 1 to 4 @xmath12 g @xmath63 . all models are shown for solar abundances . the symbols have the following meanings : star denotes ext and open circle represents the total line flux from both the core and blue - shifted components in nuc . the solid and open triangles indicate the core and blue - shifted emission line components of nuc , respectively . we shall now discuss these diagrams . figure [ shock ] shows that in two of the best optical diagrams for shock diagnostics , ( 5007/4363 vs. he ii 4686/ proposed by @xcite , @xcite , and [ ne v]@xmath33426/[ne iii]@xmath33869 vs. 5007/ proposed by @xcite ) , the positions of the nuc component lie well outside the range of model grids predicted by pure shocks and shocks + precursors . the shock + precursor models are a little more consistent with the points for the ext gas . these diagrams clearly rejects pure shock as the dominant emission line mechanism in both the nuclear and extra - nuclear regions of , as it tends to predict too low 5007 . on the other hand , ionization - bounded agn photoionization models have the opposite problem , failing to agree with the observed data by predicting too high 5007 . the presence of dust ( see 3.3 , 4.1 ) in the ext also weakens considerably any arguments that shocks play a dominant role in the main energy input mechanism . if shocks were present the dust grains would likely be destroyed ( donahue & voit 1993 ; de young 1999 ) . another argument against shocks is that the radio and emission line `` cone '' in are _ misaligned _ by as much as 38 ( hines 1999 ) , making shocks induced by the nuclear jets unlikely . also , the 5007/ 3727 line ratio ( a good indicator of the ionization parameter @xmath5 ) is observed to be 5.4 for nuc and 2.11 for ext . thus @xmath5 does not go up or remain constant in the extra - nuclear regions , as would be the case if shocks + precursors provide extra local heating ( e.g. , as in 3c 299 , @xcite ) , but actually drops , as expected from geometric dilution of the radiation field . finally , the velocity widths of the emission lines in the extension are essentially unresolved @xmath24 550 , indicating that perturbations in the gas velocity by shocks , if any , are small . in conclusion , we can rule out pure shocks as the ionizing mechanism . shock + precursors are also ruled out in the nuclear region , but it is more ambiguous whether they exist in the extension . crawford & vanderriest ( 1996 ) have suggested that shock heating may increasingly become important in the extended gas . however , we do not find sufficient evidence in our data to support a dominant role by shocks even in the extensions . more conclusive resolution between shocks and photoionization may come from additional diagnostics using uv lines ( @xcite ; @xcite ) , which are more sensitive to shocks than optical lines . as figure [ shock ] indicates , the models for the high - density ib photoionization sequence predict too high 5007 , typically giving / 16 . this has been a well - known problem for ib photoionization models , known as the `` temperature problem '' : the models predict too low a 4363/5007 ratio and thus too low electronic temperatures ( i.e. , @xcite ) . one attractive possibility that may solve this discrepancy is with the addition of mb clouds . as already mentioned , the presence of mb clouds is suggested by the higher temperature in the emitting region compared to that in the emitting region in the nucleus , since the mb clouds , where most of the originates , are expected to lie closer to the ionizing source ( @xcite ; @xcite ) . in figure [ ibmb ] , we present additional diagnostic line ratios with comparison to the ib + mb as well as shock + precursor models . the top two panels are the same as figure [ shock ] . as can be seen , the shock + precursor models clearly fail to fit all the nuc data points and miss the ext points in most diagrams . in general , the addition of mb clouds is able to resolve the two main difficulties of pure ib photoionization models : lines ratios of he ii/ and 5007/4363 . however , these sequences also have problems . for example , a common problem for all photoionization models is that [ ne v ] is overpredicted ( fig . [ ibmb ] , top right ) . note also that only models with @xmath64 are able to fit the observed he ii , but they have difficulty reproducing the high 5007/ , especially the low @xmath56 sequence ( solid curve ) . the high @xmath56 sequence ( dotted curve in the plots ) is consistent with _ both _ the observed he ii and ratios , but it can not reproduce the low - ionization lines [ o ii]/ , [ n i]/ and [ s ii]/ , which are predicted to be too strong . the low @xmath56 sequence has the opposite problem , agreeing well with these low - ionization lines but failing to explain higher ionization lines like and , as well as the weakness of he ii for the observed [ o iii]/. since it is the contribution from mb clouds that makes it possible to solve the he ii and temperature problems , and since the low - ionization lines are mostly due to the ib component , a higher ionization parameter for the ib component might be consistent with all data ( recall that @xcite used a fixed @xmath5[ib ] ) . in summary , figure [ ibmb ] shows that shocks + precursors models fails to reproduce the observed data points in nuc and , to a lesser degree , in ext . photoionization models with contribution from matter - bounded clouds are able to reproduced the data well , although no single photoionization sequence is perfect . in 3.3 , we presented polarimetric evidence for dust in the scattering regions . diagnostic line ratios also show evidence for dust in the emission line gas . one of the most notable features of the emission line spectra of is the conspicuous absence of all species of fe in ext and their strong presence in nuc ( see fig . [ nucextsp ] and table 1 ) . no other atomic species other than iron is seen to exhibit this behavior . this dramatic difference in fe line strength between nuc and ext can not be due to the lack of high ionization photons in ext , for we see clearly in ext strong high - ionization lines of [ ne v ] , which has comparable ionization potential ( 97.1 ev ) to [ fe vii ] ( 100.0 ev ) . furthermore , lower ionization fe lines like [ fe vi ] , [ fe v ] and perhaps [ fe iii ] are all completely absent from ext . coronal lines can be expected to be seen in the tenuous off - nuclear extended gas ( @xcite ) , and indeed is seen in our ext spectrum , and [ fe ix ] has been observed in the extended emission line regions of circinus ( @xcite ; @xcite ) , cygnus a ( @xcite ) and tololo 0109@xmath65383 ( murayama , taniguchi , & iwasawa 1998 ) . the peak range of the line strength distribution function in the loc models of ferguson for and overlap substantially with each other . therefore , under `` ordinary '' nlr conditions ( i.e. , solar abundances , free of dust ) we would expect to see wherever is seen . if is as strong relative to other lines as observed in the nucleus , there should be no problem detecting it in ext . in nuc , 6087 is about 3 times weaker than 3246 and 5.6 weaker than . at this level , should be even stronger than [ n i ] 5200 , or about as strong as he i 4471 , both of which are clearly seen in ext . even if it is 613 times weaker , as predicted by models of binette ( 1997 ) or ferguson ( 1997 ) , it should be at least about as strong as [ ar iv ] 4712 , 4740 or he i 5876 . somehow , virtually all gaseous fe is removed from the extension gas . with the inclusion of dust grains , is expected to be about 74 times weaker than ( @xcite ) . at this level , it would be impossible to detect with our data . this strongly suggests that there is significant amount of dust in the extension allowing gas - phase fe to deplete onto grains . this notion has been invoked to explain the emission - line spectra of seyfert galaxies ( kingdon , ferland , & feibelman 1995 ) . neon is an inert gas and not expected to be depleted ( see model calculations by @xcite ) . 6087 is a coronal line , and both ferguson ( 1997b ) and binette ( 1997 ) have suggested that coronal lines are not efficiently emitted in gas with dust . the lack of 6078 in ext therefore is consistent with the presence of dust . it is unlikely that the absence of fe lines in ext is due to the lower metallicity of the ext region , for other metals such as oxygen and nitrogen seem to be well represented . we would expect fe to show a normal abundance as there exists a well - known [ fe / h ] to [ o / h ] relation for the nebula gas observed in the solar neighborhood and galactic bulge ( e.g. , maciel 1999 ) . another element which may lend itself well to diagnostics for dust is mg since it is about as sensitive as fe to depletion . unlike the higher - ionization forbidden fe lines , is complicated by several factors . it is a resonance line so it can be strongly affected by intervening dust extinction and gas absorption . on the other hand , like other low - ionization lines such as , , it may even get strengthened by photoelectronic heating of the gas by dust grains ( @xcite ) . with these caveats , we now consider whether the resonant character of mg ii 2800 and its strong sensitivity to dust and gas absorption suggest the presence of dust as well . figure [ mgii ] shows the diagram of mg ii 2800/ 3727 vs. 5007/. the mb + ib photoionization models predict mg ii to be stronger than observed in both the nuclear and extended gas . shock + precursor models show reasonable agreement to the data , but they fail for most other lines , and have been eliminated for the nuclear emission ( see 3.4.2 , 3.4.3 ) . one way to weaken mg ii 2800 without dust is with intervening absorbing gas , containing mg ii . since mg ii is in ionized form , the most likely explanation is that the absorbing gas is in the ionized cones . we compare the predictions of the photoionization models taking into account viewing effects : depending on the viewing angle of the observer relative to the clouds , the spectrum of the resonant lines , like mg ii 2800 will be different ( villar - martin , binette & fosbury 1996 ) . if the clouds are viewed from the illuminated face , the mg ii emission will be relatively stronger than when seen from behind . in the former case , the photons find a lower column density of mg ii ( since the gas is highly ionized towards the illuminated face ) and they can escape the nebula freely . however , photons that try to leave the cloud through the `` non - illuminated '' face of the cloud will encounter a much larger column density of mg ii and will be repeatedly absorbed . in figure [ mgii ] we show this effect . the upper thick solid line shows the line ratios for a nebula viewed from the illuminated surface . the lower thick solid line shows the line ratios for a nebula seen from behind . as can be seen , the difference is large . these models show that we need to view the clouds directly from the back ( the non - illuminated side ) to reduce the mg ii emission to the low observed values . however , this extreme view is probably not the case , for if the clouds are seen from behind , the observer would be inside the ionization cone and therefore , would have a full view of a quasar or broad line radio galaxy . another possibility is that the absorbing gas is outside the ionized cones , but in this case the ionization of the gas can not be explained . thus again , the most reasonable explanation is that the emitting gas contains dust . it could be due to depletion ( mg is about as sensitive to depletion as fe ) , or to a combination of resonant scattering and dust absorption . pure resonant scattering alone can not explain the weakness of the line . this conclusion is consistent with the indication for dust from the absence of fe lines , as well as the polarized flux spectrum , which is blue compared to a composite quasar spectrum ( 3.3.2 ) . while the indication for dust in the nucleus and extended emission regions of comes as no surprise from one of the most luminous infrared emitters , this confirmation shows that the existence of dust is extensive over a large scale , and could have important implications for the origin of the extended gas . as suggested by donahue & voit ( 1993 ) for the intracluster nebular gas that they studied , the presence of dust inferred in the extension implies that ext can not be gas condensed directly out of the x - ray emitting intra - cluster medium ( icm ) . this is because dust simply can not survive in the hot icm in any appreciable amount . any dust introduced in the hot icm is expected to be destroyed long before the gas is cooled enough to be visible or gets transported to the cluster center by the cooling flow . villar - martin & binette ( 1997 ) reached a similar conclusion in their finding of dust in the extended emission - line regions of radio galaxies . since there is no evidence that the metal abundances are anomalously low ( 4.1.1 ) , the most likely candidate for the extended gas is perhaps shreds of unfortunate members of the cluster being cannibalized by . active star forming regions , either in the galaxies themselves or triggered by the interactions , are where copious amounts of dust are found . this in turn implies that the extranuclear gas is unlikely to be material expelled or outflowing from the nucleus of the central quasar . @xmath15 wfpc2 imaging of by armus ( 1999 ) does indeed show what seem like whiskers and filaments , suggestive of stripped remains from nearby cluster members . in addition , recent @xmath15 wfpc2 and nicmos imaging studies of ulirgs by surace ( 1998 ) and scoville ( 2000 ) have shown that many galaxies contain numerous circumnuclear clusters of bright young stars ( age @xmath0 10@xmath6610@xmath67 yrs ) , whose forming regions are often associated with dust . the external origin for the gas is supported by the fact that many ulirgs show evidence for interaction or merging ( e.g. , duc , mirabel & maza 1997 ; borne et al . 2000 ; @xcite ) , which has been suggested to stimulate the ulirg phase . velocity structure of the extended nebula of derived from integral field spectroscopy by crawford & vanderriest ( 1996 ) also suggests that the gas is probably external in origin . however , these authors do not favor the picture of the gas being merged with the cd galaxy , based on the low relative velocity ( @xmath0 100200 ) with respect to the nucleus . the polarizations of ulirgs with hidden broad - line regions are generally very high ( @xmath0 20%30% ) , suggesting that the inclination angle could be fairly substantial ( 35 ) . under the assumption that the lack of a well - defined v - shape polarization fan suggests that the inclination angle @xmath68 does not exceed the half - opening cone angle @xmath69 by a large amount , we can use the morphology of the polarization image ( fig . [ polmap ] ) to constrain the scattering geometry of , using the model of brown & mclean ( 1977 ) ( see also @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . analysis by hines ( 1999 ) has constrained the viewing inclination and half opening cone angle lie in the range @xmath70 and @xmath71 , based on = 20% . with our higher polarization of @xmath0 30% from the broad mg ii and imaging polarimetry , these limits are still consistent , but we can eliminate the combination of both upper limits or lower limits for @xmath68 and @xmath69 ( i.e. , @xmath72 @xmath73 , or @xmath74 @xmath75 ) , as they give too low . the same applies for the combination @xmath74 , @xmath73 , in which case our line of sight is just outside the cone , but is only 15% . agrees better with observations if @xmath68 and @xmath69 are closer to the upper and lower limits , respectively , but in this case their difference @xmath76 is too large . if we limit the half opening cone angle to be 40 constrained by the @xmath15 image of hines and armus ( 1999 ) , the difference @xmath77 is minimized to 10 for @xmath78 and @xmath79 . therefore , this appears to be the best scattering geometry for consistent with observations . our high - quality keck spectropolarimetric and imaging polarimetric data confirm the high polarization and broad balmer emission lines in the polarized flux spectra of . in addition , we also detect broad mg ii in both total and polarized flux spectra of the nuclear and extension regions , indicating the presence of a hidden quasar visible only in scattered light . the narrow - line polarizations exhibit a strong positive correlation with line width , critical density and ionization potential of the transition , indicating that the line emission arise from a radially stratified gas . the lack of a clear fan - like morphology in the polarization image suggests that our viewing angle ( @xmath80 50 ) is not far outside the ionization cone ( @xmath2 40 ) of . the emission - line spectra of are consistent with agn photoionization with contribution from both ionization - bounded and matter - bounded clouds . under the assumption that photoionization by the central hidden quasar is the main operating mechanism , should be detected in the extended gas and mg ii should be much stronger than observed . the absence of all fe lines and weakness of mg ii in the extension can be readily explained if dust is mixed with the ionized gas of the nlr . furthermore , the fact that both the nuclear and the extended continua are highly polarized with relatively blue polarized flux spectra compared to that of the average quasar also supports the existence of dust , which could efficiently scatter and bluen the light . the other possibility is that the extended gas simply is highly deficient in iron , or is metal poor . this does not seem very likely , since the strengths of the other metal lines appear to be in good agreement with solar abundance . since grains remain intact , this in turns implies that shocks are probably not important in the extra - nuclear regions . although the diagnostic diagrams offer somewhat inconclusive assessment of the shocks+precursor models for the extension region , there is no strong evidence for shocks in from the line ratios . ionization by pure shocks can be ruled out by the data . the existence of dust and lack of strong shocks in the extended gas also suggest that it is probably dismembered remnants of a cluster neighbor being disrupted or in the process of merging with the central cd galaxy . two predictions follow directly from our interpretation of the current data : the near - ir line [ fe ii ] 1.257 @xmath12 m has been observed in the nucleus of by soifer ( 1996 ) , but should be absent from the extension , confirming the absence of gaseous fe , and hence the lack of shocks and presence of dust there . likewise , [ ca ii ] 7291 , 7324 should be absent in ext spectrum , since ca is more than 10 times more sensitive to grain depletion than fe or mg ( @xcite ) . we thank p. ogle and a. putney for assistance in the observations , and l. binette for the use of his code mapping ic . we are grateful to the referee for several comments , which have improved the clarity of the paper . the w. m. keck observatory is operated as a scientific partnership between the california institute of technology and the university of california , made possible by the generous financial support of the w. m. keck foundation . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . allen , m. g. , dopita , m. a. , & tsvetanov , z. i.1998 , , 493 , 571 armus , l. , soifer , b. t. , & neugebauer , g. 1999 , in `` ultraluminous galaxies : monsters or babies '' , proc . of the ringberg workshop , 1999 , in press , astro - ph/9903332 baker , j. c. 1997 , , 286 , 23 baker , j. c. , & hunstead , r. w. 1995 , , 452 , l95 balsara , d. s. , & krolik , j. h. 1993 , , 402 , 109 barth , a. j. , tran , h. d. , brotherton , m. s. , filippenko , a. v. , ho , l. c. , van breugel , w. , antonucci , r. , & goodrich , r. w. 1999 , , 118 , 1609 binette , l. , wilson , a. s. , & storchi - bergmann , t. 1996 , , 312 , 365 binette , l. , wilson , a. s. , raga , a. , & storchi - bergmann , t. 1997 , , 327 , 909 borne , k. d. , bushouse , h. , lucas , r. a. , & colina , l. 2000 , , 529 , l77 brown , j. c. , & mclean , i. s. 1977 , , 57 , 141 brotherton , m. s. , wills , b. j. , dey , a. , van breugel , w. , & antonucci , r. 1998 , , 501 , 110 burstein , d. & heiles , c. 1984 , , 54 , 33 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , , 345 , 245 clark , n. e. , tadhunter , c. n. , morganti , r. , killen , n. e. b. , fosbury , r. a. e. , hook , r. n. , siebert , j. , & shaw , m. a. 1997 , , 286 , 558 cohen , m. h. , vermeulen , r. c. , ogle , p. m. , tran , h. d. , & goodrich , r. w. 1997 , , 484 , 193 cohen , m. h. , ogle , p. m. , tran , h. d. , goodrich , r. w. , & miller , j. s. 1999 , , 118 , 1963 condon , j. j. , huang , z .- p , yin , q. f. , & thuan , t. x. 1991 , , 378 , 65 crawford , c. s. , & vanderriest , c. 1996 , , 283 , 1003 cutri , r. m. , huchra , j. p. , low , f. j. , brown , r. l. , & vanden bout , p. a. 1994 , , 424 , l65 de robertis , m. m. , & osterbrock , d. e. 1986 , , 301 , 727 de young , d. s. 1998 , , 507 , 161 di serego alighieri , s. , cimatti , a. , fosbury , r. a. e. , & he s , r. 1997 , , 328 , 510 donahue , m. , & voit , g. m. 1993 , , 414 , l17 dopita , m. a. , koratkar , a. p. , allen , m. g. , tsvetanov , z. i. , ford , h. c. , bicknell , g. v. , & sutherland , r. s. 1997 , , 490 , 202 dopita , m. a. , & sutherland , r. s. 1995 , , 455 , 468 dopita , m. a. , & sutherland , r. s. 1996 , , 102 , 161 duc , p .- a . , mirabel , i. f. , & maza , j. 1997 , , 124 , 533 evans , a. s. , sanders , d. b. , cutri , r. m. , radford , s. j. e. , surace , j. a. , soloman , p. m. , downes , d. , & kramer , c. 1998 , , 506 , 205 feinstein , c. , macchetto , f. d. , martel , a. r. , sparks , w. b. , & mccarthy , p. j. 1999 , , 526 , 623 ferguson , w. j. , korista , k. t. , baldwin , j. a. & ferland , g. j. 1997 , , 487 , 122 ferguson , w. j. , korista , k. t. , & ferland , g. j. 1997 , , 110 , 287 ferruit p. , binette l. , sutherland r. , & pcontal e. 1997 , , 322 , 73 francis , p. j. , hewett , p. c. , foltz , c. b. , chaffee , f. h. , weymann , r. j. , & morris , s. l. 1991 , , 373 , 465 genzel , r. , lutz , d. , sturm , e. , egami , e. , kunze , d. , moorwood , a. f. a. , rigopoulou , d. , spoon , h. w. w. , sternberg , a. , tacconi - garman , l. e. , tacconi , l. , & thatte , n. 1998 , , 498 , 579 goodrich , r. w. , miller , j. s. , martel , a. , cohen , m. , tran , h. d. , ogle , p. m. , & vermeulen , r. c. 1996 , , 456 , l9 hall , p. b. , ellingson , e. , green r. f. 1997 , , 113 , 1179 heckman , t. m. , gonzalez - delgado , r. , leitherer , c. , meurer , g. r. , krolik , j. , wilson , a. s. , koratkar , a. , & kinney , a. 1997 , , 482 , 114 he s , r. , barthel , p. h. , & fosbury , r. a. e. 1993 , nature , 326 , 362 hines , d. c. , & wills , b. j. 1993 , , 415 , 82 hines , d. c. , schmidt , g. d. , smith , p. s. , cutri , r. m. , low , f. j. 1995 , , 450 , l1 hines , d. c. , schmidt , g. d. , wills , b. j. , smith , p. s. , & sowinski , l. g. 1999 , , 512 , 145 hutchings , j. b. , & neff , s. g. 1988 , , 96 , 1575 kingdon , j. , ferland , g. j. , & feibelman , w. a. 1995 , , 439 , 793 kleinmann , s. g. , hamilton , d. , keel , w. c. , wynn - williams , c. g. , eales , s. a. , becklin , e. e. , & kuntz , k. d. 1988 , , 328 , 161 low , f. j. , huchra , j. p. , kleinmann , s. g. , & cutri , r. m. 1988 , , 287 , 95 maciel , w. j. 1999 , , 351 , l49 miller , j. s. , & goodrich , r. w. 1990 , , 355 , 456 miller , j. s. , goodrich , r. w. , & mathews , w. g. 1991 , , 378 , 47 moorwood , a. f. m. , lutz , d. , oliva , e. , marconi , a. , netzer , h. , genzel , r. , sturm , e. , & de graauw , t. 1996 , , 315 , l109 murayama , t. , taniguchi , y. , & iwasawa , k. 1998 , , 115 , 460 ogle , p. m. , cohen , m. h. , miller , j. s. , tran , h. d. , fosbury , r. a. e. , & goodrich , r. w. 1997 , , 482 , l37 oke , j. b. , cohen , j. g. , carr , m. , cromer , j. , dingizian , a. , harris , f. h. , labreque , s. , lucinio , r. , schaal , w. , epps , h. , & miller , j. 1995 , , 107 , 375 oliva , e. , salvati , m. , moorwood , a. f. m. , & marconi , a. 1994 , , 288 , 457 osterbrock , d. e. 1989 , astrophysics of gaseous nebulae and active galactic nuclei , ( mill valley : university science books ) rowan - robinson , m. , broadhurst , t. , lawrence , a. , mcmahon , r. g. , lonsdale , c. j. , oliver , s. j. , taylor , a. n. , hacking , p. b. , conrow , t. , saunders , w. , ellis , r. s. , efstathiou , g. p. , & condon , j. j. 1991 , nature , 351 , 719 sanders , d. b. , soifer , b. t. , elias , j. h. , madore , b. f. , mathews , k. , neugebauer , g. , scoville , z. 1988 , , 325 , 74 sanders , d. b. , & mirabel , i. f. 1996 , , 34 , 749 scoville , n. z. 2000 , , 119 , 991 soifer , b. t. , neugebauer , g. , armus , l. , & shupe , d. l. 1996 , , 111 , 649 storchi - bergmann , t. , cid fernandes , r. , & schmitt , h. r. 1998 , , 501 , 94 surace , j. a. , sanders , d. b. , vacca , w. d. , veilleux , s. , & mazzarella , j. m. 1998 , , 492 , 116 tran , h. d. 1995 , , 440 , 565 tran , h. d. , cohen , m. h. , & goodrich , r. w. 1995 , , 110 , 2597 tran , h. d. , filippenko , a. v. , schmidt , g. d. , bjorkman , k. s. , jannuzi , b. t. , & smith , p. s. 1997 , , 109 , 489 tran , h. d. , cohen , m. h. , ogle , p. m. , goodrich , r. w. , & di serego alighieri , s. 1998 , , 500 , 660 tran , h. d. , brotherton , m. s. , stanford , s. a. , van breugel , w. , dey , a. , stern , d. , & antonucci , r. 1999 , , 516 , 85 veilleux , s. , kim , d .- , & sanders , d. b. 1999 , , 522 , 113 villar - martin , m. , binette , l. , & fosbury , r. a. e. 1996 , , 312 , 751 villar - martin , m. , & binette , l. 1997 , , 317 , 350 villar - martin , m. , tadhunter , c. , morganti , r. , axon , d. , & koekemoer , a. 1999 , , 307 , 24 wills , b. j. , wills , d. , evans , n. j. , natta , a. , thompson , k. l. , breger , m. , & sitko , m. l. 1992 , , 400 , 96 wilson , a. s. , binette , l. , & storchi - bergmann , t. 1997 , , 482 , 131l llllllll he ii 2733 & 0.0275 & 2.7 & 712 & & & & [ mg v ] 2783 & 0.07183 & 7.0 & 1449 & & & & mg iin 2798 & 0.2584 & 25.0 & 1470 & & 0.3110 & 17.4 & 840 mg iib 2798 & 0.3579 & 36 . & 14189 & & 0.71053 & 55.6 & 14800 he i 2830 + o iii 2836 & 0.08345 & 8.1 & & & 0.0324 & 1.84 & [ ar iv ] 2854 & 0.00638 & 0.52 & & & 0.0416 & 2.74 & [ ar iv ] 2868 & 0.00308 & 0.27 & & & & & [ mg v ] 2928 & 0.0176 & 1.8 & 1481 & & & & he i 2946 & 0.00995 & 0.95 & 762 & & & & o iii 3047 & 0.0355 & 3.45 & & & & & o iii 3133 & 0.1217 & 11.74 & & & 0.0367 & 2.79 & he i 3188 & 0.0434 & 4.2 & 1200 & & 0.0361 & 2.78 & he ii 3203 & 0.05546 & 5.3 & 635 & & 0.06463 & 5.09 & o iii 3312 & 0.0200 & 2.0 & & & & & 3346 & 0.2054 & 20.0 & 811 , 674 & & 0.07033 & 5.65 & [ ne v ] 3426 & 0.5339 & 53.1 & 811 , 595 & & 0.1561 & 12.7 & o iii 3444 & 0.0406 & 4.03 & 1044 & & & & [ fe vii ] 3588 & 0.0476 & 4.76 & 635 , 635 & & & & 3727 & 1.556 & 185.0 & 762 , 495 & & 3.646 & 377 . & h12 ( + h11 ) & 0.1185 & 15.6 & & & 0.0326 & 3.82 & h11 & ... & ... & & & 0.0408 & 5.1 & h10 & 0.0429 & 5.93 & & & 0.0588 & 8.08 & h9 & 0.0667 & 9.25 & & & 0.0734 & 10.2 & [ ne iii ] 3869 & 0.8060 & 121.7 & 725 , 465 & & 0.8343 & 123.3 & h8 + he i 3889 & 0.1807 & 27.9 & 600 & & 0.1918 & 29.6 & h@xmath81 3970 + [ ne iii ] 3967 & 0.3824 & 62.3 & 628 , 580 & & 0.4041 & 59.2 & he i 4026 & 0.0112 & 1.8 & & & 0.0203 & 2.9 & [ s ii ] 4071 & 0.09695 & 15.9 & 712 , 750 & & 0.1145 & 17.0 & h@xmath82 4102 & 0.2163 & 35.4 & 835 , 465 & & 0.2705 & 41.1 & he i 4144 & 0.00980 & 1.6 & & & 0.0134 & 2.05 & 4229 & 0.07158 & 12.0 & 918 , 859 & & & & h@xmath83 4340 & 0.4127 & 70.8 & 731 , 480 & & 0.4655 & 73.9 & [ o iii ] 4363 & 0.1588 & 27.1 & 567 & & 0.1343 & 22.0 & he i 4471 & 0.0359 & 6.4 & 906 , 433 & & 0.0417 & 6.40 & he ii 4686n & 0.1518 & 27.6 & 400 & & 0.1169 & 17.4 & he ii 4686b & 0.1861 & 34.0 & 5020 & & & & [ ar iv ] 4712 & 0.0150 & 2.7 & & & 0.0245 & 3.73 & [ ar iv ] 4740 & 0.0193 & 3.5 & & & 0.025 & 3.90 & h@xmath84 4861 & 1.000 & 184 . & 660 , 465 & & 1.000 & 140.0 & [ o iii ] 4959 & 3.794 & 1050 . & 687 , 495 & & 2.574 & 369.5 & [ o iii ] 5007 & 11.18 & 3541 . & 700 , 465 & & 7.701 & 1084 . & [ fe vi ] 5146 & 0.0318 & 6.36 & 567 & & & & [ fe vii ] 5159 & 0.0376 & 8.0 & 560 & & & & [ fe vi ] 5176 & 0.0289 & 5.82 & & & & & 5200 & 0.05011 & 11.0 & & & 0.1436 & 24.9 & [ fe iii + fe vii + fe ii]@xmath85 & 0.0378 & 8.3 & & & & & 5309 & 0.0361 & 8.0 & & & & & @xmath86 & 0.0225 & 5.0 & & & & & @xmath87 & 0.0436 & 9.96 & & & & & 5485 & 0.0142 & 3.26 & & & & & 5518 & 0.0117 & 2.46 & & & & & 5538 & 0.0149 & 3.14 & & & & & + [ fe vi]@xmath88 & 0.0209 & 4.72 & & & & & 5677 & 0.0153 & 3.44 & & & & & 5720 & 0.1142 & 26.5 & & & & & 5755 & 0.0137 & 3.73 & & & 0.0359 & 5.91 & he i 5876 & 0.1174 & 29.5 & 680 , 510 & & 0.1028 & 17.7 & [ fe vii ] 6087 & 0.1805 & 46.3 & 774 , 674 & & & & lccccc 3346 & 1.77 @xmath27 0.38 & 104.0 @xmath27 6.2 & 97.1 & 126.2 & @xmath89 3426 & 1.06 @xmath27 0.17 & 102.4 @xmath27 4.6 & 97.1 & 126.2 & @xmath89 3727 & 0.23 @xmath27 0.12 & 124.8 @xmath27 15.2 & 13.6 & 35.1 & @xmath90 3869 & 0.70 @xmath27 0.07 & 93.8 @xmath27 3.0 & 41.0 & 63.5 & @xmath91 4229 & 0.63 @xmath27 0.43 & 56.9 @xmath27 17.7 & 54.8 & 75.5 & 4959 & 0.70 @xmath27 0.03 & 87.7 @xmath27 1.1 & 35.1 & 54.9 & @xmath92 5007 & 0.63 @xmath27 0.03 & 85.8 @xmath27 1.3 & 35.1 & 54.9 & @xmath92 5200 & 0.90 @xmath27 0.77 & 118.0 @xmath27 24.6 & 0.0 & 14.5 & @xmath93 6087 & 1.42 @xmath27 0.71 & 68.4 @xmath27 14.4 & 100.0 & 128.3 & @xmath94

we present a new parallel supercomputer implementation of the monte - carlo method for simulating the dynamical evolution of globular star clusters . our method is based on a modified version of hnon s monte - carlo algorithm for solving the fokker - planck equation . our code allows us to follow the evolution of a cluster containing up to stars to core collapse in hours of computing time . in this paper we present the results of test calculations for clusters with equal - mass stars , starting from both plummer and king model initial conditions . we consider isolated as well as tidally truncated clusters . our results are compared to those obtained from approximate , self - similar analytic solutions , from direct numerical integrations of the fokker - planck equation , and from direct-body integrations performed on a grape-4 special - purpose computer with . in all cases we find excellent agreement with other methods , establishing our new code as a robust tool for the numerical study of globular cluster dynamics using a realistic number of stars . = cmb10 scaled 2

astro-ph9909115

the dynamical evolution of dense star clusters is a problem of fundamental importance in theoretical astrophysics , but many aspects of the problem have remained unresolved in spite of years of numerical work and improved observational data . on the theoretical side , some key unresolved issues include the role played by primordial binaries and their dynamical interactions in the overall cluster dynamics and in the production of exotic sources ( hut 1992 ) , and the importance of tidal shocking for the long - term evolution and survival of globular clusters in the galaxy ( gnedin , lee & ostriker 1999 ) . on the observational side , we now have many large data sets providing a wealth of information on blue stragglers , x - ray sources and millisecond pulsars , all found in large numbers in dense clusters ( e.g. , bailyn 1995 ; camilo 2000 ; piotto 1999 ) . although it is clear that these objects are produced at high rates through dynamical interactions in the dense cluster cores , the details of the formation mechanisms , and in particular the interplay between binary stellar evolution and dynamical interactions , are far from understood . following the pioneering work of hnon ( 1971a , b ) , many numerical simulations of globular cluster evolution were undertaken in the early 1970 s , by two groups , at princeton and cornell , using different monte - carlo methods , now known as the `` princeton method '' and the `` cornell method '' ( see spitzer 1987 for an overview of the methods ) . in the princeton method , the orbit of each star is integrated numerically , while the diffusion coefficients for the change in velocity @xmath4 and @xmath5 ( which are calculated analytically ) are selected to represent the average perturbation over an entire orbit . energy conservation is enforced by requiring that the total energy be conserved in each radial region of the cluster . the princeton method assumes an isotropic , maxwellian velocity distribution of stars to compute the diffusion coefficients , and hence does not take in to account the anisotropy in the orbits of the field stars . one advantage of this method is that , since it follows the evolution of the cluster on a dynamical timescale , it is possible to follow the initial `` violent relaxation '' phase more easily . unfortunately , for the same reason , it also requires considerably more computing time compared to other versions of the monte - carlo method . in the cornell method , also known as the `` orbit - averaged monte - carlo method '' , the changes in energy @xmath6 and angular momentum @xmath7 per unit time ( averaged over an orbit ) are computed analytically for each star . hence , the time consuming dynamical integration of the orbits is not required . in addition , since the diffusion coefficients are computed for both @xmath8 _ and _ @xmath9 , the cornell method does take in to account the anisotropy in the orbits of the stars . the `` hnon method '' is a variation of the cornell method , in which the velocity perturbations are computed by considering an encounter between pairs of neighboring stars . this also allows the local 2-d phase space distribution @xmath10 to be sampled correctly . our code is based on a modified version of hnon s method . we have modified hnon s algorithm for determining the timestep and computing the representative encounter between neighboring stars . our method allows the timestep to be made much smaller in order to resolve the dynamics in the core more accurately . we describe the basic method and our modifications in more detail below in 2 . the monte - carlo methods were first used to study the development of the gravothermal instability ( spitzer & hart 1971a , b ; hnon 1971a , b ) and to explore the effects of a massive black hole at the center of a globular cluster ( lightman & shapiro 1977 ) . in those early studies , the available computational resources limited the number of particles used in the monte - carlo simulations to @xmath11 . since this is much smaller than the real number of stars in a globular cluster ( @xmath12 ) , each particle in the simulation represents effectively a whole spherical shell containing many stars , and the method provides no information about individual objects and their dynamical interactions . more recent implementations have used up to @xmath13 particles and have established the method as a promising alternative to direct @xmath2-body integrations ( stodlkiewicz 1986 ; giersz 1998 ) . monte - carlo simulations have also been used to study specific interaction processes in globular clusters , such as tidal capture ( di stefano & rappaport 1994 ) , interactions involving primordial binaries ( hut , mcmillan , & romani 1992 ) and stellar evolution ( portegies zwart 1997 ) . however , in all these studies the background cluster was assumed to have a _ fixed structure _ , which is clearly not realistic . the main goal of our study is to perform monte - carlo simulations of cluster dynamics treating both the cluster itself and all relevant interactions self - consistently , including all dynamical interactions involving primordial binaries . this idea is particularly timely because the latest generation of parallel supercomputers now makes it possible to do such simulations for a number of objects equal to the actual number of stars in a globular cluster . using the correct number of stars in a cluster simulation ensures that the relative rates of different dynamical processes ( which all scale differently with the number of stars ) are correct . this is crucial if many different dynamical processes are to be incorporated , as we plan to do in this study . in addition to monte - carlo and @xmath2-body simulations , a new method was developed , mainly by cohn and collaborators , based on the direct numerical integration of the orbit - averaged fokker - planck equation ( cohn 1979 , 1980 ; statler , ostriker & cohn 1987 ; murphy & cohn 1988 ) . unlike the monte - carlo methods , the direct fokker - planck method constructs the ( smooth ) distribution function of the system on a grid in phase space , effectively providing the @xmath14 limit of the dynamical behavior . the original formulation of the method used a 2-d phase space distribution function @xmath10 ( cohn 1979 ) . however , the method was later reduced to a 1-d form using an isotropized distribution function @xmath15 ( cohn 1980 ) . the reduction of the method to one dimension speeded up the calculations significantly . in addition , the use of the chang & cooper ( 1970 ) differencing scheme provided much better energy conservation compared to the original 2-d method . the 1-d method provided very good results for isolated clusters , in which the effects of velocity anisotropy are small . the theoretically predicted emergence of a power - law density profile in the late stages of evolution for isolated single - component systems has been clearly verified using this method ( cohn 1980 ) . calculations that include the effects of binary interactions , including primordial binaries , have also allowed the evolution to be followed beyond core collapse ( gao 1991 ) . however , results obtained using the 1-d method showed substantial disagreement with @xmath2-body results for tidally truncated clusters , in which the evaporation rate is dramatically affected by the velocity anisotropy . ignoring the velocity anisotropy led to a significant overestimate of the evaporation rate from the cluster , resulting in shorter core - collapse times for tidally truncated clusters ( portegies zwart 1998 ) . a recent implementation of the fokker - planck method by drukier ( 1999 ) has extended the algorithm to allow a 2-d distribution function , while also improving the energy conservation . a similar 2-d method has also been developed by takahashi ( 1995 , 1996 , 1997 ) . the new implementations produce much better agreement with @xmath2-body results ( takahashi & portegies zwart 1998 ) , and can also model the effects of mass loss due to stellar evolution ( takahashi & portegies zwart 1999 ) , as well as binary interactions ( drukier 1999 ) . for many years direct @xmath2-body simulations were limited to systems with @xmath16 stars . new , special - purpose computing hardware such as the grape ( makino 1997 ) now make it possible to perform direct @xmath2-body simulations with up to @xmath17 single stars ( hut & makino 1999 ) , but the inclusion of a significant fraction of primordial binaries in these simulations remains prohibitively expensive . the large dynamic range of the orbital timescales of the stars in the cluster presents a serious difficulty for @xmath2-body simulations . the orbital timescales can be as small as the periods of the tightest binaries . the direct integration of stellar orbits is especially plagued by this effect . these difficulties are overcome using techniques such as individual integration timesteps , and various schemes for regularizing binaries ( see , e.g. , aarseth 1998 for a review ) . these short - cuts introduce specific selection effects , and complicate code development considerably . instead , in the monte - carlo methods , individual stellar orbits are represented by their constants of the motion ( energy @xmath6 and angular momentum @xmath7 for a spherical system ) and perturbations to these orbits are computed periodically on a timestep that is a fraction of the relaxation time . thus the numerical integration proceeds on the natural timescale for the overall dynamical evolution of the cluster . note also that , because of exponentially growing errors in the direct integration of orbits , @xmath2-body simulations , just like monte - carlo simulations , can only provide a statistically correct representation of cluster dynamics ( goodman 1993 ; hernquist , hut , & makino 1993 ) . a great advantage of the monte - carlo method is that it makes it particularly easy to add more complexity and realism to the simulations one layer at a time . the most important processes that we will focus on initially will be stellar evolution and mass loss through a tidal boundary . interactions of single stars with primordial binaries , binary - binary interactions , stellar evolution in binaries , and a detailed treatment of the influence of the galaxy , including tidal shocking of the cluster when it passes through the galactic disk , will be incorporated subsequently . recent improvements in algorithms and available computational resources have allowed meaningful comparisons between the results obtained using different numerical methods ( see for example the `` collaborative experiment '' by heggie 1999 ) . however , there still remain substantial unresolved differences between the results obtained using various methods . for example , the lifetimes of clusters computed recently using different methods have been found to vary significantly . lifetimes of some clusters computed using direct fokker - planck simulations by chernoff & weinberg ( 1990 ) are up to an order of magnitude shorter than those computed using @xmath2-body simulations and a more recent version of the fokker - planck method ( takahashi & portegies zwart 1998 ) . it has been found that , in many cases , the differences between the two methods can be attributed to the lack of an appropriate discrete representation of the cluster in the fokker - planck simulations . this can lead to an over - estimate of the mass - loss rate from the cluster , causing it to disrupt sooner . recently , new calibrations of the mass loss in the fokker - planck method ( takahashi & portegies zwart 1999 ) that account for the slower mass loss in discrete systems , has led to better agreement between the methods . the limitation of @xmath2-body simulations to small @xmath2 ( especially for clusters containing a large fraction of primordial binaries ) makes it particularly difficult to compare the results with fokker - planck calculations , which are effectively done for very large @xmath2 ( portegies zwart 1998 , heggie 1999 ) . this gap can be filled very naturally with monte - carlo simulations , which can be used to cover the entire range of @xmath2 s not accessible by other methods . the realization over the last 10 years that primordial binaries are present in globular clusters in dynamically significant numbers has completely changed our theoretical perspective on these systems ( see . e.g. , the review by hut 1992 ) . most importantly , dynamical interactions between hard primordial binaries and other single stars or binaries are now thought to be the primary mechanism for supporting a globular cluster against core collapse ( mcmillan , hut , & makino 1990 , 1991 ; gao 1991 ) . in addition , exchange interactions between primordial binaries and compact objects can explain very naturally the formation of large numbers of x - ray binaries and recycled pulsars in globular cluster cores ( sigurdsson & phinney 1995 ; davies & hansen 1998 ; portegies zwart 1997 ) . previously , it was thought that primordial binaries were essentially nonexistent in globular clusters , and so other mechanisms such as tidal capture and three - body encounters had to be invoked in order to form binaries dynamically during core collapse . however , these other mechanisms have some serious problems , and are much more likely to result in mergers than in the formation of long - lived binaries ( chernoff 1996 ; kochanek 1992 ; kumar & goodman 1996 ) . hubble space telescope ( hst ) observations have provided direct constraints on primordial binary fractions in clusters . the binary fraction is a key input parameter for any realistic study of cluster dynamics . for example , the recent observation of a broadened main sequence in ngc 6752 , based on hst pc images of its core , suggest that the binary fraction is probably in the range 15%38% in the inner core ( rubenstein & bailyn 1997 ) . despite the fact that binaries play a crucial role in the late phases of evolution of a cluster , the overall evolution of a binary population within a cluster , and its direct implications for the formation rate of observable binaries and blue stragglers remains poorly understood . in addition , the relative importance of binaries in a cluster , like many other physical processes , may depend on the actual size ( @xmath2 ) of the cluster . this makes it difficult to extend results obtained from smaller @xmath2-body simulations to realistic globular cluster models . when the initial primordial binary fraction is below a certain critical value , a globular cluster core can run out of binaries before the end of its lifetime , i.e. , before being evaporated in the tidal field of the galaxy ( mcmillan & hut 1994 ) . without the support of binaries , the cluster will undergo a much deeper core collapse and so - called gravothermal oscillations ( sugimoto & bettwieser 1983 ; breeden 1994 ; makino 1996 ) . at maximum contraction , the core density may increase by many orders of magnitude , leading to greatly enhanced interaction rates . our new monte - carlo code will allow us to follow the evolution of a cluster through this phase , including in detail the dynamical interactions between the @xmath18 objects in the core . of particular interest is the possibility that successive collisions and mergers of ms stars might lead to a _ runaway process_. the recent hst observations of stellar cusps in the cores of m15 ( guhathakurta 1996 , sosin & king 1997 ) and ngc 6624 ( sosin & king 1995 ) have generated renewed interest in the possibility of massive black holes in globular clusters . the most significant unresolved theoretical issue concerns the manner in which such a black hole could form in a dense cluster . one of the likely routes , which we plan to examine with our simulations , is via the collisions and mergers of main - sequence stars , leading to the runaway build - up of a massive object and its eventual gravitational collapse ( portegies zwart 1999 ) . a very significant effect of the galactic environment on a cluster is the gravitational shock heating of the cluster due to passages close to the bulge and through the disk . when a cluster passes through the galactic disk , it experiences a time - varying gravitational force that pulls the cluster toward the equatorial plane . the net effect of the shock is to induce an increase in the average energy of the stars , causing the binding energy of the cluster to decrease , and the rate of escape of stars through evaporation to increase ( chernoff , kochanek , & shapiro 1986 ) . in addition , in some cases , `` shock - induced relaxation '' can be almost as important as two - body relaxation in the overall evolution of the cluster ( gnedin , lee & ostriker 1999 ; gnedin & ostriker 1997 ) . both the energy shift and the relaxation induced by tidal shocking can be incorporated in our monte - carlo method by assuming an orbit for the cluster around the galactic center and introducing an appropriate perturbation to the energy of the stars each time the cluster passes through the disk . this can be done without adding much computational overhead to the problem , since tidal shocking only occurs twice during the orbital period of the cluster . the ability of the monte - carlo method to model such effects simultaneously with a realistic treatment of the internal dynamical evolution of the cluster makes it a very useful tool in verifying and extending previous results obtained using other methods . the star - by - star representation of the system in monte - carlo simulations makes it easy to study of the evolution of a particular population of stars within a cluster . for example , the evolution of a population of neutron stars could be followed closely , to help predict their properties and expected distributions within clusters . of particular interest are m15 and 47 tuc , which have both been the targets of several highly successful searches for pulsars ( anderson 1992 ; robinson 1995 ; camilo 2000 ) . the observed properties of pulsars in these clusters are found to be very different . the pulsars in 47 tuc are all millisecond pulsars , and most are in short - period binaries , while those in m15 are mostly single recycled pulsars with longer pulse periods . this suggests that these two clusters may provide very different dynamical environments for the formation of recycled pulsars . our basic algorithm for doing stellar dynamics is based on the `` orbit - averaged monte - carlo method '' developed by hnon ( 1971a , b ) . the method was later used and improved by stodlkiewicz ( 1982 , 1985 , 1986 ) . it has also recently been used by spurzem & giersz ( 1996 ) to follow the evolution of hard three - body binaries in a cluster with equal point - mass stars . new results using stodlkiewicz s version of the method were also presented recently by giersz ( 1998 ) . in earlier implementations of the monte - carlo method with @xmath19 , each particle in the simulation was a `` superstar , '' representing many individual stars with similar orbital properties . in our implementation , with @xmath20 , we treat each particle in the simulation as a single star . we have also modified hnon s original algorithm to allow the timestep to be made much smaller in order to resolve the dynamics in the core more accurately . in the simplest case of a spherical system containing @xmath2 point masses the algorithm can be summarized as follows . we begin by assigning to each star a mass , radius and velocity by sampling from a spherical and isotropic distribution function ( for example , the plummer model ) . once the positions and masses of all stars are known , the gravitational potential of the cluster is computed assuming spherical symmetry . the energy and angular momentum of each star are then calculated . energy and angular momentum are perturbed at each timestep to simulate the effects of two - body and three - body relaxation . the perturbations depend on each star s position and velocity , and on the density of stars in its neighborhood . the timestep should be a fraction of the relaxation time for the cluster ( which is larger than the dynamical time by a factor @xmath21 ) . the perturbation of the energy and angular momentum of a star at each timestep therefore represents the cumulative effect of many small ( and distant ) encounters with other stars . under the assumption of spherical symmetry , the cross - sections for these perturbations can be computed analytically . the local number density is computed using a sampling procedure . once a new energy and angular momentum is assigned to each star , a new realization of the system is generated by assigning to each star a new position and velocity in an orbit that is consistent with its new energy and angular momentum . in selecting a new position for each star along its orbit , each position is weighted by the amount of time the star spends around that position . using the new positions , the gravitational potential is then recomputed for the entire cluster . this procedure is then repeated over many timesteps . after every timestep , all stars with positive total energy ( cf . 2.7 ) are removed from the computation since they are no longer bound to the cluster and are hence considered lost from the cluster instantly on the relaxation timescale . the method allows stars to have arbitrary masses and makes it very easy to allow for a stellar mass spectrum in the calculations . we now describe our implementation of the monte - carlo method in detail . for completeness , we also include some of the basic equations of the method . for derivations of these equations , and a more detailed discussion of the basic method , see hnon ( 1971b ) , stodlkiewicz ( 1982 ) , and spitzer ( 1987 ) . the initial model is assumed to be in dynamical equilibrium , so that the potential does not change on the crossing timescale . this is important since the monte - carlo method uses a timestep which is of the order of the relaxation time , and hence can not handle the initial phase of `` violent relaxation '' during which the potential changes on the dynamical timescale . under the assumption of spherical symmetry , the distribution function for such an equilibrium system can be written in the form @xmath22 , where @xmath6 and @xmath7 are the energy per unit mass , and angular momentum per unit mass , @xmath23 here @xmath24 is the distance from the cluster center , @xmath25 is the radial velocity , @xmath26 is the transverse velocity , and @xmath27 is the gravitational potential . in principle , the initial distribution function @xmath28 can be arbitrary . however , in practice , computing a self - consistent potential for an arbitrary distribution function can be quite difficult . since the method requires the initial potential @xmath27 to be known , a simple initial model is usually selected so as to allow the potential to be computed quasi - analytically . common examples are the sequence of king models and the plummer model . once the number of stars @xmath2 is selected , the initial condition is constructed by assigning to each star values for @xmath24 , @xmath25 , @xmath26 , and @xmath29 , consistent with the selected model . once the positions and masses of all the stars are known , the gravitational potential @xmath30 is computed as a function of distance from the center . the energy per unit mass @xmath6 , and angular momentum per unit mass @xmath7 of each star are then computed using equations ( 1 ) and ( 2 ) . we compute the mean potential of the cluster by summing the potential due to each star , under the assumption of spherical symmetry . we use only the radial position @xmath24 of each star ( since we assume spherical symmetry , we can neglect the angular positions of the stars , to a very good approximation ) . we begin by sorting all the stars by increasing radius . then the potential at a point @xmath24 , which lies between two stars at positions @xmath31 and @xmath32 , is given by @xmath33 for any two neighboring stars at distances @xmath31 and @xmath32 , the mass contained within the radius @xmath24 remains constant for @xmath34 . hence , we can compute the potential at r , if the potentials @xmath35 and @xmath36 are known , as @xmath37 at each timestep , we store pre - computed values of @xmath38 , for each star k in the cluster . the potential at an arbitrary point @xmath24 can then be quickly computed simply by finding the index k such that @xmath39 and then using equation ( 4 ) . we now describe the process of evolving the system through one complete timestep . we simulate the effect of interactions during each timestep @xmath40 by perturbing the energy and angular momentum of each star in the cluster . the perturbations @xmath8 and @xmath9 for a star are determined by computing a single _ effective _ encounter between the star and its nearest neighbor ( in terms of distance from the center , since we assume spherical symmetry ) . during such an encounter , the two stars exchange kinetic energy , but the total energy is conserved . in the center of mass frame of the two interacting stars , the magnitude of the velocity does not change ; instead the velocity is deflected through an angle @xmath41 . in the original method described by hnon ( 1971b ) , the timestep used was a small fraction of the relaxation time for _ the entire cluster_. although the timestep computed in this way is suitable for the outer regions of the cluster , it is too large to provide an accurate representation of the relaxation in the core , especially in the later stages of cluster evolution where the relaxation time in the core can be many orders of magnitude smaller than in the outer regions . this caused the inner regions of the cluster to be under - relaxed . the limited computational resources available at that time did not permit the timestep to be made much smaller , without slowing down the computation to a crawl . the greatly increased computational power available today allows us to use a timestep that is small enough to resolve the relaxation process in the core , even for systems with @xmath42 . to provide an accurate description of the overall relaxation of the cluster , each effective encounter should give the correct mean value of the change in energy at each position . we achieve this by selecting the effective deflection angle @xmath43 for the encounter ( in the center of mass frame of the two interacting stars ) as follows . if the masses of the two stars are @xmath44 and @xmath45 , and their velocities @xmath46 and @xmath47 , respectively , then the kinetic energy changes can be written as @xmath48 where @xmath49 and @xmath50 are the changes in the velocities during the encounter . since the total kinetic energy in each encounter is conserved , the mean value of the first terms on the rhs of equations ( 5 ) and ( 6 ) must equal the mean value of the second terms ( with the opposite sign ) . this indicates that in order to get a good representation of the energy exchange between stars in the relaxation process , we must consider the mean value of @xmath51 during each timestep . the change in velocity @xmath49 during an encounter with a deflection angle @xmath41 , can be calculated from elementary mechanics as ( see , e.g. , spitzer 1987 , eq . [ 2 - 6 ] ) , @xmath52 where @xmath53 is the relative speed of the two stars before the encounter . the mean overall _ rate _ of change in the velocity @xmath54 due to many distant ( weak ) encounters of the star with other cluster stars can then be calculated by averaging over the impact parameter ( cf . spitzer 1987 , eq . [ 2 - 8 ] ) . using this , the mean change in the velocity in the time @xmath40 is given by @xmath55 where @xmath56 is the coulomb logarithm ( @xmath57 is a constant @xmath58 ; see 3.1 ) , and @xmath59 is the local number density of stars . we obtain the correct mean value of @xmath51 by equating the rhs of equations ( 7 ) and ( 8) , giving @xmath60 equation ( 9 ) relates the timestep @xmath40 to the deflection angle @xmath41 for the encounter . thus , in order to get the correct mean value of @xmath51 for the star during the time @xmath40 , we can define the _ effective _ deflection angle @xmath43 for the representative encounter , as @xmath61 in addition to using the correct mean value of @xmath51 , we can also require that its variance be correct . to compute the variance , we must calculate the mean value of @xmath62 . using equation ( 7 ) , we have @xmath63 we then use spitzer s equation ( 2 - 5 ) , and again integrate over the impact parameter to get the mean value of @xmath62 in the time @xmath40 , @xmath64 comparing equations ( 11 ) and ( 12 ) , we see that , in order to have the correct variance of @xmath51 , we should have @xmath65 consistency between equations ( 10 ) and ( 13 ) gives the relation between the number of stars in the system , and the effective deflection angle that must be used , @xmath66 this relation indicates that for large @xmath2 , the effective deflection angle must be small , while as @xmath2 decreases , close encounters become more important . if the timestep is too large , then @xmath67 is also too large , and the system is under - relaxed . hence the timestep used should be sufficiently small so as to get a good representation of the relaxation process in the cluster . in addition , the local relaxation time varies greatly with distance from the cluster center . in practice we use the shortest relaxation time in the core to compute the timestep . we first evaluate the local density @xmath68 in the core and the approximate core radius @xmath69 . we then compute the timestep @xmath40 using equation ( 10 ) and requiring that the average value of @xmath70 for the stars within the core radius @xmath71 be sufficiently small . the value of @xmath70 given by equation ( 14 ) varies only slightly between @xmath72 and @xmath73 for @xmath2 between @xmath74 and @xmath0 ( assuming @xmath75 ) . hence for all our simulations , we require that @xmath76 . equation ( 10 ) is then used to compute the effective deflection angle for all stars in the cluster . the local number density @xmath59 is computed by averaging over the nearest @xmath77 stars . we find that using a value of @xmath77 between 20 and 50 gives the best results for @xmath78 . we find that the difference in the core - collapse times obtained for various test models using values of @xmath77 between 20 and 50 is less than 1% . of course , the value of @xmath77 should not be too large so as to maintain a truly local estimate of the number density . we use a value of @xmath79 in all our calculations , which gives consistently good agreement with published results . to compute the velocity perturbation during each timestep , a single representative encounter is computed for each star , with its nearest neighbor in radius . selecting the nearest neighbor ensures that the correct local velocity distribution is sampled , and also accounts for any anisotropy in the orbits . due to spherical symmetry , selecting the nearest neighbor in radius is equivalent to selecting the nearest neighbor in 3-d , since only the velocity ( and not the position ) of the nearest neighbor is used in the encounter . following hnon s notation , we let ( @xmath24 , @xmath25 , @xmath26 ) and ( @xmath80 , @xmath81 , @xmath82 ) represent the phase space coordinates of the two interacting stars , with masses @xmath29 and @xmath83 , respectively . in addition to these parameters , the angle @xmath84 of the plane of relative motion defined by ( @xmath85 , @xmath86 ) with some reference plane is selected randomly between 0 and @xmath87 , since the distribution of field stars is assumed to be spherically symmetric . we take our frame of reference such that the @xmath88 axis is parallel to @xmath89 , and the @xmath90 plane contains @xmath91 . then the velocities of the two stars are given by @xmath92 where @xmath93 is also randomly selected between 0 and @xmath87 , since the transverse velocities are isotropic because of spherical symmetry . the relative velocity @xmath94 is then @xmath95 we now define two vectors @xmath96 and @xmath97 with the same magnitude as @xmath98 , such that @xmath96 , @xmath97 , and @xmath98 are mutually orthogonal . the vectors @xmath96 and @xmath97 are given by @xmath99 where @xmath100 . the angle @xmath84 is measured from the plane containing the vectors @xmath101 and @xmath102 . the relative velocity of the two stars after the encounter is given by @xmath103 where @xmath41 is the deflection angle computed in 2.4 . the new velocities of the two stars after the interaction are then given by @xmath104 the new radial and transverse velocities for the first star are given by @xmath105 , and @xmath106 , from which we compute the new orbital energy @xmath6 and angular momentum @xmath7 as @xmath107 , and @xmath108 . similar quantities @xmath109 and @xmath110 are also computed for the second star . once the orbits of all the stars are perturbed , i.e. , new values of @xmath6 and @xmath7 are computed for each star , a new realization of the system is generated , by selecting a new position for each star in its new orbit , in such a way that each position in the orbit is weighted by the amount of time that the star spends at that position . to do this , we begin by computing the pericenter and apocenter distances , @xmath111 and @xmath112 , for each star . the orbit of a star in the cluster potential is a rosette , with @xmath24 oscillating between @xmath111 and @xmath112 , which are roots of the equation @xmath113 see binney & tremaine ( 1987 ; 3.1 ) for a general discussion , and see hnon ( 1971b ; eqs . [ 41]-[45 ] ) for a convenient method of solution . the new position @xmath24 should now be selected between @xmath111 and @xmath112 , in such a way that the probability of finding r in an interval @xmath114 is equal to the fraction of time spent by the star in the interval during one orbit , i.e. , @xmath115 where @xmath116 is the orbital period , and @xmath117 is given by @xmath118 . unfortunately , at the pericenter and apocenter points ( @xmath111 and @xmath112 ) , the radial velocity @xmath25 is zero , and the probability distribution becomes infinite . to overcome this problem , we make a change of coordinates by defining a suitable function @xmath119 and selecting a value of s from the distribution @xmath120 we must select the function @xmath121 such that @xmath122 remains finite in the entire interval . a convenient function @xmath121 that satisfies these requirements is given by @xmath123 where @xmath124 lies in the interval -1 to 1 . we then generate a value for @xmath124 , which is consistent with the distribution @xmath122 , using the von neumann rejection technique . equation ( 26 ) then gives a corresponding value for @xmath24 which is consistent with the distribution function @xmath125 . the magnitude of the new radial velocity @xmath25 is computed using equation ( 24 ) , and its sign is selected randomly . the transverse velocity is given by @xmath126 . once a new position is selected for each star using the above procedure , the gravitational potential @xmath27 is recomputed as described in 2.3 . this completes the timestep , and allows the next timestep to be started . note that the gravitational potential used to compute new positions and velocities of the stars is from the previous timestep . the new potential can only be computed _ after _ the new positions are assigned , and it is then used to recompute the positions in the next timestep . thus the computed potential always lags slightly behind the actual potential of the system . the exact potential is known only at the initial condition . this only introduces a small systematic error in the computation , since the potential changes significantly only on the relaxation timescale . a more important source of error , especially in computing the new energies of the stars after the potential is recomputed , is the random fluctuation of the potential in the core , which contains relatively few stars , but has a high number density . since the derivative of the potential is also steepest in the core , a small error in computing a star s position in the core can lead to a large error in computing its energy . as the simulation progresses , this causes a slow but consistent leak in the total system energy . the magnitude of this error ( i.e. , the amount of energy lost per timestep ) depends partly on the number of stars @xmath2 in the system . for large @xmath2 , the grid on which the potential is pre - computed ( see 2.3 ) is finer , and the number of stars in the core is larger , which reduces the noise in the potential . the overall error in energy during the course of an entire simulation is typically of order a few percent for @xmath127 stars . in any realistic simulation , the actual energy gain or loss due to real physical processes such as stellar evolution , escape of stars through a tidal boundary , and interactions involving binaries , is at least an order of magnitude greater than this error . hence we choose not to renormalize the energy of the system , or employ any other method to artificially conserve the energy of the system , which could affect other aspects of the evolution . another possible source of error in monte - carlo simulations , which was noted by hnon ( 1971b ) is the `` spurious relaxation '' effect . this is the tendency for the system to relax because of the potential fluctuations from one timestep to the next , even in the absence of orbital perturbations due to two - body relaxation . however , this effect is significant only for simulations done with very low @xmath128 . in test calculations performed with @xmath129 and two - body relaxation explicitly turned off ( by setting the scattering angle @xmath130 in eq . [ 10 ] ) , we find no evidence of spurious relaxation . indeed hnon ( 1971b ) himself showed that spurious relaxation was not significant in his models for @xmath131 . for an isolated system , the gradual evaporation of stars from the cluster is computed in the following way . during each timestep , after the perturbations @xmath8 and @xmath9 are computed , all stars with a positive total energy ( given by eq . [ 1 ] ) are assumed to leave the cluster on the crossing timescale . they are therefore considered lost immediately on the relaxation timescale , and removed from the simulation . the mass of the cluster ( and its total energy ) decreases gradually as a result of this evaporation process . as a simple first step to take in to account the tidal field of the galaxy , we include an effective tidal boundary around the cluster , at a distance @xmath132 , where @xmath133 is the distance of the cluster from the galactic center and @xmath134 is the mass of the galaxy ( approximated as a point mass ) . the tidal radius is roughly the size of the roche lobe of the cluster in the field of the galaxy . once the initial tidal radius @xmath135 is specified , the tidal radius at a subsequent time @xmath136 during the simulation can be computed by @xmath137 . after each timestep , we remove all stars with an apocenter distance @xmath112 greater than the tidal radius , since they are lost from the cluster on the crossing timescale . as the cluster loses stars due to evaporation and the presence of the tidal boundary , its mass decreases , which causes the tidal boundary to shrink , in turn causing even more stars to be lost . the total mass loss due to a tidal boundary can be very significant , causing up to 90% of the mass to be lost ( depending on the initial model ) over the course of the simulation ( see 3.2 ) . following the convention of most previous studies , we define dynamical units so that @xmath138 = [ m_0 ] = [ -4 e_0 ] = 1 $ ] , where @xmath139 and @xmath140 are the initial total mass and total energy of the system ( hnon 1971 ) . then the units of length @xmath141 , and time @xmath142 are given by @xmath143 we see that @xmath141 is basically the virial radius of the cluster , and @xmath142 is of the order of the initial dynamical ( crossing ) time . to compute the evolution of the cluster on a relaxation timescale , we rescale the unit of time to @xmath144 , which is of the order of the initial relaxation time . using this unit of time allows us to eliminate the @xmath145 dependence of the evolution equations . the only equation that explicitly contains the evolution time is equation ( 10 ) , which relates the timestep and the effective deflection angle . in our units , equation ( 10 ) can be written as , @xmath146 = 2\pi \frac{([m_1 ] + [ m_2])^2}{[w]^3 } [ \nu ] [ \delta t ] n,\end{aligned}\ ] ] where [ q ] indicates a quantity q expressed in our simulation units . using a unit of time that is proportional to the initial relaxation time has the advantage that the evolution timescale is roughly independent of the number of stars @xmath2 once an initial model has been selected . this is only true approximately , for isolated systems of equal - mass stars , with no other processes that depend explicitly on the number of stars ( such as stellar evolution or mass segregation ) . for example , the half - mass relaxation time for the plummer model , @xmath147 is always 0.093 in our units , independent of @xmath2 . the dynamical units defined above are identical to the standard @xmath2-body units ( heggie & mathieu 1986 ) . hence to convert the evolution time from @xmath2-body time units to our monte - carlo units , we must simply multiply by a factor @xmath148 . we have implemented our monte - carlo code on the sgi / cray origin2000 parallel supercomputer at the national center for supercomputing applications ( ncsa ) , and at boston university . our parallelized code can be used to get significant speedup of the simulations , using up to 8 processors , especially for large @xmath2 simulations . this ability to perform large @xmath2 simulations will be particularly useful for doing realistic simulations of very large globular clusters such as 47 tuc ( with @xmath149 stars ) . a simulation with @xmath127 stars can be completed in approximately 1520 cpu hours on the origin2000 , which uses mips r10000 processors . for comparison , a simulation of this size would take @xmath150 months to complete using the grape-4 , which is the fastest available hardware for @xmath2-body methods . the most computationally intensive step in the simulation is the calculation of the new positions of stars . the operation involves solving for the roots of an equation ( eq . [ 22 ] ) using the indexed values of the positions of the @xmath2 stars . we find that the most efficient method to solve for the roots in this case is the simple bisection method ( e.g. , press 1992 ) , which requires @xmath151 steps to converge to the root . hence the computation of the positions and velocities also scales as @xmath151 in our method . the next most expensive operation is the evaluation of the potential at a given point @xmath24 . as described in 2.3 , this requires finding @xmath152 such that @xmath39 and then using equation ( 4 ) . this search can again be done easily using the bisection algorithm . however , since the evaluation of the potential is required several times for each star , in each timestep , it is useful to tabulate the values of @xmath152 on fine grid in @xmath24 at the beginning of the timestep . this allows the required values of @xmath152 to be found very quickly , at the minor cost of using more memory to store the table . the rest of the steps in the simulation scale almost linearly with @xmath2 . this makes the overall computation time scale ( theoretically ) as @xmath153 . in figure 1 , we show the scaling of the wall - clock time with the number of processors , and also the scaling of the overall computation time with the number of stars @xmath2 in the simulation . the overall computation time is consistent with the theoretical estimate for @xmath154 . for larger @xmath2 , the computation time is significantly higher , because of the less efficient use of cache memory and other hardware inefficiencies that are introduced while handling large arrays . for @xmath2 in the range @xmath155 , we find that the actual computation time scales as @xmath156 . we find that we can easily reduce the overall computation time by a factor of @xmath157 by using up to 8 processors . the scaling is most efficient for @xmath158 processors for simulations with @xmath159 . the scaling gets progressively worse for more than 8 processors . this is in part caused by the distributed shared - memory architecture of the origin2000 supercomputer , which allows very fast communication between the nearest 2 - 4 processors , but slower communication between the nearest 8 processors . beyond 8 processors , the communication is even slower , since the processors are located on different nodes . the most suitable architecture for implementing the parallel monte - carlo code would be a truly shared memory supercomputer , with roughly uniform memory access times between processors . our code is implemented using the message passing interface ( mpi ) parallelization library , which is actively being developed and improved . the mpi standard is highly portable , and available on practically all parallel computing platforms in use today . the mpi library is optimized for each platform and automatically takes advantage of the memory architecture to the maximum extent possible . hence we expect that future improvements in the communication speed and memory architectures will make our code scale even better . we are also in the process of improving the scaling of the code to a larger number of processors by designing a new algorithm for reducing the amount of communication required between processors . this will be described in detail in a subsequent paper , where we incorporate primordial binary interactions in our code . in this section , we describe our first results using the new monte - carlo code to compute the evolution of the plummer and king models . we explore the evolution of the plummer model in detail , and compare our results with those obtained using fokker - planck and @xmath2-body methods . we also compare core - collapse times and mass - loss rates for the series of king models ( @xmath160 ) , including a tidal radius , with similar results obtained by quinlan ( 1996 ) using a 1-d fokker - planck method . we first consider the evolution of a cluster with the plummer model ( which is a polytropic model , with index @xmath161 ; see , e.g. , binney & tremaine 1987 ) as the initial condition . perhaps the best known result for single component systems , is the expected homologous evolution of the halo , leading to the eventual development of a power - law density profile between the core and the outer halo , during the late phases of evolution . at late times the cluster evolves through a sequence of nearly self - similar configurations , with the core contracting and a power - law halo with density @xmath162 expanding out . the development of this power law has been predicted theoretically ( lynden - bell & eggleton 1980 ; heggie & stevenson 1988 ) , and verified using direct fokker - planck integrations ( cohn 1980 ) . the exponent @xmath41 is theoretically and numerically estimated to be about @xmath163 ( spitzer 1987 ) . however , since the theoretical derivations are based on an analysis of the fokker - planck equation , it is not surprising that the numerical fokker - planck integrations ( which solve the same fokker - planck equation numerically ) reproduce the theoretical exponent exactly . due to limitations in computing accurate density profiles using a small number of stars , this result has not been confirmed independently using an @xmath2-body simulation . here , we explore numerically for the first time the development of this power law using an independent method . some early results were obtained using previous versions of the monte - carlo method , but with a small number of stars @xmath19 ( duncan & shapiro 1982 ) . although the monte - carlo method can be thought of as just another way of solving the fokker - planck equation , there are significant differences between solving the equation in the continuous limit ( @xmath164 ) , as in direct fokker - planck integrations , and by using a discrete system with a finite @xmath2 as in our method . there are also many subtle differences in the assumptions and approximations made in the two methods , and even in different implementations of the same method . in figures 2a c we show the density profile of the cluster at three different times during its evolution , up to core collapse . we start with an @xmath127 isolated plummer model , and follow the evolution up to core - collapse , which occurs at @xmath165 . this simulation , performed with @xmath127 stars , took about 18 cpu hours on the sgi / cray origin2000 . in our calculations , the core - collapse time is taken as the time when the the innermost lagrange radius ( radius containing 0.3% of the total mass of the cluster ) becomes smaller than 0.001 ( in our units described in 2.8 ) , at which point the simulation is terminated . given the very rapid evolution of the core near core collapse , we find that we can determine the core - collapse time to within @xmath166 . the accuracy is limited mainly by noise in the core . the value we obtain for @xmath167 is in very good agreement with other core - collapse times between @xmath168 for the plummer model , reported using other methods . for example quinlan ( 1996 ) obtains a core collapse time of @xmath169 for the plummer model using a 1-d fokker - planck method , and takahashi ( 1993 ) finds a value of @xmath170 , using a variational method to solve the 1-d fokker - planck equation . figure 2a shows the density profile at an intermediate time @xmath171 during the evolution . the dotted line indicates the initial plummer profile . at this point in the evolution , we still see a well defined core , with the core density increased by a factor of @xmath172 compared to the initial core density . we see the power - law density profile developing , with the best - fit index @xmath173 . in figure 2b , we show the density profile just before core collapse , at @xmath174 . we see that the core density has now increased by a factor of @xmath175 over the initial core density . the power law is now clearly visible , with the best - fit index @xmath176 . finally , in figure 2c , we show the density profile at core - collapse , @xmath177 . the dashed line now indicates the _ theoretical _ power law with @xmath178 . we see that the actual density profile seems to approach the theoretical profile asymptotically as the system approaches core collapse . at this point in the evolution , the core density as measured in our simulation is about @xmath179 times greater than the initial density . in a globular cluster with @xmath180 , an average stellar mass @xmath181 , and a mean velocity dispersion @xmath182 , this would correspond to a number density of @xmath183 . note that a real globular cluster is not expected to reach such high core densities , since the formation of binaries and the subsequent heating of the core due to binary interactions become significant at much lower densities . numerical noise due to the extremely small size of the core makes it difficult to determine the core radius and density accurately at this stage . this also causes the numerical accuracy of the monte - carlo method to deteriorate , forcing us to stop the computation . thus , we find that the power - law structure of the density profile as the cluster approaches core collapse is consistent with theoretical predictions , and the power - law index approaches its theoretical value asymptotically during the late stages of core collapse . next , we look at the evolution of the lagrange radii ( radii containing constant fractions of the total mass ) , and we compare our results with those of an equivalent @xmath2-body simulation . in figure 3 , we show the evolution of the lagrange radii for an @xmath3 direct @xmath2-body integration by makino ( 1996 ) and for our monte - carlo integration with @xmath184 stars . time in the direct @xmath2-body integration is scaled to the initial relaxation time ( the standard time unit in our monte carlo method ) using equation ( 27 ) with @xmath185 ( see heggie & mathieu 1986 ; giersz & heggie 1994 ; makino 1996 ) . the agreement between the @xmath2-body and monte carlo results is excellent over the entire range of lagrange radii and time . the small discrepancy in the outer lagrange radii is caused in part by a different treatment of escaping stars in the two models . in the monte carlo model , escaping stars are removed from the simulation and therefore not included in the determination of the lagrange radii , whereas in the @xmath2-body model escaping stars are not removed . the difference is further explained by the effect of strong encounters , which is greater in the @xmath2-body simulation by a factor @xmath186 , or about 20% . in an isolated cluster , the overall evaporation rate is very low ( less than 1% of stars escape up to core collapse ) . in this regime , the escape of stars is dominated by strong interactions in the core . since the orbit - averaged fokker - planck equation is only valid when the fractional energy change per orbit is small , it does not account for strong interactions . hence , our monte - carlo simulations can not accurately predict the rate of evaporation from an isolated cluster ( see , e.g. , binney & tremaine 1987 , 8.4 ) . this problem does not occur in tidally truncated clusters , where the escape rate is much higher , and is dominated by the diffusion of stars across the tidal boundary , and not by strong interactions . in figure 4 we show the evolution of various global quantities for the system during the same simulation as in figure 3 . the virial ratio ( @xmath187 , where @xmath188 and @xmath189 are the total kinetic and potential energies of the cluster ) remains very close to 0.5 ( within 1% ) , indicating that dynamical equilibrium is maintained very well during the entire simulation . the virial ratio provides a very good measure of the quality of our numerical results , since it is not controlled in our calculations ( except for the initial model , which is constructed to be in equilibrium ) . we see that in the absence of a tidal radius , there is very little mass loss ( less than 1% ) , and hence very little energy is carried away by escaping stars . king models ( king 1966 ) have long been used to fit observed profiles of globular clusters . they usually provide a very good fit for most clusters , except for those which have reached core collapse . a king model has a well - defined , nearly constant - density core , and a `` lowered maxwellian '' velocity distribution , which represents the presence of a finite tidal radius . a king model is usually specified in terms of the dimensionless central potential @xmath190 or , equivalently , the central concentration @xmath191 , where @xmath192 is the tidal radius , and @xmath71 is the core radius . we study the evolution of the entire family of king models from @xmath193 to @xmath194 , in two different configurations . we first consider the evolution of an isolated cluster i.e. , even though the initial king model is truncated at its finite tidal radius , we do not enforce that tidal boundary during the evolution , allowing the cluster to expand indefinitely . we compute the core - collapse times for the entire sequence of king models . we then redo the calculations with a tidal boundary in place , to determine the enhanced rate of mass loss from the cluster and the final remaining mass at the time of core collapse . we compare our results for the sequence of king models with equivalent results obtained by quinlan ( 1996 ) using direct fokker - planck integrations in 1-d . in table 1 , we show the core collapse times for the various models , along with the equivalent results from quinlan ( 1996 ) . all our monte - carlo calculations were performed using @xmath127 stars . we see that the agreement in the core collapse times for isolated clusters is excellent ( within a few percent for the low-@xmath190 models , and within 10% up to @xmath195 ) . for @xmath196 , the agreement is still good , considering that the models start off in a highly collapsed state and therefore have very short core - collapse times , which leads to larger fractional errors . in figure 5 , we show the evolution of the lagrange radii for a tidally truncated king model with @xmath197 . the initial tidal radius is @xmath198 times the virial radius . in this case , the mass loss through the tidal boundary is very significant , as is seen from the evolution of the outer lagrange radii . the mass loss causes the tidal radius to constantly move inward , which further accelerates the process . figure 6 shows the evolution of the total mass and energy of the tidally truncated cluster . only 44% of the initial mass is retained in the cluster at core - collapse . also , the binding energy of the cluster is significantly lower at core - collapse , since the escaping stars carry away mass as well as kinetic energy from the cluster . in contrast , the evolution of an isolated @xmath197 king model is very much like that of the isolated plummer model described earlier , with a very low mass loss rate , and a longer core - collapse time of @xmath199 ( in excellent agreement with the value of @xmath200 computed by quinlan 1996 ) . our results for clusters with a tidal boundary show systematic differences from the 1-d fokker - planck results of quinlan ( 1996 ) . we find that the mass loss through the tidal boundary is significantly higher for the low - concentration models ( @xmath201 ) in the fokker - planck models . for the high - concentration ( @xmath202 ) models , the difference between isolated models and tidally truncated models is small , and the agreement between the methods remains very good . hence , for low @xmath190 , our models undergo core collapse at a much later time compared to the fokker - planck models , and retain more mass at core collapse . this discrepancy is caused by the 1-d nature of the fokker - planck models . in 1-d fokker - planck calculations , stars are considered lost from the cluster when their energy is greater than the energy at the tidal radius . this clearly provides an overestimate of the escape rate , since it assumes the most extended radial orbits for stars , and ignores stars on more circular orbits with high angular momentum , which would have much smaller orbits at the same energy . in contrast , in the monte - carlo method , the orbit of each star is computed using its energy _ and _ angular momentum , which allows the apocenter distance to be determined correctly . stars are considered lost only if their apocenter distances from the cluster center are greater than the tidal radius . as stars on radial orbits are removed preferentially , this creates an anisotropy within the cluster , which affects the overall evolution . the artificially high rate of mass loss in 1-d fokker - planck simulations has also been pointed out recently in comparisons with @xmath2-body results ( portegies zwart 1998 ; takahashi & portegies zwart 1999 ) . these authors show that , with appropriate modifications , the results of 2-d fokker - planck calculations can be made to agree much better with those from @xmath2-body simulations . indeed , we find that our result for the @xmath197 model with a tidal boundary ( @xmath203 , and @xmath204 ) agrees much better with that obtained using the improved 2-d fokker - planck method , which gives @xmath205 , and @xmath206 ( takahashi 1999 , private communication ) . for further comparison , and to better understand the cause of the higher mass loss in the 1-d fokker - planck calculation , we have performed a monte - carlo simulation using the same energy - based escape criterion that is used in the 1-d fokker - planck integrations . we find that using the energy - based escape criterion for @xmath197 gives @xmath207 , and @xmath208 , which agrees better with the 1-d fokker - planck result , but a significant discrepancy still remains . this is not surprising , since , even when using a 1-d escape criterion , our underlying method still remains 2-d . again , our result agrees better with the corresponding result obtained by takahashi ( 1999 , private communication ) using the energy - based escape criterion in his 2-d fokker - planck method , @xmath209 , and @xmath210 . it is reassuring to note that the differences between our 2-d results and 1-d fokker - planck results are also mirrored in the 2-d fokker - planck calculations of takahashi . since our monte - carlo method is intrinsically 2-d , it is not possible for us to do a true 1-d ( isotropic ) calculation to compare results directly with 1-d fokker - planck calculations . we have presented results obtained using our new monte - carlo code for the evolution of clusters containing @xmath211 stars , up to core collapse . we have compared our results with those of 1-d fokker - planck calculations ( quinlan 1996 ) for isolated as well as tidally truncated king models with @xmath160 . we find very good agreement for the core - collapse times of isolated king models . for tidally truncated models ( especially for @xmath201 ) , we find that the escape rate of stars in our models is significantly lower than in the 1-d fokker - planck models . this is to be expected , since the 1-d fokker - planck models use an energy - based escape criterion , which does not account for the anisotropy in the orbits of stars , and hence overestimate the escape rate . this effect is most evident in tidally truncated clusters , since stars on radial orbits are preferentially removed , while those on more circular orbits ( with the same energy ) are not . in one case ( @xmath197 ) , we have verified that our results are in good agreement with those from new 2-d fokker - planck calculations ( takahashi 1999 , private communication ) , which properly account for the velocity anisotropy , and use the same apocenter - based escape criterion as in our models . further comparisons of our results with 2-d fokker - planck calculations will be presented in a subsequent paper ( joshi , nave , & rasio 1999 ) . our detailed comparison of the evolution of the plummer model with an equivalent direct @xmath2-body simulation also shows excellent agreement between the two methods up to core collapse . our results clearly show that the monte - carlo method provides a robust , scalable and flexible alternative for studying the evolution of globular clusters . its strengths are complementary to those of other methods , especially @xmath2-body simulations , which are still prohibitively expensive for studying large systems with @xmath42 . the monte - carlo method requires more computational resources compared to fokker - planck methods , but it is several orders of magnitude faster than @xmath2-body simulations . the star - by - star representation of the system in this method makes it particularly well suited for studying the evolution of interesting sub - populations of stars within globular clusters , such as pulsars , blue stragglers , or black holes . our method also presents the interesting possibility of performing hybrid simulations that use the monte - carlo method for the bulk of the evolution of a cluster up to the core collapse phase , and then switch to an @xmath2-body simulation to follow the complex core - collapse phase during which the high reliability of the @xmath2-body method is desirable . the discreteness of the monte - carlo method , and the fact that it follows the same phase space parameters for a cluster as the @xmath2-body method , make it easy to switch from one method to the other during a single simulation . in subsequent papers , we will present results for the dynamical evolution of clusters with different mass spectra , including the effects of mass loss due to stellar evolution . we are also in the process of incorporating primordial binaries in our monte - carlo code , in order to follow the evolution in the post - core collapse phase . dynamical interactions involving binaries will be treated using a combination of direct numerical integrations of the orbits on a case - by - case basis and precomputed cross - sections . the cross - sections will be obtained from separate sets of scattering experiments as well as fitting formulae ( sigurdsson & phinney 1995 ; heggie , hut , & mcmillan 1996 , and references therein ) . our code will also incorporate a simple treatment of stellar evolution in binaries , using an extensive set of approximate recipes and fitting formulae developed recently for starlab ( portegies zwart 1995 ) . simulations of clusters containing realistic numbers of stars _ and _ binaries will allow us for the first time ever to compute detailed predictions for the properties and distributions of all interaction products , including blue stragglers ( from mergers of main - sequence stars ) , x - ray binaries and recycled pulsars ( from interactions involving neutron stars ) and cataclysmic variables ( from interactions involving white dwarfs ) . we thank piet hut and stephen mcmillan for many helpful discussions . we are grateful to jun makino and koji takahashi for kindly providing valuable data and answering numerous questions . this work was supported in part by nsf grant ast-9618116 and nasa atp grant nag5 - 8460 . f.a.r . was supported in part by an alfred p.sloan research fellowship . f.a.r . also thanks the theory division of the harvard - smithsonian center for astrophysics for hospitality . this work was also supported by the national computational science alliance under grant ast970022n and utilized the sgi / cray origin2000 supercomputer at boston university . nasa also supported this work through hubble fellowship grant hf-01112.01 - 98a awarded ( to spz ) 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 . spz is grateful to mit for its hospitality and to tokyo university for the extensive use of their grape systems .

the structure of and states inb is investigated with an orthogonality condition model ( ocm ) based on the gaussian expansion method . full levels up to the and states around the threshold (=11.1 mev ) are reproduced consistently with the experimental energy levels . it is shown that the state located around theli+ threshold has an cluster structure , whereas the and states have a shell - model - like compact structure . we found that the state does not possess an-condensate - like nature similar to the state ofc ( hoyle state ) which has a dilute-condensate structure described by a configuration with about% probability , although the monopole transition strength of the former is as large as that of the latter . we discuss the reasons why the state does not have the condensate character . on the other hand , the state just below theli+ threshold has a cluster structure which can be interpreted as a parity - doublet partner of the state . we indicate that the-mev state ( ) just above the threshold observed in theli(li,b) reaction etc . is of the dilute - cluster - gas - like , and is a strong candidate for the hoyle - analogue state which has a configuration of with about% probability from the analyses of the single - cluster motions inb . the structure property of the resonant state is analyzed with the complex scaling method . + +

1007.4125

cluster picture as well as mean - field picture is important to understand the structure of light nuclei @xcite . the hoyle state ( the @xmath26 state at @xmath27 mev in @xmath13c ) is a typical cluster state with the @xmath14 cluster structure @xcite , which appears by only @xmath28 mev above the @xmath14 threshold and is characterized by the large monopole transition rate sharing about 16 % of the energy - weighted sum rule @xcite . recently the hoyle state has been reinvestigated from the viewpoint of the @xmath9 condensation @xcite . the definitions and occurrences of the nuclear @xmath9-particle condensation are discussed in detail in ref . many theoretical works have showed that the hoyle state has a @xmath14-condensate - like structure @xcite , in which the @xmath14 particles occupy an identical @xmath29-orbit with @xmath30 probability @xcite , forming a dilute @xmath9-gas - like configuration with @xmath15 , i.e. the product state of three @xmath9 s . the structure study of @xmath31o has recently made a great advance . the six lowest @xmath32 states of @xmath31o have very nicely been reproduced up to about @xmath33 mev excitation energy , including the ground state , with the @xmath34 orthogonality condition model ( ocm ) @xcite . the ocm is a semi - microscopic cluster model , which is an approximation of rgm ( resonating group method ) and is extensively described in ref . many successful applications of ocm are reported in ref . the @xmath34 ocm calculation showed that the @xmath35 state around the @xmath34 threshold is a strong candidate of the @xmath34-particle condensate state , having a large @xmath9 condensate fraction of @xmath36 % for the @xmath9-gas - like configuration @xmath37 as well as a large component of @xmath38c@xmath39 configuration . the @xmath40 state could be called the hoyle - analogue state . the hoyle - analogue states heavier than @xmath31o are predicted to exist around their @xmath9-cluster disintegrated thresholds in self - conjugate @xmath41 nuclei @xcite . besides the @xmath42 nuclei , one can also expect cluster - gas states composed of alpha and triton clusters ( including valence neutrons etc . ) around their cluster disintegrated thresholds in @xmath43 nuclei , in which all clusters are in their respective @xmath29 orbits , similar to the hoyle state with @xmath15 . the states , thus , can be called _ hoyle - analogue _ in the non - self - conjugated nuclei . it is an intriguing subject to investigate whether or not the hoyle - analogue states exist in @xmath43 nuclei , for example , @xmath2b , composed of @xmath44 and @xmath23 clusters . the cluster structure in @xmath2b was studied about 30 years ago with the @xmath3 ocm using the harmonic oscillator basis by nishioka et al . @xcite . they found that 1 ) the @xmath45 state ( @xmath46 mev ) just below the @xmath47 threshold has a well - developed molecular - like structure of @xmath21li(g.s)+@xmath9 , where @xmath21li(g.s ) denotes the ground state of @xmath8li with the @xmath9+@xmath23 cluster structure , and 2 ) the @xmath48 ( g.s ) and @xmath49 ( @xmath50 mev ) states have shell - model - like compact structures . the model space adopted in ref . @xcite , however , was not sufficient to account simultaneously for the @xmath8li+@xmath9 cluster configuration as well as the @xmath3 gas - like configuration . in addition , limited experimental data in those days caused difficulties in giving the definite conclusion that the @xmath45 state has an @xmath3 cluster structure . recently kawabata and his collaborators have investigated the excited states of @xmath2b by using the @xmath51 reaction @xcite . they eventually concluded that the @xmath4 state at @xmath6 = 8.56 mev has an @xmath9-cluster structure . among many reasons for this conclusion , one is a large isoscalar monopole transition rate for the @xmath4 state , @xmath52 @xmath53 , which is of the similar value to that for the hoyle state in @xmath13c , @xmath54 @xmath53 . a close relation between large isoscalar monopole strengths and underlying cluster structures in excited states were discussed by the present authors et al . ( see ref . @xcite ) . according to the literature , the largeness of the isoscalar monopole transition rate for @xmath4 originates from the fact that the state has a cluster structure . another reason why the @xmath4 state has the cluster structure is that the amd ( antisymmetrized molecular dynamics ) calculation @xcite has reproduced the large monopole transition rate for the @xmath4 state and has succeeded in assigning an @xmath3 cluster structure to this state . comparing the density distribution of the @xmath4 state ( calculated by amd ) with that of the hoyle state ( together with the analysis of the expectation values of the harmonic oscillator quanta in their wave functions ) , the amd calculation @xcite claimed that the @xmath4 state has a clustering feature similar to that of the hoyle state . however , they did not study single - cluster properties such as single - cluster orbits and their occupation probabilities in the @xmath4 state . their quantities are independent on the information of the density distribution and the expectation values of the harmonic oscillator quanta in the wave function . thus , it is very important to study the single - cluster properties in order to judge whether the @xmath4 state possesses the similarity to the hoyle state with the @xmath15-like structure . as discussed in refs . @xcite , the single - cluster motions such as @xmath9 ( @xmath23 ) cluster orbits and their occupation probabilities in a nuclear state can be investigated by solving the eigenvalue equation of the single - cluster density matrix @xmath55 derived from the microscopic and/or semi - microscopic wave functions . the purpose of the present paper is to study the structure of @xmath56 and @xmath57 states in @xmath2b up to around the @xmath3 threshold . here we take the @xmath3 ocm with the gaussian expansion method ( gem ) , the model space of which is large enough to cover the @xmath3 gas , the @xmath8li+@xmath9 cluster , as well as the shell - model configurations . combining ocm and gem provides a powerful method to study the structure of light nuclei @xcite as well as light hypernuclei @xcite , because the pauli - blocking effect among the clusters is properly taken into account by ocm and gem covers an approximately complete model space @xcite . this framework can treat precisely a strong parity dependence of the @xmath58 potential : the negative - parity potential is attractive enough to make bound states ( @xmath48 and @xmath59 ) and resonant states ( @xmath60 and @xmath61 ) of @xmath8li , where s the positive - parity potential is not . this parity dependence should play an important role in producing the cluster states of @xmath2b , as will be discussed later . the single - cluster properties such as single-@xmath9-particle ( single-@xmath23-particle ) orbits and occupation probabilities are investigated to judge whether or not the @xmath18 state has a hoyle - analogue structure by solving the eigenvalue equation of the single cluster density matrix @xmath55 derived from the total wave function of @xmath2b . the resonance structure of @xmath1 states is also studied with the complex scaling method ( csm ) , since they appear around the @xmath3 threshold with @xmath62 mev . it is also interesting to explore the hoyle - analogue state with the @xmath3 structure in @xmath1 states , because one can conjecture an appearance of the product state , @xmath63 , in @xmath64 of @xmath2b ( @xmath65 comes from the spin of triton ) from the similarity of @xmath15 in the hoyle state . the structure of the positive - parity states ( isospin @xmath66 ) in @xmath2b so far has not been discussed well in amd @xcite and non - core shell model @xcite . recently several positive - parity states of @xmath2b with @xmath66 are observed with the @xmath21li(@xmath67be,@xmath68)@xmath69he reaction @xcite , the @xmath21li(@xmath8li,@xmath2b@xmath22)@xmath23 reaction @xcite , and the decay - particle measurements of @xmath9+@xmath21li and @xmath23+@xmath70be from excited @xmath2b@xmath22 states @xcite . an interesting result in their experiments is as follows : the @xmath71 state at @xmath72 mev which was identified so far as isospin @xmath73 state @xcite , is observed through the @xmath9+@xmath21li decay channel @xcite . they concluded that the @xmath19-mev state has isospin @xmath66 or alternatively it is observed at a similar energy to that with @xmath73 . therefore , it is interesting to investigate theoretically the structure of the 12.56-mev state . the present paper is organized as follows . in sec . [ sec : formulation ] we formulate the @xmath3 ocm with gem as well as csm . the results and discussion are devoted in sec . [ sec : results_discussions ] . finally we present the summary in sec . [ sec : summary ] . in this section , we present the formulation of the @xmath3 ocm with gem for @xmath2b , and give a brief formulation of csm with the @xmath3 ocm to study the resonant structures of @xmath2b . the total wave function of @xmath2b ( the total angular momentum @xmath74 and total isospin @xmath66 ) within the frame of the @xmath3 ocm is expanded in terms of the gaussian basis , @xmath75 , \label{eq : total_wf}\\ & & \phi_c^{(ij , k)}(\nu,\mu)=\left[\left[\varphi_{\ell}(\vec{r}_{ij},\nu)\varphi_{\lambda}(\vec{r}_k,\mu)\right]_l\chi_{\frac{1}{2}}(t)\right]_j,\\ & & \varphi_\ell(\vec{r},\nu)=n_\ell(\nu)r^\ell \exp(-\nu r^2 ) y_\ell ( \hat{\vec{r}}),\end{aligned}\ ] ] where we assign the cluster numbers as 1 and 2 for the two @xmath9 clusters ( spin 0 ) and 3 for @xmath23 cluster ( spin 1/2 ) . @xmath76 [ @xmath77 denotes the relative wave function of the @xmath3 system with the jacobi - coordinate system shown in fig . [ fig:1](a ) [ figs . [ fig:1](b ) and ( c ) ] . it is noted that @xmath78 is symmetric with respect to the particle - number exchange between 1 and 2 . @xmath79 is the normalization factor of the gaussian basis @xmath80 , and @xmath81 ( @xmath82 ) denotes the relative coordinate between the _ i_th and _ j_th clusters ( the _ k_th cluster and the center - of - mass coordinate of the _ i_th and _ j_th clusters ) . the angular momentum channel is presented as @xmath83_j$ ] , where @xmath84 ( @xmath85 ) denotes the relative orbital angular momentum with respect to @xmath81 ( @xmath86 ) , and @xmath87 ( @xmath88 ) is the total orbital angular momentum ( the spin of @xmath23 particle ) . the gaussian parameter @xmath89 is taken to be of geometrical progression , @xmath90 it is noted that this prescription is found to be very useful in optimizing the ranges with a small number of free parameters with high accuracy @xcite . the total hamiltonian for the @xmath3 system is presented as @xmath91 \nonumber \\ & + & v_{2\alpha t } + v_{\rm pauli } , \label{eq : hamiltonian}\end{aligned}\ ] ] where @xmath92 , @xmath93 ( @xmath94 ) and @xmath95 stand for the kinetic energy operator for the _ i_-th cluster , @xmath9-@xmath9 ( @xmath9-@xmath23 ) potential , and three - body potential , respectively , and @xmath96 ( @xmath97 ) is the coulomb potential between 2@xmath9 clusters ( @xmath9 and @xmath23 ) . the center - of - mass kinetic energy ( @xmath98 ) is subtracted from the hamiltonian . the pauli - blocking operator @xmath99 @xcite is expressed as @xmath100 which removes the pauli forbidden states between the two @xmath9-particles in @xmath29 , @xmath101 and @xmath102 states as well as those between @xmath9 and @xmath23 particles in @xmath29 , @xmath103 , @xmath101 and @xmath102 states . consequently , the ground state with the shell - model - like configuration can be described properly in the present model . the effective @xmath9-@xmath9 potential and coulomb potential , @xmath93 and @xmath96 , are constructed with the folding procedure , where we fold the modified hasegawa - nagata effective @xmath104 interaction ( mhn ) @xcite and the @xmath105 coulomb potential with the @xmath9-particle density . they reproduce the observed @xmath106 scattering phase shifts ( @xmath107- , @xmath108- and @xmath109-waves ) and the energies of the @xmath110be ground state and of the hoyle state ( @xmath111 in @xmath13c ) @xcite . concerning the effective @xmath9-@xmath112 potential and coulomb potentials , we use the folding potentials which reproduces well the low - lying energy spectra of @xmath21li [ @xmath113 ( g.s . ) , @xmath114 , @xmath115 , and @xmath116 ( including the low - energy @xmath58 scattering phase shifts ) obtained initially by nishioka et al . @xcite . the potentials are applied to the structure study of @xmath117li with the @xmath9+@xmath23+@xmath118 model and are known to reproduce well the low - lying structure of the hypernucleus @xcite . it is reminded that the @xmath58 potential has a strong parity - dependence : the odd - parity potentials are attractive to reproduce the bound states ( @xmath0 , @xmath119 ) with respect to the @xmath120 threshold , while the even - parity potentials are weakly attractive and thus no bound / resonant states with positive parity have not been observed in low - energy excitation region . the three - body potential @xmath121 is introduced phenomenologically so as to reproduce the energies of the ground state ( @xmath113 ) and the first excited positive parity state ( @xmath122 ) of @xmath2b with respect to the @xmath3 threshold . the origin of @xmath121 is considered to derive from the state dependence of the effective nucleon - nucleon interaction and an additional pauli repulsion arising from exchanging nucleons among the three clusters ( @xmath3 ) . it should be short - range , and hence only act in compact configurations . in the present paper , we take the following phenomenological three - body potential , @xmath123 where @xmath124 with the total orbital angular momentum @xmath87 represents the su(3)@xmath125(\lambda\mu)$ ] wave function with the total oscillator quanta @xmath126 ( @xmath127 ) . it is noted that the present @xmath3 model space can be classified into the su(3 ) bases with the irreducible representation , @xmath128 , with partition @xmath129=[443]$ ] , and the total wave function of @xmath2b with positive ( negative ) parity in eq . ( [ eq : total_wf ] ) can be expanded in terms of the @xmath124 bases with even ( odd ) @xmath126 values @xcite . the @xmath130 basis with @xmath131 is unique pauli allowed state for negative - parity states of @xmath2b with @xmath131 , which is equivalent to the shell - model configuration of @xmath132 . thus this su(3 ) basis becomes the main component in the ground state of @xmath2b . on the other hand , there exist only three pauli - allowed states , @xmath133 , @xmath134 and @xmath135 , for the positive - parity states with @xmath136 . therefore they correspond to the main components in the first excited states with even parity in the present model . for simplicity , the strengths of the three - body potential , @xmath137 and @xmath138 , are fixed so as to reproduce the experimental energies of the ground state of @xmath2b and the first @xmath57 state ( @xmath139 mev ) , respectively , with respect to the @xmath3 threshold . the expectation value of this three - body potential does not exceed 10 % of that of the corresponding two - body term , even for the ground state with the most compact structure , i.e. being the most sensitive to the potential . the equation of motion of @xmath2b with the @xmath3 ocm is obtained by the variational principle , @xmath140=0 , \label{eq : variational_principle}\end{aligned}\ ] ] where @xmath141 denotes the eigenenergy of @xmath2b measured from the @xmath3 threshold . the energy @xmath141 and expansion coefficients @xmath142 and @xmath143 in the total wave function shown in eq . ( [ eq : total_wf ] ) are determined by solving a secular equation derived from eq . ( [ eq : variational_principle ] ) . it is instructive to study single-@xmath9-particle ( single-@xmath23-particle ) orbits and corresponding occupation probabilities in @xmath2b . we define the single - cluster density matrices for @xmath9 and @xmath23 clusters , respectively , as @xmath144 where @xmath145 @xmath146 [ @xmath147 represents the coordinate vector of the @xmath148th @xmath9 ( triton ) cluster with respect to the center - of - mass coordinate of the @xmath3 system . the single-@xmath9-particle ( single-@xmath23-particle ) orbits and corresponding occupation probabilities are obtained by solving the eigenvalue equation of the single - cluster density matrices , @xmath149 where the eigenvalue @xmath150 ( @xmath151 ) denotes the occupation probability for the corresponding single - cluster orbit @xmath152 ( @xmath153 ) with the argument of the intrinsic coordinate of an arbitrary @xmath9 ( triton ) cluster in a nucleus measured from the center - of - mass coordinate of @xmath2b . the spectrum of the occupation probabilities provide important information on the occupancies of the single-@xmath9- ( @xmath23- ) particle orbit in @xmath2b . if the two @xmath9 particles ( one @xmath23 particle ) occupy only an single orbit , the occupation probability for this orbit becomes 100 % . the nuclear root - mean - square ( rms ) radius of @xmath2b in the present ocm is given as @xmath154^{1/2 } , \label{eq : rms}\end{aligned}\ ] ] where @xmath155 ( @xmath156 ) presents the rms distance between @xmath9 and @xmath9 ( @xmath157 and @xmath23 ) in @xmath2b . ( [ eq : rms ] ) we take into account the finite size effect of @xmath9 and @xmath23 clusters , where the experimental rms radii for the @xmath9 and @xmath23 particles are used in @xmath158 and @xmath159 , respectively . the reduced width amplitude or overlap amplitude is useful to see the degree of clustering in nucleus . in the present paper , we study the reduced width amplitudes for the @xmath21li+@xmath9 and @xmath70be+t channels , respectively , defined as @xmath160_j | \phi_j({^{11}{\rm b } } ) \right\rangle , \label{eq : rwa_alpha}\\ & & { \cal y}_{j_8(\ell_{83}\frac{1}{2})j_{83}j}^{{^8}{\rm be}-t}({r}_{83 } ) = r_{83 } \times \left\langle \left [ \frac{\delta({r^{\prime}_{83}}-r_{83})}{{r^{\prime}_{83}}^2 } \phi_{j_8}({^8}{\rm be } ) [ y_{\ell_{83}}(\hat{\vec{r}}^{\prime}_{83})\chi_{\frac{1}{2}}(t)]_{j_{83}}\right]_j | \phi_j({^{11}{\rm b } } ) \right\rangle , \label{eq : rwa_triton}\end{aligned}\ ] ] where @xmath161 ( @xmath162 ) denotes the radial part of the relative coordinate between @xmath8li and @xmath9 ( @xmath70be and @xmath23 ) . the wave function @xmath163 ( @xmath164 ) of @xmath8li ( @xmath110be ) with the total angular momentum @xmath165 ( @xmath166 ) is obtained with the @xmath120 ocm ( @xmath167 ocm ) . the spectroscopic factor @xmath168 is defined as @xmath169 ^ 2,\end{aligned}\ ] ] where @xmath170 denotes the reduced width amplitude . in the present study , we take the complex - scaling method ( csm ) for studying the resonant structures of @xmath2b . the csm is a powerful tool to investigate the resonant parameters ( energies and widths ) of the resonant states as well as the binding energies of the bound states @xcite . application of csm is easy in the framework of the @xmath3 ocm . in csm , we transform the relative coordinates of the @xmath3 model shown in fig . [ fig:1 ] , @xmath171 with @xmath172=(12,3 ) , ( 23,1 ) and ( 32,1 ) , by the operator @xmath173 as @xmath174 where @xmath175 is a scaling angle . then , the hamiltonian in eq . ( [ eq : hamiltonian ] ) is transformed into the complex - scaled hamiltonian , @xmath176 the corresponding complex - scaled schrdinger equation is expressed as @xmath177 where @xmath178 is given in eq . ( [ eq : total_wf ] ) . the eigenstates @xmath179 are obtained by solving the eigenvalue problem of @xmath180 in eq . ( [ eq : csm ] ) . in csm , all the energy eigenvalues @xmath141 of bound and unbound states are obtained on a complex energy plane , according to abc theorem @xcite . in this theorem , it is proven that the boundary condition of gamow resonances is transformed to the damping behavior at the asymptotic region . thanks to this condition , the same theoretical method as that used for the bound states can be employed to obtain the many - body resonances . for @xmath2b , the continuum states of @xmath21li+@xmath9 , @xmath70be+@xmath23 and @xmath3 channels are obtained on the branch cuts rotated with the @xmath181 dependence @xcite . on the contrary , bound states and resonances are discrete and obtainable independently of @xmath175 . thus , these discrete states are located separately from the many - body continuum spectra on the complex energy plane . one can identify the resonance poles of complex eigenvalues : @xmath182 , where @xmath183 and @xmath184 are the resonance energy measured from the @xmath3 threshold and the decay width , respectively . in order to solve the complex - scaled schrdinger equation eq . ( [ eq : csm ] ) , we expand the wave function eq . ( @xmath185 ) in terms of the gaussian basis functions shown in eq . ( [ eq : total_wf ] ) , @xmath186 . \label{eq : wf_csm_gaussian}\end{aligned}\ ] ] the expansion coefficients @xmath187 and @xmath188 together with the discrete energy spectrum @xmath141 are obtained by solving an eigenvalue problem derived from eqs . ( [ eq : csm ] ) and ( [ eq : wf_csm_gaussian ] ) @xcite . this complex scaling method with the gaussian basis functions is known to give high precision needed to maintain the numerical stability of the complex eigenvalue problem . the energy levels of @xmath0 and @xmath57 states in @xmath2b with respect to the @xmath3 threshold are shown in fig . [ fig:2 ] , which are obtained with the @xmath3 ocm using gem . in the present calculation , we found that three @xmath56 states and two @xmath57 states come out as either bound states against particle decays or resonant states . first we discuss the structures of the three @xmath0 states . the @xmath48 state at @xmath189 mev measured from the @xmath3 threshold corresponds to the ground state of @xmath2b . the calculated nuclear radius is @xmath190=2.22 fm ( see table [ tab:1 ] ) , the value of which is in correspondence with the experimental data @xmath191 fm @xcite . according to the analysis of the wave function , the main component of this state is su(3)@xmath129(\lambda,\mu)_{l}=[443](1,3)_{1}$ ] with @xmath131 harmonic oscillator quanta ( 95 % ) and its dominant angular momentum channel is @xmath192 . thus , this state has a compact shell - model - like structure , in which the two @xmath9 and one @xmath23 clusters are heavily overlapping and melting away each other ( due to antisymmetrization among nucleons ) to have a shell - model configuration . on the other hand , we found that the @xmath49 state at @xmath193 mev has also a compact shell - model - like structure with the dominant component of su(3)@xmath194 with @xmath131 and @xmath195 . the nuclear radius of this state is @xmath190=2.23 fm , the value of which is similar to that of the ground state . the energy difference between the @xmath48 and @xmath49 states is about @xmath196 mev , which is smaller than the experimental data ( @xmath197 mev ) . this feature is recovered if we take into account the su(3)@xmath198 $ ] shell model configuration in the present ocm model space ( or adopting an extended @xmath199 model space in the cluster - model terminology ) , as pointed out by nishioka et al @xcite . . calculated nuclear radii ( @xmath190 ) , @xmath200 rms distances ( @xmath155 ) and @xmath201 ones ( @xmath202 ) for the @xmath0 states of @xmath2b together with their @xmath58 and @xmath203 rms distances ( @xmath204 and @xmath205 , respectively ) . they are given in units of fm . [ cols="^,^,^,^,^,^",options="header " , ] in addition to the two @xmath0 states discussed above , the @xmath4 state appears at @xmath206 mev ( @xmath207 mev with respect to the @xmath3 threshold ) . the nuclear radius of @xmath18 is @xmath190=3.00 fm . this value is by about @xmath208 % larger than that of the ground state of @xmath2b and is consistent with the amd calculation ( @xmath190=3.0 fm ) @xcite . the rms distance between @xmath9 and @xmath9 ( between @xmath110be(@xmath44 ) and @xmath23 ) is @xmath209 fm ( @xmath210 fm ) . it is reminded that the value of @xmath211 fm in the ground state of @xmath70be with the @xmath167 ocm calculation . thus , the intra - cluster distance between the two @xmath9 s in @xmath4 is a little bit larger that that in the ground state of @xmath110be . in addition , the intra - cluster distance between the @xmath9 and @xmath23 clusters in @xmath18 is 4.15 fm ( see table [ tab:1 ] ) , which is also larger than that ( @xmath212 fm ) in the ground state of @xmath8li(@xmath120 ) . in order to study the structure of @xmath18 , it is interesting to compare the reduced width amplitudes or overlap amplitudes of the @xmath8li(g.s)+@xmath9 channel [ @xmath110be(g.s)+@xmath23 ] defined in eq . ( [ eq : rwa_alpha ] ) [ ( [ eq : rwa_triton ] ) ] for the @xmath18 and @xmath214 states . the results are shown in fig . [ fig:3 ] . one sees that 1 ) the reduced width amplitudes of the @xmath8li+@xmath9 and @xmath110be+@xmath23 channels in @xmath18 are enhanced and significantly larger than those in the ground state , in particular , in the outermost - peak region , and 2 ) the number of the nodes of their amplitudes in the former state increase by one larger than those in the latter . it is noted that the radial behaviors of the reduced width amplitudes in the ground state of @xmath2b are determined by the dominant nature of su(3)@xmath215 in its state . as for the @xmath168 factors in @xmath4 , the @xmath21li(g.s)+@xmath9 channel with the relative orbital angular momentum @xmath216 has the largest value of @xmath217 among all of the @xmath21li+@xmath9 channels , and the secondarily ( thirdly ) largest channel is the @xmath21li(g.s)+@xmath9 ( @xmath21li(@xmath218)+@xmath9 ) channel ( @xmath219 ) with @xmath220 ( 0.125 ) , while the @xmath70be(g.s)+@xmath23 channel with the relative angular momentum @xmath221 has the largest value of @xmath222 . in the @xmath48 state , the corresponding @xmath168 factors are @xmath223 , @xmath224 , @xmath225 and @xmath224 , respectively for @xmath21li(g.s)+@xmath9 ( @xmath216 ) , @xmath21li(g.s)+@xmath9 ( @xmath219 ) , @xmath21li(@xmath119)+@xmath9 ( @xmath219 ) , and @xmath70be(g.s)+@xmath23 ( @xmath221 ) . consequently , the @xmath4 state has an @xmath3 cluster structure , although the cluster structure of the @xmath9+@xmath23 part is significantly distorted from that of the ground state of @xmath21li . cc recently , kawabata et al . measured the isoscalar monopole transition rate @xmath226(e0:is ) for @xmath4 with the @xmath2b@xmath227 reaction @xcite . the experimental value is as large as @xmath228 @xmath229 , comparable to that of the hoyle state in @xmath13c ( @xmath226(e0:is)@xmath230 @xmath229 ) . the calculated result in the present model is @xmath226(e0:is)=@xmath231 @xmath229 , in good agreement with the experimental data . on the other hand , the amd calculation also reproduces the experimental data ( @xmath232(e0:is)=@xmath233 @xmath229 ) . they analyzed the density distributions of the ground state and excited states in @xmath2b and @xmath13c , and concluded that the @xmath4 state in @xmath2b is a dilute cluster state , and its features are similar to those of the hoyle state @xmath13c@xmath234 which is likely to be a gas state of @xmath14 clusters . however , it is not self - evident in the amd calculation whether the @xmath4 state possesses an @xmath9 condensate nature like the hoyle state having about @xmath16% probability of the @xmath9 particle occupied in a single @xmath107 orbit . in order to see the single - cluster properties in @xmath2b , we study the single - cluster orbits and their occupation probabilities in the @xmath4 state by solving the eigenvalue equation of the single - cluster density matrices in eqs . ( [ eq : density_alpha ] ) and ( [ eq : density_triton ] ) , and compare them with those in the hoyle state . cc + + figure [ fig:4 ] shows the occupation probabilities of the _ n_th @xmath87-wave single-@xmath9-particle ( single-@xmath23-particle ) orbits , @xmath235 ( @xmath236 ) , in the @xmath113 and @xmath45 states . in the @xmath113 state , the occupation probabilities of the single-@xmath9-particle orbits spread out in several orbits : @xmath237 , @xmath238 , @xmath239 , @xmath240 , and @xmath241 . the remaining probability , @xmath242 , belongs to the higher orbits @xmath243 with @xmath244 . these features can be understood from the fact that the @xmath113 state has the dominant configuration of su(3)@xmath129(\lambda,\mu)_{l}=[443](1,3)_{1}$ ] ( @xmath126=7 ) coupled with the spin of @xmath23-cluster . in fact , the su(3 ) wave function is expressed in terms of the harmonic oscillator basis @xmath245 , @xmath246(1,3)_{l=1 } \rangle } & = & \sqrt{\frac{64}{225}}{| ( 2s)\otimes(1p)\rangle } - \sqrt{\frac{56}{225}}{| ( 1d)\otimes(1p)\rangle } \nonumber\\ & - & \sqrt{\frac{24}{225}}{| ( 1d)\otimes(0f)\rangle } + \sqrt{\frac{81}{225}}{| ( 0g)\otimes(0f)\rangle } , \label{eq : su3_11b_l1}\end{aligned}\ ] ] where @xmath247 [ @xmath248 represents the harmonic oscillator wave function , @xmath249 and @xmath84 denoting the number of nodes and the orbital angular momentum , respectively , with respect to the relative coordinate @xmath250 between the two @xmath9 clusters ( @xmath251 between the @xmath44 and @xmath23 clusters ) . from the definition of the single-@xmath9-particle density matrix in eq . ( [ eq : density_alpha ] ) , one obtains the occupation probabilities @xmath252 as follows : @xmath237 , @xmath253 , @xmath254 , @xmath255 , and @xmath256 . these results agree well with those obtained by diagonalizing the single-@xmath9-particle density matrix as mentioned above ( see also sfig . [ fig:4 ] ) . the radial behaviors of the @xmath257- , @xmath258- and @xmath259-wave single-@xmath9-particle orbits as well as @xmath260 and @xmath261 orbits are shown in fig . [ fig:5](a ) . one sees strong oscillations in the inner regions , coming from the strong pauli - blocking effect , reflecting the nature of the compact shell - model - like structure in this state . these behaviors are quite similar to those of the ground state of @xmath13c shown in fig . [ fig:5](e ) @xcite , the structure of which is of the shell - model - like @xcite . on the other hand , the single-@xmath23-particle occupation probabilities in the @xmath214 state concentrate mainly on the two orbits : @xmath262 and @xmath263 ( see fig . [ fig:4](a ) ) . these results originate from the su(3 ) nature of the @xmath113 state . in fact , the definition of the single-@xmath23-particle density matrix in eq . ( [ eq : density_triton ] ) gives @xmath264 and @xmath265 for the su(3 ) wave function in eq . ( [ eq : su3_11b_l1 ] ) . the reason why the single-@xmath23-particle probability with odd parity is exactly zero in @xmath56 is due to the symmetric nature of the total wave function of @xmath2b in eq . ( [ eq : total_wf ] ) . in fact , the symmetric character in eq . ( [ eq : total_wf ] ) with respect to the exchange between two @xmath9 clusters causes even - parity orbital angular momentum between the two @xmath9 s and thus , only odd - parity - wave single-@xmath23-particle orbits are allowed in the negative - parity @xmath2b state . this is discussed in details in the case of @xmath13c with the @xmath14 ocm ( see ref . @xcite ) . the radial behavior of the @xmath266-wave single-@xmath23-particle orbit in fig . [ fig:5](c ) indicates a strong oscillation in the inner region , reflecting the compact shell - model - like feature of the @xmath113 state . cc + + + + as for the @xmath4 state possessing an @xmath3 cluster structure , the occupation probabilities of the single-@xmath9-particle orbits have _ no _ concentration on a single orbit : @xmath267 , @xmath268 , @xmath269 , @xmath270 , and @xmath271 ( see fig . [ fig:4 ] ) . the remaining probability , @xmath272 , belongs to the higher orbits @xmath243 with @xmath244 . one notices that @xmath273 . it is noted that there is a close relation between the occupation probability @xmath274 and the @xmath275 factor of the @xmath9 reduced width amplitudes , as discussed in ref . in fact , in the present calculation , the @xmath107-wave and @xmath108-wave @xmath168 factors of the @xmath21li+@xmath9 reduced width amplitudes in @xmath4 amount to @xmath276 and @xmath277 , respectively , the ratio of which is similar to @xmath278 . the radial behaviors of the single-@xmath9-particle @xmath257 , @xmath258 and @xmath259 orbits are shown in fig . [ fig:5](b ) . compared with the results of the ground state ( see fig . [ fig:5](a ) ) , the strong oscillation in the inner region is rather suppressed , although the outermost peak is enhanced in the @xmath18 state . for reference , we show in fig . [ fig:5](f ) the @xmath107-wave single-@xmath9-particle orbit in the hoyle state of @xmath13c @xcite ( which is a typical example of @xmath9 condensation in nuclei ) , where the @xmath9 particle occupies the @xmath107 orbit with about @xmath279 probability . in this hoyle state , one sees no nodal behaviors ( but small oscillations ) in the inner region together with a long tail , and the radial behavior is likely to be of the @xmath29-type gaussian . this gaussian behavior and its occupation probability as large as about @xmath279 are the evidence that the hoyle state has a @xmath14 condensate structure @xcite . comparing them with the case of the @xmath4 state in @xmath2b , we find remarkable qualitative differences between them : 1 ) the @xmath4 state has no concentration of occupation probability on a single-@xmath9-particle orbit and 2 ) nodal behaviors remain in the inner region in @xmath4 , indicating rather large pauli - blocking effect . these results show that the @xmath4 state does not have @xmath9-condensate nature like the hoyle state . concerning the single-@xmath23-particle occupation probabilities , the @xmath4 state has @xmath280 and @xmath281 , as shown in fig . [ fig:4 ] . the remaining probability is attributed to higher orbits @xmath243 with @xmath244 . the radial behavior of the @xmath260-wave single-@xmath23-particle orbit is shown in fig . [ fig:5](d ) . in contrast with the case of the single-@xmath9-particle orbits , one sees that the radial behavior of the @xmath260-wave single-@xmath23-particle orbit ( @xmath280 ) has no nodal behaviors but somewhat large oscillator behaviors in the inner region . this radial behavior is likely to be of the @xmath103-wave - like . the reason why the nodal behavior disappears is due to the fact that the rms distance between the center - of - mass of @xmath44 and the triton is as large as @xmath282 fm ( see table [ tab:1 ] ) , and thus the pauli blocking effect is weakened significantly to make the small oscillation in the inner region . from the above - mentioned results , the @xmath7 state could not be identified as the analogue of the hoyle state , although its state has the @xmath3 cluster structure . the essential reason why the @xmath7 state is not of the hoyle - analogue is due to the fact that the @xmath4 state is bound by @xmath283 mev with respect to the @xmath3 threshold , while the hoyle state appears by @xmath28 mev above the @xmath14 threshold , i.e. the energy level is located around the coulomb barrier produced by the clusters . this extra binding energy of the @xmath7 state with respect to the three - body threshold hinders significantly the development of the gas - like @xmath3 structure with a large nuclear radius in the state . in refs . @xcite , they discussed important requirements for the emergence of the hoyle - analogue states composing of clusters with a large radius , in which all the clusters occupy an @xmath29 orbit , and thus its dominant configuration is @xmath284 , where @xmath249 denotes the number of the clusters : 1 ) the state should be located above the constituent cluster threshold to make a dilute cluster structure with a large nuclear radius , i.e. it should appear around the coulomb barrier produced by the relevant constituent clusters , and 2 ) all of the clusters interact dominantly in @xmath107-wave each other to avoid the pauli - blocking effect among the clusters as well as reducing the effect of the centrifugal barrier among them . they are _ the necessary conditions for the appearance of the hoyle - analogue states_. these conditions are not fulfilled in the @xmath7 state as discussed in the present study . thus , the @xmath7 state does not have hoyle - analogue nature . the @xmath285 state experimentally appears at @xmath286 mev ( see fig . [ fig:6 ] ) . the lowest particle decay threshold is the @xmath8li+@xmath9 channel ( @xmath287 mev ) which is located at @xmath288 mev below the @xmath3 threshold . since the @xmath285 state appears around the @xmath8li+@xmath9 threshold , its structure may have a @xmath8li+@xmath9 cluster structure . in addition , the small excitation energy of @xmath285 in comparison with @xmath289 mev in this mass region suggests that an @xmath9-cluster correlation should play an important role in lowering the excitation energy . the positive - parity levels of @xmath2b obtained by the present @xmath3 ocm calculation are shown in fig . [ fig:6 ] ( only @xmath1 states as well as @xmath0 states are given in fig . [ fig:2 ] ) . the @xmath285 , @xmath290 and @xmath291 states are stable against any particle decays . other calculated states correspond to resonant states which are identified with use of csm , as explained in sec . [ sec : csm ] . one can see rather good correspondence to the experimental data . in the present paper , we concentrate on studying the structure of @xmath1 states . the structures of other positive - parity states will be discussed elsewhere , although we found that some even - parity states shown in fig . [ fig:6 ] have cluster structures . the calculated nuclear radius of @xmath285 is @xmath292 fm . this value is similar to the @xmath7 state which has the @xmath3 cluster structure ( see table [ tab:1 ] ) . figure [ fig:3](a ) shows the reduced width amplitude or overlap amplitude for the @xmath21li(@xmath0:g.s)+@xmath9 channel with the relative orbital angular momentum @xmath293 . one sees that the shape of the outermost peak at @xmath294 fm is similar to that of @xmath7 . the @xmath168 factor for this channel is @xmath295 . this is the largest among all of the @xmath21li+@xmath9 channels with @xmath8li@xmath296 , where the corresponding @xmath168 factors are @xmath295 , @xmath297 , @xmath298 , and @xmath299 , respectively . as for the @xmath70be+@xmath23 channel , the @xmath168 factor for @xmath70be(g.s)+@xmath23 [ @xmath70be(@xmath300)+@xmath23 ] is @xmath301 ( @xmath302 ) . the reduced width amplitude of the @xmath70be(g.s)+@xmath23 channel is shown in fig . [ fig:3](b ) . on the other hand the @xmath303 distance ( @xmath304 ) in @xmath285 is @xmath305 fm ( @xmath306 fm ) . this value is a little bit larger than that of the @xmath21li ground state ( @xmath307 fm ) obtained by the @xmath308 ocm calculation . thus , the @xmath285 state has a dominant structure of @xmath21li(g.s)+@xmath9 with @xmath266-wave relative motion . as discussed in sec . [ sec:3/2- ] , the @xmath4 state has the largest @xmath275 factor for the @xmath8li(g.s)+@xmath9 channel with @xmath107-wave relative motion . this means that the state includes significantly the structure of @xmath8li(g.s)+@xmath9 . in the non - identical two - cluster states , for example , in the @xmath31o+@xmath9 states of @xmath309ne , there appear the parity - doublet states , @xmath32 and @xmath310 , around the @xmath31o+@xmath9 threshold , in which the relative motion between @xmath31o and @xmath9 is of the @xmath107-wave and @xmath266-wave , respectively @xcite . thus , the @xmath285 and @xmath4 states of @xmath2b , which are located around the @xmath8li(g.s)+@xmath9 threshold , can be interpreted as the parity doublet in @xmath2b . cc the occupation probabilities of the single-@xmath9-particle and single-@xmath23-particle orbits for @xmath285 obtained by diagonalizing the single - cluster density matrices [ see eqs . ( [ eq : single_alpha_density ] ) and ( [ eq : single_triton_density ] ) ] are shown in fig . [ fig:7](a ) . one sees that the @xmath260 ( @xmath261 ) orbit of @xmath9-particle has the largest ( secondarily largest ) occupation probability of @xmath311 % ( @xmath208 % ) . a close relationship between the cluster occupation probabilities and @xmath168 factors ( as well as one between the single - cluster orbits and reduced width amplitudes ) is discussed in detail in refs . in fact , one can see the close relations in the present results : 1 ) the @xmath168 factor of @xmath21li(g.s)+@xmath9 with @xmath293 ( @xmath312 ) is the largest ( secondarily largest ) among all of the @xmath21li+@xmath9 channels , as discussed above , and 2 ) the radial behavior of the @xmath260 orbit in fig . [ fig:8](a ) is similar to that of the reduced width amplitude the @xmath21li(g.s)+@xmath9 channel in fig . [ fig:3](a ) . it should be reminded that the occupation probabilities satisfy with a sum rule , while the @xmath168 factors do not . on the other hand , the occupation probabilities of single-@xmath23-particle orbits spread out over @xmath107- , @xmath108- , and @xmath109-orbits . this feature is similar to the results of the @xmath168 factors for the @xmath70be+@xmath23 channels . the reason why the odd - parity @xmath23-orbit occupation probabilities are exactly zero comes from a symmetric feature of the @xmath3 cluster model with respect to the exchange of the two @xmath9 s , as mentioned above ( see also ref . the present results indicate that the @xmath285 state is not of the hoyle - analogue , in which all clusters dominantly occupy the @xmath313 orbit . the main reason why the @xmath285 state does no have a dilute structure like the hoyle state is due to the fact that @xmath285 is bound by 4.2 mev with respect to the @xmath3 threshold . this fact hinders strongly the growth of a gas - like @xmath3 structure in @xmath285 . the csm with the @xmath3 ocm is a powerful tool to identify resonant states , in particular , those around the @xmath3 threshold . figure [ fig : complex_scaling ] displays the trajectory of the energy eigenevalues obtained by solving the complex - scaled schrdinger equation in eq . ( [ eq : csm ] ) as a function of the scaled angle @xmath175 . as explained in sec . [ sec : csm ] , one can identify bound states and resonances as stationary points independently of @xmath175 . in the present study , in addition to the bound @xmath122 state , we found that the @xmath314 state appears at @xmath315 mev , ( @xmath316 mev just above the @xmath3 threshold ) , as a resonant state with the width of @xmath317 kev . it is reminded that the hoyle state ( @xmath13c@xmath234 ) appears at @xmath28 mev above the @xmath14 threshold , and has the very small width of @xmath318 ev . cc + + it is instructive to study the structure of the @xmath319 state under the bound - state approximation , because the calculated width is small . we found that the calculated nuclear radius of @xmath319 is @xmath320 fm , the value of which is significantly larger than that of @xmath321 . this means that the @xmath319 state has a dilute cluster structure . in order to study the structure further , we calculated the occupation probabilities and radial behaviors of the single-@xmath9-particle and single-@xmath23-particle orbits by diagonalizing the single - cluster density matrices defined in eqs . ( [ eq : single_alpha_density ] ) and ( [ eq : single_triton_density ] ) . the results are shown in fig . [ fig:7](b ) . we see that the occupation probability of @xmath257-wave @xmath9-orbit ( @xmath322 ) is as large as about @xmath323 , and that of @xmath257-wave @xmath23-orbit ( @xmath324 ) amounts to about @xmath325 . then the probability of the @xmath326 configuration in @xmath319 can be estimated to be @xmath327 @xmath328 65% , where the dominator denotes the total cluster number ( 3 clusters ) and the numerator is the cluster number occupied in the @xmath329 configuration , i.e. @xmath330 . the radial behaviors of the @xmath322-orbit and @xmath331-orbit are depicted in figs . [ fig:8](c ) and [ fig:8](d ) , respectively , showing a gaussian behavior with no nodes . thus , we could call this state the hoyle - analogue , where all clusters are mainly in their respective @xmath29 orbits , i.e. @xmath329 with the probability of about 65% , which is similar to the hoyle state with the main configuration of @xmath332 possessing the probability of about @xmath16% . it is reminded that the present @xmath5 state satisfies the necessary conditions of appearing the hoyle - analogue states as discussed in sec . [ sec:3/2- ] . the peak position in the @xmath322-orbit ( @xmath331-orbit ) is around 3.5 fm ( 7.0 fm ) . this result means that the triton cluster moves outside of the @xmath9 clusters . thus , the @xmath319 state has a @xmath23-halo - like structure around the two @xmath9 clusters . it should be reminded that the @xmath107-wave @xmath200 interaction is attractive enough to produce the resonant state of @xmath70be(g.s ) , but the @xmath107-wave @xmath58 interaction is weakly attractive and has no ability to produce bound / resonant states in @xmath21li . this qualitative difference makes the @xmath23-halo - like structure around the two @xmath9 clusters in the @xmath5 state . here , it is interesting to discuss the reason why the calculated width of the @xmath333 state is small ( @xmath334 kev ) , in sipte of the large excitation energy ( @xmath315 mev ) . the decay width is composed of the contributions from the two - body decays ( @xmath8li+@xmath9 and @xmath110be+@xmath23 ) and three - body decay ( @xmath3 ) . as for the three - body decay , it is suppressed strongly due to the very small phase space arising from the fact that the @xmath333 state is located at @xmath335 mev above the @xmath3 threshold . on the other hand , the @xmath70be(g.s)+@xmath23 decay is possible , because its state appears at about @xmath336 mev above the @xmath70be(g.s)+@xmath23 threshold . but , its energy level is located inside the coulomb barrier ( no centrifugal barrier for the @xmath107-wave decay of @xmath70be(g.s)+@xmath23 ) , and thus its decay is hindered . as for the @xmath21li+@xmath9 decay , the energy of @xmath333 measured from the @xmath21li(g.s:@xmath0)+@xmath9 threshold is about @xmath337 mev . in this case , the orbital angular momentum of the decaying @xmath9 particle should be @xmath266-wave , but the @xmath266-wave occupation probability of @xmath9 particle in the state is as small as @xmath338 ( see fig . [ fig:7 ] ) . thus , the @xmath266-wave decay of @xmath21li(g.s)+@xmath9 is suppressed largely due to the small occupation probability and the coulomb and centrifugal barriers . on the contrary , the @xmath107-wave decay of @xmath21li+@xmath9 is also possible , because the @xmath107-wave occupation probability of @xmath9 particle is about @xmath339 . however , the bound and/or resonant states of positive - parity @xmath21li states have not been identified experimentally so far in the low - energy region up to @xmath340 mev @xcite . this experimental fact indicates a large hindrance of the @xmath107-wave @xmath21li+@xmath9 decay with even - parity @xmath21li states . consequently the calculated width of the @xmath333 state becomes small . it is interesting to discuss the experimental situation on the @xmath57 states ( isospin @xmath66 ) of @xmath2b around the @xmath3 threshold . as mentioned in sec . [ sec : intro ] , the @xmath71 state at @xmath72 mev with @xmath341 kev ( located at @xmath342 mev above the @xmath3 threshold ) , which was identified so far as isospin @xmath73 state @xcite , has recently been observed through the @xmath9+@xmath21li decay channel @xcite . comparing the experimental results with our present theoretical results , the energy and width of the @xmath319 state obtained by the present study are in good correspondence to the experimental data . thus , the present calculated results indicate that the @xmath19-mev state could be assigned as the @xmath319 state with the hoyle - analogue structure . we have studied the structure of @xmath56 and @xmath57 states in @xmath2b with the @xmath3 ocm using gem . the model space covers to describe @xmath21li+@xmath9 and @xmath70be+@xmath23 cluster structures and @xmath3 gas - like structures including the shell - model - like structures . full levels up to @xmath4 and @xmath314 around the @xmath3 threshold are reproduced . the @xmath10 ( g.s ) and @xmath11 states are found to have the shell - model - like compact structures , the results of which are consistent with the previous cluster model analyses by nishioka et al . the @xmath7 state is characterized by the monopole transition strength as large as the @xmath26 at @xmath343 mev in @xmath13c . the present study succeeded in reproducing its excitation energy and monopole transition strength . we found that this state has an @xmath3 cluster structure with the nuclear radius of @xmath344 fm . the study of the single - cluster properties such as single - cluster orbits and occupation probabilities for @xmath45 showed that there is no concentration of the single-@xmath9 occupation probability on a single orbit and the radial part of the single-@xmath9 orbits has nodal behaviors in the inner region , illustrating rather strong pauli - blocking effect . these results are in contrast with those of the hoyle state with the dilute @xmath14-condensate - like character @xmath345 , in which the @xmath9 particle occupies a single @xmath29-orbit ( zero node ) with about @xmath16% probability . consequently the @xmath7 state could not be identified as the analogue of the hoyle state , possessing a dominant gas - like configuration of @xmath24 . the reason why the @xmath7 state is not of the hoyle - analogue can be understood from the following facts : the @xmath4 state is bound by @xmath283 mev with respect to the @xmath3 threshold , while the hoyle state is located by @xmath28 mev above the @xmath14 threshold and has a dilute @xmath14 structure . the extra binding energy of the @xmath7 state with respect to the @xmath3 threshold is likely to suppress strongly the growth of the gas - like @xmath3 structure in this state . as for the @xmath57 states , the @xmath122 state appears as a bound state at @xmath346 mev around @xmath21li+@xmath9 threshold . this low excitation energy indicates that @xmath9-type correlation should play an important role in the state . in fact , we found that the @xmath122 state with @xmath347 fm has the @xmath21li(g.s)+@xmath9 structure with @xmath266-wave relative motion , although the @xmath21li(@xmath120 ) part is rather distorted in comparison with the ground state of @xmath21li . since the @xmath4 state has the largest @xmath275 factor for the @xmath8li(g.s)+@xmath9 channel with @xmath107-wave relative motion compared with those of other @xmath8li+@xmath9 and @xmath110be+@xmath23 channels , the @xmath285 and @xmath4 states of @xmath2b can be interpreted as the parity - doublet partners each other . they are similar to the typical example of the parity doublet , @xmath348 and @xmath349 states in @xmath309ne , with the @xmath31o+@xmath9 cluster structure @xcite . in addition to @xmath285 , we found that the @xmath314 state appears as a resonant state at @xmath350 mev ( @xmath317 kev ) around the @xmath3 threshold with csm . the large radius ( @xmath351 fm ) indicates that the state has a dilute cluster structure . the analysis of the single - cluster properties showed that this state has a main configuration of @xmath352 with about @xmath25% probability . thus , we could call the @xmath314 state the hoyle - analogue with @xmath352 , which is similar to the hoyle state possessing the main configuration @xmath15 with about @xmath16% probability . it should be reminded that @xmath314 is located by @xmath335 mev above the @xmath3 threshold , while @xmath17 is bound by @xmath353 mev with respect to the three cluster threshold . the latter binding energy leads to the suppression of the development of the gas - like @xmath3 structure in @xmath17 , whereas the gas - like structure with a large nuclear radius grows up in @xmath314 because the state appears above the three - body threshold . recently , the @xmath354 state at @xmath72 mev with @xmath341 kev ( located at @xmath342 mev above the @xmath3 threshold ) was observed through the @xmath9+@xmath21li decay channel @xcite . the experimental energy and width of the @xmath19-mev state are in good correspondence to the present calculated results of the @xmath314 state . the hoyle - analogue state in @xmath2b , thus , could be assigned as the @xmath19-mev state . the present authors thank to h. horiuchi , k. ikeda , g. rpke , p. schuck , and a. tohaski for valuable discussions and encouragements . this work was partially supported by jsps ( japan society for the promotion of science ) grant - in - aid for scientific research ( c ) ( 21540283 ) and young scientists ( b ) ( 21740209 ) . t. kawabata , h. akimune , h. fujita , y. fujita , m. fujiwara , k. hara , k. hatanaka , m. itoh , y. kanada - enyo , s. kishi , k. nakanishi , h. sakaguchi , y. shimbara , a. tamii , s. terashima , m. uchida , t. wakasa , y. yasuda , h. p. yoshida , and m. yosoi , phys . b * 646 * , 6 ( 2007 ) . r. j. charity , s. a. komarov , l. g. sobotka , j. clifford , d. bazin , a. gade , jenny lee , s. m. lukyanov , w. g. lynch , m. mocko , s. p. lobastov , a. m. rogers , a. sanetullaev , m. b. tsang , m. s. wallace , r. g. t. zegers , s. hudan , c. metelko , m. a. famiano , a. h. wuosmaa , and m. j. van goethem , phys . c * 78 * , 054307 ( 2008 ) .

we present a computer assisted method for generating existence proofs and a posteriori error bounds for solutions to two point boundary value problems ( bvps ) . all truncation errors are accounted for and , if combined with interval arithmetic to bound the rounding errors , the computer generated results are mathematically rigorous . the method is formulated for-dimensional systems and does not require any special form for the vector field of the differential equation . it utilizes a numerically generated approximation to the bvp fundamental solution and green s function and thus can be applied to stable bvps whose initial value problem is unstable . the utility of the method is demonstrated on a pair of singularly perturbed model bvps and by using it to rigorously show the existence of a periodic orbit in the lorenz system . two point boundary value problems , computer assisted proofs , periodic orbits , a posteriori error analysis , singular perturbations + _ 2010 mathematics subject classification . _ primary : 37c27 , 65g20 , 34b15 ; secondary : 34b27 , 65l10 , 65l11 .

1410.0785

we propose a new computer assisted method for rigorously proving invertibility and bounding the norm of the inverse of operators of the form and the precise functions spaces in which we work will be specified later . ] @xmath1(t)=\left(v(t)-v(0)-\int_0^t a(s)v(s)ds , b_0v(0)+b_1v(1)\right),\hspace{2 mm } a(t),b_i\in\mathbb{r}^{n\times n}.\ ] ] operators of this form correspond to linear two point boundary value problems ( bvps ) and arise naturally as the frchet derivative of operators defining nonlinear bvps . bounds on the norm of the inverse of operators of this type are required in a posteriori proofs of existence and local uniqueness of nonlinear equations via the newton - kantorovich theorem @xcite . computer assisted methods have been successfully applied to many problems in differential equations and dynamical systems , including chaos in the rssler @xcite and lorenz @xcite systems , the existence of the lorenz attractor @xcite , flow ( in)stability @xcite , existence of heteroclinic @xcite , homoclinic @xcite , and periodic @xcite orbits , solution of the feigenbaum equation @xcite , and the ( in)stability of matter @xcite . generalizations to elliptic and parabolic pdes are discussed in , for example , @xcite . as it pertains to two point boundary value problems , computer assisted proofs are typically based on a fixed point theorem @xcite . in particular , various forms of the newton - kantorovich theorem were used in @xcite . a method based on differential inequalities as presented in @xcite , a method using radii polynomials is given in @xcite , and @xcite utilized chebyshev series . the methods in @xcite focused on second order equations ; @xcite worked with an equivalent integral formulation while @xcite used a finite element based method . @xcite treated an integral boundary condition . the results in @xcite utilized the theory of initial value problem ( ivp ) fundamental solutions to treat @xmath0-dimensional two point bvps , the former focusing on the periodic case . our method most closely resembles that of ref . @xcite and applies to @xmath0-dimensional systems without any special assumptions on the form of the vector field . the present method differs from @xcite in the use of the bvp fundamental solution and green s function as the primary tool , as opposed to the ivp fundamental solution . see also @xcite , which used a green s function method to solve a singular problem on an unbounded domain . a general outline of our method is as follows . 1 . generate ( non - rigorous ) approximations , @xmath2 , @xmath3 to the bvp fundamental solution and green s function , respectively , on a mesh . 2 . extend these to piecewise polynomial approximations @xmath4 and @xmath5 on the whole domain . use @xmath6 and @xmath7 to define an approximate inverse , @xmath8 , to the bvp operator @xmath9 . 4 . using estimates involving @xmath8 , prove existence of an exact inverse to @xmath9 and derive a machine computable formula that bounds the norm of @xmath10 . the use of the bvp fundamental solution makes the method presented here applicable in a range of situations for which ivp based methods are ill suited , namely when the ivp fundamental solution or its inverse has a ` large ' norm but the corresponding bvp fundamental solution and green s function has a ` moderately sized ' norm . this is often the case in singularly perturbed problems and so the effectiveness of the method is demonstrated on two such examples in section [ sec : num_tests ] . we will refer to these imprecisely defined ` large ' and ` moderate ' regimes as unstable and stable respectively , similar to the discussion in @xcite . to obtain completely rigorous results , some means of bounding the floating point rounding error is needed , typically based on the theory of interval arithmetic . there are many references from which to learn more about validated numerics and interval arithmetic , for example @xcite . results in this paper are presented so that if interval arithmetic is used for all arithmetic operations and interval valued extensions are available for the vector field of the differential equation and its derivatives then the resulting computer generated existence proofs and error bounds are mathematically rigorous ( we use the matlab package intlab @xcite to obtain rigorous bounds ) . however , in light of the additional computational cost of interval arithmetic as compared to floating point , if complete rigor is not needed then the method can just as easily be used with traditional floating point operations to obtain approximate error bounds . in section [ sec : newton ] we fix some notation and discuss a general strategy for a posteriori existence proofs via the newton - kantorovich theorem . for comparison , we outline the method of @xcite in section [ sec : ivp_method ] . development of our method begins in section [ sec : gen ] and continues in subsequent sections . background on the green s function of a bvp is given in section [ sec : greens ] . the method is tested on several sample problems in section [ sec : num_tests ] . consider a general non - linear two point bvp on @xmath11 $ ] , @xmath12 where @xmath13 is continuous , differentiable in @xmath14 , @xmath15 is continuous on @xmath16 , and @xmath17 is @xmath18 . in particular , local existence and uniqueness of the ivp is guaranteed . in order to apply functional analytic methods to this problem it is convenient to follow @xcite and define the banach spaces @xmath19,\mathbb{r}^n ) , \hspace{2 mm } c_0([0,1],\mathbb{r}^n)\equiv \{y\in x_1:y(0)=0\},\hspace{2 mm } x_2=c_0([0,1],\mathbb{r}^n)\times\mathbb{r}^n\ ] ] where the @xmath20-norm is used on @xmath21 , denoted by @xmath22 , the sup - norm @xmath23}|y(t)|$ ] is used on the function spaces , and the norm @xmath24 is used on @xmath25 . in this setting , the bvp ( [ nonlin_bvp ] ) can be put in a form more amenable to analysis as follows . define the map @xmath26 @xmath27(t)=\left(y(t)-y(0)-\int_0^tf(s , y(s))ds , g(y(0),y(1))\right).\ ] ] solutions of the bvp are in one to one correspondence with zeros of @xmath28 . its first component , valued in @xmath29,\mathbb{r}^n)$ ] , is denoted by @xmath30 and its second component , valued in @xmath21 , by @xmath31 . as shown in @xcite , @xmath28 is frchet-@xmath18 with derivative at @xmath32 given by @xmath33(t)=&v(t)-v(0)-\int_0^td_y f(s , y_0(s))v(s)ds,\\ dg^2(y_0)[v](t)=&d_{y_1}g(y_0(0),y_0(1))v(0)+d_{y_2}g(y_0(0),y_0(1))v(1).\notag\end{aligned}\ ] ] having reformulated the problem in terms of finding zeros of a @xmath18 map between banach spaces , a natural tool is the newton - kantorovich theorem ; see appendix [ newton_th ] for a precise statement . invertibility of @xmath34 along with a bound on the norm of its inverse are required to apply the theorem and these are typically the most difficult ingredients to obtain in practice . once this has been done the remainder of the estimates are relatively straightforward and have been discussed in , for example , @xcite . for this reason we will predominately focus on the invertiblility of @xmath34 i.e. on a posteriori methods for proving the existence and uniqueness to solutions of linear bvps . motivated by the previous section , consider a linear bvp on @xmath11 $ ] as given by the bounded linear operator @xmath35 defined by the formula ( [ lin_bvp ] ) , where @xmath36 is a continuous @xmath37 valued map . in theorem 1 of @xcite , the author shows how the solvability of @xmath38=(r , w)$ ] is related to the matrix of fundamental solutions , defined as the solution to the initial value problem ( ivp ) @xmath39 in our notation , this result is the following . [ thm : ivp ] let @xmath40 and @xmath41 be continuous matrix and vector valued respectively with @xmath42 and consider the two point bvp @xmath43 let @xmath44 be the corresponding matrix of fundamental solutions and define @xmath45 . then the bvp has a unique solution iff @xmath46 is nonsingular and the solution is given by @xmath47+r(t).\notag\end{aligned}\ ] ] it was also shown in @xcite how to apply this theorem to rigorously prove solvability of the bvp and derive guaranteed error bounds using numerically constructed approximations to @xmath44 and @xmath48 . because of the use of @xmath49 , this will be called the ivp fundamental solution method or simply the ivp method . a limitation of this method is that the stability properties of an ivp take center stage ; the quantities @xmath50 and @xmath51 feature critically in the error estimates derived in @xcite . unfortunately there are situations in which a bvp of interest is stable whereas the corresponding ivp is not . in such situations the bounds obtained by the ivp fundamental solution method will be very pessimistic due to the large size of @xmath50 and/or @xmath51 . as discussed in @xcite , perhaps the simplest test problem that exhibits these features is the bvp @xmath52 the corresponding ivp fundamental solution matrix and its inverse are @xmath53 these have exponentially increasing norms as a function of @xmath54 . this is in spite of the fact that the value and derivative of the exact solution @xmath55 have @xmath56 norm . even if one tries to avoid separately bounding the norms of @xmath49 and @xmath57 and take advantage of the cancellation that can occur in @xmath58 and @xmath59 , the above example is sufficient to show that this cancellation does not in general occur over the entire @xmath60 domain ; the @xmath61 component of the former quantity is given by @xmath62 which diverges at @xmath63 , @xmath64 as @xmath65 . even if cancellation does occur , it is difficult to compute @xmath44 and @xmath66 to high enough accuracy that one could take advantage of such cancellation . in summary , while the ivp fundamental solution method suffices for problems whose ivp is stable , if one is interested in bvps whose corresponding ivp is unstable then an alternative method is needed . our contribution is designed to address such situations by utilizing the bvp fundamental solution and green s function in place of the ivp fundamental solution . we are now in position to begin discussing an alternative method for producing a posteriori existence proofs and error bounds based on the bvp fundamental solution . this method will be called the bvp fundamental solution method , or bvp method for short , in contrast to the ivp method . the bvp fundamental solution is introduced in the following section , but first we discuss the function spaces that will be used in our analysis . the input to any a posteriori method consists of a certain ( numerically generated ) approximate solution . for us this will consist of function values on a mesh @xmath67 . before error bounds can be produced , these need to be extended to an approximate solution on the entire interval @xmath11 $ ] . while one could work in spaces of continuous functions by interpolating the values at the nodes , as done in the ivp method , we will find things more convenient ( and the resulting error bounds more transparent ) when this assumption is removed , both from the solution space and the matrix @xmath36 . in this section we outline the slightly generalized framework in which the equations will be formulated . to this end , fix a mesh , @xmath68 , as above and let @xmath69 be the space of functions that are piecewise continuous on this mesh . more precisely , @xmath69 is the direct sum @xmath70,\mathbb{r}^n).\ ] ] on each factor the norm will be taken to be a weighted @xmath20-norm @xmath71}|v_j(t)|_w,\hspace{2 mm } |v_j|_w\equiv |w v_j|\ ] ] where the weight , @xmath72 , is some invertible matrix . in practice , we will take @xmath72 to be diagonal and choose the diagonal elements to balance the magnitude of the errors in the different components of the solution . one could consider the generalization where the weight is allowed to vary between subintervals but this will not be explored here . @xmath69 is a banach space and the maps that send any @xmath73 to its value at any @xmath74 or at @xmath75 ( @xmath76 denots limit from above , @xmath77 limit from below ) are well defined bounded linear maps . elements of @xmath69 can be identified with elements of @xmath78 and integration or differentiation ( of piecewise @xmath18 ) elements of @xmath69 will be defined using this identification . mirroring the continuous case , ( [ x_def ] ) , define a second space @xmath79 this is a banach space under the norm @xmath80 where @xmath81 are any invertible matrices . finally , note that the @xmath82 s are closed subspaces of the @xmath83 s . for simplicity we will take @xmath84 . this will allow us to drop the subscript @xmath72 s from the above norms in the following sections . having specified the functional analytic framework , the definition of nonlinear bvps can be extended to this setting by defining @xmath85 by the same formula as in ( [ g_def ] ) . note that any zero of this map is automatically continuous , and so solutions of the generalized problem are still in 1 - 1 correspondence with solutions of the original bvp . the extended operator @xmath28 is still frchet-@xmath18 , with the same formula for the derivative ( [ df_formula ] ) . note , however , that @xmath86 is no longer continuous as a function of @xmath87 for a general @xmath88 . therefore , in the interest of applying newton - kantorovich to prove existence , we are lead to consider bounded linear maps @xmath89 defined as in ( [ lin_bvp ] ) , where the assumptions on @xmath40 are relaxed to allow for piecewise continuity on the given mesh . we are now in position to discuss the bvp fundamental solution , on which our method will be based . a bvp fundamental solution is defined as a continuous and piecewise @xmath18 solution to @xmath90 consisting of invertible matrices . recall that @xmath36 is only assumed to be piecewise continuous . analogously to the ivp fundamental solution , if a bvp fundamental solution exists then it provides an inverse to the bvp operator @xmath9 through the green s function , as defined in the following theorem . [ lin_bvp_solution ] if a bvp fundamental solution @xmath91 exists then it is unique and the unique solution to @xmath92 can be written @xmath93 where the green s function is defined as @xmath94 therefore @xmath91 provides us with an inverse of the operator @xmath9 from ( [ lin_bvp ] ) given by @xmath95 note that the green s function ( [ greens_def ] ) is not to be confused with the notation for the operator associated with a nonlinear bvp ( [ g_def ] ) . the meaning will be clear from the context . theorem [ lin_bvp_solution ] is essentially found in @xcite . the only complications beyond the presentation there is that @xmath36 is only piecewise continuous and @xmath41 must be taken to be an arbitrary continuous function that vanishes at @xmath63 , not necessarily one of the form @xmath96 for some continuous @xmath97 , as would be the case when transforming from the differential to the integral form of the bvp . one can directly verify the formula by composing ( [ f_inv_formula ] ) with ( [ lin_bvp ] ) and vice versa . we now suppose that we are given piecewise @xmath18 maps @xmath98\rightarrow\mathbb{r}^{n\times n},\hspace{2mm}\tilde g:[0,1]\times[0,1]\rightarrow\mathbb{r}^{n\times n},\ ] ] which are thought of an approximate fundamental bvp solution and approximate green s function respectively ( but without any a priori knowledge that an exact @xmath91 actually exists ) which were obtained by non - rigorous numerical means . more specifically , there are a set of nodes @xmath67 such that each @xmath99}$ ] is @xmath18 and each @xmath100\times[t_{j-1},t_j]}$ ] is @xmath18 , except when @xmath101 , in which case @xmath7 also has a discontinuity along the diagonal . we will construct @xmath7 using @xmath102 and an approximation to its inverse by replacing the corresponding exact quantities in the definition of the green s function ( [ greens_def ] ) , but this is not strictly necessary . our strategy for proving the existence of @xmath103 is to mimic the formula ( [ f_inv_formula ] ) for @xmath104 in terms of the fundamental solution but using the approximate fundamental solution instead . more precisely , replacing @xmath91 with @xmath102 and @xmath28 with @xmath7 in the formula for @xmath10 yields a bounded linear map @xmath105 . @xmath8 will be used to find conditions under which we can prove the invertiblility of @xmath9 . this will rely on the following well known result concerning perturbations of the identity , see for example @xcite . [ i_perturb_lemma ] let @xmath106 be a bounded linear map on a banach space . if @xmath107 then @xmath108 has a bounded inverse that satisfies @xmath109 in fact , a slight generalization of this result will be needed , given below . [ inverse_lemma ] let @xmath110 , @xmath111 be bounded linear maps where @xmath112 are banach spaces . if @xmath113 for @xmath114 where @xmath115 is the identity on @xmath49 then @xmath108 is surjective . if @xmath108 is also injective then @xmath108 , @xmath116 are invertible , @xmath117 by the above lemma , @xmath118 has a bounded inverse with @xmath119 in particular @xmath108 is surjective . it is injective by assumption so by the open mapping theorem it has a bounded inverse . this implies @xmath116 has a bounded inverse as well and @xmath120 @xmath121 in finite dimensions and when @xmath122 and @xmath49 have the same dimension then the hypothesis of injectivity of @xmath108 in lemma [ inverse_lemma ] is not needed but in infinite dimensions it is required in general . however , when we have the appropriate element of compactness , the situation in infinite dimensions mirrors the finite dimensional case . more specifically , recall one consequence of the fredholm alternative , see for example @xcite . let @xmath122 be a banach space and @xmath123 be a compact operator . then @xmath124 is injective iff it is surjective . if a fredholm alternative - like condition holds for @xmath9 then surjectivity will imply injectivity , and hence @xmath125 will imply the invertiblity of @xmath9 . in addition , this will provide a bound on @xmath10 per lemma [ inverse_lemma ] . these are the two ingredients that are needed in order to utilize the newton - kantorovich theorem . with slight modifications of the above we could just as well work with @xmath126 but find this less appealing for a reason we point out in appendix [ app : i - hf ] . in order to more easily apply the fredholm alternative , we reformulate the bvp operator @xmath9 in terms of a map from a single banach space to itself . define @xmath127 by @xmath128(t)=&\left(v(t)-\int_0^t a(s)(w+v(s))ds , b_0w+b_1(w+v(1))\right)\\ = & ( v(t),w)-\left(\int_0^t a(s)(w+v(s))ds,(i - b_0-b_1)w - b_1v(1)\right)\\ \equiv & ( i - k)[v , w](t).\end{aligned}\ ] ] we have @xmath38=\tilde f[v - v(0),v(0)]$ ] and so @xmath9 is surjective or injective if and only if @xmath129 is . the second component of @xmath130 has finite rank and the first is an integral operator with bounded kernel over a bounded domain and whose image consists of continuous functions , hence the compactness of @xmath130 follows from an application of the arzela - ascoli theorem . this implies that the fredholm alternative applies to @xmath129 and hence , showing that @xmath131 is sufficient to prove that @xmath9 has a bounded inverse with @xmath132 we will now discuss how to bound @xmath133 and @xmath134 . in appendix [ app : i - fh ] a formula for @xmath135 is derived . for each @xmath136 let @xmath137 . the two components of @xmath135 are given by @xmath138_1(t)\\ = & -\int_0^t\tilde \phi^\prime(s)-a(s)\tilde\phi(s)ds(w - b_1r(1))+\int_0^t(\tilde g(z^-,z)-\tilde g(z^+,z)+i)a(s)r(s)ds\notag\\ & + \sum_{j=1}^{j_t}(\tilde\phi(t_j^-)-\tilde\phi(t_{j}^+))(w - b_1r(1))+\int_0 ^ 1\left[\sum_{j=1}^{j_t}(\tilde g(t_j^-,z)-\tilde g(t_{j}^+,z))\right]a(z)r(z)dz\notag\notag\\ & -\int_0^t\int_0 ^ 1(\partial_s \tilde g(s , z)-a(s ) \tilde g(s , z))a(z)r(z)dzds,\notag\\ & ( i - fh)[r , w]_2\notag\\ = & [ i - b_0\tilde\phi(0)-b_1\tilde\phi(1)](w - b_1r(1))-\int_0 ^ 1 \left(b_0\tilde g(0,s)+b_1\tilde g(1,s)\right)a(s)r(s)ds\notag\end{aligned}\ ] ] where superscripts @xmath76 and @xmath77 denote the limits from above and below respectively . for completeness , the analogous expression for @xmath126 is given in appendix [ app : i - hf ] . the most straightforward way to use ( [ i - fh_components ] ) to bound the norm of @xmath135 is to define @xmath7 using ( [ greens_def ] ) , replacing the exact quantities with @xmath102 and its approximate inverse , and then use the submultiplicative property of induced matrix norms and bound each of the quantities in the green s function individually . this leads to an estimate whose operation count scales linearly in the number of subintervals , @xmath139 . if @xmath140 and @xmath141 are not too large over @xmath11 $ ] then this is a good strategy . however , this is often not the case , even in situations where the bvp is stable . we illustrate this phenomenon with the same test problem introduced in ( [ test_prob ] ) . consider again the simple test problem , ( [ test_prob ] ) . the fundamental solution and its inverse are @xmath142 while @xmath143 is bounded , @xmath144 diverges as @xmath65 . the general notion of bvp stability only requires the norm of the fundamental solution and of the green s function to be small , but not the inverse of the fundamental solution . see for example @xcite for details . therefore , in general we must abandon any estimate based on bounding @xmath145 and @xmath146 separately and use one based on bounding the entire green s function . the test problem ( [ test_prob ] ) does have a bounded green s function @xmath147 we emphasize that @xmath148 does not arise from a large cancellation of the computed results . rather , at each endpoint the boundary conditions annihilate the growing modes of @xmath149 . this suggests that we should not run into the same degree of numerical cancellation issues using the bvp method as opposed to the ivp method . in this section we show how to compute a sharper bound on @xmath133 by bounding the green s function as a whole . bounding the second component of @xmath135 is straightforward so we focus on the first . to this point the only assumption made about @xmath102 and @xmath7 is that they are piecewise @xmath18 . in practice , a non - rigorous numerical algorithm will provide us with approximate solution values @xmath150 and @xmath3 at the centers of the intervals @xmath151 $ ] and rectangles @xmath151\times [ t_{j-1},t_j]$ ] . for example , in our computations we obtain the @xmath150 by numerically solving the bvp and , for @xmath152 , define @xmath153 where @xmath154 is a numerical approximation to @xmath155 . on the diagonal @xmath101 two different versions of @xmath156 are needed . these are denoted by @xmath157 corresponding to the lower and upper halves of the triangle respectively . given the values at the nodes , @xmath100\times [ t_{j-1},t_j]}$ ] and @xmath158}$ ] are constructed to be the polynomials on each subinterval or rectangle that satisfy the equations @xmath159 and @xmath160 up to some order @xmath161 . formulas for the coefficients and the ode residual are given in appendix [ app : taylor ] . again , on the rectangle with @xmath101 different versions of @xmath7 are needed , one corresponding to the upper half of the triangle and one to the lower , constructed from @xmath162 and @xmath163 respectively . the resulting @xmath102 and @xmath7 will be piecewise smooth but not continuous . this is the reason we generalized the solution space in section [ sec : gen ] to allow for discontinuities . we emphasize that in the absence of jump discontinuities at the mesh points , this method reduces to a piecewise taylor model ivp method for @xmath91 , see for example @xcite . while well suited for initial value problems , such methods perform extremely poorly for bvps with an unstable ivp such as the examples in section [ sec : num_tests ] . therefore , having accurate @xmath164 s as inputs is crucial to the success of the method ; the taylor series simply ensure sufficiently small error over the interior of each subinterval of the mesh . using appendix [ app : taylor ] we can write @xmath165 on each @xmath166 $ ] , and similarly for @xmath7 , and break the integrals in ( [ i - fh_components ] ) into sums of integrals over the subintervals of the mesh to obtain @xmath167\|a\|_j\|r_j\|_j\notag\\ & + \sum_{j=1}^nh_j\|(i+e_j(z))\left(\tilde g^+_{j , j}-\tilde g^-_{j , j}\right)(i+f_j(z))+i\|_j \|a\|_j\notag\\ & + \sum_{j=1}^{n-1}|(i+e_j(t_j))\psi_j-(i+e_{j+1}(t_j))\psi_{j+1}|(1+|b_1|)\notag\\ & + \sum_{j=1}^n2\frac{(h_j/2)^{m+1}}{m+1}(1+|b_1|)\|r_j\psi_j\|_j,\notag\end{aligned}\ ] ] @xmath168 where the subscript @xmath169 on @xmath170 indicates that the supremum is taken only over @xmath171 $ ] . @xmath172 and @xmath173 are defined in ( [ ej_def ] ) and ( [ fj_def ] ) respectively . in bounding @xmath174 on @xmath171 $ ] a means for analytically bounding the error term in the taylor series expansion of @xmath36 is needed . this can be achieved from an interval matrix valued extension for @xmath36 and its derivatives . to bound the polynomial terms in the @xmath174 , @xmath172 , and @xmath173 s we use the simple method of summing the bounds on the norms of each monomial . finally , to use ( [ f_inv_bound ] ) to bound the norm of @xmath10 a machine computable bound on @xmath134 is needed . @xmath175.\notag\end{aligned}\ ] ] the leading order term in @xmath139 of the complexity of computing these quantities by the above formulae is @xmath176 where @xmath0 is the dimension of the ode system . in particular , it does not depend on @xmath177 , and so from the point of view of complexity , the degree of @xmath102 can be set as high as is necessary for the taylor series errors over the subintervals to be negligible , making the boundary conditions and jump terms the main sources of error . parallelizing the computation of these bounds is straightforward . + these results are summarized in the following theorem . [ main_theorem ] let @xmath69 and @xmath178 be defined by ( [ y1_def ] ) and ( [ y2_def ] ) respectively and consider the two point bvp @xmath38=(r , w)$ ] defined by @xmath89 , @xmath179(t)=\left(v(t)-v(0)-\int_0^t a(s)v(s)ds , b_0v(0)+b_1v(1)\right)\ ] ] where @xmath180\rightarrow \mathbb{r}^{n\times n}$ ] is piecewise @xmath181 on the mesh @xmath182 for some @xmath183 and @xmath184 . let @xmath102 and @xmath7 be constructed as discussed in section [ sec : sharper_bound ] . define @xmath105 by @xmath185(t)&=\tilde\psi(t)(w - b_1r(1))+r(t)+\int_0 ^ 1 \tilde g(t , s)a(s)r(s)ds.\end{aligned}\ ] ] if @xmath186 , as computed by ( [ i - fh_bound2_1 ] ) and ( [ i - fh_bound2_2 ] ) , then @xmath103 exists , is bounded , and @xmath187 for the linear test problems presented in the following section , it will be useful to replace the newton - kantorovich theorem with a simpler and more explicit a posteriori error bound . suppose we have proven @xmath186 so that @xmath9 is invertible . let @xmath188 be the exact solution of @xmath38=(r , w)$ ] and suppose we are given an approximate solution @xmath189 . then by ( [ f_inv_bound ] ) , @xmath190\|\leq \|f^{-1}\|\|f[\tilde v]-(r , w)\|\leq\frac{\|h\|}{1-\alpha}\|f[\tilde v]-(r , w)\|.\ ] ] suppose @xmath189 is piecewise continuous on the mesh @xmath67 and that @xmath36 and @xmath41 are piecewise @xmath181 , hence on each subinterval @xmath171 $ ] we have taylor series with remainders @xmath191 the components of the residual in ( [ inhomo_err ] ) can then be bounded . @xmath192-(r , w))_1\|\leq\notag\\ & \sup_{t\in[t_{j-1},t_j]}|\tilde v(t)-\tilde v(0)-\sum_{k=1}^{j-1}\int_{t_{k-1}}^{t_k } a_k(s)\tilde v(s)ds-\int_{t_{j-1}}^{t } a_j(s)\tilde v(s)ds - r_j(t)|\label{int_line}\\ & + \sum_{k=1}^j \frac{h_k^{m+1}}{m+1}\|r^a_k\|_k\|\tilde v\|_k+h_j^m\|r_j^r\|_j,\notag\\ & |(f[\tilde v]-(r , w))_2|\leq |b_0\tilde v(0)+b_1\tilde v(1)-w|\notag\end{aligned}\ ] ] where , as above , @xmath170 denotes the sup norm over @xmath171 $ ] . similar to the approximate fundamental solution , given values @xmath193 at the nodes , @xmath194 is constructed on @xmath171 $ ] to be the polynomial @xmath195 this causes the degree @xmath196 through @xmath177 terms in ( [ int_line ] ) to vanish . we do nt have the freedom to choose the zero order term ; it is the analog of the jump terms from ( [ i - fh_components ] ) . analytically evaluating the integrals in ( [ int_line ] ) , the bound on the first component of the residual becomes @xmath197-(r , w))_1\|\\ & \leq\max_j \bigg[|\tilde v_{j-1}-\tilde v_0-\sum_{k=1}^{j-1}\int_{t_{k-1}}^{t_k}a_k(s)\tilde v(s)ds - r(t_{j-1})|\\ & + \sum_{l = m}^{2m}\frac{h_j^{l+1}}{l+1}\sum_{i=\max\{0,l - m\}}^{\min\{m , l\}}|a_ic_{l - i}|+\sum_{k=1}^j \frac{h_k^{m+1}}{m+1}\|r^a_k\|_k\|\tilde v\|_k+h_j^m\|r_j^r\|_j\bigg],\\ & \text{where } \int_{t_{k-1}}^{t_k}a_k(s)\tilde v(s)ds=\sum_{l=0}^{2m}\frac{h_k^{l+1}}{l+1}\sum_{i=\max\{0,l - m\}}^{\min\{m , l\}}\!\!\!\!\!a_ic_{l - i}.\end{aligned}\ ] ] the bound is now in a form that can easily programmed for automated rigorous evaluation using interval arithmetic or estimation in traditional floating point . we also note that the cost is just @xmath198 as opposed to the @xmath199 cost of bounding @xmath200 . in ( [ i - fh_bound2_1 ] ) and ( [ h_bound ] ) bounds were given on @xmath133 and @xmath134 that involve finitely representable objects and finite operations with which we can perform validated computations , namely the piecewise polynomials @xmath4 and @xmath5 and the interval valued extension of @xmath36 and its derivatives . for rigorous bounds , we implemented this using the matlab interval analysis package intlab @xcite by creating simple routines that perform algebraic operations with polynomials whose coefficients are interval matrices . with such functionality , the code that produces the desired bound on @xmath133 is a simple translation of ( [ i - fh_bound2_1 ] ) and ( [ i - fh_bound2_2 ] ) . in this section , the applicability and accuracy of the bvp method is assessed using a pair of singularly perturbed test problems , taken from @xcite , as well as a nonlinear example , the existence of a periodic orbit in the lorenz system . to obtain usable bounds , even the exponentially small parts of the approximate fundamental solution need to be resolved with the same relative accuracy as the larger portions , otherwise the loss of accuracy will prevent the growing modes from being annihilated when forming the approximate green s function . for problems with quickly decaying modes , the absolute tolerance of the non - rigorous solver must be set near the underflow limit in order to achieve this . this is done in the first two test problems . tests were performed using a uniform mesh . there is likely room for improvement by intelligently selecting the mesh but we did not employ such methods here . in table [ table : tests ] the mesh size was chosen by adding points until the error approximately stabilized . our first test problem is a singularly perturbed airy equation @xmath201 which is a model of the quantum mechanical wave function near a turning point in the potential , here at @xmath202 . the exact solution is a linear combination of airy functions and is shown in figure [ fig : airy_sol ] , where @xmath203 . the solution exhibits dense oscillations for @xmath204 and a boundary layer near @xmath205 . we used the bvp method in the cases @xmath206 . we used @xmath207 with the aim of making the taylor series and approximate inverse errors negligible , allowing us to focus on how the error in the initial approximation supplied to the bvp method translates into the bound produced by the method . if we let @xmath208 be the @xmath209th component of the @xmath169th jump error from ( [ int_line ] ) then the weight used in this test has diagonal elements given by @xmath210 this choice prevents overestimation of the error in certain components when they have widely different scales and makes a large difference in the quality of the bounds , up to several orders of magnitude ; it was often the difference between the existence test @xmath125 passing or failing . since the jump errors can be computed in @xmath198 operations prior to the main @xmath199 computation , this is an effective and practical way to improve the estimates . the test results are shown in table [ table : tests ] . the fourth column gives the error in solution values on the mesh that are provided to the bvp method , the fifth column is the bound on @xmath211 computed by the method , and the final column is the computed bound on the absolute value of the error in @xmath188 . the a posteriori bvp method does overestimate the error by several orders of magnitude , but the error bounds are still practical . the ivp fundamental solution method would not be usable in this situation , as the fundamental solution matrix has a norm @xmath212 for @xmath213 . the use of an adaptive weighted norm is crucial for the success of the method and greatly improves the error bounds . without the weight , the existence test @xmath125 fails even at @xmath214 . the bvp method fails on this problem for @xmath215 , as @xmath91 becomes so ill conditioned that its decaying components can not be resolved to sufficient relative accuracy due to underflow . this point will reappear in the subsequent example as well ; the exponentially decaying modes of @xmath91 ( which tend to be the least ` interesting ' features ) cause the most trouble for the bvp method . in other words , while a large norm for the inverse fundamental solution does not impact bvp stability , over / underflow issues mean that it is still the main limiting factor of the bvp method in its current form . however , the problem is much less severe than ill conditioning in the ivp method . . for @xmath204 the system is oscillatory ( left ) and it has a boundary layer near @xmath205 ( right).,height=226 ] |c |c |c |c|c|c| ' '' '' example & @xmath216 & n & input error & @xmath133 & solution error bound + ' '' '' turning point & @xmath217 & @xmath218 & @xmath219 & @xmath220&@xmath221 + ' '' '' & @xmath222 & @xmath223 & @xmath224 & @xmath225 & @xmath226 + ' '' '' & @xmath227 & @xmath228 & @xmath229 & @xmath230&@xmath231 + ' '' '' potential barrier & @xmath222 & @xmath232 & @xmath233 & @xmath234 & @xmath235 + ' '' '' & @xmath227 & @xmath228 & @xmath236 & @xmath237 & @xmath238 + as a second test , consider a quantum mechanical potential well problem @xmath239 where we take @xmath240 . the solution is oscillatory on @xmath241\cup[3/4,1]$ ] and exponentially decaying as one moves towards the center of @xmath242 $ ] as seen in figure [ fig : potential ] ( right pane ) for @xmath203 . we used @xmath207 and the same formula for @xmath72 as in the prior example . again , the method fails once the decaying modes start to underflow , which occurs for @xmath215 . ( right).,height=226 ] for this example we do nt have an analytical formula for the exact solution , so the input error reported for in table [ table : tests ] for this problem is obtained by comparing approximate solution values after refining the mesh . in the first two examples , the supremum norm of the exact solution was relatively consistent between the different parameter values , but that is not true for the current example , where @xmath243 for @xmath214 and is single digits @xmath203 . this explains why the absolute error is so similar for the cases tested despite the difference in @xmath216 . as our final example , we study a non - linear problem , the existence of periodic solutions to the lorenz system . scaling the independent variable by the ( unknown ) period @xmath244 , the bvp is @xmath245 @xmath246 in order to fix the initial point on the periodic orbit an additional boundary condition is needed , which we take to be @xmath247 let @xmath248 denote the vector field and boundary conditions respectively and @xmath28 be the corresponding bvp operator , defined as in ( [ g_def ] ) . in @xcite , computer aided proof techniques were used to show existence of periodic orbits to the lorenz system . in particular , in @xcite existence was shown for parameter values @xmath249 and initial conditions and period @xmath244 close to @xmath250 note that our initial point on the orbit , defined by ( [ phase_cond ] ) , differs from the above initial conditions . we will test our method on this same orbit , shown in the @xmath251-@xmath14 plane in figure [ fig : lorenz_orbit ] . + -@xmath14 plane.,height=226 ] a taylor series approximation to the solution centered at @xmath252 can be derived @xmath253 @xmath254 where @xmath255 , @xmath256 , @xmath257 , and @xmath244 are obtained from numerical approximations to the periodic orbit at the centers of the mesh intervals @xmath151 $ ] . the jacobian of the lorenz system vector field @xmath258 can be taylor expanded @xmath259 combined with theorem [ main_theorem ] , this can be used to compute upper bounds on the values of the parameters @xmath260 , @xmath130 , @xmath261 that appear in the newton - kantorovich theorem , see appendix [ newton_th ] . the trivial weight @xmath262 was used in this test . denote the piecewise polynomial approximate solution by @xmath256 , let @xmath263 , and @xmath8 be its approximate inverse as in theorem [ main_theorem ] . using @xmath207 , an equally spaced mesh of size @xmath264 , and a domain @xmath265 , @xmath266 , our method provides a rigorous existence proof with the following parameters @xmath267 in other words , a solution exists within a @xmath20-norm ball of radius @xmath268 about the approximate solution and the solution is unique within the ball of radius @xmath269 . the computation was done in matlab using the interval arithmetic package intlab @xcite on a 2ghz intel core i7 and took 142 seconds . this includes the time to compute the approximate solution @xmath256 and approximate bvp fundamental solution @xmath102 . these were computed using the matlab routines ode15s and bvp5c with absolute and relative error tolerances of @xmath270 . we have presented an a posteriori method for proving existence to and deriving rigorous error bounds for two point boundary value problems on a finite interval , summarized in theorem [ main_theorem ] . the method applies to general @xmath0-dimensional systems without any assumption on the form of the vector field . it uses a ( non - rigorous ) approximation to the green s function of the bvp to generate the bounds and can be evaluated using interval arithmetic for mathematically rigorous results or in traditional floating point for faster generation of approximate bounds . because the method is based on the green s function of a bvp rather than the fundamental solution to an ivp , the method is applicable to cases where the bvp is stable but the corresponding ivp is unstable . an adaptively chosen weight matrix was used in some of the norms . this improves the quality of the final @xmath78 error bounds by several orders of magnitude in many of the tests . in section [ sec : num_tests ] we tested the bvp method on a pair of singularly perturbed linear problems that exhibit such problematic features as dense oscillation and boundary layers . the method is successful in proving existence and producing reasonable and usable error bounds , even when the relevant perturbation parameter becomes moderately small . we have also successfully applied it to rigorously prove the existence of a periodic orbit in the lorenz system . in the cases tested , the main limiting factor on the success of the method is the extreme ill conditioning of the approximate bvp fundamental solution @xmath91 as the parameter that controls the singular perturbation is made smaller . although @xmath91 itself has moderate sized norm in the tests , it is close to being singular . the bvp method requires computation of the fundamental solution and its inverse to sufficient _ relative accuracy _ ; poorly resolved decaying modes will spoil the method and the method always fails once the underflow limit is reached . this commonly occurs for very stiff problems and so , while the bvp method works for moderately stiff problems as we have shown , it fails in the very stiff limit . this issue could likely be addressed in a brute force manner by employing an interval or floating point arithmetic package that allows for exponent values of substantially larger magnitude than are allowed in standard floating point arithmetic , but we did not employ such a technique here . for convenience , in this appendix we state a version of the newton - kantorovich theorem that was used above . [ newton ] let @xmath122 , @xmath49 be banach spaces , @xmath271 be open and convex , @xmath272 be differentiable , and @xmath273 in @xmath274 . let @xmath275 and @xmath276 have bounded inverse @xmath108 . suppose @xmath277 , @xmath278 , and @xmath279 . set @xmath280 suppose @xmath281 . then the newton iteration is well defined , lies in @xmath282 and converges to @xmath283 , a solution of @xmath284 . this is the unique solution in @xmath285 . in the above , @xmath261 is computed by bounding @xmath286 . in this appendix a formula is derived for @xmath135 . the goal is decompose it into several terms , each of which obviously vanishes when @xmath6 and @xmath7 equal the exact fundamental solution @xmath91 and green s function @xmath28 , respectively . the calculation is long but relatively straightforward . we outline the key steps below . first , using the definitions of @xmath8 and @xmath9 and taking @xmath287 the two components equal @xmath288_1(t)\\ = & \tilde\phi(0)(w - b_1r(1))+\int_0 ^ 1 \tilde g(0,s)a(s)r(s)ds-\tilde\phi(t)(w - b_1r(1))\\ & -\int_0 ^ 1 \tilde g(t , s)a(s)r(s)ds+\int_0^ta(s)\tilde\phi(s)ds(w - b_1r(1))+\int_0^ta(s)r(s)ds\\ & + \int_0^ta(s)\int_0 ^ 1 \tilde g(s , z)a(z)r(z)dzds,\\ & ( i - fh)[r , w]_2\\ = & [ i - b_0\tilde\phi(0)-b_1\tilde\phi(1)](w - b_1r(1))-\int_0 ^ 1 \left(b_0\tilde g(0,s)+b_1\tilde g(1,s)\right)a(s)r(s)ds.\end{aligned}\ ] ] one can check that @xmath289 it is apparent that each term vanishes for the exact solution . as for the first component , we introduce @xmath290 and @xmath291 to obtain terms that vanish when the ode is satisfied @xmath288_1(t)\\ = & \tilde\phi(0)(w - b_1r(1))+\int_0 ^ 1 \tilde g(0,s)a(s)r(s)ds-\tilde\phi(t)(w - b_1r(1))-\int_0 ^ 1 \tilde g(t , s)a(s)r(s)ds\\ & + \int_0^t-\tilde\phi^\prime(s)+a(s)\tilde\phi(s)ds(w - b_1r(1))+\int_0^t\tilde\phi^\prime(s)ds(w - b_1r(1))+\int_0^ta(s)r(s)ds\\ & + \int_0^t\int_0 ^ 1(-\partial_s \tilde g(s , z)+a(s ) \tilde g(s , z))a(z)r(z)dzds+\int_0^t\int_0 ^ 1\partial_s \tilde g(s , z)a(z)r(z)dzds.\end{aligned}\ ] ] now integrate by parts , taking into account the discontinuities of @xmath292 and @xmath293 at @xmath294 as well as @xmath293 at @xmath295 . after substantial simplification we arrive at @xmath296_1(t)\\ = & -\int_0^t\tilde \phi^\prime(s)-a(s)\tilde\phi(s)ds(w - b_1r(1))+\int_0^t(\tilde g(z^-,z)-\tilde g(z^+,z)+i)a(s)r(s)ds\notag\\ & + \sum_{j=1}^{j_t}(\tilde\phi(t_j^-)-\tilde\phi(t_{j}^+))(w - b_1r(1))+\int_0 ^ 1\left[\sum_{j=1}^{j_t}(\tilde g(t_j^-,z)-\tilde g(t_{j}^+,z))\right]a(z)r(z)dz\notag\\ & -\int_0^t\int_0 ^ 1(\partial_s \tilde g(s , z)-a(s ) \tilde g(s , z))a(z)r(z)dzds\notag\end{aligned}\ ] ] where @xmath297 is the smallest index with @xmath298 and the superscripts @xmath76 and @xmath77 denote limits from above and below respectively . the exact greens function satisfies @xmath299 and is continuous away from the boundary . the exact fundamental solution @xmath91 is continuous as well . therefore , if @xmath102 and @xmath7 equal the exact fundamental solution @xmath91 and green s function @xmath28 then the jump terms in ( [ i_fh1_simp ] ) vanish . the remaining terms are zero since @xmath91 and @xmath28 satisfy the ode . therefore we have succeeded in putting @xmath135 into a form where each term is manifestly zero when using the exact solution . this suggests that a reasonable bound on @xmath133 might be obtained by bounding each of these terms individually , as done in section [ sec : sharper_bound ] . a similar procedure to appendix [ app : i - fh ] yields a formula for @xmath126 . for @xmath300 @xmath301(t)\notag\\ = & \int_0 ^ 1\left(\frac{\partial}{\partial s}\tilde g(t , s)+\tilde g(t , s)a(s)\right)\left(v(0)+\int_0^s a(r)v(r)dr\right)ds\\ & + ( \tilde g(t,0)-\tilde\phi(t)b_0)v(0)-(\tilde g(t,1)+\tilde\phi(t)b_1)\left(v(0)+\int_0 ^ 1 a(s)v(s)ds\right)\\ & -\sum_{j=1}^{n-1}\left(\tilde g(t , t_j^-)-\tilde g(t , t_{j}^+)\right)\left(v(0)+\int_0^{t_{j } } a(r)v(r)dr\right)\\ & -\left(\tilde g(t , t^-)-\tilde g(t , t^+)-i\right)\left(v(0)+\int_0^t a(r)v(r)dr\right)\end{aligned}\ ] ] where @xmath302 and @xmath303 . the most significant difference is that the jump terms now involve the second argument of @xmath7 . in practice we found that the bounds obtained using @xmath135 to be tighter than those using @xmath126 and so we have based the bvp method on @xmath135 , but the other choice could have been made and may be preferable in some cases . as discussed above , the data for the bvp method consists of sets @xmath150 and @xmath3 of approximations to the fundamental solution @xmath91 and green s function @xmath28 at the centers of the intervals @xmath151 $ ] and rectangles @xmath151\times [ t_{j-1},t_j]$ ] . these need to be extended to @xmath18 functions on each interval or rectangle . this is done by letting @xmath304 , @xmath305 be the polynomials of degree @xmath183 that satisfies the odes @xmath306 up to order @xmath161 . the functions @xmath307}=(i+e_i(t))\tilde\phi_i,\hspace{2 mm } \tilde g(t , s)|_{[t_{i-1},t_i]\times[t_{j-1},t_j]}=(i+e_i(t))\tilde g_{i , j}(i+f_j(s))\ ] ] are then approximate solutions to the odes satisfied by @xmath91 and @xmath28 . note that on the rectangle with @xmath101 two different versions are needed , one corresponding to the upper half of the triangle and one to the lower , constructed from @xmath162 and @xmath163 respectively , where @xmath308 recursive formulas for the coefficients are given below . we suppose the coefficient matrix @xmath36 is @xmath309 and write the taylor series with remainder about @xmath310 as @xmath311 define @xmath172 , @xmath173 on @xmath171 $ ] by @xmath312 where @xmath310 denotes the midpoint of @xmath171 $ ] . note that these quantities do not involve the remainder term @xmath313 . with these , the ode residuals are @xmath314 this work was conducted with government support under and awarded by dod , air force office of scientific research , national defense science and engineering graduate ( ndseg ) fellowship , 32 cfr 168a . d. ambrosi , g. arioli and h. koch , a homoclinic solution for excitation waves on a contractile substratum , _ siam journal on applied dynamical systems _ , * 11 * ( 2012 ) , 15331542 , url http://dx.doi.org/10.1137/12087654x . g. arioli and h. koch , existence and stability of traveling pulse solutions of the fitzhugh nagumo equation , _ nonlinear analysis : theory , methods & applications _ , * 113 * ( 2015 ) , 51 70 , url http://www.sciencedirect.com/science/article/pii/s0362546x14003137 . u. ascher and l. petzold , _ computer methods for ordinary differential equations and differential - algebraic equations _ , society for industrial and applied mathematics ( siam , 3600 market street , floor 6 , philadelphia , pa 19104 ) , 1998 , url http://books.google.com/books?id=2ixovtfcl74c . b. a. coomes , h. koc cak and k. j. palmer , rigorous computational shadowing of orbits of ordinary differential equations , _ numerische mathematik _ , * 69 * ( 1995 ) , 401421 , url http://dx.doi.org/10.1007/s002110050100 . y. eidelman , v. milman and a. tsolomitis , _ functional analysis : an introduction _ , graduate studies in mathematics , american mathematical society , 2004 , url http://books.google.com/books?id=bzugkyoaqocc . o. fogelklou , w. tucker and g. kreiss , a computer - assisted proof of the existence of traveling wave solutions to the scalar euler equations with artificial viscosity , _ nonlinear differential equations and applications nodea _ , * 19 * ( 2012 ) , 97131 , url http://dx.doi.org/10.1007/s00030-011-0120-7 . o. fogelklou , w. tucker , g. kreiss and m. siklosi , a computer - assisted proof of the existence of solutions to a boundary value problem with an integral boundary condition , _ communications in nonlinear science and numerical simulation _ , * 16 * ( 2011 ) , 1227 1243 , url http://www.sciencedirect.com/science/article/pii/s1007570410003734 . g. folland , _ real analysis : modern techniques and their applications _ , pure and applied mathematics : a wiley series of texts , monographs and tracts , wiley , 2013 , url http://books.google.com/books?id=wi4fawaaqbaj . z. galias and p. zgliczyski , computer assisted proof of chaos in the lorenz equations , _ physica d : nonlinear phenomena _ , * 115 * ( 1998 ) , 165 188 , url http://www.sciencedirect.com/science/article/pii/s0167278997002339 . a. hungria , j .- lessard and j. d. m. james , rigorous numerics for analytic solutions of differential equations : the radii polynomial approach , url http://archimede.mat.ulaval.ca/jplessard/publications_files/final_draft.pdf . l. jaulin , _ applied interval analysis : with examples in parameter and state estimation , robust control and robotics _ , no . v. 1 in applied interval analysis : with examples in parameter and state estimation , robust control and robotics , springer london , 2001 , url http://books.google.com/books?id=zg0qxkyue_ac . l. kantorovich and g. akilov , _ functional analysis in normed spaces _ , international series of monographs in pure and applied mathematics , pergamon press ; [ distributed in the western hemisphere by macmillan , new york ] , 1964 , url http://books.google.com/books?id=g_tqaaaamaaj . j. lee and l. greengard , a fast adaptive numerical method for stiff two - point boundary value problems , _ siam journal on scientific computing _ , * 18 * ( 1997 ) , 403429 , url http://dx.doi.org/10.1137/s1064827594272797 . lessard , j. mireles james and c. reinhardt , computer assisted proof of transverse saddle - to - saddle connecting orbits for first order vector fields , _ journal of dynamics and differential equations _ , * 26 * ( 2014 ) , 267313 , url http://dx.doi.org/10.1007/s10884-014-9367-0 . lessard and c. reinhardt , rigorous numerics for nonlinear differential equations using chebyshev series , _ siam journal on numerical analysis _ , * 52 * ( 2014 ) , 122 , url http://dx.doi.org/10.1137/13090883x . m. a. mccarthy and r. a. tapia , computable a posteriori l-error bounds for the approximate solution of two - point boundary value problems , _ siam journal on numerical analysis _ , * 12 * ( 1975 ) , pp . 919937 , url http://www.jstor.org/stable/2156420 . r. moore , _ methods and applications of interval analysis _ , studies in applied and numerical mathematics , society for industrial and applied mathematics , 1979 , url http://books.google.com/books?id=wyjd2-r2zmgc . k. nagatou , n. yamamoto and m. nakao , an approach to the numerical verification of solutions for nonlinear elliptic problems with local uniqueness , _ numerical functional analysis and optimization _ , * 20 * ( 1999 ) , 543565 . m. t. nakao , solving nonlinear parabolic problems with result verification . part i : one - space dimensional case , _ journal of computational and applied mathematics _ , * 38 * ( 1991 ) , 323 334 , url http://www.sciencedirect.com/science/article/pii/037704279190179n . m. t. nakao , a numerical verification method for the existence of weak solutions for nonlinear boundary value problems , _ journal of mathematical analysis and applications _ , * 164 * ( 1992 ) , 489 507 , url http://www.sciencedirect.com/science/article/pii/0022247x92901292 . n. nedialkov , k. jackson and g. corliss , validated solutions of initial value problems for ordinary differential equations , _ applied mathematics and computation _ , * 105 * ( 1999 ) , 21 68 , url http://www.sciencedirect.com/science/article/pii/s0096300398100838 . a. takayasu , s. oishi and t. kubo , numerical existence theorem for solutions of two - point boundary value problems of nonlinear differential equations , _ nonlinear theory and its applications , ieice _ , * 1 * ( 2010 ) , 105118 . w. tucker , the lorenz attractor exists , _ comptes rendus de lacadmie des sciences - series i - mathematics _ , * 328 * ( 1999 ) , 1197 1202 , url http://www.sciencedirect.com/science/article/pii/s076444429980439x . j. van den berg , j. mireles - james , j. lessard and k. mischaikow , rigorous numerics for symmetric connecting orbits : even homoclinics of the gray scott equation , _ siam journal on mathematical analysis _ , * 43 * ( 2011 ) , 15571594 , url http://dx.doi.org/10.1137/100812008 . j. b. van den berg , c. m. groothedde and j. f. williams , rigorous computation of a radially symmetric localized solution in a ginzburg landau problem , _ siam journal on applied dynamical systems _ , * 14 * ( 2015 ) , 423447 , url http://dx.doi.org/10.1137/140987973 . y. watanabe , m. plum and m. t. nakao , a computer - assisted instability proof for the orr - sommerfeld problem with poiseuille flow , _ zamm - journal of applied mathematics and mechanics / zeitschrift fr angewandte mathematik und mechanik _ , * 89 * ( 2009 ) , 518 . n. yamamoto , a numerical verification method for solutions of boundary value problems with local uniqueness by banach s fixed - point theorem , _ siam journal on numerical analysis _ , * 35 * ( 1998 ) , pp . 20042013 , url http://www.jstor.org/stable/2587285 .

new advances in nano sciences open the door for scientists to study biological processes on a microscopic molecule - by - molecule basis . recent single - molecule biophysical experiments on enzyme systems , in particular , reveal that enzyme molecules behave fundamentally differently from what classical model predicts . a stochastic network model was previously proposed to explain the experimental discovery . this paper conducts detailed theoretical and data analyses of the stochastic network model , focusing on the correlation structure of the successive reaction times of a single enzyme molecule . we investigate the correlation of experimental fluorescence intensity and the correlation of enzymatic reaction times , and examine the role of substrate concentration in enzymatic reactions . our study shows that the stochastic network model is capable of explaining the experimental data in depth . .

1209.6210

in a chemical reaction , the number of molecules involved can drastically vary from millions of moles a forest devastated by a fire to only a few reactions in a living cell . while most conventional chemical experiments were designed for a large ensemble in which only the average could be observed , chemistry textbooks tend to explain what really happens in a reaction on a molecule - by - molecule basis . this extrapolation certainly requires the homogeneity assumption : each molecule behaves in the same way , so the average also represents individual behavior . to verify this assumption , the kinetic of a single molecule must be directly observed , which requires rather sophisticated technology not available until the 1990s . since then , the development of nanotechnology has enabled scientists to track and manipulate molecules one by one . a new age of single - molecule experiments began [ @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ] . such experiments offer a greatly amplified view of single - molecular dynamics over considerably _ long _ time periods from seconds to hours , a time scale that far exceeds what can be achieved by computer based molecular dynamic simulation ( even with a super computer , molecular dynamic simulation can not reach beyond milliseconds ) . the single - molecule experiments also provide detailed information on the intermediate transition steps of a biological process not available in traditional experiments . not surprisingly , these experiments reveal the stochastic nature of nanoscale particles long masked by ensemble averages : rather than remain rigid , those particles undergo dramatic conformation change driven by external thermal motion . future development in this area will provide us a deeper understanding of biological processes [ such as molecular motors , @xcite ] and accelerate new technology development [ such as single - molecule gene sequencing , @xcite ] . among bio - molecules , enzymes play an important role : by lowering the energy barrier between the reactant and product , they ensure that many life essential processes can be effectively carried out in a living cell . an aspiration of bioengineers is to artificially design and produce new and efficient enzymes for specific use . studying and understanding the mechanism of existing enzymes , therefore , remains one of the central topics in life science . according to the classical literature , the kinetic of an enzyme is described by the michaelis menten mechanism [ @xcite ] : an enzyme molecule @xmath0 could bind with a reactant molecule @xmath1 , which is referred to as a substrate in the chemistry literature ( hence the symbol @xmath1 ) , to form a complex @xmath2 . the complex can either dissociate to enzyme and substrate molecules or undergo a catalytic process to release the product @xmath3 . the enzyme then returns to the original state @xmath0 to start another catalytic circle . this process is typically diagrammed as@xmath4}{\rightleftarrows}}_{k_{-1}}}{\mathit{es}}\stackrel { k_{2}}{\rightarrow}e^{0}+p,\qquad e^{0}\stackrel{\delta}{\rightarrow}e,\ ] ] where @xmath5 $ ] is the substrate concentration ( @xmath6 is the release state of the enzyme ) , @xmath7 is the association rate per unit substrate concentration , @xmath8 and @xmath9 are , respectively , the dissociation and catalytic rate , and @xmath10 is the returning rate . all the transitions are memoryless in the michaelis menten scheme , so the whole process can be modeled as a continuous - time markov chain consisting of three states @xmath0 , @xmath2 and @xmath6 for an enzyme molecule . a recent single - molecule experiment [ @xcite ] conducted by the xie group at harvard university ( department of chemistry and chemical biology ) studied the enzyme @xmath11-galactosidase ( @xmath11-gal ) , which catalyzes the breakdown of the sugar lactose and is essential in the human body [ @xcite , @xcite ] . in the experiment a single @xmath11-gal molecule is immobilized ( by linking to a bead bound on a glass coverslip ) and immersed in buffer solution of the substrate molecules . this setup allows @xmath11-gal s enzymatic action to be continuously monitored under a fluorescence microscope . to detect the individual turnovers , that is , the enzyme s switching from the @xmath0 state to the @xmath6 state , careful design and special treatment were carried out ( such as the use of photogenic substrate resorun-@xmath11-d - galactopyranoside ) so that once the experimental system was placed under a laser beam the reaction product and _ only _ the reaction product was fluorescent . this setting ensures that as the @xmath11-gal enzyme catalyzes substrate molecules one after another , a strong fluorescence signal is emitted and detected only when a product is released , that is , only when the reaction reaches the @xmath12 stage in ( [ mm ] ) . recording the fluorescence intensity over time thus enables the experimental determination of individual turnovers . a sample fluorescence intensity trajectory from this experiment is shown in figure [ fig1 ] . high spikes in the trajectory are the results of intense photon burst at the @xmath12 state , while low readings correspond to the @xmath0 or @xmath2 state . the time lag between two adjacent high fluorescence spikes is the enzymatic turnover time , that is , the time to complete a catalytic circle . $ ] ( 20 , 100 and 380 micromolar ) . right column : the autocorrelations predicted by the classical michaelis menten model . under the michaelis menten model , the turnover time autocorrelation should be zero and the intensity autocorrelations should decay exponentially and decay faster under larger concentration . all contradict the experimental findings . ] examining the experimental data , including the distribution and autocorrelation of the turnover times as well as the fluorescence intensity autocorrelation , researchers were surprised that the experimental data showed a considerable departure from the michaelis menten mechanism . section [ sec2 ] describes the experimental findings in detail . figure [ fig2 ] illustrates the discrepancy between the experimental data and the michaelis menten model in terms of the autocorrelations . the left two panels show the experimentally observed fluorescence intensity autocorrelation and turnover time autocorrelation under different substrate concentrations @xmath5 $ ] . the right two panels show the corresponding autocorrelation patterns predicted by the michaelis menten model . comparing the bottom two panels , we note that under the classical michaelis menten model the turnover time autocorrelation should be zero ( hence the horizontal line at the bottom - right panel ) , which clearly contradicts the experimental result on the left . from the top two panels we note that under the michaelis menten model the fluorescence intensity autocorrelation should decay _ exponentially _ and should decay _ faster _ with _ larger _ substrate concentration , but the experimental result shows the opposite : the intensity autocorrelations decay _ slower _ with _ larger _ substrate concentration , and they do not decay exponentially . to explain the experimental puzzle , a new stochastic network model was introduced [ @xcite , @xcite ] , and it was shown that the stochastic network model well explained the experimental distribution of the turnover times . the autocorrelation of successive turnover times and the correlation of experimental fluorescence intensity , however , were not investigated in the previous articles . this paper further explores the stochastic network model , concentrating on the correlation structure of the turnover times and that of the fluorescence intensity . the rest of the paper is organized as follows . section [ sec2 ] reviews the preceding work , including the experiment observation and the new stochastic network model . section [ sec3 ] analytically calculates the turnover time autocorrelation and the fluorescence intensity autocorrelation based on the stochastic network model . these analytical results give an explanation of the multi - exponentially decay pattern of the autocorrelation functions . section [ sec4 ] discusses how to fit the experiment data within the framework of the stochastic network model . the paper ends in section [ sec5 ] with a summary and some concluding remarks . under the classical michaelis menten model ( [ mm ] ) , an enzyme molecule behaves as a three - state continuous - time markov chain with the generating matrix ( infinitesimal generator)@xmath13 & k_{1}[s ] & 0 \cr k_{-1 } & -(k_{-1}+k_{2 } ) & k_{2 } \cr \delta & 0 & -\delta}.\ ] ] we can readily draw two properties from this continuous - time markov chain model . the density function of the turnover time , the time that it takes the enzyme to complete one catalytic cycle ( i.e. , to go from state @xmath0 to state @xmath6 ) , is@xmath14}{2p}\bigl(e^{-(q - p)t}-e^{-(q+p)t}\bigr),\ ] ] where @xmath15+k_{2}+k_{-1})^{2}/4-k_{1}k_{2}[s]}$ ] and @xmath16+k_{2}+k_{-1})/2 $ ] . the successive turnover times have no correlation . the first proposition implies that the density of turnover time is almost an exponential , since the term @xmath17 easily dominates the term @xmath18 for most values of @xmath19 ; see @xcite for a proof . the second proposition is a consequence of the markov property : each turnover time , which is a first passage time , is independently and identically distributed . the third property concerns the autocorrelation of the fluorescence intensity . as we have seen in figure [ fig1 ] , the experimentally recorded fluorescence intensity consists of high spikes and low readings . the high peaks correspond to the release of the fluorescent product ( when the enzyme is at the state @xmath6 ) , whereas the low readings come from the background noise . we can thus think of the fluorescence intensity reading as a record of an on off system : @xmath6 being the on state , @xmath0 and @xmath2 being the off states . the autocorrelation function of the fluorescence intensity is proportional to @xmath20))$ ] . the proof of the proposition will be given in corollary [ cor4 ] . this proposition says that under the michaelis menten model the intensity autocorrelation decays exponentially and faster with larger substrate concentration @xmath5 $ ] . the results from the single - molecule experiment on @xmath11-gal [ @xcite ] contradict all three properties of the michaelis menten model : \(1 ) the empirical distribution of the turnover time does not exhibit exponential decay ; see @xcite for a detailed explanation . \(2 ) the experimental turnover time autocorrelations are far from zero , as seen in figure [ fig2 ] . \(3 ) the experimental intensity autocorrelations decay neither exponentially nor faster under larger concentration . see figure [ fig2 ] . we believe these contradictions are rooted in the molecule s _ dynamic _ conformational fluctuation . an enzyme molecule is not rigid : it experiences constant changes and fluctuations in its three - dimensional shape and configuration due to the entropic and atomic forces at the nano scale [ @xcite , @xcite ] . although for a large ensemble of molecules , the ( nanoscale ) conformational fluctuation is buried in the macroscopic population average , for a single molecule the conformational fluctuation can be much more pronounced : different conformations could have different chemical properties , resulting in time - varying performance of the enzyme , which can be studied in the single - molecule experiment . the following stochastic network model [ @xcite ] was developed with this idea:@xmath21}{\rightleftarrows}}_{k_{-11 } } } { \mathit{es}}_{1 } \stackrel{k_{21}}{\rightarrow } p+e_{1}^{0},\qquad e_{1}^{0}\stackrel{\delta_{1}}{\rightarrow}e_{1 } , \nonumber\\ & & \hphantom{s}\downarrow\uparrow\hspace*{37pt}\downarrow\uparrow\hspace*{30pt}\hphantom{0 } \downarrow\uparrow\qquad\hspace*{10pt}\hphantom{0}\cdots\nonumber\\ & & s+e_{2 } { \mathop{\stackrel{k_{12}[s]}{\rightleftarrows}}_{k_{-12 } } } { \mathit{es}}_{2 } \stackrel{k_{22}}{\rightarrow } p+e_{2}^{0},\qquad e_{2}^{0}\stackrel { \delta_{2}}{\rightarrow}e_{2},\\ & & \hphantom{s}\hspace*{1.5pt}\cdots\hspace*{37pt}\cdots\hspace*{33.5pt}\hphantom{0 } \cdots\qquad\hspace*{10.5pt}\hphantom{0}\cdots\nonumber\\[-8pt ] & & \hphantom{s}\downarrow\uparrow\hspace*{37pt}\downarrow\uparrow\hspace*{30pt}\hphantom{0 } \downarrow\uparrow\qquad\hspace*{10pt}\hphantom{0}\cdots\nonumber\\ & & s+e_{n } { \mathop{\stackrel{k_{1n}[s]}{\rightleftarrows}}_{k_{-1n } } } { \mathit{es}}_{n } \stackrel{k_{2n}}{\rightarrow } p+e_{n}^{0},\qquad e_{n}^{0}\stackrel { \delta_{n}}{\rightarrow}e_{n}.\nonumber\end{aligned}\ ] ] this is still a markov chain model but with @xmath22 states instead of three . the enzyme still exists as a free enzyme @xmath0 , an enzyme substrate complex @xmath2 or a returning enzyme @xmath6 , but it can take @xmath23 different conformations indexed by subscripts in each stage . at each transition , the enzyme can either change its conformation within the same stage ( such as @xmath24 or @xmath25 ) or carry out one chemical step , that is , move between the stages ( such as @xmath26 , @xmath27 or @xmath28 ) . since only the product @xmath3 is fluorescent in the experiment , in model ( [ 2by2 ] ) any state @xmath29 is an on - state , and the others are off - states . consequently , the turnover time is the traverse time between any two on - states @xmath30 and @xmath31 . to fully specify the model , we need to stipulate the transition rates . for @xmath32 , we use @xmath33 , @xmath34 and @xmath35 to denote , respectively , the transition rates of @xmath24 , @xmath25 and @xmath36 . @xmath37 $ ] , @xmath38 , @xmath39 and @xmath40 are , respectively , the transition rates of @xmath41 , @xmath27 , @xmath28 and @xmath42 . define @xmath43 , @xmath44 and @xmath45 to be square matrices:@xmath46_{n\times n},\qquad \mathbf{q}_{bb}=[\beta _ { ij}]_{n\times n},\qquad \mathbf{q}_{cc}=[\gamma_{ij}]_{n\times n},\ ] ] where @xmath47 , @xmath48 and @xmath49 . they correspond to transitions among the @xmath50 states , among the @xmath51 states and among the @xmath29 states , respectively . define diagonal matrices@xmath52,k_{12}[s],\ldots , k_{1n}[s]\},\nonumber\\ \mathbf{q}_{ba}&=&\operatorname{diag}\ { k_{-11},k_{-12},\ldots , k_{-1n}\},\nonumber\\[-8pt]\\[-8pt ] \mathbf{q}_{bc } & = & \operatorname{diag}\{k_{21},k_{22},\ldots , k_{2n}\},\nonumber\\ \mathbf{q}_{ca}&=&\operatorname{diag}\{\delta_{1},\delta_{2},\ldots,\delta _ { n}\}. \nonumber\end{aligned}\ ] ] they correspond to transitions between the different stages . the generating matrix of model ( [ 2by2 ] ) is then@xmath53 under this new model , the distribution of the turnover time , the correlation of turnover times and the correlation of the fluorescence intensity can be analyzed and compared with experimental data . this paper studies the autocorrelation of turnover time and the autocorrelation of fluorescence intensity . in the chemistry literature , the term `` equilibrium '' often refers to the state in which all the macroscopic quantities of a system are time - independent . for the microscopic system studied in single - molecule experiments , macroscopic quantities , however , are meaningless , and microscopic parameters never cease to fluctuate . nonetheless , for a micro - system , one can talk about dynamic equilibrium in the sense that the _ distribution _ of the state quantities become time - independent , that is , they reach the stationary distribution . the single - molecule enzyme experiment that we consider here falls into this category , since the enzymatic reactions happen quite fast . we cite the following lemma [ lemma 3.1 of @xcite ] , which gives the stationary distribution of the markov chain ( [ 2by2 ] ) : [ lem1]let @xmath54 be the process evolving according to ( [ 2by2 ] ) . suppose all the parameters @xmath55 , and @xmath35 are positive . then @xmath54 is ergodic . let the row vectors @xmath56 , @xmath57 , and @xmath58 denote the stationary distribution of the entire network . up to a normalizing constant , they are determined by@xmath59 where the matrices@xmath60 ^{-1 } , \\ \mathbf{m } & = & [ \mathbf{q}_{bb}-\mathbf{q}_{bc}-(\mathbf{q}_{bb}-\mathbf{q}_{ba}-\mathbf{q}_{bc})\mathbf{q}_{ab}^{-1}\mathbf{q}_{aa } ] ^{-1}.\end{aligned}\ ] ] under the stochastic network model ( [ 2by2 ] ) , a turnover event can start from any state @xmath50 and end in any @xmath30 . it follows that the _ overall _ distribution of all the turnover times is characterized by a mixture distribution with the weights given by the _ stationary _ probability of a turnover event s starting from @xmath50 . the following lemma , based on lemma 3.4 of @xcite , provides the stationary probability . [ lem2]let @xmath61 be a row vector , @xmath62 , where @xmath63 denotes the stationary probability of a turnover event s starting from state @xmath50 . then up to a normalizing constant , @xmath61 is the nonzero solution of@xmath64 the enzyme turnover event occurs one after another . each can start from any @xmath50 and end in any @xmath30 . the next turnover may start from @xmath65 ( @xmath66 ) when the system exits the @xmath6 stage from @xmath31 . to calculate the correlation between turnover times , it is necessary to find out the probabilities of all these combinations and the expected turnover times . we introduce the following notation . let @xmath67 and @xmath68 denote the first passage time of reaching the set @xmath69 from @xmath50 and @xmath51 , respectively . let @xmath70 and @xmath71 be the probability that a turnover event , starting , respectively , from @xmath50 and @xmath51 , ends in @xmath30 . let @xmath72 denote the probability that , after the previous turnover ends in @xmath29 , a new turnover event starts from @xmath73 . finally , let @xmath74 and @xmath75 be the first passage time of reaching the state @xmath30 from @xmath50 and @xmath51 , respectively . for the values of @xmath76 and @xmath77 , we cite the following lemma [ corollary 3.3 of @xcite ] . [ lem3]let the vectors @xmath78 and @xmath79 denote the mean first passage times . then they are given by@xmath80 where the matrices @xmath81 and @xmath82 are given by@xmath83 ^{-1 } , \\ \mathbf{r } & = & [ \mathbf{q}_{bb}-\mathbf{q}_{ba}-\mathbf{q}_{bc}-\mathbf { q}_{ba}(\mathbf{q}_{aa}-\mathbf{q}_{ab})^{-1}\mathbf{q}_{ab } ] ^{-1}.\end{aligned}\ ] ] for the probabilities @xmath70 , @xmath71 and @xmath72 , we have the following lemma : [ lem4]let @xmath84 , @xmath85 and @xmath86 be probability matrices @xmath87_{n\times n}$ ] , @xmath88_{n\times n}$ ] and @xmath89_{n\times n}$ ] . then they are given by@xmath90 for the expectation of @xmath74 and @xmath75 , we have the following . [ lem5]let @xmath91_{n\times n}\quad \mbox{and } \quad\mathbf{e}_{bc}=[p_{{\mathit{es}}_{i}e_{j}^{0}}e(t_{{\mathit{es}}_{i}e_{j}^{0}})]_{n\times n}\ ] ] be two @xmath92 matrices . then they are given by@xmath93 we defer the proofs of lemmas [ lem4 ] and [ lem5 ] to the . let @xmath94 denote the @xmath95th turnover time . the next theorem , based on lemmas [ lem1 ] to [ lem5 ] , obtains the autocorrelation of the successive turnover times . we defer its proof to the . [ theo1]the covariance between the first turnover and the @xmath96th turnover ( @xmath97 ) is given by@xmath98\bolds{\mu}_{a},\ ] ] where @xmath99 . the matrix @xmath100 is the product of two transition - probability matrices , so it is a stochastic matrix . given that all the states in the stochastic network model communicate with each other , @xmath100 is also irreducible , and all its entries are positive . according to the perron frobenius theorem [ @xcite ] , such a matrix has eigenvalue one with simplicity one , and the absolute values of the other eigenvalues are strictly less than one . we therefore obtain the following corollary of theorem [ theo1 ] . [ cor1]suppose that @xmath100 is diagonalizable : @xmath101 where the diagonal matrix @xmath102 consists of the eigenvalues of @xmath100 with @xmath103 ; the columns , @xmath104 , of matrix @xmath105 are the corresponding right eigenvectors ; and the rows , @xmath106 , of @xmath107 are the corresponding left eigenvectors . then we have@xmath108 where @xmath109 . although the matrix @xmath100 may have complex eigenvalues , these complex eigenvalues and corresponding eigenvectors always appear as conjugate pairs so that the imaginary parts in ( corrt ) cancel each other . as a result , we could treat all @xmath110 and @xmath111 as if they were real numbers . theorem [ theo1 ] , along with corollary [ cor1 ] , provides an explanation of why the correlation of turnover times is not zero . at first sight , it seems to contradict the memoryless property of a markov chain . what actually happens is that the state must be explicitly specified for the memoryless property to hold ( i.e. , one needs to exactly specify whether an enzyme is at state @xmath112 or @xmath113 ) , whereas in the single - molecule experiment we only know whether the system is in an `` on '' or `` off '' state ( e.g. , one only knows that the enzyme is in one of the on - states @xmath114 ) . when there are multiple states , this aggregation effect leads to incomplete information that prevents the independence between successive turnovers ; consequently , each turnover time carries some information about its reaction path , which is correlated with the reaction path of the next turnover , resulting in the correlation between successive turnover times . corollary [ cor1 ] also states that since @xmath103 , the autocorrelation is a mixture of exponential decays . thus , depending on the relative scales of the eigenvalues , the actual decay might be single - exponential when one eigenvalue dominates the others or multi - exponential when several major eigenvalues jointly contribute to the decay . in most enzymatic reactions , including the one we study , the enzyme returns very quickly to restart a new cycle once the product is released [ @xcite ] . those enzymes are called fast - cycle - reset enzymes . to model this fact , we let @xmath40 @xmath115 , the transition rate from @xmath29 to @xmath50 , go to infinity . then any enzyme in state @xmath29 will always return to state @xmath50 instantly , and the related transition probability matrix @xmath86 , defined in ( [ pmat ] ) , becomes the identity matrix . in the single - enzyme experiments , the raw data are the time traces of fluorescence intensity , as shown in figure [ fig1 ] . the time lag between two adjacent high fluorescence spikes gives the enzymatic turnover time . the fluorescence intensity reading , however , is subject to detection error : the error caused by the limited time resolution @xmath116 of the detector . starting from time 0 , the detector will only record intensity data at multiples of @xmath116 : @xmath117 the intensity reading at time @xmath118 is actually the total number of photons received during the period of @xmath119 . thus , the detection errors of turnover time are roughly @xmath116 . when the successive reactions occur slowly , the average turnover time is much longer than @xmath116 , and the error is negligible . but when the reactions happen very frequently , the average turnover time becomes comparable to @xmath116 , and this error can not be ignored . in fact , when the substrate concentration is high enough , the enzyme will reach the `` on '' states so frequently that most of the intensity readings are very high , making it impossible to reliably determine the individual turnover times . under this situation , it is necessary to directly study the behavior of the raw intensity reading . there are two main sources of the photons generated in the experiment : the weak but perpetual background noise and the strong but short - lived burst . the number of photons received from two different sources can be modeled as two independent poisson processes with different rates . we can use the following equation to represent @xmath120 , the intensity recorded at time @xmath19:@xmath121 where @xmath122 and @xmath123 represent the total number of photons received due to the burst and background noise , respectively , within a length @xmath124 subinterval of @xmath125 ; @xmath126 is the total time that the enzyme system spends at the `` on '' states ( any @xmath29 ) within the time interval @xmath125 . @xmath122 and @xmath123 are independent poisson processes with rates @xmath127 and @xmath128 , respectively . with this representation , we have the following theorem , whose proof is deferred to the . [ theo2]the covariance of the fluorescence intensity is @xmath129 where @xmath130 are the nonzero eigenvalues of the generating matrix @xmath131 defined in ( [ q ] ) , and @xmath132 are constants only depending on @xmath131 . since @xmath133 is a semi - stable matrix [ @xcite ] , it follows that the real parts of all @xmath134 @xmath135 are negative . for a real matrix , the complex eigenvalues along with their eigenvectors always appear in conjugate pairs ; thus , the imaginary parts cancel each other in ( [ covi ] ) and only the real parts are left . therefore , we know according to theorem [ theo2 ] that the covariance of intensity will decay multi - exponentially . a fast - cycle - reset enzyme jumps from state @xmath29 to @xmath50 with little delay . a short burst of photons is released during the enzyme s short stay at @xmath29 . for fast - cycle - reset enzymes , the behavior of the whole system can be well approximated by an alternative system , where only states @xmath50 and @xmath51 @xmath136 exist : the transition rates among the @xmath0 s , among the @xmath2 s , and from @xmath50 to @xmath51 are exactly the same as in the original system , but the transition rate from @xmath51 to @xmath50 is changed from @xmath38 to @xmath137 , since once a transition of @xmath28 occurs , the enzyme quickly moves to @xmath50 . we can thus think of lumping @xmath29 and @xmath50 together to form the alternative system , which has generating matrix@xmath138 @xmath139 is also a negative semi - stable matrix with @xmath140 eigenvalues , one of which is zero . the following theorem details how well the eigenvalues of @xmath139 approximate those of @xmath131 . [ theo3]assume @xmath141 , where @xmath142 are fixed constants , while @xmath10 is large . let @xmath143 @xmath144 denote the nonzero eigenvalues of @xmath139 , then for each @xmath143 , there exists an eigenvalue @xmath130 of @xmath131 such that@xmath145 the other @xmath23 eigenvalues of @xmath131 satisfy@xmath146 the proof is deferred to the . this theorem says that for fast - cycle - reset enzymes with large @xmath10 , the first @xmath147 nonzero eigenvalues of @xmath131 can be approximated by the eigenvalues of @xmath139 , while the other @xmath23 eigenvalues @xmath148 of @xmath131 are of the same order of @xmath10 . since all the eigenvalues have negative real parts , according to ( [ covi ] ) , the terms associated with @xmath148 decay much faster so their contribution can be ignored . thus , we have the following results for the intensity autocorrelation . [ cor3]for fast - cycle - reset enzymes ( @xmath149 ) , @xmath150 where @xmath143 are the nonzero eigenvalues of matrix @xmath139 defined in ( [ k ] ) . [ cor4]for the classic michaelis menten model , where , @xmath151 & k_{1}[s ] \cr k_{2}+k_{-1 } & -k_{2}-k_{-1}}.\ ] ] the only nonzero eigenvalue is @xmath152)$ ] . we thus have , for fast - cycle - reset enzymes,@xmath153)(t-\delta t)}.\ ] ] we have shown in the preceding sections that the autocorrelation of turnover times and the correlation of intensity follow@xmath154 before applying these equations to fit the experimental data , the following problems must be addressed . first , we know so far that the decay patterns must be multi - exponential , but we do not yet know how the eigenvalues are related to the rate constants ( @xmath155 , @xmath38 , @xmath39 , etc . ) and the substrate concentration @xmath5 $ ] , which is the only adjustable parameter in the experiment . second , we do not know the expressions of the coefficients ( @xmath111 and @xmath132 ) . third , we do not know the number of distinct conformations @xmath23 . we only know that it must be large : each enzyme consists of hundreds of vibrating atoms , and , as a whole , it expands and rotates in the 3-dimensional space within the constraint of chemical bonds . we next address these questions before fitting the experimental data . in the enzyme experiments , the transition rates are intrinsic properties of the enzyme and the enzyme substrate complex ; they are not subject to experimental control . the only variable subject to experimental control is the concentration of the substrate molecules @xmath5 $ ] . the higher the concentration , the more likely that the enzyme molecule could bind with a substrate molecule to form a complex . this is why the association rate @xmath156 $ ] ( the rate of @xmath41 ) is proportional to the concentration . the experiments were repeated under different concentrations , resulting in different decay patterns of the autocorrelation functions as in figure [ fig2 ] . a successful theory should be able to explain the relationship between concentration and autocorrelation decay pattern . the concentration only affect the transition rates between @xmath50 and @xmath157 , which are denoted by @xmath158 in ( [ qsub ] ) . define @xmath159 , which is independent of @xmath5 $ ] ; then @xmath160\mathbf{\tilde{q}}_{ab}$ ] . to delineate the relationship between @xmath5 $ ] and the autocorrelation decay pattern , we next simplify the generating matrices . below are four scenarios that we will consider . each of the scenarios guarantees the classical michaelis menten equation , a hyperbolic relationship between the reaction rate and the substrate concentration,@xmath161}{[s]+c } \qquad\mbox{with some constant } c,\ ] ] which was observed in both the traditional and single - molecule enzyme experiments [ see @xcite , @xcite and @xcite for detailed discussion ] . each scenario has its own biochemical implications . _ scenario _ 1 . there are no or negligible transitions among the @xmath50 states , that is , @xmath162 for @xmath32 . _ scenario _ 2 . there are no or negligible transitions among the @xmath163 states , that is , @xmath164 for @xmath32 . scenarios 1 and 2 correspond to the so - called slow fluctuating enzymes ( whose conformation fluctuates slowly over time ) . _ scenario _ 3 . the transitions among the @xmath50 states are much faster than the others , that is , @xmath165 and the scale @xmath166 is much larger than other transition rates . _ scenario _ 4 . the transitions among the @xmath51 states are much faster than the others , that is , @xmath167 and the scale @xmath166 is much larger than other transition rates . scenarios 3 and 4 correspond to the so - called fast fluctuating enzymes ( whose conformations fluctuate fast ) . in the previous work [ @xcite ] , there are two other scenarios , which can also give rise to the hyperbolic relationship ( hyper ) : primitive enzymes , whose dissociate rate is much larger than their catalytic rate [ @xcite , @xcite , @xcite ] , and conformational - equilibrium enzymes , whose energy - barrier difference between dissociation and catalysis is invariant across conformations [ @xcite ] . but our analysis based on those two scenarios does not lead to any meaningful conclusion , so we omit them here . based on the four scenarios , we have the following theorem for autocorrelation of turnover times . [ theo4]for enzymes with fast cycle reset , the transition probability matrix governing the autocorrelation of turnover times is@xmath168 its eigenvalues @xmath110 , under the four different scenarios , satisfy the following : \1 . @xmath110 do not depend on @xmath5 $ ] , the substrate concentration . thus , the autocorrelation decay should be similar for all concentrations . @xmath110 depend on @xmath5 $ ] hyperbolically . more precisely , if we use @xmath169)$ ] @xmath115 to emphasize the dependence of the eigenvalues on @xmath5 $ ] , we have@xmath170)=\frac{1}{1-(1-\lambda_{i}^{-1}(1))/[s]}.\ ] ] thus , the autocorrelation decay should be slower under larger concentration . 3 4 . the nonone eigenvalues are of order @xmath171 , so the autocorrelation should decay extremely fast for all concentrations . this theorem tells us that for fast fluctuation enzymes ( scenarios 3 or 4 ) , the turnover time correlation tends to be zero . intuitively , this is because the fast fluctuation enzymes prefer conformation fluctuation rather than going through the binding - association - catalytic path that leads to the product , so in a single turnover event , the enzyme undergoes intensive conformation changes , which effectively blurs the information on the reaction path carried by the turnover time , resulting in zero correlation . under scenario 1 , the autocorrelation decay pattern does not vary when the concentration changes . this is because when the enzyme does not fluctuate , it goes from @xmath50 to @xmath51 directly , and the change of concentration consequently does not alter the distribution of the reaction path . thus , the correlation between turnover times does not depend on the concentration . the result from scenarios 1 , 3 or 4 contradicts the experimental finding : correlation exists between the turnover time and is stronger under higher concentration ( see figure [ fig2 ] ) . only scenario 2 fully agrees with the experiments , suggesting that the enzyme substrate complex ( @xmath51 ) does not fluctuate much . this is supported by recent single - molecule experimental findings [ @xcite , @xcite , @xcite ] where slow conformational fluctuation in the enzyme substrate complexes were observed . we now consider the intensity autocorrelation under each of the four scenarios . we write @xmath172 , where @xmath173 , and @xmath174 , where @xmath175 . for scenarios 1 and 2 , we assume that both the enzyme and the enzyme substrate complex fluctuate slowly : @xmath33 and @xmath176 @xmath177 are negligible , but the sums @xmath178 and @xmath179 are not . furthermore , we assume that in scenario 1 the enzyme fluctuation is much slower than the enzyme substrate complex fluctuation ( so @xmath180 and @xmath181 in scenario 1 ) , and in scenario 2 the enzyme substrate complex fluctuation is much slower ( so @xmath182 and @xmath183 scenario 2 ) . [ theo5]for enzymes with fast cycle reset , the matrix governing the intensity autocorrelation is @xmath139 . its eigenvalues and the autocorrelation decay , under the four different scenarios , satisfy the following : 1 ( @xmath180 and @xmath184 ) . the autocorrelation decay is slower under lower concentration , and the dominating eigenvalues are given by@xmath185k_{1i}-\beta_{ii } + k_{-1i}+k_{2i})+\sqrt{([s]k_{1i}-\beta_{ii } + k_{-1i}+k_{2i})^{2}+4\beta_{ii } [ s]k_{1i } } \bigr).\ ] ] 2 ( @xmath182 and @xmath186 ) . the autocorrelation decay is faster under lower concentration , and the dominating eigenvalues are given by@xmath187k_{1i}-\alpha_{ii } + k_{-1i}+k_{2i})\\ & & \hspace*{9pt}{}+\sqrt { ( [ s]k_{1i}-\alpha_{ii } + k_{-1i}+k_{2i})^{2}+4\alpha _ { ii}(k_{-1i}+k_{2i})}\bigr).\end{aligned}\ ] ] \3 . the autocorrelation decay does not depend on the concentration . \4 . the autocorrelation decay is slower under lower concentration . the proof of the theorem is given in the . our results of the dependence of turnover time autocorrelation and fluorescence intensity autocorrelation on the substrate concentration show that in order to have slower decay under higher substrate concentration ( as seen in figure fig2 ) , fluctuation of both the enzyme and the enzyme substrate complex can not be fast ; furthermore , the fluctuation of the enzyme substrate complex needs to be slower than the fluctuation of the enzyme . in summary , each of the four scenarios yields a different autocorrelation pattern , but only the one under scenario 2 matches the experimental finding . therefore , we will focus on scenario 2 from now on . to simplify the coefficients @xmath111 and @xmath132 and to address the number of distinct conformations @xmath23 , we adopt the idea in the previous work [ @xcite , @xcite ] by utilizing a continuous limit . first , we let @xmath188 and in this way model the transition rates as continuous variables with certain distributions . consequently , we treat the eigenvalues also as continuous variables . second , we assume that all the coefficients ( @xmath111 and @xmath132 ) are proportional to the probability weight of the conjugate eigenvalues . this assumption is partly based on the fact that all the observed experimental correlations are positive . with these two assumptions , the covariance can be represented by@xmath189 where @xmath190 and @xmath191 are the corresponding distribution functions . @xmath192 and @xmath193 are functions of the transition rates . since the transition rates are always positive , a natural choice is to model the transition rates as either constants or following gamma distributions . in the previous work [ @xcite ] on the stochastic network model , the association rate @xmath7 and dissociation rate @xmath8 are modeled as constants while the catalytic rate @xmath9 follows a gamma distribution @xmath194 . we adopt them in our fitting . ) and ( [ kappa ] ) . right : the corresponding distributions of the eigenvalues @xmath195 and @xmath196 . ] we know from section [ sec41 ] that scenario 2 matches the experimental finding , so we take @xmath182 and @xmath183 . then the eigenvalue @xmath192 ( based on theorem [ theo4 ] and its proof in the ) is given by@xmath197k_{1}k_{2})},\ ] ] and the eigenvalue @xmath193 ( based on theorem [ theo5 ] ) is@xmath198k_{1}+\alpha^{\ast}+k_{-1}+k_{2})\nonumber\\[-8pt]\\[-8pt ] & & \hspace*{9pt}{}+\sqrt{([s]k_{1}+\alpha^{\ast}+k_{-1}+k_{2})^{2}-4\alpha^{\ast } ( k_{-1}+k_{2})}\bigr),\nonumber\end{aligned}\ ] ] where @xmath199 stands for a generic @xmath200 ( since we are taking the continuous version ) . for the distribution of @xmath199 ( i.e. , the distribution of @xmath200 ) , we note that , first , its support should be the positive real line , and , second , @xmath201 is a sum of many random variables @xmath33 from a common distribution , so we expect that the distribution of @xmath202 should be infinitely divisible . these two considerations lead us to assume a gamma distribution @xmath203 for @xmath202 . the data available to us include the intensity correlation under three concentrations : @xmath5=$ ] 380 , 100 and 20 @xmath204 ( micro molar ) , and turnover time autocorrelation under two concentrations @xmath5=$ ] 100 and 20 @xmath205 . we calculated eigenvalues based on ( [ lam ] ) and ( [ kappa ] ) , where @xmath7 and @xmath8 are constants , @xmath199 and @xmath9 follow distributions @xmath203 and @xmath194 , respectively . the parameters of interests are @xmath206 and @xmath207 . the best fits are found through minimizing the square distance between the theoretical and observed values . the parameters are estimated as follows : @xmath208 @xmath209 , @xmath210 @xmath211 , @xmath212 , @xmath213 @xmath211 , @xmath214 , and @xmath215 @xmath211 ( @xmath124 stands for second ) . figure [ fig3 ] shows the fitting of the autocorrelation functions and the distributions of the eigenvalues . figure [ fig3 ] shows that our model gives a good fit to the turnover time autocorrelation and an adequate fit to the fluorescence intensity autocorrelation , capturing the main trend in the intensity autocorrelation . the distributions of the eigenvalues in the right panels clearly indicate that higher substrate concentration corresponds to larger eigenvalues , which are then responsible for the slower decay of the autocorrelations . our model thus offers an adequate explanation of the observed decay patterns of the autocorrelation functions . the stochastic network model tells us why the decay must be multi - exponential . it further explains why the decay is slower under higher substrate concentration . our consideration of the different scenarios also provides insight on the enzyme s conformational fluctuation : slow fluctuation , particularly of the enzyme substrate complex , gives rise to the experimentally observed autocorrelation decay pattern . in this article we explored the stochastic network model previously developed to account for the empirical puzzles arising from recent single - molecule enzyme experiments . we conducted a detailed study of the autocorrelation function of the turnover time and of the fluorescence intensity and investigate the effect of substrate concentration on the correlations . our analytical results show that ( a ) the stochastic network model gives multi - exponential autocorrelation decay of both the turnover times and the fluorescence intensity , agreeing with the experimental observation ; ( b ) under suitable conditions , the autocorrelation decays more slowly with higher concentration , also agreeing with the experimental result ; ( c ) the slower autocorrelation decay under higher concentration implies that the fluctuation of the enzyme substrate complex should be slow , corroborating the conclusion from other single - molecule experiments [ @xcite , @xcite , @xcite ] . in addition to providing a theoretical underpinning of the experimental observations , the numerical result from the model fits well with the experimental autocorrelation as seen in section [ sec4 ] . some problems remain open for future investigation : \(1 ) when we discussed the dependence of intensity autocorrelation on substrate concentration in section [ sec41 ] , we approximated the fluctuation transition matrix with its diagonal entries . this simple approximation provides useful insight into the decay pattern under different concentration . a better approximation that goes beyond the diagonal entries is desirable . it might lead to a better fitting to the experimental data . \(2 ) we used gamma distribution to model the transition rates . this is purely statistical . can it be derived from a physical angle ? if so , the connection not only will lead to better estimation , but also provides new insight into the underlying mechanism of the enzyme s conformation fluctuation . \(3 ) we used the continuous limit @xmath188 to do the data fitting so that the number of parameters reduces from more than @xmath22 to a manageable six . obtaining the standard error for the estimates is open for future investigation . the main difficulties are the lack of tractable tools to approximate the standard error of the autocorrelation estimates and the challenge to carry out a monte carlo estimate ( @xmath23 needs to be quite large for an ad hoc monte carlo simulation , but such an @xmath23 will bring back a large number of unspecified parameters ) . single - molecule biophysics , like many newly emerging fields , is interdisciplinary . it lies at the intersection of biology , chemistry and physics . owing to the stochastic nature of the nano world , single - molecule biophysics also presents statisticians with new problems and new challenges . the stochastic model for single - enzyme reaction represents only one such case among many interesting opportunities . we hope this article will generate further interest in solving biophysical problems with modern statistical methods ; and we believe that the knowledge and tools gained in this process will in turn advance the development of statistics and probability . proof of lemma [ lem4 ] using the first - step analysis , we have@xmath216 where@xmath217 only the diagonal elements of @xmath218 are negative , and its row sums are either 0 or negative . thus , @xmath219 is a stable matrix [ @xcite ] , which always has an inverse . thus , we have @xmath220 for @xmath72 , similarly , we have@xmath221 proof of lemma [ lem5 ] for @xmath222 , when the first - step analysis is applied , the first - step probability should be conditioned on the exit state @xmath30 , that is , @xmath223 returns to @xmath65 first @xmath224 exit at @xmath225 returns to @xmath65 first@xmath226 . thus , we have the following equation : @xmath227\frac{p_{{\mathit{es}}_{i}e_{j}^{0}}}{p_{e_{i}e_{j}^{0}}}e(t_{{\mathit{es}}_{i}e_{j}^{0}})+\sum_{k\neq i}\alpha _ { ik}\frac{p_{e_{k}e_{j}^{0}}}{p_{e_{i}e_{j}^{0}}}e(t_{e_{k}e_{j}^{0}})\biggr)\\ & & { } \bigg /\biggl ( k_{1i}[s]+\sum_{k\neq i}\alpha_{ik}\biggr ) .\end{aligned}\ ] ] similar expression can be derived for @xmath228 . together we have@xmath229 proof of theorem [ theo1 ] @xmath230 . the first term @xmath231 can be expressed as@xmath232 that is , the system starts the first turnover event from @xmath50 , ends it in @xmath30 , then starts the second from @xmath65 , repeats this procedure for @xmath233 times , and finally starts the last turnover from @xmath234 . note that @xmath235_{n\times n}=(\mathbf{p}_{ac}\mathbf { p}_{ca})^{m-2}$ ] . thus , using the matrices defined in lemmas [ lem1 ] to [ lem5 ] , we have@xmath236 applying the results of lemmas [ lem4 ] and [ lem5 ] and the facts that @xmath237 and @xmath238 , we can finally arrange the covariance as@xmath98\bolds{\mu}_{a}.\ ] ] proof of corollary [ cor1 ] we only need to prove that @xmath239 and @xmath61 are , respectively , the right and left eigenvectors of @xmath100 associated with the eigenvalue 1 . the first is a direct consequence of the fact that @xmath100 is a stochastic matrix . the second can be verified by observing that@xmath240 through ( [ eqv ] ) . proof of theorem [ theo2 ] in ( [ i ] ) , the second term @xmath241 represents the independent background noise during period @xmath242 . thus,@xmath243-e[n_{t}(t_{\mathrm{on}}(t))]e[n_{0}(t_{\mathrm{on}}(0 ) ) ] \\ & = & \nu^{2}[e(t_{\mathrm{on}}(t)t_{\mathrm{on}}(0))-e(t_{\mathrm{on}}(t))e(t_{\mathrm{on}}(0))].\end{aligned}\ ] ] let @xmath244 be the set of all possible states . let @xmath245 be the process evolving according to ( [ 2by2 ] ) . let @xmath246 be the equilibrium probability of state @xmath95 and @xmath247 be the transition probability from state @xmath95 to state @xmath248 after time @xmath124 . we have@xmath249 & & \hspace*{78pt}{}\times e\bigl(t_{\mathrm{on}}(t)|x_{t-\delta t}=k , x_{t}=l\bigr ) \\[-2pt ] & & \qquad=\sum_{i , j , k , l\in\mathcal{s}}\pi_{i}p_{ij}(\delta t)p_{kl}(\delta t)e\bigl(t_{\mathrm{on}}(0)|x_{-\delta t}=i , x_{0}=j\bigr ) \\[-2pt ] & & \hspace*{32pt}\qquad\quad{}\times e\bigl(t_{\mathrm{on}}(t)|x_{t-\delta t}=k , x_{t}=l\bigr)\ { p_{jk}(t-\delta t)-\pi _ { k}\}.\end{aligned}\ ] ] the probability transition matrix @xmath250_{3n\times3n}$ ] is the matrix exponential of the generating matrix ( [ 2by2 ] ) : @xmath250_{3n\times 3n}=\exp(\mathbf{q}t)$ ] . zero is an eigenvalue of @xmath131 with right eigenvector @xmath239 and left eigenvector @xmath251 , the stationary distribution . assume @xmath131 is diagonalizable . let @xmath130 , @xmath252 , denote the other eigenvalues , and @xmath253 and @xmath254 be the corresponding right and left eigenvectors . we have @xmath255 therefore , we can rewrite@xmath256 to prove theorem [ theo3 ] , we need the following two useful lemmas on the eigenvalues of a matrix . [ lema ] let @xmath257\in m_{n}$ ] , where @xmath258 is the set of all complex matrices . let @xmath259 $ ] be given and define @xmath260 and @xmath261 as the deleted row and column sums of @xmath262 , respectively,@xmath263 then , ( 1 ) all the eigenvalues of @xmath262 are located in the union of @xmath23 discs@xmath264 ( 2 ) furthermore , if a union of @xmath265 of these @xmath23 discs forms a connected region that is disjoint from all the remaining @xmath266 discs , then there are precisely @xmath265 eigenvalues of @xmath262 in this region . [ lemb ] let @xmath262 and @xmath267 be matrices with elements satisfying @xmath268 . if @xmath269 is a simple eigenvalue ( i.e. , an eigenvalue with multiplicity 1 ) of @xmath262 , then for matrix @xmath270 , where @xmath271 is sufficiently small , there will be a eigenvalue @xmath272 of @xmath270 such that@xmath273 furthermore , if we know that one eigenvector of @xmath262 associated with @xmath274 is @xmath275 , then there is an eigenvector @xmath276 of @xmath270 associated with @xmath277 such that@xmath278 note that since dividing a matrix by a constant only changes the eigenvalues with the same proportion , the condition that the entries of @xmath262 and @xmath267 are bounded by 1 can be relaxed to that the entries of @xmath262 and @xmath267 are bounded by a finite positive number . proof of theorem [ theo3 ] according to lemma [ lema ] , all the eigenvalues of @xmath131 must lie in the union of discs centered at @xmath279 with radii defined by ( [ radius ] ) . if we take @xmath280 in ( [ radius ] ) , then the first @xmath23 discs corresponding to the diagonal entries of @xmath281 have centers @xmath282 and radii @xmath283 ; the second @xmath23 discs corresponding to the diagonal entries of @xmath284 have centers @xmath282 and radii @xmath282 ; the third @xmath23 discs corresponding to the diagonal entries of @xmath285 have centers @xmath286 and radii @xmath283 . thus , for @xmath10 large enough , the union of the first @xmath140 discs does not overlap with the union of the last @xmath23 discs , so we know from lemma [ lema ] that @xmath131 has @xmath140 eigenvalues with order @xmath283 in the union of the first @xmath140 discs and @xmath23 other eigenvalues with order @xmath286 in the union of the last @xmath23 discs . for the @xmath23 eigenvalues with order @xmath286 , consider the following two matrices:@xmath287 , \\ \mathbf{z}&=&\left [ \matrix { \mathbf{q}_{aa}-\mathbf{q}_{ab } & \mathbf{q}_{ab } & \mathbf{0 } \cr \mathbf{q}_{ba } & \mathbf{q}_{bb}-\mathbf{q}_{ba}-\mathbf{q}_{bc } & \mathbf{q}_{bc } \cr \mathbf{0 } & \mathbf{0 } & \mathbf{q}_{cc } } \right ] .\end{aligned}\ ] ] we have @xmath288 . zero is an eigenvalue of @xmath289 with multiplicity @xmath140 , and the other @xmath23 eigenvalues of @xmath289 are @xmath290 . for large @xmath10 , according to lemma [ lemb ] , there exists @xmath23 eigenvalues of @xmath291 that satisfy@xmath292 that is,@xmath293 now for the @xmath147 nonzero eigenvalues of @xmath131 with order @xmath294 , they are the solutions of@xmath295 for large @xmath10 , the matrix @xmath296 is invertible , since it is strictly diagonal dominated . we can decompose the determinant as@xmath297 where@xmath298 .\ ] ] therefore , @xmath130 , @xmath299 , is also the eigenvalue of the matrix@xmath300 = \mathbf{k}+\mathbf{s}\ ] ] with@xmath301 .\ ] ] we note that @xmath302 , where @xmath303 . since @xmath304 is of the order @xmath286 and @xmath130 is of the order @xmath283 , the entries of @xmath305 are of the order @xmath306 , so are the entries of @xmath307 . applying lemma [ lemb ] to @xmath308 tells us that for each @xmath130 there must be an eigenvalue @xmath143 of @xmath139 , which has the property that@xmath309 proof of theorem [ theo4 ] we know from lemma [ lem1 ] that @xmath310 ^{-1}.\ ] ] _ scenario _ 1 . when @xmath180 , @xmath311 , so the eigenvalues and eigenvectors of @xmath312 have nothing to do with @xmath5 $ ] . _ scenario _ 2 . when @xmath182 , @xmath313}\mathbf{q}_{bc}^{-1}(\mathbf{q}_{ba}+\break\mathbf { q}_{bc})\mathbf{\tilde{q}}_{ab}^{-1}\times\mathbf{q}_{aa}$ ] . thus , if @xmath312 has eigenvalue @xmath314 when @xmath5=1 $ ] , then for general @xmath5 $ ] , @xmath315 has eigenvalue @xmath170)=\frac{1}{1-(1-\lambda_{i}^{-1}(1))/[s]}.\ ] ] _ scenario _ 3 . we write @xmath165 , where @xmath316 is large . then @xmath315 is@xmath317.\end{aligned}\ ] ] suppose the eigenvalues of @xmath318 are @xmath319 . then according to lemma [ lemb ] , the eigenvalues of @xmath320 are@xmath321 thus , the eigenvalues of @xmath315 are@xmath322 namely , all the nonone eigenvalues of @xmath315 are of order @xmath171 . _ scenario _ 4 . using an identical method as in scenario 3 , we can show that all the nonone eigenvalues of @xmath315 are of order @xmath323 . proof of theorem [ theo5 ] the matrix @xmath139 can be written as@xmath324 we thus know from lemma [ lemb ] that the eigenvalues of @xmath139 can be approximated by the eigenvalues of @xmath325 . if @xmath326 is invertible , then@xmath327 we know that any eigenvalue @xmath193 of @xmath325 must make the second determinant on the right - hand side zero . this determinant only involves diagonal matrices , so we have@xmath328k_{1i}(k_{2i}+k_{-1i})}{\kappa -\alpha_{ii}+[s]k_{1}}.\ ] ] if @xmath329 is not invertible , then there is at least one @xmath248 so that @xmath330 $ ] . but it can be verified that in order to make @xmath193 an eigenvalue of @xmath325 , there must exit another @xmath331 such that ( [ root ] ) holds for this @xmath193 . therefore , any eigenvalue must be a root of ( [ root ] ) . equation ( [ root ] ) has two negative roots for each @xmath95 , but we only need to consider the root closer to 0 , since it dominates the decay . _ scenario _ 1 . the root is@xmath185k_{1i}-\beta_{ii } + k_{-1i}+k_{2i})+\sqrt{([s]k_{1}-\beta_{ii } + k_{-1i}+k_{2i})^{2}+4\beta_{ii } [ s]k_{1}}\bigr),\ ] ] which is monotone decreasing in @xmath5 $ ] . _ scenario _ 2 . @xmath333 . the root is@xmath187k_{1i}-\alpha_{ii}+k_{-1i}+k_{2i})\\ & & \hspace*{9pt}{}+\sqrt { ( [ s]k_{1}-\alpha_{ii}+k_{-1i}+k_{2i})^{2}+4\alpha _ { ii}(k_{-1i}+k_{2i})}\bigr),\end{aligned}\ ] ] which is monotone increasing in @xmath5 $ ] . _ scenario _ 3 . @xmath334 , where @xmath316 is large . following the same method as we used in the proof of theorem [ theo3 ] , we can show that @xmath23 eigenvalues of @xmath139 are of the order @xmath335 and they will not contribute much to the correlation . the other eigenvalues governing the decay pattern can be approximated by the eigenvalues of @xmath336 , which do not depend on the concentration @xmath5 $ ] . _ scenario _ 4 . @xmath337 , where @xmath316 is large . using the same method as in the proof of theorem [ theo3 ] , we can show that the dominating eigenvalues of @xmath139 can be approximately by the eigenvalues of @xmath281 . since we know that @xmath338 , the eigenvalues of @xmath281 is approximately @xmath339k_{1i } $ ] , which is monotone decreasing in @xmath5 $ ] . the authors thank the xie group at the department of chemistry and chemical biology of harvard university for sharing the experimental data .

let be a graph whose each component has order at least 3 . let for some integer be an improper edge coloring of ( where adjacent edges may be assigned the same color ) . if the induced vertex coloring defined by ( where the indicated sum is computed in and denotes the set of all edges incident to ) results in a proper vertex coloring of , then we refer to such a coloring as an improper twin-edge coloring . the minimum for which has an improper twin-edge coloring is called the improper twin chromatic index of and is denoted by . + it is known that , unless and in this case . in this paper , we first give a short proof of this result and we show that if admits an improper twin-edge coloring for some positive integer , then admits an improper twin-edge coloring for all ; we call this the monotonicity property . in addition , we provide a linear time algorithm to construct an improper twin edge coloring using at most colors , whenever a-vertex coloring is given . then we investigate , to the best of our knowledge the first time in literature , the complexity of deciding whether or , and we show that , just like in case of the edge chromatic index , it is np - hard even in some restricted cases . lastly , we exhibit several classes of graphs for which the problem is polynomial . modular chromatic index , twin edge - vertex coloring , odd - even color classes , np - hardness

1601.02267

throughout this paper all graphs are simple with no loops and multiple edges . let @xmath0 be a graph . we denote the vertex set and the edge set of @xmath0 , by @xmath17 and @xmath18 , respectively ( or simply @xmath19 and @xmath20 when no ambiguity may occur ) . for @xmath21 , @xmath6 denotes the set of all edges incident to @xmath7 and @xmath22 denotes the ( open ) neighborhood of @xmath7 in @xmath0 ; when no ambiguity may occur we just denote @xmath23 . we denote respectively by @xmath24 , @xmath25 and @xmath26 the complete graph of order @xmath27 , the path or order @xmath27 and the cycle of order @xmath27 . @xmath8-clique _ is a set of @xmath8 vertices inducing a complete graph and an independent set is a set of vertices pairwise not linked by an edge ( i.e. , inducing a subgraph with only isolated vertices ) . @xmath28 will denote the complete bipartite graph with parts of size @xmath29 and @xmath30 . if @xmath31 , then it is star graph and the vertex of the part of size 1 is called center . a _ ( proper ) @xmath8-vertex coloring _ of @xmath0 is an assignment of @xmath8 colors to the vertices of @xmath0 such that no two adjacent vertices have the same color ( possibly some colors are not used ) . given such a @xmath8-vertex coloring , a color class is the set of vertices of the same color ; it is an independent set . a graph having a @xmath8-vertex coloring is said @xmath8-vertex colorable . the _ chromatic number _ @xmath32 of @xmath0 is the minimum @xmath8 for which @xmath0 is @xmath8-vertex colorable . a _ @xmath8-edge coloring _ of @xmath0 is an assignment of @xmath8 colors to the edges of @xmath0 ( possibly some colors are not used ) ; it is said to be _ proper _ if two edges with a common endpoint do not have the same color , and _ improper _ if it is not necessarily proper . note that the distinction between proper and improper colorings exists as well for vertex colorings but since we will not use it ( formally since all our vertex colorings are proper ) we specify proper / improper for edge colorings only . the minimum @xmath8 such that @xmath0 admits a proper @xmath8-edge coloring is the _ chromatic index _ of @xmath0 denoted by @xmath33 . integers will be used to label colors , which will give us the opportunity to compute operations . a _ walk _ in @xmath0 is a sequence @xmath34 , whose terms are alternately vertices and edges , such that @xmath35 , for @xmath36 . the number @xmath8 is called the length of the walk . a walk is called a _ closed walk _ if @xmath37 . the reader is referred to @xcite for all definitions in graph theory which are not defined here . for any integer @xmath2 , we will denote by @xmath38 the additive group over @xmath39 . for @xmath40 in @xmath5 , @xmath41 in @xmath5 is also denoted @xmath42 . for any two integers , we will denote @xmath43 if @xmath44 . in the literature , we can find a number of results about proper or improper edge colorings _ inducing _ a coloring for vertices by mapping , for any vertex @xmath7 , the multiset of colors of the edges adjacent to @xmath7 to a vertex color . such a mapping is called _ vertex - coloring mapping _ ; see @xcite for variations of such colorings . given a vertex - coloring mapping , an edge - coloring is called _ vertex - distinguishing _ if the colors induced by the edge coloring are different for every vertex . the edge - coloring is called _ neighbor - distinguishing _ if only adjacent vertices have different induced colors , see e.g. @xcite and @xcite . neighbor - distinguishing edge colorings are also called _ vertex coloring edge partitions _ in @xcite . chapter 13 of @xcite is devoted to distinguishing colorings . these notions strongly vary with respect to the vertex coloring mapping , i.e. , the way the vertex colors are induced from the edge colors . for instance , in @xcite , the color of the vertex @xmath7 is directly the multiset of the colors in @xmath6 while in @xcite , the color of a vertex is obtained as the sum of the colors of the edges incident to it and the related parameter is called _ sum chromatic number_. this definition gives rise to the `` 1 - 2 - 3 conjecture '' which states that for every connected graph @xmath0 of order at least 3 , there exists a neighbor - distinguishing edge coloring of @xmath0 using only 3 colors . a variation of the sum chromatic number obtained using abelian groups is called the _ group sum chromatic number _ and studied in @xcite ; it is the smallest value @xmath45 such that taking any abelian group @xmath46 of order @xmath45 , there exists a function @xmath47 such that the sums of edge colors properly color the vertices i.e. , with different colors on adjacent vertices . another popular vertex coloring mapping uses the specific group @xmath5 , where @xmath8 is at least the number of colors used for the edges . if the edges are colored using colors between 0 and @xmath48 , seen as elements of @xmath5 , then one assigns to a vertex @xmath7 the sum in @xmath5 of the colors of the edges in @xmath6 . more formally , let @xmath0 be a graph the components of which have order at least 3 . let @xmath1 for some integer @xmath2 be a @xmath8-edge coloring of @xmath0 . we then define @xmath49 as follows : @xmath50 if @xmath51 is a ( proper ) @xmath8-vertex coloring , then it will be called the _ vertex coloring induced by @xmath45 _ and @xmath45 is said to _ induce _ @xmath51 . a proper edge coloring @xmath45 inducing a vertex coloring @xmath51 is called a _ proper twin @xmath8-edge coloring _ of @xmath0 . this concept was defined in @xcite where it has been shown that any graph with components of size at least 3 has at least one such coloring . then , the minimum @xmath8 for which @xmath0 has a proper twin @xmath8-edge coloring is called the _ proper twin chromatic index of _ @xmath0 and is denoted by @xmath52 . similarly , an improper ( i.e. not necessarily proper ) edge coloring @xmath45 inducing a vertex coloring @xmath51 is called an _ improper twin @xmath8-edge coloring _ of @xmath0 . the difference between both notions is only whether one deals with proper or improper edge colorings , but the related vertex coloring is always required to be proper . since a proper edge coloring is improper as well , the existence of improper twin @xmath8-edge colorings for some @xmath8 comes from the existence of proper ones . the minimum @xmath8 for which @xmath0 has an improper twin @xmath8-edge coloring is called the _ improper twin chromatic index _ of @xmath0 and is denoted by @xmath9 . the same parameter was already studied under the name of _ modular chromatic index _ , denoted by @xmath53 , in @xcite and @xcite . note that @xmath54 and @xmath52 are not defined for graphs having a component which is @xmath55 . following the terminology of @xcite we call _ nice _ a graph without a component of size 2 . however , the definitions are valid for graphs with isolated vertices by considering that an empty sum is null and then defining @xmath56 for all isolated vertices @xmath7 and all twin @xmath8-edge colorings , @xmath2 . all the graphs considered from now on will be supposed to be nice . by definition , for every nice graph @xmath0 we have : @xmath57 . in this paper , we deal with improper twin @xmath8-edge colorings of @xmath0 . note that although they are very close , the group sum chromatic number and the improper twin chromatic index are not the same . in fact , for a graph @xmath0 , @xmath58 does not exceed the group sum chromatic number of @xmath0 . here is an example where the group sum chromatic number is greater than the improper twin chromatic index . consider the disjoint union of @xmath59 and @xmath55 and make one vertex of @xmath59 adjacent to one vertex of @xmath55 and call the resultant graph by @xmath0 . note that @xmath0 has order 6 with degree sequence : 4,3,3,3,2,1 . the group sum chromatic number of @xmath0 is 5 since this graph is ugly ( see definition 1.1 of @xcite ) and ugly graphs have group sum chromatic number of @xmath60 ( see theorem 1.2 of @xcite ) . however , the improper twin chromatic index of @xmath0 is 4 by theorem [ thm : mainfromjcmcc ] reminded below . let us conclude the definitions with a concrete example of improper twin edge coloring . let @xmath61 be the petersen graph . since @xmath61 contains an odd cycle , @xmath62 . the assignment of numbers to @xmath63 given in figure [ example ] shows that @xmath64 . let us now first underline that some natural properties of the classical edge and vertex colorings do not hold for the improper twin edge coloring problem . [ rem0 ] the notions of improper twin edge coloring and induced vertex coloring strongly depend on the labels assigned to the colors . in particular they are not stable under a reordering of the colors . consider indeed the star graph @xmath65 ; the 2-vertex coloring assigning 0 to the center and 1 to the two other vertices is induced by the edge coloring assigning 1 to both edges while the 2-vertex coloring assigning 1 to the center and 0 to the two other vertices is not induced by any improper twin edge coloring . actually this 2-edge coloring is the unique improper 2-edge coloring . similarly , considering a @xmath66 , the edge coloring assigning color 1 to two adjacent edges and 0 to the third edge is an improper 2-edge coloring while the edge coloring assigning 0 to two adjacent edges and 1 to the last edge is not . [ rem1 ] unlike the usual vertex / edge coloring problems , there are some graphs @xmath0 with a subgraph @xmath67 such that @xmath68 . indeed , using theorem [ thm : mainfromjcmcc ] ( see below ) , we consider @xmath69 and @xmath70 with @xmath71 ; then @xmath72 and @xmath73 . taking @xmath74 and @xmath75 raises infinitely many examples where taking a subgraph increases the improper twin chromatic index . [ rem2 ] unlike the usual edge coloring problem , there are some graphs @xmath0 with an improper twin @xmath8-edge coloring @xmath45 such that the restriction of @xmath45 to @xmath76 for some edge @xmath77 is not necessarily an improper twin @xmath8-edge coloring of @xmath78 . indeed , a triangle admits an improper twin 3-edge coloring with colors 0 , 1 and 2 on its three edges , however , removing the edge of color 2 leaves a graph with an edge coloring which does not induce a proper vertex coloring . note however that removing an edge colored with 0 obviously leaves an improper twin @xmath8-edge coloring in the remaining graph . [ rem3 ] unlike the usual vertex / edge coloring problems , an improper twin @xmath8-edge coloring is not necessarily an improper twin @xmath13-edge coloring for @xmath14 . this is due to the @xmath79 or @xmath80 operations respectively used in the definition of twin @xmath13- and twin @xmath8-edge coloring . consider indeed a star with a center vertex @xmath81 and four pending vertices @xmath82 . the edge coloring assigning color 1 to all the four edges is an improper twin 2-edge coloring but not an improper twin 3-edge coloring . in this paper we investigate the improper twin edge chromatic index of graphs and we consider , to the best of our knowledge , the first time , the complexity of a neighbor - distinguishing edge coloring problem . the following summarizes the main contribution of @xcite and @xcite : ( @xcite ) [ thm : mainfromjcmcc ] let @xmath0 be a connected graph of order at least 3 . then @xmath83 if and only if @xmath84 and every proper @xmath32-vertex coloring of @xmath0 results in color classes of odd size ; and @xmath85 in all other cases . it should be noted that this result is obtained as a combination of several results from @xcite and @xcite . in section [ sec : general ] , we provide a shorter and more compact proof of theorem [ thm : mainfromjcmcc ] . actually we obtain a more precise result ( see theorem [ thm : final ] ) describing a constructive way to build in linear time an improper twin edge coloring using @xmath8 or @xmath15 colors , whenever a @xmath8-vertex coloring is provided . in @xcite it is shown that for non - bipartite connected graphs @xmath0 and odd @xmath86 , any @xmath13-vertex coloring can be induced by an improper @xmath13-edge coloring . we give an alternative proof of this result ( proposition [ oddc ] ) and give a necessary and sufficient condition for a @xmath13-vertex coloring with even @xmath13 being induced by an improper twin @xmath13-edge coloring in a non - bipartite connected graph ( proposition [ equal ] ) . moreover , these two results , together with theorem [ bip ] allow us to establish the following _ monotonicity _ property of the improper twin chromatic index for graphs : if a graph @xmath0 admits an improper twin @xmath8-edge coloring for some positive integer @xmath8 , then it admits an improper twin @xmath13-edge coloring for all @xmath14 ( theorem [ thm : at least k ] ) . note that following remark [ rem3 ] , unlike the classical coloring problems , this property is not straightforward for the improper twin edge coloring problem . the characterization of graphs having @xmath87 emphasizes the closely related problem of deciding whether all color classes are of odd size in all optimal colorings . in section [ sec : complexity ] , we use this new problem to show that , interestingly enough , just like in the case of usual edge chromatic index which can only take two possible values , it is np - complete to decide whether @xmath88 even when the graph @xmath0 is restricted to have @xmath32 equal to its maximum clique size and in the bounded degree case . finally , in section [ sec : poly ] , we give first examples of graph classes for which this question can be decided in polynomial time . these classes include bipartite graphs , split graphs , co - chordal graphs and co - comparability graphs with bounded chromatic number . the contribution of this section is two - fold . first , we provide a short and compact proof of the bounds @xmath89 together with a description of the graphs having @xmath90 . second , we prove the monotonicity of the improper twin chromatic index . we first settle both the monotonicity and the bounds for bipartite graphs . more precisely , we show that for every nice bipartite graph @xmath0 and every positive integer @xmath91 , @xmath0 admits an improper twin @xmath13-edge coloring . given a connected graph @xmath92 , a @xmath13-vertex coloring @xmath93 and an improper twin edge coloring @xmath45 using colors in @xmath94 , we will say that @xmath45 _ almost induces @xmath93 _ if there is a vertex @xmath95 such that @xmath96 . in this case @xmath7 is called the _ defective vertex_. note that if @xmath45 induces @xmath93 then it almost induces it and any vertex can be chosen as defective . moreover if @xmath45 is almost inducing @xmath93 , either it induces it or there is a unique possible defective vertex . the following lemma plays a crucial role in our approach . [ lem : tree ] given connected graph @xmath97 , a vertex @xmath13-coloring @xmath93 of @xmath0 ( @xmath98 ) , a vertex @xmath95 and a spanning tree @xmath99 , there is an improper twin edge coloring @xmath45 that almost induces @xmath93 with @xmath7 as defective vertex . moreover @xmath100 . we direct the edges of @xmath99 from the root to the leaves using @xmath7 as the root . we set @xmath101 . then , starting from the deepest level of @xmath99 and going up to the root , color the edges @xmath77 of @xmath99 with colors @xmath102 in @xmath103 to ensure that for every vertex @xmath104 , the color of @xmath105 corresponds to the sum in @xmath94 of the colors of edges of @xmath99 incident to @xmath105 . since @xmath106 for the edges @xmath107 , it means that @xmath108 . in the configuration described in lemma [ lem : tree ] , we will say that @xmath45 is _ associated with @xmath99 and @xmath7_. note that this notion is not unique : several different improper twin @xmath13-edge colorings can be associated with the same tree @xmath99 and @xmath7 . [ bip ] if @xmath109 is a nice bipartite graph with vertex parts @xmath110 and @xmath111 , then the following statements hold : 1 . @xcite @xmath112 if and only if @xmath0 does not have any connected component with both parts of the bipartition of odd size . 2 . for any integer @xmath91 the graph @xmath0 admits an improper twin @xmath13-edge coloring . without loss of generality , suppose that @xmath0 is connected with at least 3 vertices . for a vertex @xmath113 consider a spanning tree @xmath99 of @xmath0 and direct the edges from the root of the leaves using @xmath7 as the root . set @xmath114 for all vertices @xmath115 in the same part of the bipartition as @xmath7 and @xmath116 for all vertices in the other part . using lemma [ lem : tree ] , we consider an improper twin @xmath13-edge coloring @xmath45 of @xmath0 associated with @xmath99 and @xmath7 . @xmath117 consider @xmath118 . without loss of generality we assume that @xmath119 is even and then , we chose @xmath120 . we then have : @xmath121 where the first summation is the sum of an even number of 1 s and the second summation is the sum of 0 s . on the other hand , by construction of @xmath93 we have : @xmath122 using relations [ eq : bip ] and [ eq : bip2 ] we deduce @xmath123 . so , @xmath124 and @xmath45 is an improper twin 2-edge coloring of @xmath0 . to prove the other direction , we proceed by contradiction : assume that @xmath119 and @xmath125 are both odd and we have an improper twin 2-edge coloring of @xmath0 . since @xmath0 is connected , @xmath126 is the unique bipartition of @xmath0 which should be also induced by the improper twin 2-edge coloring . by double - counting the color of all edges as previously , we obtain a contradiction to the fact that both @xmath119 and @xmath125 are odd . + @xmath127 we assume now that @xmath7 is of degree at least 2 in @xmath0 ( @xmath0 is nice ) and consider @xmath99 such that @xmath7 is of degree at least 2 in @xmath99 . let @xmath128 be the colors ( among @xmath129 ) of the edges of @xmath99 incident to @xmath7 ( @xmath130 ) . if @xmath131 , then @xmath45 is an improper twin @xmath13-edge coloring of @xmath99 that almost induces @xmath93 . now , if @xmath132 , then we consider two cases . if @xmath133 , then change the color of edge @xmath134 from @xmath135 to @xmath136 and the color of edge @xmath137 from @xmath138 to @xmath139 for some @xmath140 to obtain an improper twin @xmath13-edge coloring @xmath141 . @xmath142 and @xmath143 ( note that @xmath144 ) . if @xmath145 , then change the color of edge @xmath134 from @xmath135 to @xmath146 and change the color of edge @xmath137 from @xmath138 to @xmath147 to obtain an improper twin @xmath13-edge coloring @xmath141 . we have @xmath148 and @xmath149 . note that in the case @xmath127 , the induced @xmath13-vertex coloring has at most 3 vertices with a color greater than 1 . the following is a key lemma for the establishment of the monotonicity property and the bounds for non - bipartite case in a constructive way . [ odd - cycle ] let @xmath92 be a connected non - bipartite graph . let @xmath93 be a @xmath8-vertex coloring of @xmath0 almost induced by an improper twin @xmath8-edge coloring @xmath45 with defective vertex @xmath7 . for any @xmath150 such that @xmath151 , the @xmath8-vertex coloring obtained by replacing @xmath152 with @xmath153 is induced by an improper twin @xmath8-edge coloring . since @xmath154 , the graph @xmath0 contains an odd cycle , say @xmath155 . since @xmath0 is connected , there exists a path between @xmath7 and @xmath155 , say @xmath61 . we construct an odd closed walk @xmath156 passing by @xmath7 as follows : traverse @xmath61 starting at @xmath7 and go to @xmath155 and then traverse @xmath155 and come back to @xmath7 via @xmath61 . note that each edge of @xmath61 is traversed twice . given @xmath150 , add alternately @xmath157 to @xmath102 , for each @xmath77 in @xmath156 starting from @xmath7 and consider the vertex coloring induced by this new edge coloring @xmath141 . all vertices @xmath158 have the same color since the summation of the colors of edges incident to @xmath105 is the same ( @xmath159 ) . vertices outside @xmath156 are not affected at all and the color @xmath152 is changed into @xmath160 . this new vertex coloring is a proper vertex coloring since the color @xmath161 is not used in @xmath22 and it is induced by @xmath141 . from now on , we deal with the non - bipartite case . we first settle the bounds and the monotonicity for all improper twin edge colorings with odd number of colors . lemmas [ lem : tree ] and [ odd - cycle ] give an alternative proof of the following result . [ oddc]@xcite if @xmath0 is a nice graph without bipartite component and @xmath13 is an odd positive integer such that @xmath162 , then any @xmath13-vertex coloring of @xmath0 with colors in @xmath163 can be induced by an improper twin @xmath13-edge coloring of @xmath0 . we assume without loss of generality that @xmath0 is connected since otherwise the same proof applies to every connected component ( note that if the hypotheses hold for @xmath0 they hold for each component ) . let @xmath164 be a @xmath13-vertex coloring of @xmath0 with colors in @xmath163 with @xmath165 . we show that there is an improper twin @xmath13-edge coloring @xmath45 of @xmath0 satisfying @xmath166 for all vertices @xmath167 . consider a spanning tree @xmath99 of @xmath0 and direct the edges from the root of the leaves using @xmath168 as the root . using lemma [ lem : tree ] , we consider an improper @xmath13-edge coloring @xmath45 of @xmath0 associated with @xmath99 and @xmath7 . it satisfies @xmath169 . let @xmath170 be the colors ( among @xmath103 ) of the edges in @xmath171 incident to @xmath168 . we have @xmath172 . if @xmath173 , then @xmath164 is induced by the improper twin @xmath13-edge coloring @xmath45 . otherwise , since @xmath8 is odd ( @xmath174 ) , the equation @xmath175 has a solution @xmath150 . lemma [ odd - cycle ] ensures that we can modify @xmath45 into @xmath141 such that @xmath176 ( replacing @xmath177 by @xmath178 does not change the vertex coloring @xmath164 ) . it completes the proof . note that the non - bipartite condition in proposition [ oddc ] is required . in particular we can not improve theorem [ bip ] ii ) in the sense that , for bipartite components , only some @xmath13-vertex colorings are induced by an improper twin @xmath13-edge twin coloring . indeed , by theorem [ bip ] i ) a 2-vertex coloring of the star graph @xmath179 can not be induced by an improper twin 2-edge coloring . moreover , for any even @xmath180 , the @xmath29-vertex coloring of the star graph @xmath181 assigning color 0 to the center and different colors from 1 to @xmath182 for the other vertices is not induced by an improper twin @xmath29-edge coloring since @xmath183 . for any odd @xmath184 , the @xmath29-vertex coloring of the star graph @xmath181 assigning color 1 to the center and different colors in @xmath185 for the other vertices is not induced by an improper twin @xmath29-edge coloring since @xmath186 . the following proposition will enable us to obtain the monotonicity of the improper twin edge coloring for any number of colors ( theorem [ thm : at least k ] ) and the bounds on @xmath54 for graphs with even chromatic number at least 4 . [ equal ] let @xmath0 be a nice non - bipartite connected graph and @xmath13 an even number such that @xmath187 . then , a @xmath13-vertex coloring @xmath164 with colors in @xmath188 is induced by an improper twin @xmath13-edge coloring if and only if @xmath189 is even . @xmath190 denote @xmath191 and @xmath192 and let @xmath164 be a @xmath13-vertex coloring of @xmath0 with colors in @xmath188 such that @xmath13 and @xmath193 are even . as in the proof of proposition [ oddc ] we consider a spanning tree @xmath99 of @xmath0 and direct the edges from the root of the leaves using @xmath168 as the root . using lemma [ lem : tree ] , we consider an improper @xmath13-edge coloring @xmath45 of @xmath0 associated with @xmath99 and @xmath168 . it satisfies @xmath169 . if @xmath194 , then the proof is complete . else we claim that @xmath195 is even . indeed , by definition of @xmath45 and denoting by @xmath196 the set of edges in @xmath171 not adjacent to @xmath168 we have : @xmath197 since @xmath189 is even , this implies @xmath198 is even and consequently @xmath195 is even as well . we set @xmath199 so @xmath200 and use lemma [ odd - cycle ] to modify @xmath45 in @xmath141 such that @xmath176 . it completes the proof of the first implication . @xmath201 conversely , suppose that @xmath10 . consider an improper twin edge coloring @xmath45 using the colors @xmath202 . this induces a vertex coloring @xmath164 . we have @xmath203 now , since @xmath8 is even , the proof is complete . now , we are ready to show the monotonicity property of the improper twin edge chromatic index for the general case . [ thm : at least k ] let @xmath0 be a nice graph . if @xmath0 admits an improper twin @xmath8-edge coloring ( @xmath2 ) , then @xmath0 admits an improper twin @xmath13-edge coloring for all @xmath14 . we assume without loss of generality that @xmath0 is connected since otherwise the same proof applies to every connected component . the assertion vacuously holds for @xmath204 for all graphs . if @xmath0 is bipartite , the assertion holds by theorem [ bip ] . assume now @xmath0 is not bipartite and @xmath205 . if @xmath13 is odd , then the assertion holds by proposition [ oddc ] . assume @xmath13 is even . obviously , @xmath206 . consider a proper vertex coloring of @xmath0 , say @xmath207 . if the sum @xmath189 is even then proposition [ equal ] allows to conclude . otherwise , consider the maximum color used in @xmath164 , say @xmath208 and change the color of exactly one vertex with color @xmath208 to @xmath209 and call @xmath210 the resultant proper coloring . clearly , @xmath211 is even and apply proposition [ equal ] to conclude the proof . let us finally note that lemma [ odd - cycle ] and its consequence proposition [ equal ] give a comprehensive and direct proof of theorem [ thm : mainfromjcmcc ] in the case where @xmath32 is even of size at least 4 , as stated below . the other cases are immediate consequences of theorem [ bip ] and proposition [ oddc ] for which we gave alternative simple proves as well . moreover , as stated in theorem [ thm : final ] the arguments can be turned into a linear - time algorithm . + if @xmath0 is bipartite , the statement holds by theorem [ bip ] . for a non - bipartite connected graphs @xmath0 of order at least 3 , if @xmath32 is odd then the statement holds by proposition [ oddc ] . to complete the proof of theorem [ thm : mainfromjcmcc ] , it remains to show that the statement holds for @xmath32 even . let us show this under two cases . * case 1 : @xmath212*. then , for some positive integer @xmath213 , there are @xmath214 colors with an even label and the same holds for colors of odd labels . consider a @xmath32-vertex coloring of @xmath0 , say @xmath164 . if all color classes have odd size , then @xmath215 is even and by proposition [ equal ] ( @xmath0 is not bipartite ) we are done . similarly , if all classes have even size we are done as well . thus assume that we have @xmath32 color classes @xmath216 in which , without loss of generality , @xmath217 and @xmath218 have different parity . now , color @xmath219 , arbitrarily using the colors @xmath220 . we have two possibilities for coloring @xmath221 and @xmath222 by two colors @xmath223 and @xmath224 , and obviously for one of them , the total sum of colors of vertices are even and proposition [ equal ] completes the proof . * case 2 : @xmath225*. then there is an odd number of odd colors and the same odd number of even colors in a @xmath32-vertex coloring of @xmath0 . we will show that if there is a color class of even size in some @xmath32-vertex coloring of @xmath0 then @xmath10 , and otherwise , that is if all color classes in all @xmath32-vertex colorings of @xmath0 are of odd size , then @xmath16 . + consider a @xmath32-vertex coloring of @xmath0 , say @xmath164 . if all color classes have even size , then obviously @xmath215 is even and by proposition [ equal ] we are done . if @xmath164 has two color classes , say without loss of generality @xmath222 and @xmath226 whose cardinalities are of different parity , then proposition [ equal ] allows to conclude exactly as in part ( i ) . now , assume that all color classes in all @xmath32-vertex coloring of @xmath0 are of odd sizes . then @xmath227 is odd for all @xmath32-vertex colorings @xmath164 of @xmath0 and by proposition [ equal ] , @xmath228 . now , @xmath60 is an odd number and by the method used in the proof of proposition [ oddc ] , one can obtain an improper twin edge coloring of @xmath0 with @xmath60 colors , implying that @xmath16 . finally , the following theorem shows that the arguments can be turned into a linear - time algorithm which constructs an improper twin @xmath8 or @xmath229-edge coloring whenever a @xmath8-vertex coloring is given . [ thm : final ] there is a @xmath230 algorithm that computes , for a nice graph @xmath92 and a @xmath8-vertex coloring of @xmath0 , @xmath231 : + i ) if @xmath232 and there is at least one connected component with an odd number of vertices of each color , an improper twin @xmath229-edge coloring ; + ii ) else an improper twin @xmath8-edge coloring . given the graph @xmath92 the connected components can be determined in @xmath230 ; we can assume @xmath0 is connected and apply , for a non connected graph , the same algorithm on each connected component . there is a @xmath230 algorithm deciding whether @xmath0 is bipartite and , in this case , computing its bipartition . note first that operations in @xmath5 can be performed in @xmath233 time and @xmath234 . in particular , if @xmath8 is odd , determining @xmath115 such that @xmath235 requires for instance the euclidean algorithm @xcite . + lemma [ lem : tree ] ( including the determination of the spanning tree @xmath99 ) gives a @xmath230 algorithm to compute , for a @xmath13-vertex coloring @xmath93 of @xmath0 and @xmath95 an improper @xmath13-edge coloring that almost induces @xmath93 with @xmath7 as defective vertex . indeed , it requires ( 1 ) computing a spanning tree @xmath99 , ( 2 ) assigning the color 0 to edges outside the trees and ( 3 ) computing the colors of the tree - edges proceeding from the leaves to the root . all these operations can be respectively performed in @xmath236 and @xmath230 . + suppose that @xmath0 is bipartite . + if @xmath237 and one part of @xmath0 is of even size or if @xmath238 , we use the method described in the proof of theorem [ bip ] to compute an improper twin @xmath8-edge coloring inducing @xmath93 . it requires an application of lemma [ lem : tree ] with the right choice of the root @xmath7 ( either of degree 2 if @xmath238 or in an even part of the bipartition if @xmath237 ) . this choice requires @xmath239 operations . if @xmath237 and both parts of the bipartition are odd , then we apply the previous method replacing @xmath8 by 3 . + let us now suppose @xmath92 is not bipartite ( of course @xmath238 ) . + lemma [ odd - cycle ] gives a @xmath230 strategy to modify a given @xmath8-edge coloring accordingly . it requires determining an odd - cycle @xmath155 , a path @xmath61 from @xmath7 to @xmath155 and performing modifications along the related closed walk @xmath156 . these operations require respectively @xmath230 , @xmath230 and @xmath240 operations . + if @xmath8 is odd , then we revisit the proof of proposition [ oddc ] to compute an improper twin @xmath8-edge coloring . it requires one application of lemma [ lem : tree ] and eventually one equation in @xmath5 and one application of lemma [ odd - cycle ] . if @xmath241 or @xmath232 , @xmath242 , then we use similar arguments as in the proof of theorem [ thm : mainfromjcmcc ] , replacing @xmath32 with @xmath8 . if @xmath241 or if @xmath232 and at least one color class of even size , then determining an improper twin @xmath8-edge coloring requires , using the method in proposition [ equal ] , an application of lemma [ lem : tree ] and , eventually , one application of lemma [ odd - cycle ] . finally , if @xmath232 and all color classes are of odd size , then we use the case where @xmath8 is odd considering the @xmath8-vertex coloring as a @xmath229-vertex coloring . since the chromatic number of planar graphs is bounded by 4 @xcite , we trivially have the following : let @xmath0 be a nice connected planar graph , then @xmath243 . this section is devoted to some complexity results for the problem of deciding , given a graph @xmath0 , whether @xmath10 . we restrict ourselves to the class of instances for which a clique of size @xmath32 as well as a @xmath32-vertex coloring are given , which allows as to guarantee the polynomial recognition of an instance , a usual assumption in the area of np - hardness results . more formally we consider the following problem : + improper twin @xmath8-edge coloring + input : a graph @xmath0 with a clique of size @xmath8 and a @xmath8-vertex coloring of @xmath0 . + question : is @xmath10 ? + first , we notice that from the definition of improper twin @xmath8-edge coloring , an instance @xmath0 has @xmath244 and this is equal to the size of a largest clique of @xmath0 . this problem is clearly in np since , given a yes " answer with an improper twin @xmath32-edge coloring of @xmath0 , it takes linear time to check whether the @xmath32-vertex coloring induced by this improper edge coloring is proper . in addition , we will show in the sequel that the following problem , called all odd optimal @xmath8-vertex colorings is conp - complete . this implies by theorem [ thm : final ] that improper twin @xmath8-edge coloring is np - complete for @xmath245 ; indeed these are complementary problems in the sense that the answer is yes " for one of them if and only if the answer is no " for the other . + all odd optimal @xmath8-vertex colorings + input : a graph @xmath0 with a clique of size @xmath8 and a @xmath8-vertex coloring of @xmath0 . + question : are all color classes of odd size in all @xmath32-vertex colorings of @xmath0 ? + we need the following result for our reduction : [ lem : gadget ] let @xmath246 be the graph in figure [ fig : gadget ] ( a ) and consider a 3-vertex coloring of @xmath246 with colors f , t and r where vertices @xmath40 and @xmath110 are colored either @xmath99 or @xmath247 . then the following hold : 1 . in any such 3-vertex coloring of @xmath246 , where @xmath110 is colored @xmath99 , at least one among @xmath40 should be colored @xmath99 and among @xmath248 , there are @xmath249 and @xmath250 colors . 2 . similarly , in any such 3-vertex coloring of @xmath246 , where @xmath110 is colored @xmath247 , at least one among @xmath40 should be colored @xmath247 and among @xmath248 , there are @xmath251 and @xmath250 colors . we only prove statement ( i ) since ( ii ) is obtained similarly . let @xmath164 be a @xmath252-vertex coloring of @xmath246 , where @xmath253 and each one of @xmath254 is in @xmath255 . then at least one of @xmath40 has color @xmath99 , since otherwise ( that is if all of @xmath40 are colored @xmath247 ) we necessarily have @xmath256 , implying @xmath257 . now , @xmath258 has all three colors in its neighbourhood , hence it is impossible to color @xmath258 with one of the three colors @xmath259 and @xmath260 . besides , @xmath253 implies that we have either @xmath261 or @xmath262 . in either case , among @xmath263 there are exactly one @xmath99 , one @xmath247 and one @xmath260 . therefore , there are in total exactly @xmath249 and @xmath250 colors among @xmath264 . and the instance @xmath0.,width=415 ] [ conp ] for all @xmath238 , all odd optimal @xmath8-vertex colorings is conp - complete . first , all odd optimal @xmath8-vertex colorings is in conp , since one can verify the correctness of a no " certificate in linear time ; in other words , given a @xmath8-vertex coloring with at least one color class of even size , one can check whether it is a vertex coloring using at most @xmath8 colors and whether it contains a color class with even number of vertices . in the sequel , we show that 3-sat can be polynomially reduced to the complement of all odd optimal @xmath8-vertex colorings with @xmath265 . let @xmath266 be an instance of 3-sat with @xmath27 variables and @xmath267 clauses . we assume without loss of generality that @xmath27 is even ( if not , add a variable that appears in no clause ) . consider a truth assignment ( for instance where all variables are true ) and let @xmath268 and @xmath269 be respectively the numbers of true and false clauses . denote also by @xmath270 and @xmath271 the corresponding sets of clauses , respectively . assume without loss of generality that both @xmath268 and @xmath269 are even ( otherwise just double each clause in @xmath266 ) . we construct an instance @xmath0 of the complement of all odd optimal @xmath8-vertex colorings with @xmath265 as follows ( see figure [ fig : gadget ] ( b ) ) . for every variable , consider an edge between vertices @xmath272 and @xmath273 representing the variable and its negation . introduce two vertices @xmath274 and @xmath275 representing respectively clauses in @xmath270 and clauses in @xmath271 . for every clause , consider a clause gadget @xmath246 as in figure [ fig : gadget ] ( a ) , whose @xmath40 vertices are identified with the variable vertices representing the tree literals ( variables or their negations ) in the related clause , and whose @xmath110 vertex is identified with @xmath274 if it is a clause in @xmath270 and with @xmath275 otherwise . finally , add a vertex @xmath260 adjacent to all variable vertices and to @xmath274 and @xmath275 . clearly , this instance @xmath0 can be constructed in polynomial time in the size of the 3-sat instance @xmath266 . moreover , it is @xmath252-vertex colorable ; color vertex @xmath260 by color @xmath260 , @xmath274 by @xmath99 and @xmath275 by @xmath247 , color also all true literals by @xmath99 and all false literals by @xmath247 . now , one can check that this 3-vertex coloring can be extended to the clause gadgets corresponding to clauses in @xmath270 for all possible cases with 1 , 2 or 3 true literals per clause . similarly , it can be extended to the clause gadgets corresponding to clauses in @xmath271 and thus having all variable vertices colored @xmath247 . indeed , the only 3-vertex coloring of the three literals and @xmath274 ( or @xmath275 ) with colors @xmath99 and @xmath247 which can not be extended to a 3-vertex coloring of the clause gadget is when the three literals are all assigned the color @xmath247 and @xmath274 is assigned @xmath99 , or when the three literals are all assigned the color @xmath99 and @xmath275 assigned @xmath247 ; clearly , this never occurs in the 3-vertex coloring that we defined . let @xmath266 be a 3-sat instance and @xmath0 a graph constructed as described above . then , @xmath266 is satisfiable if and only if @xmath0 has a 3-vertex coloring with @xmath276 and variable vertices colored with @xmath99 or @xmath247 , where at least one color class has even size . consider a truth assignment satisfying all clauses of @xmath266 . then @xmath0 can be 3-colored with @xmath259 and @xmath260 as follows . color @xmath277 and all vertices representing true literals by @xmath99 . color the vertex @xmath260 by color @xmath260 . now , one can check that for all possible cases with 1 , 2 or 3 true literals per clause , this coloring can be extended to a 3-vertex coloring of @xmath0 . moreover , by lemma [ lem : gadget ] and by the fact that in any such 3-vertex coloring of @xmath0 there are exactly @xmath27 variable vertices assigned color @xmath99 , in this 3-vertex coloring there are in total @xmath278 vertices colored by @xmath99 , which is an even number ( remind that @xmath279 and @xmath269 are even and @xmath280 ) . now , consider a 3-vertex coloring @xmath164 of @xmath0 with @xmath277 and all variable vertices are colored with @xmath99 or @xmath247 , where there is at least one color class of even size . note that in any such 3-vertex coloring of @xmath0 , there are @xmath27 variable vertices of color @xmath99 and @xmath27 variable vertices of color @xmath247 . if @xmath274 and @xmath275 are colored with different colors , say by @xmath99 and @xmath247 , respectively , then the numbers of occurrences of @xmath259 and @xmath260 are , respectively @xmath281 and @xmath282 , where @xmath280 . the other case , where @xmath283 and @xmath284 yields the following numbers of occurrences for @xmath259 and @xmath260 , respectively : @xmath285 and @xmath282 . it can be seen that all these numbers are odd , a contradiction . therefore @xmath286 , and , using lemma [ lem : gadget ] , a truth assignment satisfying all clauses can be obtained by giving the value true to the literals having the same color as @xmath274 and @xmath275 . since 3-sat is np - complete @xcite , deciding whether a given graph with a clique of size 3 and a 3-vertex coloring admits an optimal vertex coloring ( with 3 colors ) with at least one color class of even size is np - complete . it follows that all odd optimal @xmath8-vertex colorings is conp - complete for @xmath287 . the conp - completeness of all odd optimal @xmath8-vertex colorings for @xmath288 is obtained simply by adding a clique of size @xmath289 completely linked to the graph @xmath0 in the above reduction ; clearly , the @xmath289 vertices of this clique should receive colors other than @xmath259 and @xmath260 and thus will not affect the above facts . let @xmath0 be an instance of improper twin @xmath8-edge coloring such that @xmath290 ( which can be checked in constant time ) . now , it follows from theorem [ thm : final ] that the answer to all odd optimal @xmath8-vertex colorings for @xmath0 is yes " if and only if the answer to improper twin @xmath8-edge coloring is no " . in addition , we have already noticed that improper twin @xmath8-edge coloring is in np . therefore improper twin @xmath8-edge coloring and all odd optimal @xmath8-vertex colorings are complementary problems and theorem [ conp ] and theorem [ thm : final ] imply the following : for all @xmath291 , @xmath290 , improper twin @xmath8-edge coloring is np - complete . it is polynomial for all other values of @xmath8 . the proof of theorem [ conp ] opens some natural questions about restricted hard and polynomial cases . we give first results in this direction that underline some open questions for future research . the first natural particular case is the class of graphs of bounded degree . the next proposition shows that , for any fixed @xmath238 , all odd optimal @xmath8-vertex colorings is still hard in these graphs and consequently so does improper twin @xmath8-edge coloring if @xmath291 , @xmath290 . [ prop : bounded_degree ] for all @xmath238 , all odd optimal @xmath8-vertex colorings is conp - complete in graphs of maximum degree @xmath292 . + for @xmath291 , @xmath290 improper twin @xmath8-edge coloring is np - complete in graphs of maximum degree @xmath292 . for the sake of simplicity we denote @xmath293 . we show that for @xmath238 , all odd optimal @xmath8-vertex colorings polynomially reduces to the same problem in graphs of maximum degree at most @xmath294 . consider a graph @xmath0 of maximum degree @xmath295 as an instance of all odd optimal @xmath8-vertex colorings for some @xmath238 . if @xmath296 , then we use the following gadget to construct a graph @xmath297 where the number of vertices of degree @xmath295 is one less ( eventually 0 ) than in @xmath0 and there is no vertex of larger degree , such that @xmath0 is a @xmath298-instance for all odd optimal @xmath8-vertex colorings if and only if @xmath297 is . let @xmath299 . let @xmath300 . so , @xmath29 is even and @xmath301 . @xmath302 is defined as follows : 1 . consider @xmath29 independent copies of @xmath303-cliques @xmath304 . [ enum : s]add an independent set @xmath305 of size @xmath306 such that @xmath45 is odd and @xmath307 . @xmath308 if @xmath309 and @xmath310 if @xmath311 . 3 . for @xmath312 , link by an edge every vertex of @xmath313 with vertices @xmath314 . 4 . for @xmath312 , partition @xmath315 into @xmath208 parts of size at most @xmath316 and link by an edge vertex @xmath317 with all vertices of part @xmath318 , @xmath319 . within this construction it is straightforward to verify that in any @xmath8-vertex coloring of @xmath302 , all vertices of @xmath305 are of the same color and each of the other colors is used exactly once in each clique @xmath320 . and how it replaces a vertex @xmath115 of degree 6 . @xmath321.,title="fig : " ] and how it replaces a vertex @xmath115 of degree 6 . @xmath321.,title="fig : " ] given the graph @xmath0 of maximum degree @xmath295 , if @xmath296 , we consider a vertex @xmath115 of degree @xmath295 and number its neighbors @xmath322 . then we transform @xmath0 as follows . we first remove @xmath115 as well as all its adjacent edges ; then we add a copy of @xmath302 and link by an edge @xmath323 with @xmath324 , @xmath325 , so as to create a matching @xmath326 of size @xmath295 between @xmath305 and the neighborhood of @xmath115 . we denote by @xmath297 the resulting graph ( see figure [ fig : bound ] ) . we evaluate the degree of vertices of @xmath302 in the graph @xmath297 : vertices of the cliques @xmath313 , @xmath327 are of degree at most @xmath328 and vertices of @xmath305 are ( taking into account one possible edge of @xmath326 ) of degree at most @xmath329 . since @xmath330 , @xmath297 has one less vertex of degree @xmath331 . using the property of @xmath8-vertex colorings of @xmath302 , every @xmath8-vertex coloring of @xmath0 with @xmath115 of color @xmath81 can be transformed into a @xmath8-vertex coloring of @xmath297 with @xmath305 colored @xmath81 and moreover , the number of vertices colored @xmath81 is increased by @xmath332 and the number of vertices in all other colors is increased by @xmath29 . conversely every @xmath8-vertex coloring of @xmath297 with @xmath305 colored @xmath81 can be transformed into a @xmath8-vertex coloring of @xmath0 with @xmath332 vertices less colored @xmath81 and all other color classes have @xmath29 less vertices . since @xmath29 and @xmath332 are both even , @xmath0 is a @xmath298-instance for all odd optimal @xmath8-vertex colorings if and only if @xmath297 is . the construction from @xmath0 to @xmath297 can be performed in polynomial time and by repeating it while there are still vertices of degree greater than @xmath294 we show that all odd optimal @xmath8-vertex colorings polynomially reduces to the same problem in graphs of maximum degree @xmath294 . theorem [ thm : final ] allows to conclude the proof . note that the result of proposition [ prop : bounded_degree ] for @xmath333 could be obtained by using the gadgets @xmath334 , @xmath335 , used in @xcite , page 86 ; however this gadget can not be ( easily ) generalized for larger @xmath8 . in particular , proposition [ prop : bounded_degree ] for @xmath333 and for @xmath336 respectively implies that all odd optimal @xmath252-vertex colorings is still conp - complete in graphs of maximum degree 4 and improper twin @xmath337-edge coloring is np - complete in graphs of maximum degree 8 . it leaves open the cases with lower degrees . in this section we exhibit some graph classes for which the problems under consideration are polynomial . we will say that a vertex coloring is _ complete _ if all colors are maximal independent sets . equivalently , it means that all colors appear in the closed neighborhood of every vertex . [ rem : maximal ] + ( 1 ) if for a graph @xmath0 all color classes have an odd size in all @xmath32-vertex colorings , then all @xmath32-vertex colorings of @xmath0 are complete ; + ( 2 ) all @xmath32-vertex colorings of @xmath0 are complete if and only if for all connected component @xmath67 of @xmath0 , all @xmath32-vertex colorings of @xmath67 are complete ( in particular @xmath338 ) . + ( 3 ) if all @xmath32-vertex colorings of @xmath0 are complete , then every vertex is of degree at least @xmath339 . \(1 ) if a graph @xmath0 has a @xmath32-vertex coloring such that one color class is not a maximal independent set , then if all colors are of odd size , it is possible to transfer one vertex from one color to a non - maximal color class so as to create two color classes of even size . the first polynomial case below , compared to proposition [ prop : bounded_degree ] , emphasizes that , for bounded degree graphs , there is still a gap of value @xmath341 for the maximum degree between hard cases and polynomial cases . all odd optimal @xmath8-vertex colorings is polynomial in graphs of maximum degree @xmath48 . the graph is a @xmath298-instance if and only if it is the disjoint union of an odd number of @xmath8-cliques . consider a graph @xmath0 of maximum degree @xmath48 , with @xmath342 . we distinguish three kinds of connected components of @xmath0 : @xmath8-cliques ( we denote by @xmath29 their number ) , odd cycles of length at least 5 ( we denote by @xmath30 their number ) and other components ( we denote by @xmath213 their number ) . using brook s theorem @xcite , all components that are neither a @xmath8-clique nor an odd cycle of length at least 5 are @xmath303-vertex colorable . so , if @xmath343 , remark [ rem : maximal ] allows to conclude that @xmath0 has a @xmath32-vertex coloring with at least one color class of even size . odd cycles of size at least 5 are either @xmath303-vertex colorable as well if @xmath242 or have a @xmath8-vertex coloring which is incomplete if @xmath333 ( one can ensure that one color is of size 1 ) . in both cases , if @xmath344 , remark [ rem : maximal ] also ensures that @xmath0 has a @xmath32-vertex coloring with at least one color of even size . the only case where all color classes are of odd size in all @xmath8-vertex colorings is if @xmath345 and @xmath29 is odd . we recall that a graph is a _ split graph _ if its vertex set can be decomposed into a clique and an independent set . also , a graph @xmath0 is called _ perfect _ if every induced subgraph @xmath297 of @xmath0 has a ( maximum ) clique of size @xmath346 . it is well known ( see e.g. @xcite ) that a split graph @xmath0 can be decomposed into a maximum clique @xmath347 and an independent set @xmath305 . since split graphs are perfect , we have @xmath348 . consider a split graph @xmath0 decomposed into a maximum clique @xmath347 of size @xmath342 and an independent set @xmath305 of size @xmath45 . then all color classes are of odd size in all @xmath8-vertex colorings if and only if every vertex in the independent set has degree @xmath303 and the degree of every vertex in the clique is of the opposite parity as the number of vertices @xmath349 . note first that if @xmath350 , then the condition holds and in this case all color classes are singletons . suppose now @xmath351 . using remark [ rem : maximal ] , a necessary condition to have all color classes of odd size in all @xmath8-vertex colorings is that the degree of every vertex in @xmath305 is equal to @xmath48 ( it can not be more since @xmath347 is a maximum clique ) . then there is a unique @xmath8-vertex coloring ( up to the permutation of colors ) ; each color class has odd size if and only if every vertex in @xmath347 is not adjacent to an even number of vertices in @xmath305 . it follows that for every vertex @xmath352 the number of adjacent vertices in @xmath305 is of the same parity as @xmath45 ; moreover , @xmath115 is adjacent to all other @xmath48 vertices in @xmath347 , this is equivalent to the fact that the degree of @xmath115 is of the opposite parity as the number of vertices @xmath349 . [ pro : cochordal ] consider a hereditary class of graphs @xmath46 for which all maximal independent sets and the chromatic number can be computed in polynomial time . then all odd optimal @xmath8-vertex colorings and minimum improper twin edge coloring are polynomial in @xmath46 . for any graph @xmath353 , to decide whether @xmath0 admits an optimal coloring with at least one color class including an even number of vertices , we proceed in two phases . first enumerate all maximal independent sets with an even number of vertices and test for each one whether the graph obtained by removing its vertices has chromatic number @xmath339 . if it is the case for at least one independent set of even size , then this independent set is a color of even size in at least one optimal vertex coloring . else consider all maximal independent sets @xmath354 with odd number of vertices and for every @xmath355 remove @xmath356 and test whether the remaining graph has chromatic number @xmath339 . if it is the case for some @xmath354 , then there is also an optimal coloring with a color class of even size . suppose now that the two phases failed to find a coloring with an even color class . from the first phase , we know that every complete @xmath32-vertex coloring has all color classes of odd size . note finally that , if there is an incomplete optimal coloring , then by transferring some vertices from other colors to a non - maximal color class we can construct an optimal coloring where one color is a maximal independent set minus one vertex . this color class can not be of odd size if the first phase failed . if the second phase also failed , it can not be of even size neither , and consequently all optimal colorings are complete and thus all color classes are of odd size in all optimal colorings . we conclude by theorem [ thm : final ] . a graph is _ chordal _ ( or triangulated ) if every cycle of size greater than 3 has a _ chord _ , that is an edge between two non consecutive vertices in the cycle . a graph is co - chordal if its complementary is chordal . it is known that in chordal graphs all maximal cliques can be computed with complexity @xmath230 ( @xcite ) . since co - chordal graphs are perfect , proposition [ pro : cochordal ] applies . we finally consider the class of co - comparability graphs that includes for instance interval graphs and permutation graphs ( see @xcite for definitions of these classes ) . a graph is a _ graph if it has a transitive orientation of its edges , that is an orientation of edges such that , for every two adjacent edges @xmath357 and @xmath358 , if @xmath357 is oriented from @xmath115 to @xmath359 and @xmath358 is oriented from @xmath359 to @xmath360 , then @xmath361 is an edge and it is oriented from @xmath115 to @xmath360 . a graph is a co - comparability graph if its complement is a comparability graph . a co - comparability graph @xmath92 has a _ co - comparability order _ that is an order of vertices @xmath362 such that @xmath363 , if @xmath364 and @xmath365 , then @xmath366 @xcite . for any @xmath8-vertex coloring ( @xmath367 ) we will call the _ type _ of this coloring the list @xmath368 $ ] of the sizes of the different colors ( we may have @xmath369 for some @xmath81 ) . of course @xmath370 . in order to avoid useless repetitions of types due to the order of colors , we systematically write a type with colors ordered in non - decreasing order of their size . we consider a co - comparability order @xmath362 and process by dynamic programming . for a vertex @xmath371 , @xmath81 will be called its _ rank_. we denote by @xmath372 a fictive vertex considered as isolated . for any @xmath373 , we denote by @xmath374 the subgraph of @xmath0 induced by @xmath375 . the number @xmath99 of possible types of @xmath32-vertex colorings of @xmath374 is bounded by @xmath376 , which is polynomial in @xmath27 since @xmath32 is bounded . for any coloring of type @xmath377 $ ] we denote the colors @xmath378 of size @xmath379 , respectively . we define its _ state _ as the list @xmath380 $ ] , where @xmath381 is the vertex of maximum rank in color @xmath382 if @xmath383 and @xmath384 if @xmath369 . denoting by @xmath385 the number of possible states of @xmath32-vertex colorings of @xmath374 , we have @xmath386 . note that , using the key property of the co - comparability order , for any independent set @xmath387 of @xmath374 , @xmath388 , with @xmath371 ( @xmath389 ) the vertex of maximum rank in @xmath387 , @xmath390 is independent if and only if @xmath391 and @xmath371 are not adjacent . thus , from all possible states of @xmath374 we can compute in polynomial time all possible states in @xmath392 . starting from the states @xmath393 $ ] we can compute in polynomial time all states of @xmath32-vertex colorings of @xmath0 . the set of all possible types can be trivially deduced . these simple examples of polynomial cases allow us to underline some specific situations leading to a polynomial algorithm . these algorithms are all based on an enumeration of all possible color sizes . cliques or connected bipartite graphs are uniquely colorable , which makes all odd optimal @xmath8-vertex colorings trivial . the case of split graphs also reduces to the case of a uniquely colorable graph . proposition [ pro : cochordal ] exhibits a situation where we can exhaustively look for one of the colors while in the case of proposition [ pro : cocomp ] the sizes of all color classes can be computed in polynomial time . remark [ rem : maximal ] gives a completely different approach that does not require the explicit values of the sizes of color classes . it would be very interesting to find other classes where similar approaches can be applied as well as cases inducing completely new methods . in terms of graph classes , these examples underline open cases we leave for future research . let us mention in particular the complexity in perfect graphs and in planar graphs as well as the case of interval graphs or permutation graphs with unbounded chromatic number .

in this paper we describe a procedure for calculating the number of regular octahedrons that have vertices with coordinates in the set . as a result , we introduce a new sequence in _ the online encyclopedia of integer sequences _ ( a178797 ) and list the first one hundred terms of it . we adapt the method appeared in which was used to find the number of regular tetrahedra with coordinates of their vertices in . the idea of this calculation is based on the theoretical results obtained in . a new fact proved here helps increasing the speed of all the programs used before . the procedure is put together in a series of commands written for maple . [ section ] [ theorem ] proposition [ theorem ] corollary [ theorem ] lemma [ theorem ] definition [ theorem ] problem [ theorem ] conjecture [ theorem ] example [ theorem ] remark # 1#1 # 1#1 # 1*#1 *

1007.1655

in this note we complete the work begun in the sequence of papers @xcite , @xcite-@xcite about equilateral triangles , regular tetrahedra , cubes , and regular octahedrons all with vertices having integer coordinates . very often we will refer to this property by saying that the various objects are in @xmath1 . strictly speaking these geometric objects are defined as being more than the set of their vertices that determines them , but for us here , these are just the vertices . so , for instance , an equilateral triangle is going to be a set of three points in @xmath2 for which the euclidean distances between every two of these points are the same . the main purpose of the paper is to take a close look at the regular octahedrons in @xmath2 . the simplest example of a regular octahedron with integer coordinates for its vertices is @xmath3 , [ 1 , 0 , 1 ] , [ 1 , 1 , 0 ] , [ 1 , 1 , 2 ] , [ 1 , 2 , 1 ] , [ 2 , 1 , 1]\},\ ] ] which can be obtained from the usual unit cube in @xmath4 , multiplying the vertices by a factor of two and then taking the coordinates of the centers of the new faces . it turns out that this procedure gives all such octahedrons as shown in @xcite : [ fig1 ] @xmath5 @xmath6 [ oldtheorem ] every regular octahedron in @xmath2 is the dual of a cube that can be obtained ( up to a translation with a vector with integer coordinates ) by doubling a cube in @xmath1 . referring to figure [ fig1 ] ( b ) , we showed that if the regular octahedron @xmath7 is in @xmath2 , then so is the cube @xmath8 and vice versa . this defines a one - to - one correspondence between the classes of cubes ( invariant under integer translations ) and the classes regular octahedra ( invariant under integer translations ) in @xmath2 . in @xcite we determined a sequence of irreducible cubes , one from each of the classes of cubes invariant under integer translations and cube symmetries . for each one of these cubes we can construct as before a regular octahedron , obtaining this way a sequence of irreducible regular octahedrons . in @xcite , we improved and adapted the earlier code for counting all cubes with vertices in @xmath9 and extended the sequence a098928 . in this paper , the usual techniques and ideas are going to be the same except some counting procedure that is very efficient in comparison to what we had before . we are going to treat this in the general case so , let us suppose that these objects can be either equilateral triangles , regular tetrahedrons , cubes or regular octahedrons with vertices in @xmath2 . for such an object , say @xmath10 , we can translate it , within @xmath2 , to @xmath11 that is in the positive octant and in such way each plane of coordinates contains at least one vertex of @xmath11 . let us denote by @xmath12 the number of objects in @xmath13 obtained by applying to @xmath11 all 48 possible symmetries of the cube @xmath13 . these symmetries are generated in the following way : first we have symmetries with respect to the middle planes and compositions , for example @xmath14 in a total of eight including the identity , then each one of these is coupled with one of the six permutations of the variables ( @xmath15 ) . these transformations form a group isomorphic with the group of all @xmath16 orthogonal matrices having entries @xmath17 and it is also known as the group of symmetries of a cube or of a regular octahedron . it is isomorphic to @xmath18 . we are going to denote this group by @xmath19 although it is usually known under the name of ( extended ) octahedral group and denoted simply by @xmath20 . if we think of @xmath12 as the cardinality of @xmath21 which is , by the first theorem of isomorphism of groups , the same as the cardinality of the group factor @xmath22 , where @xmath23 is the subgroup of @xmath19 of those symmetries that leave @xmath11 invariant . the structure of subgroups of @xmath19 is known and for each divisor of 48 there is a subgroup of that order . hence , we expect @xmath12 to be in the set @xmath24 and most of the time to be @xmath25 since an arbitrary object @xmath11 in @xmath13 is unlikely to be invariant under any of the symmetries of @xmath26 . then , we denote by @xmath27 , the cardinality of the set of all the objects counted in @xmath12 and their ( all possible ) integer translations that leave the resulting objects in @xmath13 . also , we denote by @xmath28 the objects counted in @xmath27 which are in @xmath29 . finally , let us denote by @xmath30 the number of objects counted in @xmath28 which are in @xmath31 . then , we found a formula that gives the number of objects obtained from @xmath10 , under all symmetries and translation that leaves the resulting object in @xmath32 , @xmath33 . this fact has been essentially proved in theorem 2.2 in @xcite . the formula that gives this number is @xmath34 let us suppose that the object @xmath10 can be squeezed within a box of dimensions @xmath35 ( @xmath36 ) , i.e. up to symmetries and translations , @xmath10 can be transformed to @xmath37 fitting snugly into @xmath38 we can similarly consider all eight reflections compatible with the box @xmath39 of the form @xmath40 let us denote the group of these transformations by @xmath41 . we notice that each one of these transformation leaves the object @xmath11 inside the box @xmath39 . from case to case , depending of what the values @xmath42 , @xmath43 and @xmath44 are , we may have the result of some or all of the permutation transformations applied to @xmath37 still in @xmath39 . hence , we will denote by @xmath45 the cardinality of the set @xmath46 ( { \cal o ' } ) \in b_{m , n , p}|s_1\in { \cal s}_b , s_2\in { \cal s}_6\}.\ ] ] let us look at an example . suppose @xmath10 ( equal with @xmath11 ) is the equilateral triangle given by its vertices : @xmath47 , [ 5 , 7 , 0 ] , [ 7 , 0 , 1]\}.\ ] ] we observe that @xmath48 . then one can check that @xmath49 is the collection of eight triangles @xmath50 , [ 2 , 7 , 0 ] , [ 7 , 2 , 2]\ } , \{[0 , 0 , 1 ] , [ 2 , 7 , 2 ] , [ 7 , 2 , 0]\ } , \{[0 , 2 , 0 ] , [ 5 , 7 , 2 ] , [ 7 , 0 , 1]\},\\ \\ \{[0 , 5 , 0 ] , [ 5 , 0 , 2 ] , [ 7 , 7 , 1]\ } , \{[0 , 5 , 2 ] , [ 5 , 0 , 0 ] , [ 7 , 7 , 1]\ } , \{[0 , 7 , 1 ] , [ 2 , 0 , 0 ] , [ 7 , 5 , 2]\ } , \{[0 , 7 , 1 ] , [ 2 , 0 , 2 ] , [ 7 , 5 , 0]\ } , \end{array}}\ ] ] so @xmath51 . it turns out that @xmath52 , @xmath53 , @xmath54 and @xmath55 . formula ( [ oldformula ] ) becomes @xmath56 it turns out the this factorization is not accidental and the following alternative to ( [ oldformula ] ) is true . [ coj ] given @xmath10 , one of the objects mentioned before , and @xmath39 the smallest box containing a translation of @xmath10 ( @xmath36 ) , we let @xmath57 , @xmath58 , and @xmath59 then the number of distinct objects in the cube @xmath60 ( @xmath61 ) , obtained from @xmath10 by all possible integer translations and symmetries is equal to @xmath62 . the case @xmath63 implies @xmath64 and @xmath65 because there is no room to shift the orbit @xmath66 inside of @xmath67 . the formula follows from ( [ oldformula ] ) . let us look into the case @xmath68 and @xmath69 . we begin by observing that each integer translation of the box @xmath39 in all possible ways inside @xmath60 will give @xmath70 more copies of @xmath10 . there is no overlap between these copies because neither one of them can be inside of two distinct translations of @xmath39 . this is due to the minimality of @xmath42 , @xmath43 and @xmath44 . we get @xmath71 such copies by counting all possible translations . since @xmath42 , @xmath43 and @xmath44 are all distinct , the box @xmath39 can be positioned first with the biggest of its dimensions along one of the directions given by the axis of coordinates , that is three different ways , and for each such position the next largest dimension can be positioned along the two remaining directions . the minimality of @xmath42 , @xmath43 and @xmath44 makes the six different situations generate distinct objects . this explains the factor of six that appears in ( [ newformula ] ) for this situation . in the last case , the box @xmath39 has two of its dimensions the same , so there are only three possibilities to arrange the box before one translates it . to see that we get all possible translates and symmetries of @xmath10 by this counting , we can start with one copy @xmath11 . construct the minimum box around it . in terms of its position and dimensions , we know in what of the six or three cases we are . we transform it into the standard standard position , @xmath39 , and look at the corresponding object , @xmath72 . the transformations involved form a group of transformations generated by the permutations of the coordinates , the reflections into the axes and integer translations . every transformation in this group , say @xmath73 with @xmath74 a permutation , @xmath75 a reflection or a composition of reflections and @xmath76 a translation , which satisfies @xmath77 determines a representation @xmath78 with @xmath79 , @xmath80 as in the definition of @xmath70 . this can be done by taking @xmath81 and @xmath82 . this is true again because of the minimality of the box @xmath39 , i.e. there is only one integer translation that takes a reflected box @xmath83 into @xmath39 . this new way of counting is more efficient from a computational point of view because @xmath84 is simply no bigger than 48 , as opposed to the previous situation when @xmath27 , @xmath28 and @xmath30 could turn out to be big numbers and so the number of iterations for computing them would be also large . roughly speaking , this counting factors out fast the problem with the integer translations . as an example , let us consider @xmath85 , [ 0 , 3 , 4 ] , [ 1 , 4 , 0 ] , [ 3 , 0 , 4 ] , [ 4 , 1 , 0 ] , [ 4 , 4 , 3]\}.\ ] ] the minimal box here is @xmath86 and after rotating @xmath87 in all possible ways ( figure [ fig2 ] ( b ) ) we get @xmath88 . [ fig2 ] @xmath89 @xmath90 the idea of calculations is basically the same as in @xcite , in which we have constructed a list of irreducible cubes that are used to generate all the other cubes in @xmath60 . here , we are using theorem [ oldtheorem ] , to construct a similar list of irreducible regular hexahedrons . as expected , an _ irreducible _ regular hexahedron is one whose coordinates can not be obtained from a strictly smaller one with vertices in @xmath2 by integer dilations and translations . one simple consequence of theorem [ oldtheorem ] is the next corollary . [ sideofoctahedrons ] the sides of every irreducible regular octahedron in @xmath2 are of the form @xmath91with @xmath92 . we used the same way of finding the parameterizations of the equilateral triangles as in @xcite : [ fig3 ] @xmath93 @xmath94 @xmath95 @xmath96 where @xmath97 and @xmath98 is a suitable solution of @xmath99 that makes all the numbers in ( [ paramtwo ] ) integers . the sides - lengths of @xmath100 are equal to @xmath101 and the triangle can be completed to a regular tetrahedron ( in space ) with integer coordinates if and only if @xmath102 for some @xmath103 . related to this fact we have the following proposition . [ numberofro ] for a prime @xmath104 , the number of irreducible regular octahedrons in @xmath2 ( up to translations and symmetries ) having side lengths equal to @xmath105 , is at most @xmath106 , where @xmath107 with @xmath108 and @xmath109 defined as in @xcite . the exact number of such objects is yet a big mystery to us . we wrote the code using the maple software and so we took advantage of the build in commands available for a various number of functions . the beginning is pretty standard : * step 1 . * the next three procedures calculate all possible values of @xmath110 , then the parameters @xmath42 , @xmath43 , and finally the normal vector @xmath111 . the values of @xmath110 are determined by using a characterization theorem for the quadratic form involved here , i.e. all values of @xmath110 less than @xmath43 such that @xmath110 is of the form a product of primes of the form @xmath112 , @xmath113 or @xmath114 . .... > kvalues:=proc(n ) local i , j , k , l , a , p , q , r , m , mm ; l:={};mm:=floor((n+1)/2 ) ; for i from 2 to mm do a:=ifactors(2*i-1 ) ; k:=nops(a[2]);r:=0 ; for j from 1 to k do m:=a[2][j][1 ] ; p:=m mod 3 ; if m=3 then r:=1 fi ; if r=0 and p=2 then r:=1 fi ; od ; if r=0 then l:=l union { 2*i-1};fi ; od ; l:=l union { 1 } ; l:=convert(l , list ) ; end : .... next , we are interested in the solutions @xmath116 of the equation @xmath102 , which are primitive in the sense that @xmath117 , @xmath118 , @xmath119 , and @xmath120 . we apply this procedure to only those @xmath110 s which are the output of @xmath121 . .... > listofmn:=proc(k ) local a , b , i , nx , x , m , n , l ; x:=[isolve(k^2=m^2-m*n+n^2 ) ] ; nx:=nops(x ) ; l:= { } ; for i from 1 to nx do if lhs(x[i][1])=m then a:=rhs(x[i][1 ] ) ; b:=rhs(x[i][2 ] ) ; else b:=rhs(x[i][1 ] ) ; a:=rhs(x[i][2 ] ) ; fi ; if gcd(a , b)=1 and a>=0 and b>0 and 2*a < b then l:=l union { [ a , b]};fi ; od;l ; end : .... a prime of the form @xmath112 has a unique primitive decomposition as described , for example , @xmath122\}$ ] since @xmath123 and @xmath124 is a prime . in general , the number of solutions is equal to @xmath125 where @xmath126 is the number of distinct factors of @xmath110 which are of the form @xmath112 . this choice of @xmath42 and @xmath43 helps identify the irreducible octahedrons : all the other solutions @xmath116 of the equation @xmath102 lead to the same regular octahedron . we included the case @xmath127 to obtain the output @xmath128\}$ ] for @xmath129 which is necessary later on . in the next procedure , we find only the solutions @xmath111 of @xmath130 that satisfy @xmath131 and @xmath132 . the number of such solutions can be calculated precisely in terms of the prime factorization of @xmath133 ( see @xcite ) . .... abcsol:=proc(d ) local i , j , k , m , u , x , y , sol , cd;sol:= { } ; for i from 1 to d do u:=[isolve(3*d^2-i^2=x^2+y^2)];k:=nops(u ) ; for j from 1 to k do if rhs(u[j][1])>=i and rhs(u[j][2])>=i then cd:=gcd(gcd(i , rhs(u[j][1])),rhs(u[j][2 ] ) ) ; if cd=1 then sol:=sol union { sort([i , rhs(u[j][1]),rhs(u[j][2])])};fi;fi ; od ; od;convert(sol , list ) ; end : .... for @xmath134 , for instance , the procedure _ abcsol _ gives the four solutions : [ 1 , 5 , 29 ] , [ 7 , 17 , 23 ] , [ 11 , 11 , 25 ] and [ 13 , 13 , 23 ] . we observe that two of them have the property that @xmath135 , in which case , we know ( see @xcite ) that the formulae ( [ paramtwo ] ) simplify quite a bit . * step 2 . * the next seven procedures implement the new way of finding @xmath136 and @xmath137 which appear in the parametrization ( [ paramtwo ] ) . it is followed by the calculation of the of the parametrization of equilateral triangles ( [ paramtwo ] ) and only one regular octahedron is constructed based on theorem [ oldtheorem ] . this octahedron is then translated minimally into the positive octant and the minimal cube containing it is computed . .... > uniquedecomposition:=proc(p ) local out , s , out1,out2 ; if p=2 then out:=[1,1 ] ; fi ; if p>2 then s:=isolve(p = x^2 + 3*y^2 ) ; out1:=abs(rhs(s[1][1]));out2:=abs(rhs(s[1][2 ] ) ) ; if out1 ^ 2 + 3*out2 ^ 2=p then out:=[out1,out2 ] ; else out:=[out2,out1];fi ; fi ; out ; end : .... for @xmath140 this procedure has the needed output @xmath141 $ ] . as an example , @xmath142 $ ] since @xmath143 is a prime and @xmath144 . the next procedure is calculating the factorization in @xmath145 $ ] of a number of the form @xmath146 . .... > factorovereisensteinintegers:=proc(u , v ) local i , n , m , k , a , x , f1,f2,g , y , y1,y2,l , nn , mm , uu , vv ; a:=sqrt(3)*i;nn:=gcd(u , v);uu:=u / nn;vv:=v / nn ; n:=uu^2 + 3*vv^2;x:=uu+vv*a;m:=ifactors(n);k:=nops(m[2 ] ) ; for i from 1 to k do f1:=m[2][i][1 ] ; f2:=m[2][i][2 ] ; if f1>2 then g:=uniquedecomposition(f1 ) ; else g:=[1,1 ] ; f2:=1 ; fi ; y:=expand(rationalize(x/(g[1]+a*g[2 ] ) ) ) ; y1:=re(y);y2:=type(y1,integer ) ; if y2=true then l[i]:=[g[1]+a*g[2],f2 ] ; else l[i]:=[g[1]-a*g[2],f2 ] ; fi ; od;[nn , seq(l[i],i=1 .. k),expand(nn*product(l[ii][1]^l[ii][2],ii=1 .. k))];end : .... @xmath149 perhaps one word of caution is necessary at this point . the decomposition in general is not unique in the usual sense since @xmath150 . as in the example shown , we go for the second representation when something like this happens . we are interested in this decomposition because it provides suitable values for @xmath136 and @xmath137 that we find next . this turns out to be the greatest common divisor between @xmath151 and @xmath152 with @xmath153 , @xmath154 and @xmath97 ( see @xcite ) . .... > findgcd:=proc(a , b , q ) local a , i , j , f , f1,f2,common , m , qq , fac , nfac , s , rs , p ; a:=sqrt(3)*i ; f:=factorovereisensteinintegers(a , b);m:=nops(f)-1 ; common:=gcd(f[1],2*q ) ; qq:=2*q / common^2 ; p:=common ; fac:=ifactors(qq);nfac:=nops(fac[2 ] ) ; for i from 1 to nfac do f1:=fac[2][i][1];f2:=fac[2][i][2 ] ; if f1=2 then f2:=1;fi ; s:=uniquedecomposition(f1 ) ; for j from 1 to m do if s[1]+a*s[2]=f[j][1 ] then p:=p*(s[1]+a*s[2])^(min(f[j][2],f2 ) ) ; fi ; if s[1]-a*s[2]=f[j][1 ] then p:=p*(s[1]-a*s[2])^(min(f[j][2],f2 ) ) ; fi ; od ; od;p:=expand(p);rs:=[re(p),im(p)/sqrt(3 ) ] ; [ rs , a*rs[1]+3*b*rs[2 ] mod 2*q , a*rs[2]-b*rs[1 ] mod 2*q ] ; end : .... in the case we have seen before , where @xmath134 , if we take @xmath155 , @xmath156 and @xmath157 , then we have @xmath158 , @xmath159 and @xmath160 . the procedure above gives @xmath161 and @xmath162 and checks that @xmath163 ( mod @xmath152 ) and @xmath164 ( mod @xmath152 ) . these two conditions are enough to insure that the expressions in ( [ paramtwo ] ) , all give integer values for the coordinates of @xmath165 and @xmath166 . in the next procedure we use the @xmath136 and @xmath137 determined earlier and construct an irreducible regular octahedron in @xmath2 , given a vector @xmath111 and the possible values for @xmath116 in the decomposition of @xmath102 . we are using simple formulae which can be derived easily from figure 1(b ) , thinking that the point @xmath167 is the origin , @xmath168 $ ] , @xmath169 $ ] and @xmath170 $ ] are the points @xmath171 , @xmath172 and @xmath173 . then the other three vertices are simply given by the addition of every two of these there vectors . although there are six possible equilateral triangles that one may start with in this construction , one can see that in the end , essentially the same regular octahedron ( up to symmetries and translations ) is obtained . .... > findpar:=proc(a , b , c , mm , nn ) local q , d , r , s , rs , a , b , k , mx , nx , my , ny , mz , nz , mu , nu , mv , nv , mw , nw , u , v , w , x , y , z , t , r1,r2,dd , e , f ; q:=a^2+b^2;k:=sqrt(mm^2-mm*nn+nn^2);d:=sqrt((a^2+b^2+c^2)/3);a:=a*c;b:=b*d ; rs:=findgcd(a ,- b , q);r:=rs[1][2 ] ; s:=rs[1][1 ] ; mx:=-(d*b*(3*r+s)+a*c*(r - s))/(2*q);nx:=-(r*a*c+d*b*s)/q ; my:=(d*a*(3*r+s)-b*c*(r - s))/(2*q);ny:=-(r*b*c - d*a*s)/q;mz:=(r - s)/2;nz:=r ; mu:=nx;mv:=ny;mw:=nz;nu:=nx - mx;nv:=ny - my ; nw:=nz - mz ; u:=mu*mm - nu*nn;v:=mv*mm - nv*nn;w:=mw*mm - nw*nn ; x:=mx*mm - nx*nn;y:=my*mm - ny*nn;z:=mz*mm - nz*nn ; r1:=[(x+u-2*a*k)/3,(v+y-2*b*k)/3,(z+w-2*c*k)/3 ] ; r2:=[(x+u+2*a*k)/3,(v+y+2*b*k)/3,(z+w+2*c*k)/3 ] ; if r1[1]=floor(r1[1 ] ) then t:=[[u , v , w],[x , y , z],r1 ] ; else t:=[[u , v , w],[x , y , z],r2 ] ; fi ; dd:=[t[1][1]+t[2][1],t[1][2]+t[2][2],t[1][3]+t[2][3 ] ] ; e:=[t[1][1]+t[3][1],t[1][2]+t[3][2],t[1][3]+t[3][3 ] ] ; f:=[t[2][1]+t[3][1],t[2][2]+t[3][2],t[2][3]+t[3][3 ] ] ; [ t[1],t[2],t[3],dd , e , f ] ; end : .... since for @xmath174 we get @xmath175\}$ ] , we checked to see what regular octahedron is obtained for @xmath176 : [ 2301 , -1421 , -880 ] , [ 880 , -2301 , 1421 ] , [ 2401 , 100 , 1521 ] , [ 3181 , -3722 , 541 ] , [ 4702 , -1321 , 641 ] , and [ 3281 , -2201 , 2942 ] . since the set @xmath177 has 336 elements in it , there are essentially at most 337 irreducible regular octahedra in @xmath2 with side lengths equal to @xmath178 as we pointed out in proposition [ numberofro ] . next , we have a short function for subtracting two vectors @xmath179 and @xmath180 . in order to compare various octahedrons it is easier if they are all translated to the positive octant in such a way each plane of coordinates contains at least a vertex of the octahedron . this is accomplished with the next routine . .... > tmttopqoctahedron:=proc(t ) local i , a , b , c , v , c ; a:=min(t[1][1],t[2][1],t[3][1],t[4][1],t[5][1],t[6][1 ] ) ; b:=min(t[1][2],t[2][2],t[3][2],t[4][2],t[5][2],t[6][2 ] ) ; c:=min(t[1][3],t[2][3],t[3][3],t[4][3],t[5][3],t[6][3 ] ) ; v:=[a , b , c];c:={subtrv(t[1],v),subtrv(t[2],v),subtrv(t[3],v ) , subtrv(t[4],v),subtrv(t[5],v),subtrv(t[6],v ) } ; end : .... so , for instance , the octahedron mentioned earlier of sides lengths @xmath178 becomes : [ 2401 , 1521 , 3822 ] , [ 3822 , 2401 , 1521 ] , [ 2301 , 0 , 1421 ] , [ 1521 , 3822 , 2401 ] , [ 0 , 1421 , 2301 ] , and [ 1421 , 2301 , 0 ] . the smallest cube @xmath181 ^ 3 $ ] containing an octahedron positioned in the positive octant as specified earlier is computed in the following procedure . .... > mscofmoctahedron:=proc(q ) local a , b , c , t ; t:=convert(q , list ) ; a:=max(t[1][1],t[2][1],t[3][1],t[4][1],t[5][1],t[6][1 ] ) ; b:=max(t[1][2],t[2][2],t[3][2],t[4][2],t[5][2],t[6][2 ] ) ; c:=max(t[1][3],t[2][3],t[3][3],t[4][3],t[5][3],t[6][3 ] ) ; max(a , b , c ) ; end : .... * step 3 . * in our construction of these octahedrons we end up with essentially the same octahedron if we proceed from a different face of it . to eliminate the possibility of counting an octahedron twice or more than one time , we would like to have a way of distinguishing between octahedra and so an invariant to translation and symmetries , like the side lengths , will be good . such an invariant is the set of @xmath110 values which are given by the four pairs of opposite ( parallel ) faces of a regular octahedron . we must have @xmath182 so , knowing @xmath183 and the @xmath184 s will give us @xmath185 s . the set @xmath186 is clearly and invariant to translations and the symmetries we have talked about at the beginning of the paper . hence , two octahedra with different sets of k - values will be essentially different . we determine these k - values from the first three points given in the procedure @xmath187 . the following calculation finds @xmath188 , given @xmath111 such that @xmath130 . .... > fourkvalues:=proc(t ) local n1,n2,n3,n4,x , length ; length:=distance(t[1],t[2])/sqrt(2 ) ; n1:=unitvector(addvec(t[1],t[2],t[3 ] ) ) ; n2:=unitvector(addvec(t[1],t[2],multbyfactorv(t[3],-3 ) ) ) ; n3:=unitvector(addvec(t[1],t[3],multbyfactorv(t[2],-3 ) ) ) ; n4:=unitvector(addvec(t[3],t[2],multbyfactorv(t[1],-3 ) ) ) ; { length / n1,length / n2,length / n3,length / n4 } ; end : .... .... > orbitbox1:=proc(t ) local i , k , t1,a , b , c , x , t2,t3,t4,t5,t6,t7,t8,t9,t10,t11,t12,t13,t14,t15,t16,t17,t18 , t19,t20,t21,t22,t23,t24,s , q ; q:=convert(t , list ) ; a:=max(q[1][1],q[2][1],q[3][1],q[4][1],q[5][1],q[6][1 ] ) ; b:=max(q[1][2],q[2][2],q[3][2],q[4][2],q[5][2],q[6][2 ] ) ; c:=max(q[1][3],q[2][3],q[3][3],q[4][3],q[5][3],q[6][3 ] ) ; t1:=t ; t2:={seq([q[i][1],q[i][2],c - q[i][3]],i=1 .. 6 ) } ; t3:={seq([q[i][1],b - q[i][2],q[i][3]],i=1 .. 6 ) } ; t4:={seq([a - q[i][1],q[i][2],q[i][3]],i=1 .. 6 ) } ; t5:={seq([a - q[i][1],b - q[i][2],q[i][3]],i=1 .. 6 ) } ; t6:={seq([a - q[i][1],q[i][2],c - q[i][3]],i=1 .. 6 ) } ; t7:={seq([q[i][1],b - q[i][2],c - q[i][3]],i=1 .. 6 ) } ; t8:={seq([a - q[i][1],b - q[i][2],c - q[i][3]],i=1 .. 6 ) } ; s:={t1,t2,t3,t4,t5,t6,t7,t8 } ; s ; end : .... .... > orbitbox:=proc(t , p ) local s , q , tt , a , b , c , m , ss , i , d , u , v , w , y , mm , nn , x , k ; q:=convert(t , list ) ; a:=max(q[1][1],q[2][1],q[3][1],q[4][1],q[5][1],q[6][1 ] ) ; b:=max(q[1][2],q[2][2],q[3][2],q[4][2],q[5][2],q[6][2 ] ) ; c:=max(q[1][3],q[2][3],q[3][3],q[4][3],q[5][3],q[6][3 ] ) ; d:=max(a , b , c);u:=d - a;v:=d - b;w:=d - c;y:={u , v , w};k:=p - d ; tt[1]:={seq([q[i][3],q[i][2],q[i][1]],i=1 .. 6 ) } ; tt[2]:={seq([q[i][2],q[i][3],q[i][1]],i=1 .. 6 ) } ; tt[3]:={seq([q[i][1],q[i][3],q[i][2]],i=1 .. 6 ) } ; tt[4]:={seq([q[i][2],q[i][1],q[i][3]],i=1 .. 6 ) } ; tt[5]:={seq([q[i][3],q[i][1],q[i][2]],i=1 .. 6 ) } ; s:=orbitbox1(t ) ; for i from 1 to 5 do s:=s union orbitbox1(tt[i ] ) ; od ; s:=convert(s , list);m:=nops(s);ss:= { } ; for i from 1 to m do if s[i][1][1]<=a and s[i][2][1]<=a and s[i][3][1]<=a and s[i][4][1]<=a and s[i][5][1]<=a and s[i][6][1]<=a and s[i][1][2]<=b and s[i][2][2]<=b and s[i][3][2]<=b and s[i][4][2]<=b and s[i][5][2]<=b and s[i][6][2]<=b and s[i][1][3]<=c and s[i][2][3]<=c and s[i][3][3]<=c and s[i][4][3]<=c and s[i][5][3]<=c and s[i][6][3]<=c and s[i][1][1]>=0 and s[i][2][1]>=0 and s[i][3][1]>=0 and s[i][4][1]>=0 and s[i][5][1]>=0 and s[i][6][1]>=0 and s[i][1][2]>=0 and s[i][2][2]>=0 and s[i][3][2]>=0 and s[i][4][2]>=0 and s[i][5][2]>=0 and s[i][6][2]>=0 and s[i][1][3]>=0 and s[i][2][3]>=0 and s[i][3][3]>=0 and s[i][4][3]>=0 and s[i][5][3]>=0 and s[i][6][3]>=0 then ss:=ss union { s[i ] } ; fi ; od ; mm:=nops(ss ) ; nn:=(k+u+1)*(k+v+1)*(k+w+1 ) ; if nops(y)=3 then x:=6*mm*nn;fi ; if nops(y)=2 then x:=3*mm*nn;fi ; if nops(y)=1 then x:=mm*nn;fi ; [ x , ss , nops(ss),[u , v ] ] ; end : .... the next two routines calculate the orbit of an octahedron within its minimal cube @xmath13 . this orbit has at most 48 elements and it is needed in the process of elimination of octahedrons that have already appeared in the construction . in comparison with the previous orbit , it is bigger and invariant to all the symmetries . .... > orbit1octahedron:=proc(t ) local i , k , t1,a , b , c , x , t2,t3,t4,t5,t6,t7,t8,t9,t10,t11,t12,t13,t14 , t15,t16,t17,t18,t19,t20,t21,t22,t23,t24,s , q , d , a1,b1,c1 ; q:=convert(t , list ) ; d:=mscofmoctahedron(t ) ; t1:=t ; t2:={seq([q[k][2],q[k][3],q[k][1]],k=1 .. 6 ) } ; t3:={seq([q[k][1],q[k][3],q[k][2]],k=1 .. 6 ) } ; t4:={seq([q[k][1],q[k][2],d - q[k][3]],k=1 .. 6 ) } ; t5:={seq([q[k][2],q[k][3],d - q[k][1]],k=1 .. 6 ) } ; t6:={seq([q[k][1],q[k][3],d - q[k][2]],k=1 .. 6 ) } ; t7:={seq([q[k][1],d - q[k][2],q[k][3]],k=1 .. 6 ) } ; t8:={seq([q[k][2],d - q[k][3],q[k][1]],k=1 .. 6 ) } ; t9:={seq([q[k][1],d - q[k][3],q[k][2]],k=1 .. 6 ) } ; t10:={seq([d - q[k][1],q[k][2],q[k][3]],k=1 .. 6 ) } ; t11:={seq([d - q[k][2],q[k][3],q[k][1]],k=1 .. 6 ) } ; t12:={seq([d - q[k][1],q[k][3],q[k][2]],k=1 .. 6 ) } ; t13:={seq([q[k][1],d - q[k][2],d - q[k][3]],k=1 .. 6 ) } ; t14:={seq([q[k][2],d - q[k][3],d - q[k][1]],k=1 .. 6 ) } ; t15:={seq([q[k][1],d - q[k][3],d - q[k][2]],k=1 .. 6 ) } ; t16:={seq([d - q[k][1],d - q[k][2],q[k][3]],k=1 .. 6 ) } ; t17:={seq([d - q[k][2],d - q[k][3],q[k][1]],k=1 .. 6 ) } ; t18:={seq([d - q[k][1],d - q[k][3],q[k][2]],k=1 .. 6 ) } ; t19:={seq([d - q[k][1],q[k][2],d - q[k][3]],k=1 .. 6 ) } ; t20:={seq([d - q[k][2],q[k][3],d - q[k][1]],k=1 .. 6 ) } ; t21:={seq([d - q[k][1],q[k][3],d - q[k][2]],k=1 .. 6 ) } ; t22:={seq([d - q[k][1],d - q[k][2],d - q[k][3]],k=1 .. 6 ) } ; t23:={seq([d - q[k][2],d - q[k][3],d - q[k][1]],k=1 .. 6 ) } ; t24:={seq([d - q[k][1],d - q[k][3],d - q[k][2]],k=1 .. 6 ) } ; s:={t1,t2,t3,t4,t5,t6,t7,t8,t9,t10,t11,t12,t13,t14,t15,t16 , t17,t18,t19,t20,t21,t22,t23,t24 } ; s ; end : > orbitoctahedron:=proc(t ) local s , q , t1 ; q:=convert(t , list ) ; t1:={seq([q[k][3],q[k][2],q[k][1]],k=1 .. 6 ) } ; s:=orbit1octahedron(t ) union orbit1octahedron(t1 ) ; s ; end : .... * step 5 . * at this point we are ready to build a list of minimal , irreducible octahedrons . with each entry we keep , in order , @xmath133 , the value @xmath42 which is the size of the minimal cube @xmath192 , the six vertices of the octahedron , the @xmath110-values and the corresponding vector @xmath111 . in the end , another list is created in a new procedure , to get the list of all corresponding reducible cubes needed in a calculation for a given dimension @xmath43 . the construction is a little complicated because there are octahedrons for which all @xmath110-values are different of @xmath193 . .... > extendlist:=proc(n , n , l::list , mm , nn , orb::array ) local d , i , sol , nsol , nel , ttpcube , orb::array , ll::list , tnel , nl::list , cio , o , po , k , kvalues , m , exception , int , nn ; nn:=floor((n+1)/2 ) ; orb:=array(1 .. 2*nn+1 ) ; nel:=nops(l ) ; ll:=l ; k:=sqrt(mm^2-mm*nn+nn^2 ) ; m:=n*k ; if m<=n then sol:=abcsol(n);nsol:=nops(sol ) ; tnel:=nel;orb:=orb ; for i from 1 to nsol do o:=findpar(sol[i][1],sol[i][2],sol[i][3],mm , nn);po:=tmttopqoctahedron(o ) ; kvalues:=fourkvalues([o[1],o[2],o[3 ] ] ) ; exception:=evalb(1 in kvalues ) ; if k=1 or exception = false then int:=orbitoctahedron(po ) intersect orb[m ] ; cio:=nops(int ) ; if cio=0 then orb[m]:=orb[m ] union orbitbox(po , n)[2 ] ; d:=mscofmoctahedron(po ) ; nl[tnel+1]:=[n , d , po , kvalues , sol[i ] ] ; tnel:=tnel+1 ; fi ; fi ; od ; ll:=[seq(l[i],i=1 .. nel),seq(nl[j],j = nel+1 .. tnel ) ] ; fi;ll , orb ; end : .... .... > extendlistupton:=proc(n ) local i , j , k , l , kv , kvn , nn , l , orb::array , mn , nmn , e , n , ii ; kv:=kvalues(n);kvn:=nops(kv ) ; nn:=floor((n+1)/2 ) ; l:=[];orb:=array(1 .. 2*nn+1 ) ; for ii from 1 to 2*nn+1 do orb[ii]:= { } ; od ; for i from 1 to kvn do k:=kv[i];mn:=listofmn(k);nmn:=nops(mn ) ; for j from 1 to nmn do for l from 1 to nn+1 do n:=2*l-1 ; e:=extendlist(n , n , l , mn[j][1],mn[j][2],orb ) ; l:=e[1 ] ; for ii from 1 to 2*nn+1 do orb[ii]:=e[2][ii ] ; od ; od ; od ; od ; l;end : .... .... > l:=extendlistupton(100 ) : extendlistuptonmultiples:=proc(n , l ) local i , j , x , lc , m , mm , d , dd , c , cc , ll ; m:=nops(l);i:=1;ll:= { } ; while i<=m do d:=l[i][2 ] ; if d<=n then mm:=floor(n / d);c:=l[i][3];j:=2 ; while j<=mm do cc:=multbyfactoroctahedron(c , j ) ; dd:=d*j;lc:=nops(ll ) ; ll:=ll union { [ l[i][1]*j , dd , cc , l[i][4 ] ] } ; j:=j+1 ; od ; fi ; i:=i+1 ; od ; convert(ll , list ) ; end : .... > addupnew:=proc(n , l , ll ) local i , j , k , nc , mm , m , d , c , cc , x , dd , nt ; nc:=0 ; m:=nops(l ) ; i:=1 ; while i<=m do d:=l[i][2 ] ; if d<=n then x:=orbitbox(l[i][3],n)[1 ] ; nc:=nc+x ; fi ; i:=i+1 ; od ; m:=nops(ll ) ; i:=1 ; while i<=m do d:=ll[i][2 ] ; if d<=n then x:=orbitbox(ll[i][3],n)[1 ] ; nc:=nc+x ; fi ; i:=i+1 ; od ; nc;end : .... no : = [ [ 1 , 0 ] , [ 2 , 1 ] , [ 3 , 8 ] , [ 4 , 32 ] , [ 5 , 104 ] , [ 6 , 261 ] , [ 7 , 544 ] , [ 8 , 1000 ] , [ 9 , 1696 ] , [ 10 , 2759 ] , [ 11 , 4296 ] , [ 12 , 6434 ] , [ 13 , 9352 ] , [ 14 , 13243 ] , [ 15 , 18304 ] , [ 16 , 24774 ] , [ 17 , 32960 ] , [ 18 , 43223 ] , [ 19 , 55976 ] , [ 20 , 71752 ] , [ 21 , 90936 ] , [ 22 , 113973 ] , [ 23 , 141312 ] , [ 24 , 173436 ] , [ 25 , 210960 ] , [ 26 , 254587 ] , [ 27 , 305000 ] , [ 28 , 364406 ] , [ 29 , 432824 ] , [ 30 , 511421 ] , [ 31 , 600992 ] , [ 32 , 702556],[33 , 817200 ] , [ 34 , 946131 ] , [ 35 , 1090392 ] , [ 36 , 1251238 ] , [ 37 , 1430072 ] , [ 38 , 1629391 ] , [ 39 , 1850064 ] , [ 40 , 2094276 ] , [ 41 , 2363616 ] , [ 42 , 2659813 ] , [ 43 , 2984600 ] , [ 44 , 3341660 ] , [ 45 , 3731720 ] , [ 46 , 4156689 ] , [ 47 , 4618480 ] , [ 48 , 5119292 ] , [ 49 , 5661600 ] , [ 50 , 6248705 ] , [ 51 , 6882808 ] , [ 52 , 7568126 ] , [ 53 , 8306520 ] , [ 54 , 9104339 ] , [ 55 , 9962320 ] , [ 56 , 10888762 ] , [ 57 , 11882896 ] , [ 58 , 12949661 ] , [ 59 , 14090952 ] , [ 60 , 15311286 ] , [ 61 , 16613736 ] , [ 62 , 18001975 ] , [ 63 , 19479680 ] , [ 64 , 21052826 ] , [ 65 , 22724576 ] , [ 66 , 24500175 ] , [ 67 , 26383240 ] , [ 68 , 28387456 ] , [ 69 , 30510616 ] , [ 70 , 32758963 ] , [ 71 , 35136544 ] , [ 72 , 37656214 ] , [ 73 , 40317328 ] , [ 74 , 43125329 ] , [ 75 , 46085496 ] , [ 76 , 49207224 ] , [ 77 , 52493112 ] , [ 78 , 55954267 ] , [ 79 , 59592272 ] , [ 80 , 63415296 ] , [ 81 , 67428832 ] , [ 82 , 71642127 ] , [ 83 , 76059704 ] , [ 84 , 80701546 ] , [ 85 , 85565064 ] , [ 86 , 90662451 ] , [ 87 , 95997360 ] , [ 88 , 101592122 ] , [ 89 , 107443264 ] , [ 90 , 113561009 ] , [ 91 , 119951832 ] , [ 92 , 126644136 ] , [ 93 , 133629672 ] , [ 94 , 140916757 ] , [ 95 , 148513712 ] , [ 96 , 156444624 ] , [ 97 , 164706400 ] , [ 98 , 173308509 ] , [ 99 , 182260568 ] , [ 100 , 191575248 ] ] r. guy , _ unsolved problems in number theory _ , springer - verlag , 2004 e. grosswald , _ representations of integers as sums of squares _ , springer verlag , new york , 1985 . i. j. schoenberg , _ regular simplices and quadratic forms _ , j. london math . soc . 12 ( 1937 ) 48 - 55 .

rotating waves are periodic solutions in equivariant dynamical systems . their precession frequency changes with parameters and it may change sign , passing through zero . when this happens , the dynamical system is very sensitive to imperfections that break the symmetry and the waves may become trapped by the imperfections , resulting in steady solutions that exist in a finite region in parameter space . this is the so - called pinning phenomenon . in this study , we analyze the breaking of the symmetry in a dynamical system close to a hopf bifurcation whose frequency changes sign along a curve in parameter space . the problem is very complex , as it involves the complete unfolding of high codimension . a detailed analysis of different types of imperfections indicates that a pinning region surrounded by infinite - period bifurcation curves appears in all cases . complex bifurcational processes , strongly dependent on the specifics of the symmetry breaking , appear very close to the intersection of the hopf bifurcation and the pinning region . scaling laws of the pinning region width , and partial breaking of to , are also considered . previous and new experimental and numerical studies of pinned rotating waves are reviewed in light of the new theoretical results .

1206.1643

dynamical systems theory plays an important role in many areas of mathematics and physics because it provides the building blocks that allow us to understand the changes many physical systems experience in their dynamics when parameters are varied . these building blocks are the generic bifurcations ( saddle - node , hopf , etc . ) that any arbitrary physical system experiences under parameter variation , regardless of the physical mechanisms underlying the dynamics . when one single parameter of the system under consideration is varied , codimension - one bifurcations are expected . if the system depends on more parameters , higher codimension bifurcations appear and they act as organizing centers of the dynamics . the presence of symmetries changes the nature and type of bifurcations that a dynamical system may undergo . symmetries play an important role in many idealized situations , where simplifying assumptions and the consideration of simple geometries result in dynamical systems equivariant under a certain symmetry group . bifurcations with symmetry have been widely studied @xcite . however , in any real system , the symmetries are only approximately fulfilled , and the breaking of the symmetries , due to the presence of noise , imperfections and/or other phenomena , is always present . there are numerous studies of how imperfect symmetries lead to dynamics that are unexpected in the symmetric problem , e.g. @xcite . however , a complete theory is currently unavailable . one observed consequence of imperfections in systems that support propagating waves is that the waves may become trapped by the imperfections ( e.g. , see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in these various examples , the propagation direction is typically biased . however , a more recent problem has considered a case where a rotating wave whose sense of precession changes sign is pinned by symmetry - breaking imperfections @xcite . we are unaware of any systematic analysis of the associated normal form dynamics for such a problem and this motivates the present study . when a system is invariant to rotations about an axis ( invariance under the @xmath0 symmetry group ) , @xmath0-symmetry - breaking hopf bifurcations result in rotating waves , consisting of a pattern that rotates about the symmetry axis at a given precession frequency without changing shape . this frequency is parameter dependent , and in many problems , when parameters are varied , the precession frequency changes sign along a curve in parameter space . what has been observed in different systems is that in the presence of imperfections , the curve of zero frequency becomes a band of finite width in parameter space . within this band , the rotating wave becomes a steady solution . this is the so - called pinning phenomenon . it can be understood as the attachment of the rotating pattern to some stationary imperfection of the system , so that the pattern becomes steady , as long as its frequency is small enough so that the imperfection is able to stop the rotation . this pinning phenomenon bears some resemblance to the frequency locking phenomena , although in the frequency locking case we are dealing with a system with two non - zero frequencies and their ratio becomes constant in a region of parameter space ( a resonance horn ) , whereas here we are dealing with a single frequency crossing zero . in the present paper , we analyze the breaking of @xmath0 symmetry in a dynamical system close to a hopf bifurcation whose frequency changes sign along a curve in parameter space . the analysis shows that breaking @xmath0 symmetry is much more complex than expected , resulting in a bifurcation of high codimension ( about nine ) . although it is not possible to analyze in detail such a complex and high - codimension bifurcation , we present here the analysis of five different ways to break @xmath0 symmetry . this is done by introducing into the normal form all the possible terms , up to and including second order , that break the symmetry , and analyzing each of these five terms separately . three of these particular cases have already been analyzed in completely different contexts unrelated to the pinning phenomenon @xcite . in the present study , we extract the common features that are associated with the pinning . in all cases , we find that a band of pinning solutions appears around the zero frequency curve from the symmetric case , and that the band is delimited by curves of infinite - period bifurcations . the details of what happens when the infinite - period bifurcation curves approach the hopf bifurcation curve are different in the five cases , and involve complicated dynamics with several codimension - two bifurcations occurring in a small region of parameter space as well as several global bifurcations . interest in the present analysis is two - fold . first of all , although the details of the bifurcational process close to the zero - frequency hopf point are very complicated and differ from case to case , for all cases we observe the appearance of a pinning band delimited by infinite - period bifurcations of homoclinic type that , away from the small region of complicated dynamics , are snic bifurcations ( saddle - node on an invariant circle bifurcation , e.g. see * ? ? ? secondly , some of the scenarios analyzed are important _ per se _ , as they correspond to the generic analysis of a partial breaking of the @xmath0 symmetry , so that after the introduction of perturbations , the system still retains a discrete symmetry ( the @xmath2 case is analyzed in detail ) . the paper is organized as follows . in section [ sec_perfect_hopf ] the properties of a hopf bifurcation with @xmath0 symmetry with the precession frequency crossing through zero are summarized , and the general unfolding of the @xmath0 symmetry breaking process is discussed . the next sections explore the particulars of breaking the symmetry at order zero ( [ sec_epsilon ] ) , one ( [ sec_bar_z ] ) and two ( [ sec_quadratic ] ) . sections [ sec_bar_z ] and [ sec_z3 ] are particularly interesting because they consider the symmetry - breaking processes @xmath3 and @xmath4 which are readily realized experimentally . section [ sec_general ] extracts the general features of the pinning problem from the analysis of the specific cases carried out in the earlier sections . section [ sec_experiments ] presents comparisons with experiments and numerical computations in two real problems in fluid dynamics , illustrating the application of the general theory developed in the present study . finally , in [ conclusions ] , conclusions and perspectives are presented . the normal form for a hopf bifurcation is @xmath5 where @xmath6 is the complex amplitude of the bifurcating periodic solution , @xmath7 is the bifurcation parameter , and @xmath8 and @xmath9 are functions of @xmath7 and generically at the bifurcation point ( @xmath10 ) both are different from zero . it is the non - zero character of @xmath8 that allows one to eliminate the quadratic terms in @xmath6 in the normal form . this is because the normal form @xmath11 satisfies ( e.g. , see * ? ? ? * ) @xmath12 where @xmath13 is a low order polynomial that captures the dynamics in a neighborhood of the bifurcation point . if @xmath14 , this equation becomes an identity and @xmath13 can not be simplified . the case @xmath14 is a complicated bifurcation and it depends on the details of the double - zero eigenvalue of the linear part @xmath15 of @xmath13 ; as @xmath16 is complex , the matrix of @xmath15 using the real coordinates @xmath17 is a real @xmath18 matrix . if @xmath15 is not completely degenerate , that is @xmath19 then we have the well - studied takens bogdanov bifurcation , whereas the completely degenerate case , @xmath20 is a high - codimension bifurcation that has not been completely analyzed . if the system has @xmath0 symmetry , it must also satisfy @xmath21 where @xmath1 is the discrete symmetry retained by the bifurcated solution . when the group @xmath1 is generated by rotations of angle @xmath22 about an axis of @xmath23-fold symmetry , as is usually the case with @xmath0 , then the group is also called @xmath24 . equations and are completely equivalent and have the same implications for the normal form structure . advancing in time is the same as rotating the solution by a certain angle ( @xmath25 ) ; the bifurcated solution is a rotating wave . therefore , if @xmath8 becomes zero by varying a second parameter , we still have the same normal form , due to , with @xmath8 replaced by a small parameter @xmath26 : @xmath27 the hopf bifurcation with @xmath0 symmetry and zero frequency is , in this sense , trivial . introducing the modulus and phase of the complex amplitude @xmath28 , the normal form becomes @xmath29 where @xmath30 , and let us assume for the moment that @xmath31 and @xmath32 are positive . the bifurcation frequency in is now the small parameter @xmath26 . the bifurcated solution @xmath33 exists only for @xmath34 , and has amplitude @xmath35 and frequency @xmath36 . the limit cycle @xmath33 becomes an invariant set of steady solutions along the straight line @xmath37 ( labeled l in figure [ zerohopf ] ) where the frequency of @xmath33 goes to zero ; the angle between l and the hopf bifurcation curve ( the horizontal axis @xmath10 ) is @xmath38 . the bifurcation diagram and a schematic of the bifurcations along a one - dimensional path is also shown in figure [ zerohopf ] . the bifurcation point @xmath39 , labeled zf ( zero - frequency hopf point ) in figure [ zerohopf]@xmath40 , is a codimension - two bifurcation . it coincides with the generic hopf bifurcation , except that it includes a line l along which the bifurcated solution has zero frequency . symmetry and zero frequency ; part @xmath40 shows the bifurcation diagram , where the thick lines are bifurcation curves , and part @xmath41 shows the bifurcations along the path a shown in @xmath40 . the fixed point curve is labeled with the signs of its eigenvalues . in regions ii@xmath42 and ii@xmath43 the limit cycles , born at the hopf bifurcations h@xmath42 and h@xmath43 , rotate in opposite senses . l is the line where the limit cycle becomes an invariant curve of fixed points . ] assuming @xmath44 , we can simplify by scaling @xmath6 so that @xmath45 ; we will write @xmath46 which helps simplify subsequent expressions . the case @xmath31 and @xmath32 both positive , which we will analyze in detail in the following sections , corresponds to one of the fluid dynamics problems that motivated the present analysis . for other signs of @xmath31 and @xmath32 , analogous conclusions can be drawn . it is of particular interest to consider the subcritical case @xmath47 as it corresponds to the other fluid dynamics problem analyzed here . by reversing time and changing the sign of @xmath7 and @xmath26 , we obtain exactly the same normal form but with the opposite sign of @xmath31 and @xmath32 . by changing the sign of @xmath48 and @xmath26 , we obtain with the opposite sign of @xmath32 . therefore , all possible cases corresponding to different signs of @xmath31 and @xmath32 can be reduced to the case where @xmath31 and @xmath32 are both positive . if the @xmath0 symmetry in the normal form is completely broken , and no symmetry remains , then the restrictions imposed on the normal form by disappear completely and all the terms in @xmath6 and @xmath49 missing from will reappear multiplied by small parameters . this means that the normal form will be @xmath50 where additional cubic terms have been neglected because we assume @xmath44 and that @xmath51 will be dominant . as the @xmath52 are complex , we have a problem with 12 parameters . additional simplifications can be made in order to obtain the so - called hypernormal form ; this method is extensively used by @xcite , for example . unfortunately , many of the simplifications rely on having some low - order term in the normal form being non - zero with a coefficient of order one . for example , if @xmath53 , it is possible to make @xmath9 real by using a time re - parametrization . in our problem , all terms up to and including second order are zero or have a small coefficient , and so only a few simplifications are possible . these simplifications are an infinitesimal translation of @xmath6 ( two parameters ) , and an arbitrary shift in the phase of @xmath6 ( one parameter ) . using these transformations the twelve parameters can be reduced to nine . in particular , one of either @xmath54 or @xmath55 can be taken as zero and the other can be made real . by rescaling @xmath6 , we can make @xmath9 of modulus one , as in . a complete analysis of a normal form depending on nine parameters , i.e. a bifurcation of codimension of about nine , is completely beyond the scope of the present paper . in the literature , only codimension - one bifurcations have been completely analyzed . most of the codimension - two bifurcations for ode and maps have also been analyzed , except for a few bifurcations for maps that remain outstanding @xcite . a few codimension - three and very few codimension - four bifurcations have also been analyzed @xcite , but to our knowledge , there is no systematic analysis of bifurcations of codimension greater than two . in the following sections , we consider the five cases , @xmath56 to @xmath55 , separately . a combination of analytical and numerical tools allows for a detailed analysis of these bifurcations . we extract the common features of the different cases when @xmath57 , which captures the relevant behavior associated with weakly breaking @xmath0 symmetry . in particular , the @xmath58 case exhibits very interesting and rich dynamics that may be present in some practical cases when the @xmath0 symmetry group is not completely broken and a @xmath2 symmetry group , generated by the half - turn @xmath59 , remains . some general comments can be made here about these five cases , which are of the form @xmath60 for integers @xmath61 , excluding the case @xmath62 which is @xmath0 equivariant and so @xmath63 can be absorbed into @xmath7 and @xmath26 . by changing the origin of the phase of @xmath6 , we can modify the phase of @xmath63 so that it becomes real and positive . then , by re - scaling @xmath6 , time @xmath64 , and the parameters @xmath7 and @xmath26 as @xmath65 we obtain with @xmath66 , effectively leading to codimension - two bifurcations in each of the five cases . we expect complex behavior for @xmath67 , when the three parameters are of comparable size , while the effects of small imperfections breaking @xmath0 will correspond to @xmath68 . from now on @xmath66 will be assumed , and we can restore the explicit @xmath63-dependence by reversing the transformation . three of the five normal forms have been analyzed in the literature ( discussed below ) , focusing on the regions where @xmath7 , @xmath26 and @xmath63 are of comparable size ; here we will also consider what happens for @xmath68 which is particularly important for the pinning phenomenon . the normal forms corresponding to the @xmath56 , @xmath58 and @xmath69 cases have already been analyzed in contexts completely different to the @xmath0 symmetry - breaking context considered here . the context in which these problems were studied stems from low - order resonances in perturbed hopf problems . @xcite studied time - periodic forcing near a hopf bifurcation point , analyzing the problem using the poincar stroboscopic map . the normal forms corresponding to the 1:1 , 1:2 and 1:3 strong resonances coincide with the normal forms we present below for cases with only the @xmath56 , @xmath58 and @xmath69 terms retained in , respectively . later , motivated by a problem of a nonlinear oscillator with damping and quasi - periodic driving , a series of papers extended the strong resonances results of @xcite by studying the semi - global bifurcations for periodically and quasi - periodically perturbed driven damped oscillators near a hopf bifurcation ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . the other two cases we consider , with only the @xmath54 or the @xmath55 terms retained in , do not appear to have been studied previously . they fall outside of the context in which the other three were studied because they do not correspond to any canonical resonance problem . we should point out that within the resonance context , the three cases studied would not make sense to consider in combination ( they correspond to completely distinct frequency ratios and so would not generically occur in a single problem ) . in contrast , within the context motivating our study , all five cases correspond to different ways in which the @xmath0 symmetry of a system may be broken , and in a physical realization , all five could co - exist . in the following sections , we present a detailed analysis of all five cases . the normal form to be analyzed is with @xmath70 and @xmath66 : @xmath71 this case has been analyzed in @xcite . it is convenient to introduce coordinates @xmath72 in parameter space , rotated an angle @xmath38 with respect to @xmath73 , so that the line l becomes the new coordinate axis @xmath74 . many features , that are symmetric with respect to l will then simplify ; e.g. the distance to the bifurcation point along l is precisely @xmath75 . @xmath76 where @xmath77 , @xmath78 . figure [ uv_and_epsilon_sn]@xmath40 shows the relationship between the two coordinate systems . [ cols="^,^ " , ] another feature we have found in the five scenarios discussed above is that the snic@xmath79 bifurcations , where the stable limit cycle disappears on entering the pinning region away from the origin ( @xmath39 ) , become homoclinic ( or heteroclinic if the @xmath0 symmetry is not completely broken ) collisions between the stable limit cycle and the saddle point that is born in the saddle - node bifurcations sn@xmath79 that now take place not on the limit cycle but very close to it . there is a codimension - two global bifurcation , termed snichom in the bifurcation diagrams discussed above , where the snic curve , the saddle - node curve and the homoclinic collision curve meet . the scaling laws of the periods when approaching a homoclinic or a snic bifurcation are different , having logarithmic or square root profiles : @xmath80 where @xmath81 is the positive eigenvalue of the saddle , and @xmath82 a constant . so the question is what happens with the scaling of the period close to the snichom bifurcation ? we have analized in detail the @xmath83 case , being the other cases very similar . we refer to the figures [ imphopf_regions ] and [ snic_het ] from a practical point of view , close to but before the snichet@xmath42 point , the saddle - node bifurcation sn@xmath42 is very closely followed by the heteroclinic collision of the limit cycle @xmath84 with the saddles @xmath85 and @xmath86 , and it becomes almost indistinguishable from the snic@xmath42 bifurcation . we have numerically computed the period of @xmath84 at @xmath87 for decreasing @xmath7 values approaching sn@xmath42 in the range @xmath88 $ ] , for @xmath89 . figure [ snic_versus_hom]@xmath40 shows both fits using the values of the period over the whole computed range . the log fit overestimates the period while the square - root fit underestimates it , and this underestimate gets larger as the heteroclinic collision is approached . figure [ snic_versus_hom]@xmath41 again shows both fits , but now using values close to the collision for the log fit and values far away from the collision for the square - root fit . both fits are now very good approximations of the period in their corresponding intervals , and together cover all the values numerically computed . when the interval between the sn@xmath42 bifurcation and the heteroclinic collision ( in figure [ snic_versus_hom ] , @xmath90 and @xmath91 , respectively ) is very small , it can not be resolved experimentally ( or even numerically in an extended systems with millions of degrees of freedom , as is the case in fluid dynamics governed by the three - dimensional navier stokes equations ) . in such a situation the square - root fit looks good enough , because away from the sn@xmath42 point , the dynamical system just feels the ghost of the about to be formed saddle - node pair and does not distinguish between whether the saddle - node appears on the limit cycle or very close to it . however , if we are able to resolve the very narrow parameter range between the saddle - node formation and the subsequent collision with the saddle , then the log fit matches the period in this narrow interval much better . due to the presence of two very close bifurcations ( het and sn@xmath42 ) , the scaling laws become cross - contaminated , and from a practical point of view the only way to distinguish between a snic and a homoclinic collision is by computing or measuring periods extremelly close to the infinite - period bifurcation point . we can also see this from the log fit equation in ; when both bifurcations are very close , @xmath81 , the positive eigenvalue of the saddle , goes to zero ( it is exactly zero at the saddle - node point ) , so the log fit becomes useless , except when the periods are very large . when the interval between the sn@xmath42 bifurcation and the homoclinic collision is very small , it can not be resolved experimentally ( or even numerically in an extended systems with millions of degrees of freedom , as is the case in fluid dynamics governed by the three - dimensional navier stokes equations ) . in such a situation , the square - root fit appears to be good enough , because away from the sn@xmath42 point , the dynamical system just feels the ghost of the about - to - be - formed saddle - node pair and does not distinguish between whether the saddle - node appears on the limit cycle or just very close to it . however , if we are able to resolve the very narrow parameter range between the saddle - node formation and the subsequent collision with the saddle , then the log fit matches the period in this narrow interval much better . the bifurcations that a limit cycle can undergo have been an active subject of research since dynamical systems theory was born . even in the case of isolated codimension - one bifurcations , a complete classification was not completed until fifteen years ago @xcite , when the blue - sky catastrophe was found . the seven possible bifurcations are : the hopf bifurcation , where a limit cycle shrinks to a fixed point , and the length of the limit cycle reduces to zero . three bifurcations where both the length and period of the limit cycle remain finite : the saddle node of cycles ( or cyclic fold ) , the period doubling and the neimark - sacker bifurcations . two bifurcations where the length remains finite but the period goes to infinity : the collision of the limit cycle with an external saddle forming a homoclinic loop , and the appearance of a saddle - node of fixed points on the limit cycle ( the snic bifurcation ) . finally , there is the blue - sky bifurcation where both the length and period go to infinity , corresponding to the appearance of a saddle - node of limit cycles transverse to the given limit cycle . the seven bifurcations are described in many books on dynamical systems , e.g. @xcite ; they are also described on the web page http://www.scholarpedia.org/article/blue-sky_catastrophe maintained by a. shilnikov and d. turaev . of the seven bifurcations , only four ( hopf , cyclic fold , homoclinic collision and snic ) are possible in planar systems , as is the case in the present study , and we have found the four of them in the different scenarios explored . we have also found a number of codimension - two bifurcations of limit cycles . for these bifurcations a complete classification is still lacking , and it is interesting to list them because some of the bifurcations obtained are not very common . the codimension - two bifurcations of limit cycles associated to codimension - two bifurcations of fixed points can be found in many dynamical systems texbooks , and include takens bogdanov bifurcations ( present in almost all cases considered here ) and the bautin bifurcation ( in the @xmath63 case ) . codimension - two bifurcations of limit cycles associated only to global bifurcations are not so common . we have obtained five of them , that we briefly summarize here . pfgl : : a gluing bifurcation with the saddle point undergoing a pitchfork bifurcation ; it may happen in systems with @xmath2 symmetry ; see [ sec_bar_z ] . cfhom : : a cyclic - fold and a homoclinic collision occurring simultaneously ; see [ sec_bar_z ] . cfhet : : a cyclic - fold and a heteroclinic collision occurring simultaneously ; see [ sec_epsilon ] . snichom : : a snic bifurcation and a homoclinic collision occurring simultaneously ; see [ sec_quadratic ] . snichet : : a double snichom bifurcation occurring simultaneously in @xmath2 symmetric systems ; see [ sec_bar_z ] . the snichom bifurcation is particularly important in our problem because it separates the two possible scenarios upon entering the pinning region : the stable limit cycle outside may disappear in a homoclinic collision or a snic bifurcation . experiments in small aspect - ratio taylor couette flows have reported the presence of a band in parameter space where rotating waves become steady non - axisymmetric solutions ( a pinning effect ) via infinite - period bifurcations @xcite . previous numerical simulations , assuming @xmath0 symmetry of the apparatus , were unable to reproduce these observations @xcite . recent additional experiments suggest that the pinning effect is not intrinsic to the dynamics of the problem , but rather is an extrinsic response induced by the presence of imperfections that break the @xmath0 symmetry of the ideal problem . additional controlled symmetry - breaking perturbations were introduced into the experiment by tilting one of the endwalls @xcite . @xcite conducted direct numerical simulations of the navier stokes equations including the tilt of one endwall by a very small angle . those simulations agree very well with the experiments , and the normal form theory developed in this paper provides a theoretical framework for understanding the observations . a brief summary of those results follows . taylor couette flow consists of a fluid confined in an annular region with inner radius @xmath92 and outer radius @xmath93 , capped by endwalls a distance @xmath94 apart . the endwalls and the outer cylinder are stationary , and the flow is driven by the rotation of the inner cylinder at constant angular speed @xmath95 ( see figure [ schemtc ] for a schematic ) . the system is governed by three parameters : @xmath96 where @xmath26 is the kinematic viscosity of the fluid . the system is invariant to arbitrary rotations about the axis , @xmath0 symmetry , and to reflections about the mid - height , a @xmath2 symmetry that commutes with @xmath0 . in both the experiments and the numerical simulations , the radius ratio was kept fixed at @xmath97 . @xmath98 and @xmath99 were varied , and these correspond to the parameters @xmath7 and @xmath26 in the normal forms studied here . m0.02@m0.45 @m0.02@m0.4 @xmath40 & experimental regimes found in the small aspect - ratio taylor - couette problem , with a pinning region , adapted from @xcite . @xmath41 numerically computed rotating wave from @xcite , found in regions ii@xmath79.,title="fig : " ] & @xmath41 & experimental regimes found in the small aspect - ratio taylor - couette problem , with a pinning region , adapted from @xcite . @xmath41 numerically computed rotating wave from @xcite , found in regions ii@xmath79.,title="fig : " ] for small @xmath98 , below the curve pf in figure [ pinning_tc]@xmath40 , the flow is steady , axisymmetric and reflection symmetric , consisting of two taylor vortices @xcite . the @xmath2 reflection symmetry is broken in a pitchfork bifurcation along the curve pf , and a pair of steady axisymmetric one - vortex states that have a jet of angular momentum emerging from the inner cylinder boundary layer near one or other of the endwalls is born . both are stable , and which is realized depends on initial conditions . the only symmetry of these symmetrically - related solutions is @xmath0 . the inset in figure [ schemtc ] shows the azimuthal velocity associated with the state with the jet near the top . these steady axisymmetric one - vortex states are stable in region i. there are other flow states that are stable in this same region . for example , above the dashed curve sn in figure [ pinning_tc]@xmath40 , the two - taylor - vortex state becomes stable and coexists with the one - vortex states . however , the two - vortex and the one - vortex states are well separated in phase space and the experiments and numerics we describe below are focused on the one - vortex state . on increasing @xmath98 , the one - vortex state suffers a hopf bifurcation that breaks the @xmath0 symmetry and a rotating wave state emerges with azimuthal wave number @xmath100 . figure [ pinning_tc]@xmath41 shows an isosurface of axial angular momentum , illustrating the three - dimensional structure of the rotating wave . for slight variations in aspect ratio , the rotating wave may precess either prograde ( in region ii@xmath43 above h@xmath43 ) or retrograde ( in region ii@xmath42 above h@xmath42 ) with the inner cylinder , and in between a pinning region iii is observed . this is observed even with a nominally perfect experimental system , i.e. with the @xmath0 symmetry to within the tolerances in constructing the apparatus . the hopf bifurcation is supercritical around the region where the precession frequency changes sign . however , for smaller aspect ratios the hopf bifurcation becomes subcritical at the bautin point ba in figure [ pinning_tc]@xmath40 . m0.02@m0.43 @m0.02@m0.43 @xmath40 & the experimental results of @xcite with the natural imperfections of their system , and @xmath41 the numerical results of @xcite with a tilt of @xmath101 on the upper lid . the dotted curve in both is the numerically determined hopf curve with zero tilt.,title="fig : " ] & @xmath41 & the experimental results of @xcite with the natural imperfections of their system , and @xmath41 the numerical results of @xcite with a tilt of @xmath101 on the upper lid . the dotted curve in both is the numerically determined hopf curve with zero tilt.,title="fig : " ] various different experiments in this regime have been conducted in the nominally perfect system , as well as with a small tilt of an endwall @xcite . figure [ bif_diagram_exp]@xmath40 shows a bifurcation diagram from the laboratory experiments of @xcite . these experiments show that without an imposed tilt , the natural imperfections of the system produce a measurable pinning region , and that the additional tilting of one endwall increases the extent of the pinning region . tilt angles of the order of @xmath101 are necessary for the tilt to dominate over the natural imperfections . figure [ bif_diagram_exp]@xmath41 shows a bifurcation diagram from the numerical results of @xcite , including a tilt of one of the endwalls of about @xmath101 , showing very good agreement with the experimental results . included in figures [ bif_diagram_exp]@xmath40 and @xmath41 is the numerically computed bifurcation diagram in the perfect system , shown as dotted curves . the effects of imperfections are seen to be only important in the parameter range where the hopf frequency is close to zero and a pinning region appears . it is bounded by infinite - period bifurcations of limit cycles . the correspondence between these results and the normal form theory described in this paper is excellent , strongly suggesting that the general remarks on pinning extracted from the analysis of the five particular cases are indeed realized both experimentally and numerically . these two studies @xcite are the only cases we know of where quantitative data about the pinning region are available . yet , even in these cases the dynamics close to the intersection of the hopf curve with the pinning region , that according to our analysis should include complicated bifurcational processes , has not been explored either numerically or experimentally . this is a very interesting problem that deserves further exploration . m0.02@m0.36 @m0.02@m0.36 @xmath40 & schematic of the bautin bifurcation , including the path l of frequency zero bifurcated states in the @xmath0 perfect system , shown in a two parameter space @xmath102 with @xmath103 a global measure of the solution . also shown is the projection of the saddle - node surface on parameter space . @xmath41 schematic of the rotating convection apparatus , with the streamlines of the basic state shown in the inset.,title="fig : " ] & @xmath41 & schematic of the bautin bifurcation , including the path l of frequency zero bifurcated states in the @xmath0 perfect system , shown in a two parameter space @xmath102 with @xmath103 a global measure of the solution . also shown is the projection of the saddle - node surface on parameter space . @xmath41 schematic of the rotating convection apparatus , with the streamlines of the basic state shown in the inset.,title="fig : " ] + up to now , we have considered the zero - frequency hopf problem in the context of a supercritical hopf bifurcation . however , in the taylor couette example discussed in the previous section , the zero frequency occurs quite close to a bautin bifurcation , at which the hopf bifurcation switches from being supercritical to subcritical , and a natural question is what are the consequences of the zero - frequency occurring on a subcritical hopf bifurcation . the normal form theory for the behavior local to the hopf bifurcation carries over by changing the direction of time and the sign of the parameters @xmath7 and @xmath26 as discussed before , but then both the limit cycle and the pinned state are unstable and not observable in a physical experiment or direct numerical simulation . the limit cycle becomes observable as it undergoes a saddle - node of limit cycles ( a cyclic fold ) bifurcation at the fold associated with the bautin bifurcation ( see a schematic in figure [ schemrc]@xmath31 ) , and we expect that the pinned state does likewise with a saddle - node of fixed points bifurcation along the same fold . in this subsection , we identify a rotating convection problem where precisely this occurs @xcite , and conduct new numerical simulations by introducing an @xmath0 symmetry - breaking bifurcation that produces a pinning region on the upper branch of the subcritical hopf bifurcation . the rotating convection problem consists of the flow in a circular cylinder of radius @xmath104 and height @xmath94 , rotating at a constant rate @xmath8 rad / s . the cold top and hot bottom endwalls are maintained at constant temperatures @xmath105 , where @xmath106 is the mean temperature and @xmath107 is the temperature difference between the endwalls . the sidewall has zero heat flux . figure [ schemrc]@xmath41 shows a schematic of the flow configuration . using the boussinesq approximation that all fluid properties are constant except for the density in the gravitational and centrifugal buoyancy terms , and using @xmath94 as the length scale , @xmath108 as the time scale , and @xmath107 as the temperature scale , the governing equations written in the rotating frame of reference are : @xmath109 where @xmath110 is the velocity field in cylindrical coordinates @xmath111 in the rotating frame , @xmath112 is the kinematic pressure ( including gravitational and centrifugal contributions ) , @xmath113 the unit vector in the vertical direction @xmath6 , and @xmath114 is the radial vector in cylindrical coordinates . instead of the non - dimensional temperature @xmath115 , we have used the temperature deviation @xmath116 with respect to the conductive profile , @xmath117 , as is customary in many thermal convection studies . there are five non - dimensional independent parameters : @xmath118 where @xmath119 is the coefficient of volume expansion , @xmath120 is the gravitational acceleration , @xmath121 is the thermal diffusivity , and @xmath26 is the kinematic viscosity . the boundary conditions for @xmath122 and @xmath116 are : @xmath123 for any @xmath124 , the system is not invariant to the so - called boussinesq symmetry corresponding to invariance to a reflection @xmath125 about the half - height @xmath126 , whose action is @xmath127 . the system is only invariant under rotations about the axis of the cylinder , the @xmath0 symmetry . the governing equations have been solved using a second - order time - splitting method combined with a pseudo - spectral method for the spatial discretization , utilizing a galerkin fourier expansion in the azimuthal coordinate @xmath128 and chebyshev collocation in @xmath129 and @xmath6 . the details are presented in @xcite . we have used @xmath130 , @xmath131 and @xmath132 spectral modes in @xmath129 , @xmath128 and @xmath6 and a time - step @xmath133 thermal time units in all computations . we have checked the spectral convergence of the code using the infinity norm of the spectral coefficients of the computed solutions . the trailing coefficients of the spectral expansions are at least five orders of magnitude smaller than the leading coefficients . in order to compute the zero - frequency line l in the subcritical region of the bautin bifurcation , where the fixed points and limit cycles involved are unstable , we have used arclength continuation methods for fixed points and for rotating waves adapted to our spectral codes @xcite . m0.02@m0.38 @m0.02@m0.38 @xmath40 & bifurcation curves for @xmath134 , @xmath135 and @xmath136 , where h@xmath137 are the segments of the hopf bifurcation with negative and positive frequency , the bautin point ba is where the hopf bifurcation switches from super- to subcritical and the cyclic - fold bifurcation curve cf emerges . the lines l@xmath138 and l@xmath139 are the loci where the rotating wave has zero frequency on the lower and upper branches of the cyclic fold . the rectangle around l@xmath139 corresponds to figure [ rcpinning]@xmath31 . @xmath41 the frequency along the hopf bifurcation h@xmath137 with the bautin point and the point zf , where the sense of precession changes , marked as open symbols.,title="fig : " ] & @xmath41 & bifurcation curves for @xmath134 , @xmath135 and @xmath136 , where h@xmath137 are the segments of the hopf bifurcation with negative and positive frequency , the bautin point ba is where the hopf bifurcation switches from super- to subcritical and the cyclic - fold bifurcation curve cf emerges . the lines l@xmath138 and l@xmath139 are the loci where the rotating wave has zero frequency on the lower and upper branches of the cyclic fold . the rectangle around l@xmath139 corresponds to figure [ rcpinning]@xmath31 . @xmath41 the frequency along the hopf bifurcation h@xmath137 with the bautin point and the point zf , where the sense of precession changes , marked as open symbols.,title="fig : " ] figure [ rcbifs]@xmath40 shows the parameter region of interest in this convection problem . in the region of high froude number ( region i ) we have a stable steady solution , consisting of a single axisymmetric convective roll where the warm fluid moves upwards close to the axis ( due to the rotation of the container ) , and returns along the sidewall , as illustrated in figure [ schemrc]@xmath41 ; the inset shows streamlines of the flow in this base state , that is @xmath0-equivariant with respect to rotations about the cylinder axis . this base state looses stability when the froude number @xmath140 decreases , in a hopf bifurcation along the curves h@xmath79 . the bifurcation is supercritical for @xmath141 and subcritical for higher @xmath142 ; the change from supercritical to subcritical happens at the codimension - two bautin bifurcation point ba , at @xmath143 . the bifurcated limit cycle , a rotating wave with azimuthal wave number @xmath144 , is unstable , but becomes stable at the cyclic fold curve cf ( a saddle - node bifurcation of limit cycles ) . this curve cf originates at the bautin point ba . there are other flow states that are stable in this same region @xcite ; these additional states are well separated in phase space and the numerics we describe below are focused on the base state and the @xmath144 bifurcated rotating wave . figure [ rcbifs]@xmath41 shows the computed frequency of the limit cycle along the hopf bifurcation curve . this frequency is negative along h@xmath42 and positive along h@xmath43 , and is zero at the zf ( zero frequency ) point . at this point we have precisely the scenario discussed in the present paper : a flow ( the base state ) with @xmath0 symmetry undergoing a hopf bifurcation that has zero frequency at that point . figure [ rcbifs]@xmath40 also includes the line l in parameter space where the frequency of the bifurcated states is zero . this curve has been computed using continuation methods since the zero - frequency state is unstable in the lower part ( l@xmath138 ) of the saddle - node cf , and therefore can not be obtained via time evolution . the zero frequency state becomes stable upon crossing the saddle - node curve cf and moving to the upper part l@xmath139 of the saddle - node cf , and becomes observable both experimentally and by numerical simulations advancing the navier - stokes equations in time . m0.02@m0.4 @m0.02@m0.4 @xmath40 & snic bifurcation curves in @xmath145 space bounding the pinning region for @xmath134 , @xmath135 and @xmath136 . the region shown is the rectangle in figure [ rcbifs]@xmath31 , and the dashed line is the l@xmath139 curve in the perfect case . @xmath41 the period of the rotating wave c3 as it approaches the snic bifurcations at @xmath146 and other parameters as in @xmath40.,title="fig : " ] & @xmath41 & snic bifurcation curves in @xmath145 space bounding the pinning region for @xmath134 , @xmath135 and @xmath136 . the region shown is the rectangle in figure [ rcbifs]@xmath31 , and the dashed line is the l@xmath139 curve in the perfect case . @xmath41 the period of the rotating wave c3 as it approaches the snic bifurcations at @xmath146 and other parameters as in @xmath40.,title="fig : " ] + in order to break the @xmath0 symmetry and see if a pinning region appears , an imperfection has been introduced , in the form of an imposed linear profile of temperature at the top lid : @xmath147 where @xmath63 is a measure of the symmetry breaking . this term completely breaks the rotational symmetry of the governing equations , and no symmetry remains . figure [ rcpinning]@xmath40 shows that the line l becomes a band of pinned solutions , steady solutions with frequency zero , as predicted by the normal form theory presented in this paper . we can also check the nature of the bifurcation taking place at the boundary of the pinning region . figure [ rcpinning]@xmath41 shows the variation of the period of the limit cycle approaching the pinning region . it is an infinite period bifurcation , and the square root fit ( shown in the figure ) works better than the logarithmic fit . we estimate that the bifurcation is a snic bifurcation , as the normal form theory presented predicts it should be sufficiently far from the zero frequency point zf . m0.02@m0.4 @m0.02@m0.4 @xmath40 & ) for @xmath134 , @xmath135 , @xmath136 , @xmath148 and @xmath146 . @xmath40 is the symmetric solution without imperfection ( @xmath149 ) , and @xmath41 is a pinned solution with an imperfection @xmath150 . there are 20 quadratically spaced contours in the interval @xmath151 $ ] , with blue ( red ) for the cold ( warm ) fluid.,title="fig : " ] & @xmath41 & ) for @xmath134 , @xmath135 , @xmath136 , @xmath148 and @xmath146 . @xmath40 is the symmetric solution without imperfection ( @xmath149 ) , and @xmath41 is a pinned solution with an imperfection @xmath150 . there are 20 quadratically spaced contours in the interval @xmath151 $ ] , with blue ( red ) for the cold ( warm ) fluid.,title="fig : " ] + figure [ temp_contours ] shows contours of the temperature at a horitzontal section at mid height ( @xmath126 ) for the symmetric system ( @xmath149 ) in figure [ temp_contours]@xmath40 , and for the system with an imperfection @xmath150 , corresponding to a maximum variation of temperature of 5%@xmath107 at the top lid , in figure [ temp_contours]@xmath41 . the parameter values for both snapshots are @xmath148 and @xmath146 , inside the pinning region in figure [ rcpinning]@xmath40 . the pinned solution [ temp_contours]@xmath41 has broken the @xmath0 symmetries , is a steady solution , and we can see that one of the three arms of the solution is closer to the wall than the other two . the attachment of the solution to the side wall , due to the imperfection at the top lid , results in the pinning phenomenon . the aim of this paper has been to provide a general dynamical systems description of the pinning phenomenon which is observed in systems possessing two ingredients : slowly traveling or rotating waves and imperfections . the description boils down to the unfolding of a hopf bifurcation in an @xmath0 equivariant system about the point where the hopf frequency is zero . this turns out to be a very complicated problem due to the degeneracies involved , but by considering all of the low - order ways in which @xmath0 symmetry may be broken near a zero frequency hopf bifurcation , we can identify a number of general features which are common to all scenarios , and hence can be expected to be found in practice . these are that the curve of zero frequency splits into a region in parameter space of finite width that scales with the strength of the imperfection , and this region is delimited by snic bifurcations . in the very small neighborhood of the zero frequency hopf bifurcation point , where the snic curves and the hopf curve approach each other , the dynamics is extremely complicated , consisting in a multitude of codimension - two local bifurcations and global bifurcations . the details depend on the particulars of the imperfection , but all of these complications are very localized and are not resolvable in any practical sense . we provide two examples in canonical fluid dynamics to illustrate both the pinning phenomenon and the use of the theory to describe it . these are a taylor - couette flow in which the hopf bifurcation is supercritical and a rotating rayleigh - bnard flow where the hopf bifurcation is subcritical . this work was supported by the national science foundation grants dms-05052705 and cbet-0608850 , the spanish government grants fis2009 - 08821 and bes-2010 - 041542 , and the korean science and engineering foundation wcu grant r32 - 2009 - 000 - 20021 - 0 . l@l@ + name & description + sn@xmath152 & saddle - node ( also called fold ) bifurcations + h@xmath152 & hopf bifurcations + pf@xmath79 & pitchfork bifurcations + cf & cyclic fold : two limit cycles are born simultaneously + l , l@xmath153 & limit cycle becoming a family of fixed points + hom@xmath152 , hom & homoclinic collision of a limit cycle with a saddle + het@xmath152 & heteroclinic collision of a limit cycle with saddles + snic@xmath152 & saddle - node appearing on a limit cycle + glu & gluing bifurcation two limit cycles collide with a saddle + l@l@ + name & description + cusp@xmath79 & cusp bifurcations + tb@xmath79 , tb & takens - bogdanov bifurcations + dpf@xmath79 & degenerate pitchfork zero cubic term + ba & bautin bifurcation degenerate hopf with zero cubic term + pfgl & simultaneous gluing gl and pitchfork pf bifurcations + cfhom & simultaneous cyclic - fold cf and homoclinic collision hom + cfhet@xmath79 & simultaneous cyclic - fold cf and heteroclinic collision hom + snichom@xmath152 & simultaneous snic and homoclinic collision + snichet@xmath152 & simultaneous snic and heteroclinic collision + the fixed points of the normal form are given by @xmath154 , i.e. @xmath155=1,\ ] ] resulting in the cubic equation @xmath156 , where @xmath157 . this equation always has a real solution with @xmath158 , and in some regions in parameter space may have three solutions . the curve separating these behaviors is a curve of saddle - node bifurcations , where a couple of additional fixed points are born . this saddle - node curve is given by @xmath159 ; from these equations we can obtain @xmath73 as a function of @xmath160 . in order to describe the curve it is better to use the rotated reference frame @xmath72 , where the @xmath75-axis coincides with the line l , introduced in ( see also figure [ uv_and_epsilon_sn]@xmath40 ) . the saddle - node curve is given by @xmath161 where @xmath162 is the double root of the cubic equation @xmath163 . the third root @xmath164 is given by @xmath165 . if there is any point where the three roots coincide ( i.e. a cusp bifurcation point , where two saddle - node curves meet ) , it must satisfy @xmath166 . there are two such points cusp@xmath79 , given by @xmath167 and @xmath168 , dividing the saddle - node curve into three branches : sn@xmath169 , joining cusp@xmath43 and cusp@xmath42 , and unbounded branches sn@xmath43 and sn@xmath42 starting at cusp@xmath43 and cusp@xmath42 respectively , and becoming asymptotic to the line l. along sn@xmath169 , @xmath170 , while along sn@xmath43 and sn@xmath42 , @xmath171 . at the cusp points , the three roots coincide and their common value is @xmath172 . a better parametrization of the saddle - node curve is obtained by introducing @xmath173 , so that now @xmath174 , cusp@xmath79 corresponds to @xmath175 and @xmath176^{2/3}},\quad \rho_0=\big(\frac{4}{1 + 3s^2}\big)^{2/3},\quad \rho_2=\big(\frac{1 + 3s^2}{4}\big)^{1/3}.\ ] ] using cartesian coordinates @xmath16 in we obtain @xmath177 where we have set @xmath66 . the jacobian of the right - hand side of is given by @xmath178 the invariants of the jacobian are given by @xmath179 a hopf bifurcation takes place iff @xmath180 and @xmath181 . when @xmath180 , @xmath182 . the fixed points satisfying @xmath180 are given by @xmath163 and @xmath183 , resulting in the curve t in parameter space @xmath184 that can be parametrized as @xmath185 for @xmath186 , @xmath10 and @xmath187 , and the curve is asymptotic to the @xmath10 axis , the hopf curve for @xmath149 . along the t curve , the determinant @xmath188 is given by @xmath189 resulting in two hopf bifurcation curves ( when @xmath181 ) : @xmath190 the end points of these curves have @xmath191 , and are takens bogdanov bifurcation points tb@xmath79 . they are precisely on the saddle - node curve , where both curves are tangent . the coordinates of the four points cusp@xmath79 and tb@xmath79 are : @xmath192 as tb@xmath79 are on the saddle - node curve , its @xmath193 parameter , according to , can be computed . the result is @xmath194 , with @xmath195 . therefore tb@xmath196sn@xmath169 , closer to cusp@xmath43 than to cusp@xmath42 ; tb@xmath197sn@xmath42 if @xmath198 , tb@xmath197sn@xmath169 when @xmath199 and tb@xmath200cusp@xmath42 for @xmath201 . the fixed points of the normal form are given by @xmath154 . one solution is @xmath202 ( named @xmath203 ) . the other fixed points are the solutions of @xmath204 resulting in the bi - quadratic equation @xmath205 , whose solutions are @xmath206 and every @xmath207 admits two solutions , differing by @xmath208 ( they are related by the symmetry @xmath2 , @xmath209 discussed above ) . introducing a new phase @xmath210 , @xmath211 where @xmath212 $ ] because @xmath213 , we immediately obtain : @xmath214 with the other solution being @xmath215 ; @xmath210 is a function of @xmath73 while @xmath38 is a fixed constant . we have obtained two pairs of @xmath2 symmetric points , @xmath216 and @xmath217 , and @xmath218 and @xmath219 . the jacobian of the right - hand side of is given by @xmath220 the invariants of the jacobian are the trace @xmath115 , the determinant @xmath188 and the discriminant @xmath221 . they are given by @xmath222 the eigenvalues of the jacobian matrix in terms of the invariants are @xmath223 . for example , a hopf bifurcation takes place iff @xmath180 and @xmath224 . for the fixed points @xmath225 and @xmath226 , where @xmath227 , we obtain @xmath228 as a result , @xmath229 and @xmath85 and @xmath86 never experience a hopf bifurcation . after some computations , @xmath230 results in the ellipse @xmath231 centered at the origin , contained between the straight lines sn@xmath43 and sn@xmath42 and passing through the points @xmath232 , the ends of the horizontal diameter of the circle @xmath233 . this ellipse is tangent to sn@xmath43 and sn@xmath42 at the points @xmath234 . the condition @xmath224 is only satisfied on the elliptic arc from @xmath235 to @xmath236 with @xmath34 ; along this arc @xmath237 and @xmath238 undergo a hopf bifurcation . we have assumed that @xmath31 and @xmath32 are both positive . the properties of the ellipse are : @xmath239 where @xmath240 . the eccentricity @xmath241 is given by @xmath242 the jacobian evaluated at the three points tb@xmath43 , tb@xmath42 and tb is : @xmath243 the three matrices have double - zero eigenvalues and are of rank one , so the three of them correspond to takens bogdanov bifurcations . the state that bifurcates at the tb point is @xmath237 , without any symmetry , so that it is an ordinary takens bogdanov bifurcation , although the @xmath2 symmetric state @xmath238 also bifurcates at the same point in parameter space ( but removed in phase space ) at another ordinary takens bogdanov bifurcation . the state that bifurcates at the tb@xmath79 points is @xmath203 . this state is @xmath2 symmetric , and so these are takens bogdanov bifurcations with @xmath2 symmetry . the jacobian evaluated at the two points dpf@xmath79 is @xmath244 the corresponding eigenvalues are @xmath245 and @xmath246 for dpf@xmath43 , and @xmath247 and @xmath246 for dpf@xmath42 . both points are pitchfork bifurcations , and in order to determine if they are degenerate , their normal form needs to be computed in order to verify that the cubic term is zero . however , since in a degenerate pitchfork bifurcation a curve of saddle - node bifurcations emerges that is tangent to the pitchfork bifurcation curve , from figure [ bif_fixed_points ] it is immediate apparent that both dpf@xmath79 are degenerate pitchfork bifurcations . the three cases , and can be dealt with by considering the normal form @xmath248 where @xmath249 for the @xmath250 case ( [ sec_z2 ] ) , @xmath251 for the @xmath252 case ( [ sec_zbarz ] ) and @xmath253 for the @xmath254 case ( [ sec_z3 ] ) . the fixed points , other than the trivial solution @xmath203 ( @xmath202 ) , in the three cases are given by the biquadratic equation @xmath255 , with solutions @xmath256 @xmath257 the phases @xmath48 of the @xmath256 fixed points can be recovered from @xmath258 for @xmath259 the solution is unique ; for @xmath144 the solutions come in triples , differing by @xmath22 . it is convenient to use the phase space coordinates adapted to line l , introduced in ( see also figure [ uv_and_epsilon_sn]@xmath40 ) . in terms of these coordinates , @xmath260 , and the fixed points @xmath256 exist only in the interior of the parabola @xmath261 , whose axis is the line l. on the parabola , these points are born in saddle - node bifurcations . in order to explore additional bifurcations of these points , we compute the jacobian matrix of the normal form , @xmath262 whose trace and determinant , for the @xmath256 points , are easily computed : @xmath263 therefore @xmath264 , and for @xmath265 ( @xmath266 ) only @xmath237 ( @xmath85 ) may undergo a hopf bifurcation . _ the @xmath254 case ( [ sec_z3 ] ) . _ here @xmath253 and @xmath267 . @xmath85 is a saddle , but @xmath237 undergoes a hopf bifurcation when @xmath180 . the condition @xmath180 for @xmath237 gives @xmath268 by squaring and simplifying , we obtain the ellipse @xmath269 which is tangent to the line @xmath10 at the origin , with its center at @xmath270 , and whose elements are : @xmath271 where @xmath240 . this ellipse has much in common with the one found in the @xmath83 case , and the eccentricity @xmath241 is given by the same expression . for @xmath272 , the ellipse is located in the interior of the parabola of saddle - nodes , for @xmath273 it becomes tangent to the parabola at a single point , and for @xmath274 it becomes tangent at the two points @xmath275 these are the points tb@xmath276 in figure [ quadratic_barzz]@xmath41 . only the points on the elliptic arc h@xmath169 joining these two points satisfy , and along this arc @xmath237 undergoes a hopf bifurcation . _ the @xmath252 case ( [ sec_zbarz ] ) . _ here @xmath251 and @xmath277 , so there are no hopf bifurcations . moreover , @xmath278 , so it is always stable and @xmath279 , so it is a saddle . the only exception is when @xmath202 , and this only happens at @xmath39 , the degenerate high - codimension point at the origin . _ the @xmath250 case ( [ sec_z2 ] ) . _ here @xmath249 , and @xmath280 is zero on the line @xmath10 inside the parabola . on this line h@xmath169 , @xmath237 undergoes a hopf bifurcation , and the points of contact with the parabola have @xmath281 so they are takens bogdanov bifurcations ( see figure [ quadratic_bifs]b ) . the hopf and takens bogdanov bifurcations are degenerate , as will be discussed in [ degenerate_case ] . in the @xmath250 case , numerical simulations of the normal form show that the hopf bifurcation h@xmath169 and the takens bogdanov points tb@xmath79 are degenerate . this can also be found by direct computation . let us work out the details for the tb@xmath42 point . the coordinates of tb@xmath42 in parameter space are @xmath282 , where @xmath256 are born in a saddle - node bifurcation , and the fixed points are given by @xmath283 in order to obtain the normal form corresponding to the takens bogdanov point , a translation of the origin plus a convenient rescaling of @xmath6 and time is made : @xmath284 substituting in results in @xmath285 in order to obtain the normal form , we introduce the real variables @xmath286 @xmath287 so that the linear part of the ode is transformed into jordan form , and we obtain @xmath288 now we can reduce the quadratic and cubic terms to normal form by an appropriate near - identity quadratic transformation @xmath289 . @xcite gives explicitely the normal form coefficients up to and including third order , in terms of the coefficients of the original ode ( in the form [ deg_tb ] ) ; a nice summary is also given in @xcite . using this explicit transformation , we obtain @xmath290 and the @xmath291 term in the normal form of the takens bogdanov bifurcation is missing , resulting in a degenerate case , the so - called cusp case , of codimension three . the unfolding of this degenerate case has been analyzed in detail in @xcite . note that the ode is hamiltonian at least up to order three , which helps to explain the continuous family of periodic orbits obtained in the interior of the homoclinic loop in figure [ quadratic_bifs]@xmath41 . 2008 survey of strong normal - internal k : l resonances in quasi - periodically driven oscillators for l = 1 , 2 , 3 . in _ spt 2007 : symmetry and perturbation theory , otranto 2007 _ ( ed . g. gaeta , r. vitolo & s. walcher ) , pp . world scientific . 2001 semi - local analysis of the k : 1 and k : 2 resonances in quasi - periodically forced systems . in _ global analysis of dynamical systems _ w. broer , b. krauskopf & g. vegter ) , pp . iop publishing ltd .

using hubble space telescope ( hst ) photometry , we age - date 59 supernova remnants ( snrs ) in the spiral galaxy m31 and use these ages to estimate zero - age main sequence masses ( ) for their progenitors . to accomplish this , we create color - magnitude diagrams ( cmds ) and employ cmd fitting to measure the recent star formation history ( sfh ) of the regions surrounding cataloged snr sites . we identify any young coeval population that likely produced the progenitor star , then assign an age and uncertainty to that population . application of stellar evolution models allows us to infer thefrom this age . because our technique is not contingent on identification or precise location of the progenitor star , it can be applied to the location of any known snr . we identify significant young star formation around 53 of the 59 snrs and assign progenitor masses to these , representing a factor of increase over currently measured progenitor masses . we consider the remaining 6 snrs as either probable type ia candidates or the result of core - collapse progenitors that have escaped their birth sites . in general , the distribution of recovered progenitor masses is bottom heavy , showing a paucity of the most massive stars . if we assume a single power law distribution , , we find a distribution that is steeper than a salpeter imf ( ) . in particular , we find values of outside the range to be inconsistent with our measured distribution at 95% confidence . if instead we assume a distribution that follows a salpeter imf up to some maximum mass , we find that values of are inconsistent with the measured distribution at 95% confidence . in either scenario , the data suggest that some fraction of massive stars may not explode . the result is preliminary and requires more snrs and further analysis . in addition , we use our distribution to estimate a minimum mass for core collapse between 7.0 and 7.8 m .

1210.6353

both theoretically and observationally , core - collapse supernova ( ccsn ) explosions are linked with the death of massive stars . however , because precise measurements of ccsn progenitor masses are scarce , the mapping between explosion scenario and the progenitor mass distribution is less clear . the most common method of progenitor mass determination is direct imaging , in which one identifies the progenitor star in multi - band pre - explosion imaging , fits a spectral energy distribution to the photometry , and verifies the star is gone once the sn has faded . this methodology is ideal given appropriate data , but it is limited to contemporary sne that have pre - explosion hubble space telescope ( hst ) or deep ground - based imaging . as a result , only @xmath125 sne have any constraint on their progenitor masses , and half of these are only upper limits @xcite . a significant increase in the number of progenitor masses could allow us a powerful window into further understanding supernovae . direct imaging has created a prototypical picture of massive star death . type ii - p sne are assumed to be created by red supergiant ( rsg ) stars with intact hydrogen envelopes , while more exotic sne may be created by higher mass stars . however , a number of questions remain to be answered regarding supernova physics . @xcite identify what they term to be the ` red supergiant problem , ' an observed lack of progenitors of @xmath0@xmath116 - 30 m@xmath7that we would expect to explode as type ii - p sn . a variety of channels have been proposed to account for these missing explosions , including direct black hole formation or the stars exploding as different sn types . binarity of the progenitor system can also play a significant role in sn explosions . @xcite show that numbers of type ibc and iib explosions are significantly underestimated if single - star explosions are assumed to be the only sn channel . these issues are difficult to address without further progenitors to analyze . unfortunately , while direct - imaging of the progenitor star would be ideal , there are constraints that limit the frequency with which it may be applied . first , direct imaging is limited by the sn rate : roughly one per century in a large spiral galaxy @xcite . second , direct imaging requires that the host galaxy be close enough to resolve the progenitor star , an effective limit of @xmath120 mpc @xcite . third , preexisting sub - arcsecond images of the site must exist , necessitating archival hst images with either the wide field planetary camera 2 ( wfpc-2 ) or the advanced camera for surveys ( acs ) . finally , even with the existence of appropriate images , there is no guarantee that the precursor will be identified . for example , @xcite summarized 20 progenitor detections in type ii - p explosions . only five progenitors were directly observed , and an additional two which fell on compact star clusters were well constrained . the remainder had no detection , but were given upper bounds based on the maximum luminosity with which they could escape detection ( see also @xcite ) . in short , while direct imaging is ideal given appropriate data , it suffers from prerequisites that limit the number of opportunities in which it may be applied . it is clear that development of an independent and complementary technique would be of interest . in this paper we make use of an alternative technique , stellar population analysis , to infer progenitor masses . the technique involves examining a color - magnitude diagram ( cmd ) of the surrounding population of stars to infer an age and mass for the progenitor star . the advantage of our methodology is that we are not reliant on individual identification and photometry of the specific progenitor star , leaving us free to apply our method to cataloged supernova remnants ( snrs ) in addition to directly observed sne . this ability drastically increases the number of progenitor masses we may find , allowing us to make a more complete measurement of the underlying distribution of progenitor masses . our method offers no direct way to probe the binarity of the system , which is relevant given the role binarity likely plays in certain ccsn scenarios . however , our estimate of the age of the surrounding stellar population is not affected by the particular details of a given progenitor system . our method also offers no way to determine the type of ccsn explosion we are observing , so for snr analysis we are not able to provide a direct link between sn explosion type and progenitor mass . however , we will be able to comment on the ranges of zams masses that are producing ccsn explosions . stellar population analysis has been used many times historically to analyze the characteristics of progenitors . many groups have carried out age dating of stellar clusters coincident with supernovae and derived corresponding stellar masses @xcite . @xcite examined the sfh map of the lmc published by @xcite . they estimate ages and masses for the four most recent snr produced by ccsne in the lmc . in @xcite , we employed a technique identical to the one used in this paper to estimate the age of the progenitor of ngc 300 ot2008 - 1 . we found that the star was likely to have formed 8 - 13 myr ago , corresponding to a @xmath0 of 12 - 17 m@xmath7 . finally , in @xcite we applied the same technique as this paper to archival images of sn 2011dh , finding a most likely age of 17@xmath8 myr and a @xmath0 of 13@xmath9 m@xmath7 . clearly there is a long list of successful applications of stellar population analysis leading to constraints on progenitor physical properties . in this paper we examine 59 snrs in the spiral galaxy m31 , listed in table [ tab_snr ] . in 2 we outline the methods used to determine star formation histories ( sfhs ) , ages , and masses for the progenitor stars . in 3 we present the results of the analysis . we find that 53 snr display recent star formation , and we derive masses for these . we also present four example cases that are representative of the overall sample . finally , in 4 we examine the distribution of progenitor masses . we search for indications of a minimum mass , which we find to be between 7.0 and 7.8 m@xmath7 , and note that the distribution is more bottom - heavy than a salpeter imf distribution . the basic procedure for deriving a mass estimate for each progenitor is as follows . we first use coordinates from three snr catalogs as explained in 2.2 , which we assume are accurate to within a few arcseconds and unbiased to any one type of sn . we then search for any appropriate hst fields that contain the snr in question . we perform photometry on the stars in each field , and create cmds of the region within @xmath150 pc of the snr coordinates , assuming a distance modulus of @xmath10 @xcite . we use cmd fitting to measure the sfh of each region . for each region that displays recent sf , we use the age of this recent sf to associate an age with the progenitor star . finally , we apply stellar evolution models to convert the age of the progenitor to the @xmath0 of the progenitor . our method requires several assumptions to produce results . above all , we must assume that some of the stars surrounding the snr are in fact coeval with the progenitor star . there are several pieces of evidence suggesting this assumption is reasonable . first , 90% of stars form in clusters with sizes of @xmath11 pc @xcite . these stars are expected to stay spatially associated for timescales greater than the lifetimes of ccsne progenitors , even for dissolving , unbound clusters . second , theoretical predictions are that stars will stay spatially associated on scales of @xmath1100 pc for 100 myr @xcite . observational constraints support this @xcite , as do simulations by @xcite who found that @xmath185% of ccsne will explode within 100 pc of their birth site . in addition , even if a progenitor would travel farther , we would still expect to see a fraction of the coeval young stellar population which is sufficient for age dating . the precise number of interest is the fraction of young stars that will stay within a certain distance in the lifetime of a core collapse supernova progenitor . the question is a complex one , and we know of no research that quantifies it precisely at this time . we conclude that it is a reasonable assumption that most of the young stellar population around our observed snrs is coeval with their progenitors . based on the analysis in @xcite , we adopt a value of @xmath150 pc for the radius of our star selection annulus . our method does not assume any information about the type of sn that exploded . we assume that snrs associated with recent star formation are ccsne , but do not distinguish individual subtypes . for ccsne , the particular type of supernova has no effect on our mass determination process . however , the progenitors of thermonuclear type ia sne are likely to arise from older stellar populations which may vary considerably more in age than the coeval populations of massive stars . while there exists considerably discussion in the literature as to the precise nature of type ia progenitor systems and their distribution in age ( see @xcite and references therein ) , we would expect to see type ia progenitors ranging from ages of a few hundred myr to several gyr . since these timescales are comparable to or significantly greater than the dynamical timescale of m31 , our methodology of identifying and analyzing coeval stars is ineffective at measuring the precise ages of type ia progenitors . our method does have the risk of misinterpreting a type ia snr as a ccsnr . in their volume limited sample ( d @xmath11 60 mpc ) , @xcite find that type ia sn compose 24% of observed sn . however , there are several qualitative reasons as to why one would expect a smaller ia / ccsn fraction in our survey . first , there is some discussion as to whether such a volume - limited survey may underestimate the fraction of ccsn due to their faintness compared to type ia sn ( see discussions in @xcite ) . this potential bias suggests that 24% could be considered an upper limit . in addition , the areas of m31 with greater hst coverage are primarily star forming regions , where we would naively expect ccsn to be more common , increasing the relative fraction of ccsn . thus the actual fraction of type ia snr we expect in our sample is probably somewhat less than 24% . moreover , it is likely that some type ia snr can be identified by their lack of an associated young stellar population . @xcite examined four ia sites in the lmc and found that three displayed a significant lack of recent star formation . out of the 59 examined , we found 6 snr displayed no recent sf in their surrounding stellar populations . finally , we have examined what effect additional hidden ia contamination will have on our data . the procedure we used is explained in 4.3 . we found that additional ia contamination up to a total fraction of @xmath125% has very little effect on the overall distribution observed in the sample . to summarize , we argue that there are several qualitative reasons to assume that the effective ia fraction in our sample is smaller than 24% . in addition , our method offers a means to identify and remove some snr that are likely ia progenitors . finally , our overall results are largely insensitive to inclusion of additional ia contamination up to the 24% observed in the volume - limited survey from @xcite . another possibility for snrs with no coincident young sf is that they are the result of high - velocity progenitor stars that have left their birth sites . as a test of this possibility , we examined any site with no observed recent sf in galex fuv data . we found that all 6 sites with no recent sf are still relatively close ( @xmath150 to 100 pc ) to galex fuv sources . if we assume that the progenitor star resulted from the most nearby fuv source for each zero - sf snr , then a 50 myr old star would only require velocities of a few km / s to reach the snr site from the fuv source site , well within reasonable velocities for runaway stars . higher mass stars would of course require higher velocities to leave the fuv source in their shorter lifetimes . this analysis suggests that our study is unable to differentiate between type ia snr and snr resulting from high - velocity stars . we note , however , that a null - result for progenitor mass will not affect the distribution of progenitor masses we measure . it will simply reduce our effective sample size . finally , our method is highly contingent on the accuracy of the stellar evolution models used to model our observed cmds . because the models are theoretical , they generally do not have easily quantifiable uncertainties . the wide array of available model sets tend to systematically differ in bolometric luminosity and temperature , but are consistent within about 0.2 mag in bolometric luminosity and 0.02 dex in log temperature . we therefore quantify the uncertainty of our results due to the models used by including random shifts in the bolometric luminosity and temperature of the models as part of our monte carlo ( mc ) tests ( see also @xcite ) . we detail the process further in 2.5 . there are only a few extensive catalogs of snrs in m31 . we use the catalogs of @xcite , @xcite , and @xcite . all three make identifications based on [ sii]-to - h@xmath4 ratios , and @xcite make additional use of morphology and ob star associations . @xcite identify three confidence levels based on approximate levels of contamination , although they note that the confidence levels are somewhat subjective . we only perform analysis on their first and second confidence candidates . the catalogs are not mutually exclusive , and the coverage areas of the three catalogs overlap in several areas , causing several remnants to be identified in both catalogs . to avoid treating a double - identified remnant as two separate remnants , we identify all remnants from two separate catalogs in which the 50 pc areas overlap and only select coordinates from one catalog . in these duplicate cases , we adopted the coordinates first from @xcite , then from @xcite if @xcite did not identify a remnant in that position . on occasion , a single catalog identified two remnants in which these 50 pc regions overlapped , but we still performed analysis on both candidates in such a case . in addition , we cross - referenced the snr catalogs with @xcite and eliminated two , k567 and k884 , which were found to not be snr based on their optical / x - ray properties . for our analysis , we require that the stellar population surrounding the snr be imaged by either the advanced camera for surveys ( acs ) or the wide - field planetary camera 2 ( wfpc2 ) instruments on hst in at least two broadband filters . we also require our 50% completeness limits ( see 2.3 ) for a given cmd to be at least 24.5 magnitude in f475w , f555w , or f606w , whichever represents the blue filter ( all qualifying fields of sufficient depth were either f475w , f555w , or f606w vs. f814w ) . the location of the ms turnoff is very similar in color in these filter sets , so adopting a similar magnitude cut in all three is reasonable . our comparisons between snrs imaged multiple times at varying depths indicate that simply reaching the ms turnoff for a given cmd is insufficient to correctly determine the age . cmd fitting is highly sensitive to densities of stars at given ages , requiring a well - sampled stellar population at the age of interest . in general , we found that our methodology applied to shallower data had a tendency to miss older bursts of sf which still may have resulted in ccsn . the available 2-filter broadband data for m31 happens to be distributed in two groups , including many shallow fields ( blue depth @xmath12 ) and many deep fields ( blue depth @xmath13 ) , with very few in between . the choice of our depth cut causes us to reject five snr that have only been imaged with shallow wfpc-2 data . in table [ tab_snr ] we list the snr for which our criteria is met , along with the corresponding hst fields . in cases where the regions in question were imaged multiple times , we selected whichever set of images had the greatest number of stellar detections . in fig . [ spatial ] we chart the locations of the examined remnants on a star - subtracted image of m31 in h@xmath4 . those remnants colored red represent probable ia candidates , where no young sf was found . there is a clear grouping of ccsn candidates along the star forming arms of m31 . we performed resolved stellar photometry using the photometry pipeline developed for the acs nearby galaxy treasury program @xcite . this pipeline uses the dolphot stellar photometry package @xcite to fit the well - characterized acs point spread function to all of the point sources in the images . we then converted fluxes to vega magnitudes using the standard zero - points and aperture corrections from the acs handbook . we assess photometric errors and completeness using fake star tests . at least @xmath14 tests are performed by inserting fake stars of known color and magnitude into the data one at a time and blindly attempting to recover them with the same software . both fake star tests and photometry were performed on the full hst fields . cmd fitting is a powerful tool for measuring star formation histories @xcite . to estimate star formation histories of the regions surrounding the snr , we used the software package match @xcite . match works by creating many model cmds based on theoretical isochrones for a variety of ages and metallicities . a linear combination of these model cmds are then fit to the observed cmd . we use the models of @xcite and @xcite both for cmd fitting and estimation of @xmath0 . for each field , we selected stars from our photometry catalog within a 15 radius around the snr coordinates , which equates to a physical size of @xmath150 pc at an assumed distance modulus of @xmath10 @xcite . this is consistent with the spatial correlation discussed in 2.1 . we account for photometric errors through the use of fake star tests . we selected fake stars in a region @xmath12.5 times the radius of the real star annulus to ensure that at least a few thousand recovered fake stars are included for each snr . we defined lower magnitude limits as the point at which fake star completeness dipped below 50% . match requires a variety of parameters to generate and fit cmds . we assumed a salpeter imf , @xmath15 @xcite , and a binary fraction of 0.35 . @xcite demonstrated that varying the imf value from -2.0 to -2.7 or varying the binary fraction from 0.2 to 0.5 had no effect on the epoch assigned to a recent burst . we also varied the imf from -1.3 to -3.3 for select regions and found no significant difference in relative fractions of sf for various epochs . although absolute amplitudes of sf did change , only relative amplitudes have an effect on our method of determining ages . note that the imf and binary fraction are used by match purely for purposes of populating the models . we assume nothing about either value with regards to a potential ccsn progenitor system or the overall progenitor mass distribution . match produces fits in logarithmic age bins . we adopted 71 age bins increasing in .05 increments from 6.60 ( 4 myr ) up to 10.10 ( 12.5 gyr ) . we are unable to fit for ages younger than this due to the lack of isochrones at younger ages . match will interpret any sf from a population younger than the 6.60 to 6.65 ( 4 to 4.5 myr ) bin as being included in this bin . thus our youngest bin actually includes all sf from present times back to 4.5 myr . as a result , we may only quote an upper age limit for any sf found in this youngest bin . in fig . [ isochrones ] we plot the isochrones for every - other age bin from 6.60 through 8.00 in f555w vs. f555w - f814w . fitting metallicity with match is not viable given the very weak dependence of the optical colors of the upper main sequence on metallicity . we therefore constrained the metallicity to a spread of @xmath10.15 dex and to increase with the lifetime of the galaxy , which consistently produced best - fit metallicities of solar . this value is consistent with the known gas - phase metallicity of m31 @xcite . finally , we binned our cmds in units of 0.3 in magnitude and 0.15 in color . this binning accounted for the fact that some fields had few upper ms stars , allowing us to reduce the impact of this paucity on our fits . however , we still maintain a number of bins significantly larger than the number of free parameters used in the fitting . we experimented with using finer binning and found no change in our age determinations within our uncertainties . recent sfhs are very sensitive to the treatment of reddening . our most simplistic treatment of reddening is to assume that all reddening is due to a galactic foreground , search over a specified range of reddening values for a best fit , and apply this value to the cmd as a whole . however , we frequently found significant differential reddening ( @xmath16 ) across our small snr - centered regions , which manifests as an increased width to the ms and red clump . star forming regions will typically have extensive amounts of dust and gas , often not uniformly distributed , and as such there is no reason to believe that a single reddening value would correctly describe the region as a whole . match by default allows for 0.5 mag of differential reddening for populations below 40 myr , after which it falls linearly down to 0.0 mag at 100 myr . in addition , the user has the ability to add additional full - field differential reddening across the cmd . we almost always found that the default treatment of reddening was inadequate to account for all the reddening present in the regions . as an example of how ignoring differential reddening can lead to an erroneous result , we consider the snr bw-102 . the photometry is quite deep , extending to magnitudes of @xmath17 and @xmath18 . in the left panel of fig . [ bw102_nodav ] we plot the observed cmd of the region , as well as the best - fit model generated by match in the background in greyscale . in the right panel , we plot the cumulative sfh as a fraction of total sfh in the past 50 myr , assuming a fixed differential reddening of @xmath19 mags affecting only the young stars . we found a best - fit reddening value of @xmath20 . all the star formation is concentrated in the youngest time bin ( @xmath11 4.5 myr ) , corresponding to a lower limit @xmath0 of 52 m@xmath7 . however , several qualities of the cmd would cause us to be suspicious of this result . the main sequence appears very dim , and appears to contain lack the young , bright , blue stars that the model predicts at @xmath21 . we would not expect this given a result of very extensive young star formation . looking beyond the main sequence , we see that the red clump is extended over nearly 2 magnitudes in f475w . indeed , the model very poorly models the stars at @xmath22 , @xmath23thus not only was the default value of differential reddening insufficient to model the entire cmd , the treatment of differential reddening as something unique to the young stellar population is clearly not reflected in the reality of the cmd . for regions such as this , it is necessary to apply differential reddening across the cmd . we must then define the amount of differential reddening to add . to understand our technique for determining the amount of @xmath24 to use , we must expand on the differential reddening model used by match . match applies differential reddening to the model cmds by applying a top - hat distribution along the reddening line . the low end of the top - hat is defined by a foreground value of @xmath25 for which match fits . the width of the distribution is then defined by the specified value provided by the user . when applying additional differential reddening , fit values tended to improve until we reached unphysical values of @xmath16 the only true constraint we may apply is that this minimum value of @xmath25 must be at least equivalent to the milky way foreground value . @xcite find a foreground reddening value to m31 of @xmath26 . while this value is likely to change slightly over the angular size of m31 , the changes will clearly be small compared to the magnitude of differential reddening internal to m31 . our solution to cmds such as bw-102 is to increase the width of the differential reddening distribution applied to all ages until the best - fit value of @xmath25 match finds reaches the foreground value . we then use this differential reddening distribution to measure the sfh and derive the age of the sn progenitor . in fig . [ bw102dav ] we plot the sfh results for a variety of differential reddening values , increasing from @xmath27 up to @xmath28 magnitudes , in increments of 0.1 . note that this differential reddening is in addition to the default young @xmath19 , leading to up to 2.0 magnitudes of differential reddening on the very youngest stellar populations . we found that a value of @xmath28 was the point at which the overall reddening value dropped to the expected foreground value for bw-102 . for this value of differential reddening , the sfh corresponded to an age of 36 myr and a mass of @xmath19 m@xmath7 , a much more reasonable result given the observed cmd . note that the sfh reaches this result ( within uncertainty ) before the best - fit @xmath29 value reaches foreground , suggesting that our result is not highly contingent on the precise value of foreground reddening assumed . in the left panel of fig [ bw102 ] we plot the observed cmd with the best - fit model plotted in greyscale in the background , created using the above @xmath24 procedure . in the right panel , we plot the cumulative sfh of the region for the past 50 myr , with error bars from the monte carlo analysis described in 2.5 we found essentially all fields suffered from differential reddening in excess of the default values used by match . when no extra differential reddening was included in the models , many regions without obvious young star formation were best fit by high overall reddening values and included star formation in the youngest few time bin . the issue arises from faint main sequence stars revealed by deep photometry . without inclusion of differential reddening , these stars are fit by the uniform foreground reddening value for the region , which is typically a high value ( e.g. @xmath30 ) . using this erroneously high value , the stars are corrected to be brighter and bluer than they actually are , leading match to mistake them as evidence of young star formation . by allowing for differential reddening , we can incorporate the full range of different reddening values across the field . for our final procedure , we increase the amount of full - cmd differential reddening until we find the best fit @xmath25 reaches the foreground value , @xmath31 . we then adopt this value of differential reddening for subsequent analysis of that particular snr . our ultimate goal is to estimate the mass of the progenitor . we do this by identifying the burst of recent star formation from which the progenitor star most likely originated , allowing us to assign an age to each progenitor star . ideally , each sfh would have a single isolated burst , making it easy to associate a single age to the young stellar population . in many cases , however , multiple bursts of varying rates and durations appear in the young star formation history . as a result , we examine the cumulative star formation history of each cmd as a fraction of the total recent star formation . when examined as such , the fraction of star formation in each age bin corresponds to the probability that the progenitor will be of that age . because we are only interested in recent sf , we examine only the most recent 50 myr of sf ( see 4.1 for an explanation of this limit ) . our results suffer from both random uncertainties due to the sampling of the cmd and systematic errors due to differences between the theoretical isochrones and observations . for example , if the models were consistently redder than the observed stars , we would generally find younger ages when these models are applied to real data . the random errors are generally highly dependent on the number of upper main sequence stars in the field , with higher numbers providing a tighter constraint on the data ( although more upper main sequence stars may also simply indicate a younger population ) . see @xcite for further discussion of this point . to analyze our uncertainties , we performed a series of monte carlo ( mc ) realizations of the data ( see also @xcite ) . in each mc run , we re - sampled the observed cmd to account for the poisson errors of the stars . in addition , to estimate how systematic model differences can impact the results , we add in random shifts to the models for each run . we used random shifts of @xmath32 in temperature and @xmath33 in bolometric luminosity . these systematic offsets are much larger than uncertainty in the distance modulus , thereby incorporating both effects into our overall uncertainty analysis . uncertainty in the age of the recent star formation manifests itself in two distinct ways . the first is uncertainty measured by the mc analysis performed above , while the second is due to the width of the intrinsic spread of the burst across multiple age bins . we estimate the latter of these as the difference between the median age and the ages where 16% and 84% of star formation has occurred . we estimate the former of these as the rms difference between the median age of the best fit and the median ages of sf from the mc tests . we then add these differences in quadrature to assign a confidence interval to each result . thus for each progenitor we determine a median age , as well as ( potentially asymmetric ) uncertainties about this median . finally , we make the assumption that we are unable to determine the age of sf to a greater precision than the age bins in which we have measured this sf . we thus round the age range to the age of the next isochrone out from the median . we perform this step for both younger and older star formation , always rounding away from the median . the ages of these isochrones determine our final uncertainties . in order to use this method of analysis , we must define the maximum age of star formation which may produce core - collapse progenitors . as explained in 4.1 , we use our distribution to estimate a minimum mass for core - collapse between 7.8 and 7.0 m@xmath7 . as a result , we adopt 50 myr as the maximum age of a core - collapse progenitor and only perform the above mass estimation analysis over the most recent 50 myr of sf . the @xcite and @xcite models specify a maximum mass for an isochrone at a given age . more massive stars will have already died off . thus the isochrones we have identified as bounding our confidence interval can be linked directly to values for @xmath0 . our value for the median progenitor mass comes from interpolating the final isochrone masses between isochrones to the median age value . this is necessary because our median age wo nt line up exactly with a defined isochrone . in fig . [ massplot ] we plot the @xcite and @xcite isochrones for final isochrone mass vs. age for metallicities of z = 0.004 , 0.008 , 0.019 , and 0.030 . the models produce very similar age to mass conversions regardless of the assumed metallicity . the vast majority of our regions produce best fit metallicities of approximately solar . as a result , we adopt the z = 0.019 isochrone for mass determinations . note that masses change very quickly for younger populations , while masses for older populations change at a much slower rate . this means that our results for less massive progenitors will naturally be more precise . note that we have neglected systematic uncertainties in the age - to - mass conversion process for our individual progenitor results . this leads to very small error bars on some progenitor masses , especially those at older ages , where mass does nt change significantly as a function of age . our mass distributions are not sensitive to this model - dependent systematic uncertainty , as shown by similarity between the median progenitor distribution and the probability distribution ( see 4.2 ) . however , uncertainties are almost certainly underestimated for some progenitors , especially at the low - mass end . to estimate the magnitude of this systematic uncertainty , we compared the solar metallicity isochrones of @xcite to the @xcite isochrones used for our analysis . we found that for a @xmath150 myr lifetime star , the @xcite isochrones find a maximum mass of 6.7 m@xmath7 , compared to the 7.3 m@xmath7predicted for the @xcite isochrones . at a lifetime of @xmath122 myr , the youngest stars tested by @xcite , the maximum masses for the different isochrone sets are 9.9 m@xmath7and 10.9 m@xmath7respectively . these suggest systematic uncertainties of around 0.5 - 1.0 m@xmath7for the age - mass conversion process . again , we do not include these systematic uncertainties in the reported values in table 2 ; the reader should keep this in mind when interpreting the results of any individual progenitor star in our study . in table [ tab_results ] we list results for all 59 snr analyzed . we tabulate the designation , mass , age , number of main sequence stars ( defined as @xmath34 , @xmath35 , and @xmath36 ) , total stellar mass , and additional full - field @xmath16 applied . in general the number of ms stars may be taken as an indication of confidence in the answer , but older populations will also generally have fewer ms stars . as a result , this mapping is very approximate . the total stellar mass listed is highly dependent on the precise value selected for the imf , and as such should not be taken as an exact measurement . it is intended only to compare relative amplitudes of star formation between various snr regions . we observe six fields that have no significant recent sf , and identify these as likely either type ia snr or snr resulting from runaway progenitor stars . they are not included in analysis of the distribution of recovered progenitor masses . to examine any underlying biases in our data selection , we plot the recovered ages against the 50% completeness magnitudes in fig . [ depthcompare ] . the top panel displays the bluer filter for each particular cmd , while the bottom is for the redder ( f814w in all cases ) . following application of our depth requirement , we find no evidence of any correlation between recovered age and depth of data . the top panel of fig . [ mscompare ] plots the recovered ages as a function of the number of ms stars detected . the bottom panel plots recovered ages as a function of @xmath16 used . note that these values only include full - field @xmath16 , not the 0.5 mag of @xmath16 included for all young populations . no correlation is observed in either comparison , which suggests that our treatment of differential reddening from 2.4 does not bias us towards a specific age . while an in - depth discussion of all snr is prohibitively long , it is worth exploring a representative sample of the various cases observed . in general , fields with fewer main sequence stars will tend to have larger uncertainties , and we list our results in table [ tab_results ] in order of the number of these stars . however , this generalization is not absolute , and this ordering should be taken as approximate . finally , some of the regions display no young star formation , which we list in a separate category . for illustrative purposes , we subjectively identify four classes of results . below we examine four representative snr for each case . in the left panel of fig . [ k376 ] we display the color - magnitude diagram for the region around k376 , plotted in red . we also plot our best - fit model cmd in the background in greyscale , with darker regions corresponding to a larger expected stellar density . the data we used for k376 are from hst project 12055 ( brick 9 , field 14 ) , and included acs images in f475w and f814w . the photometry is fairly deep , with 50% completeness limits at f475w=27.1 and f814w=25.8 . we find 6874 stars in the 50 pc region around the snr , 350 of which are ms stars , making k376 a very well populated cmd . a qualitative inspection of the cmd indicates a bright , blue upper main sequence , indicative of a significant young stellar population . the cmd displays a great deal of differential reddening , as indicated by a large spread in the red clump . we used a full - field value of @xmath37 , in addition to @xmath19 for young stars , to model this cmd . match predictably finds that this cmd displays significant young star formation : we find the field well fit by a single burst of star formation at 12@xmath381 myr , corresponding to a mass estimate of 16@xmath381 m@xmath7 . the older burst at @xmath140 myr is of significantly lower prominence compared to the young burst . the result is well - constrained due to the large number of stars in the field . similar snr include 2 - 020 , 2 - 024 , k934 , k947 , bw-60 , bw-69 , bw-74 . [ k180 ] displays the cmd and sfh for the region surrounding k180 . the data for k180 are from hst project 12073 ( brick 2 , field 11 ) , and includes acs images in f475w and f814w . the photometry is quite deep , with 50% completeness limits of f475w=27.2 and f814w=25.9 respectively . we find 3950 stars in the cmd , 162 of which are ms stars . we used full - field value of @xmath39 , in addition to @xmath19 for young stars , to model this cmd . the upper main sequence is not nearly as bright or prominent as that of k376 , and as such we would predict an older population with a lower mass progenitor star . match agrees , finding the population is fit best by a single burst of star formation at 33@xmath382 myr , corresponding to a mass of 8.8@xmath380.2 m@xmath7 . as above , because of the large number of stars in the field , the mass is fairly well constrained . similar snr include 1 - 006 , 1 - 008 , 1 - 009 , 1 - 010 , 2 - 025 , 2 - 044 , 2 - 046 , 2 - 048 , 2 - 050 , k516 , k526a , k574 , k594 , bw-18 , bw-20 , bw-31 , bw-32 , bw-76 , bw-81 , bw-84 , bw-86 , bw-102 , and bw-110 . [ k891 ] displays the cmd and sfh for the region surrounding k891 . the data for k891 are from hst project 12055 , including acs images in f475w and f814w . the photometry is again quite deep , with 50% completeness limits of f475w=27.6 and f814w=26.6 respectively . the cmd has 1441 total stars , including 149 ms stars . we used a full - field value of @xmath40 , in addition to @xmath19 for young stars , to model this cmd . the cmd for k891 does nt appear to comprise a single uniform population , and indeed has stars spread out at a wide variety of ages . match finds two distinct star formation bursts : one in our youngest age bin at less than 4.4 myr , and another in an older bin at @xmath132 myr . the error bars also indicate that the relative prominence of these bursts is such that we may not favor one burst over the other . while we find a median mass of @xmath19 m@xmath7 , the uncertainties allow this result to range over all almost masses past the minimum mass for core - collapse . k891 is an example of a snr where we have reasonable confidence in our answer , but this answer allows no constraint on the parameters we ultimately wish to measure . similar snr include 2 - 049 , k446 , k497 , k525a , k527a , k856a , k908 , k956a , bw-11 , bw-39 , bw-44 , bw-61 , bw-65 , bw-66 , bw-71 , bw-77 , bw-82 , bw-89 , bw-105 , and bw-106 . [ 2 - 028 ] displays the cmd and sfh for the region surrounding 2 - 028 . the data for 2 - 028 is also from hst project 10273 , and includes acs images in f555w and f814w . the photometry has more shallow 50% limits than the cmds above , with f555w=25.7 and f814w=25.4 respectively . the cmd has only 374 stars , a smaller number of detections than typical , including 53 ms stars . we used a full - field value of @xmath41 , in addition to @xmath19 for young stars , to model this cmd . the cmd clearly has a very sparse and dim main sequence , leading us to predict that match would find little to no young star formation . indeed , match finds the youngest star formation as occurring at older than 90 myr , well beyond the age of stellar populations that we would expect to produce progenitors of core - collapse sne . we classify this snr as the result of either a type ia sn or a runaway star that has left the young stellar population at its birth site . we do not use any results in this category as for purposes of analyzing the mass distribution . similar snr include 2 - 016 , 2 - 021 , 2 - 026 , bw-19 , and bw-36 . our method measures the prominence of a burst as a fraction of the total recent sf . this approach requires that we define the period of time that we consider as `` recent '' sf . specifically , we must identify the age range where star formation can produce stars massive enough to result in ccsne . theoretical arguments and observational evidence point to a minimum mass necessary for progenitors to undergo core - collapse to a neutron star . stars below this mass are generally assumed to leave behind white dwarf stars , producing no sn explosion . thus one may constrain this cross - over mass by either measuring the maximum mass from which a star may create a white dwarf , or the minimum mass necessary for a star to explode . measurements of white dwarfs have defined a lower limit on this minimum mass of 6.3 - 7.1 m@xmath7@xcite , corresponding to an age of between @xmath155 and 63 myr , and direct progenitor mass measurements has have converged on a value of 8@xmath381 m@xmath7@xcite , corresponding to an age of between @xmath133 and 55 myr . in our observed progenitor mass distribution , we would expect to see the following behavior : above the minimum mass for core - collapse , we expect the distribution of progenitor masses to follow the imf , assuming that the recent sfr is approximately constant . below the minimum mass , the inferred progenitor mass should have no physical connection to the ccsne process and should reflect random sampling of the sfr at @xmath4250 myr , producing an essentially flat distribution in inferred progenitor mass . in the left panel of fig . [ minmass ] , we plot the distribution of progenitor masses for a variety of assumed minimum masses . we vary the minimum mass from 9.6 m@xmath7to 6.0 m@xmath7(our chosen masses are mapped using the @xmath0from @xcite and @xcite isochrones ) . we find that for assumed minimum mass greater than 8.1 m@xmath7 , the distribution increases until the assumed minimum mass is reached . for assumed minimum mass values below 8.1 m@xmath7 , there is a peak between 7.5 and 8.5 m@xmath7 , below which the number of progenitors drops , suggesting a minimum mass in this range . to find the actual minimum mass , we lower the assumed minimum mass until the measured minimum mass no longer reflects this assumed value . we note the amplitude of the peak in the distribution is greatest for an assumed minimum of 7.3 m@xmath7 . to quantify the location of this peak , we calculate the derivative of the number of progenitors as a function of mass . we assume that the maximum value of this derivative occurs at the minimum mass a star undergoes core - collapse , as this value identifies the beginning of the peak . in the right panel of fig . [ minmass ] we plot the location of this maximum against our assumed minimum mass . we note that above an assumed minimum mass of 7.3 m@xmath7 , the peak value traces the assumed minimum mass . for 7.3 m@xmath7 and below , however , the peak value is always around @xmath17.5 m@xmath7 . if we assume our uncertainties are at least the width of the mass bins , the we find a minimum mass for core - collapse between 7.0 and 7.8 m@xmath7 . this range is consistent with the observational measurements of @xcite and @xcite . we therefore have adopted the 44.7 to 50 myr ( 7.7 to 7.3 m@xmath7 ) bin as the oldest included for our final results . in the left panel of fig . [ diffhist ] we plot the histogram of median progenitor masses , restricted to 7 m@xmath7 and above . in the right panel of fig . [ diffhist ] we plot the cumulative fraction of progenitor masses . unless otherwise noted , we assume a maximum mass of 120 m@xmath7 , although the choice of this value at high masses is essentially irrelevant to the overall distributions given the rarity of extremely massive stars . qualitatively , the observed distribution shows a lack of the most massive stars when compared to a salpeter imf ( @xmath2 , where @xmath3 ) . we performed a kolomogorov - smirnov ( ks ) test , assuming a single power law distribution . we found values of @xmath4 outside the range @xmath5 inconsistent with the measured distribution at 95% confidence . alternatively , we may consider a model distribution that is a salpeter imf ( @xmath3 ) up to some maximum mass , which we may vary . we found that this model was inconsistent with the data at 95% confidence at assumed maximum masses @xmath43 m@xmath7 . however , precise determination of this value is difficult in our survey due to the intrinsic rarity of massive stars and the mass spacing in our isochrines ( see fig . [ massplot ] ) . rather , this value represents the sort of mass range in which one must consider ccsn possible in order to maintain a salpeter imf . in either scenario , the full distribution of measured masses suggests that some fraction of massive stars are not exploding as ccsn . this result has interesting implications for ccsn physics . a wide variety of sn channels have been explored both theoretically and observationally in the literature . theoretical predictions have explored the possibility of direct black hole formation beyond a certain mass threshold somewhere around @xmath125 m@xmath7 . the manifestation of such events in an overall mass distribution would be an observed lack of progenitors beyond the mass threshold , essentially a more bottom - heavy imf than that of all massive stars . while the reality is likely something more complicated than a well - defined threshold between ccsn and black hole formation ( many different scenarios likely combine to produce a complicated mass distribution ) , the qualitative effect will be that which we observe in our distribution . @xcite first identified the red supergiant problem , an observed lack of type iip progenitors between 16 and 30 m@xmath7 . many solutions have been proposed to explain the problem ( see @xcite and references therein ) . in addition , the recent sn 2012aw may fall in this mass range @xcite , suggesting the possibility that type iip progenitors do exist in this range and have simply not yet been observed in sufficient number . we identify six progenitors with median progenitor masses between 16 and 30 m@xmath7 , although the uncertainties on many of these are large . if our progenitors were sampled uniformly from a salpeter imf , we would expect to find @xmath110 progenitors ( 20% ) in the mass range from 16 to 30 m@xmath7 . thus while we do nt observe a complete lack of progenitors in the specified mass range , we do observe fewer than we would expect given a salpeter imf distribution . finally , while we assumed our snr catalogs constituted a complete , unbiased sample , the possibility exists that selection effects in the catalogs lead to the lower end of the progenitor mass distribution being sampled more heavily . in particular , extremely massive progenitors are likely associated with strong h ii regions , where identification of snrs is a more difficult observational task . it is possible snrs of this type are systematically undersampled in the survey . we attempted to quantify both the imf slope and minimum mass using markov chain monte carlo methods , but found that the data could not produce meaningful constraints beyond those determined by our more simplistic techniques . we believe the chief reason for this is the size of errors due to differential reddening of the fields . part of the problem is treatment of differential reddening as a top - hat distribution . in addition , the inclination of m31 contributes to these high differential extinction values . the application of this technique to a less inclined galaxy would be of benefit in this respect . finally , the simple addition of more progenitor mass estimates would allow us to better constrain our analysis . we are currently performing identical analysis on an additional @xmath165 snr in m33 in pursuit of these final two points . type ia snrs coincident with star - forming regions could in principle affect our mass distribution . to test the possibility of additional type ia contamination beyond the 11% observed , we examined galex fuv fluxes at the sites of all snr in our sample . we assumed that sites with the lowest fuv flux corresponded to possible older type ia sites which happened to be coincident with a small amount of recent sf . the eight lowest flux snr ( not including the 6 with no recent sf ) were bw-18 , bw-69 , 1 - 006 , 1 - 010 , 2 - 024 , 2 - 050 , k891 , and k956a . from table [ tab_results ] , these additional snrs in general have fewer ms stars and less total sf then most snrs in the sample . we found that after removing these snr , our observed distribution now ranged from @xmath44 , which is consistent with our earlier measurement . this suggests that additional type ia contamination has little effect on our overall result , and still results in an imf that is steeper than salpeter ( -2.35 ) . using resolved hst photometry , we have analyzed the stars surrounding 59 snr in m31 . using cmd fitting , we calculate a sfh within a 50 pc radius of each snr . we find that 53 of the snr regions display significant evidence of recent star formation , which we use to age - date the progenitor star . the remaining six regions display no recent star formation , and we consider them either possible type ia candidates or the result of massive runaway progenitor stars . we examine the distribution of progenitor masses for our ccsn candidates and find a lack of massive stars compared to a standard salpeter imf ( @xmath2 , where @xmath3 ) . if a uniform single imf is assumed , we find values for @xmath4 outside the range @xmath45 inconsistent with the measured distribution at 95% confidence . alternatively , if we consider a distribution that is a salpeter imf up to some maximum mass , we place an upper limit on the maximum mass allowed at @xmath46 m@xmath7 . we also estimate a minimum mass for core collapse of between 7.0 and 7.8 m@xmath7 , which is both greater than the maximum mass for white dwarf collapse @xcite and consistent with direct progenitor measurements @xcite . is supported in part by funding from the mary gates endowment . z.g.j . , b.f.w . , and j.j.d . are supported in part by go-12055 . is supported in part by an nsf astronomy and astrophysics postdoctoral fellowship under award ast-0802315 . this work is based on observations made with the nasa / esa hubble space telescope , obtained from the data archive at the space telescope science institute . support for this work was provided by nasa through hubble fellowship grant 51273.01 awarded to k.m.g . by the space telescope science institute . stsci is operated by the association of universities for research in astronomy , inc . under nasa contract nas 5 - 26555 . map of m31 . snrs in red correspond to those with no young sf detected . we consider these as possible type ia locations , or as the results of runaway stars . we note that a large amount of our ccsn candidates fall along the star forming arms of m31 . this is clearly expected ( ccsn candidates will tend to be found in regions of recent sf ) , but it also indicates the constraints imposed on our sample by the locations of archival hst data ( star - forming regions are targeted significantly more than other areas).,scaledwidth=80.0% ] + 1 - 006 & 10.6318 & 41.1005 & pos-33 & 10273 & wfpc-2 & f555w=24.5 , f814w=23.4 + 1 - 008 & 10.7675 & 41.6031 & pos-21 & 10273 & acs & f555w=26.6 , f814w=26.0 + 1 - 009 & 10.7975 & 41.6256 & pos-23 & 10273 & acs & f555w=25.6 , f814w=25.3 + 1 - 010 & 10.7979 & 41.4853 & pos-18 & 10273 & wfpc-2 & f555w=24.9 , f814w=23.7 + 2 - 016 & 10.3196 & 40.9554 & g-87 & 6671 & wfpc-2 & f555w=25.2 , f814w=23.9 + 2 - 020 & 10.4508 & 41.1138 & g-104 & 10260 & acs & f606w=25.6 , f814w=23.5 + 2 - 021 & 10.4773 & 40.7866 & g-119 & 6671 & wfpc-2 & f555w=26.2 , f814w=25.0 + 2 - 024 & 10.5958 & 41.0036 & pos-29 & 10273 & acs & f555w=25.8 , f814w=25.1 + 2 - 025 & 10.6448 & 40.9688 & pos-29 & 10273 & acs & f555w=25.9 , f814w=25.1 + 2 - 026 & 10.6698 & 41.0447 & pos-30 & 10273 & acs & f555w=26.0 , f814w=25.2 + 2 - 028 & 10.7376 & 40.9698 & pos-41 & 10273 & acs & f555w=25.7 , f814w=25.4 + 2 - 044 & 11.1989 & 41.4654 & b08-f10 & 12075 & acs & f475w=27.3 , f814w=26.0 + 2 - 046 & 11.2928 & 41.5993 & b12-f17 & 12071 & acs & f475w=27.5 , f814w=26.1 + 2 - 048 & 11.3094 & 41.6033 & b12-f17 & 12071 & acs & f475w=27.5 , f814w=26.1 + 2 - 049 & 11.3254 & 41.8683 & b15-f08 & 12056 & acs & f475w=27.4 , f814w=26.0 + 2 - 050 & 11.3662 & 41.8698 & b15-f07 & 12056 & acs & f475w=27.4 , f814w=26.1 + + bw-11 & 10.2467 & 40.6081 & g-76 & 11081 & wfpc-2 & f606w=26.2 , f814w=24.9 + bw-18 & 10.3988 & 41.1155 & g-104 & 10260 & acs & f606w=26.1 , f814w=24.9 + bw-19 & 10.5408 & 40.9472 & pos-27 & 10273 & acs & f555w=26.1 , f814w=25.5 + bw-20 & 10.5433 & 40.8644 & pos-26 & 10273 & acs & f555w=26.0 , f814w=25.0 + bw-31 & 10.7317 & 40.9956 & pos-41 & 10273 & acs & f555w=26.4 , f814w=25.4 + bw-32 & 10.7329 & 40.9717 & pos-41 & 10273 & acs & f555w=25.7 , f814w=24.6 + bw-36 & 10.7725 & 41.3750 & g-205 & 10260 & acs & f606w=25.5 , f814w=24.5 + bw-39 & 10.7933 & 41.6282 & pos-23 & 10273 & acs & f555w=25.8 , f814w=25.4 + bw-44 & 10.8779 & 41.6882 & pos-24 & 10273 & acs & f555w=26.5 , f814w=25.0 + bw-60 & 11.0883 & 41.9018 & b15-f12 & 12056 & acs & f475w=28.0 , f814w=26.8 + bw-61 & 11.1054 & 41.3501 & b06-f10 & 12105 & acs & f475w=27.3 , f814w=25.9 + bw-65 & 11.1488 & 41.4227 & b06-f04 & 12105 & acs & f475w=27.3 , f814w=25.9 + bw-66 & 11.1550 & 41.8666 & b15-f17 & 12056 & acs & f475w=27.5 , f814w=26.2 + bw-69 & 11.1825 & 41.9645 & b17-f18 & 12059 & acs & f475w=27.6 , f814w=26.5 + bw-71 & 11.1958 & 41.4886 & b08-f04 & 12075 & acs & f475w=27.5 , f814w=26.0 + bw-74 & 11.2129 & 41.4847 & b08-f04 & 12075 & acs & f475w=27.1 , f814w=25.8 + bw-76 & 11.2267 & 41.5121 & b08-f04 & 12075 & acs & f475w=27.3 , f814w=25.9 + bw-77 & 11.2269 & 41.5306 & b08-f04 & 12075 & acs & f475w=27.3 , f814w=25.9 + bw-81 & 11.2858 & 41.6101 & b12-f17 & 12071 & acs & f475w=27.4 , f814w=26.1 + bw-82 & 11.2892 & 41.8523 & b15-f08 & 12056 & acs & f475w=27.6 , f814w=26.3 + bw-84 & 11.3162 & 41.6561 & b12-f11 & 12071 & acs & f475w=27.4 , f814w=26.0 + bw-86 & 11.3387 & 41.6668 & b12-f11 & 12071 & acs & f475w=27.3 , f814w=26.0 + bw-89 & 11.3650 & 41.9036 & b15-f01 & 12056 & acs & f475w=27.5 , f814w=26.3 + bw-102 & 11.4675 & 42.1618 & b21-f11 & 12055 & acs & f475w=27.6 , f814w=26.8 + bw-105 & 11.6296 & 41.9886 & b18-f03 & 12108 & acs & f475w=27.6 , f814w=26.6 + bw-106 & 11.6417 & 42.1804 & b21-f08 & 12055 & acs & f475w=28.0 , f814w=27.0 + bw-110 & 11.6896 & 42.2183 & b21-f01 & 12055 & acs & f475w=27.7 , f814w=27.0 + + k180 & 10.9186 & 41.1814 & b02-f11 & 12073 & acs & f475w=27.2 , f814w=25.9 + k376 & 11.0850 & 41.5804 & b09-f14 & 12057 & acs & f475w=27.1 , f814w=25.8 + k446 & 11.1298 & 41.3572 & b06-f04 & 12105 & acs & f475w=27.3 , f814w=26.0 + k497 & 11.1562 & 41.4133 & b06-f10 & 12105 & acs & f475w=27.3 , f814w=26.0 + k516/bw-67 & 11.1699 & 41.4145 & b06-f04 & 12105 & acs & f475w=27.4 , f814w=26.0 + k525a & 11.1820 & 41.4372 & b08-f10 & 12075 & acs & f475w=27.3 , f814w=26.0 + k526a & 11.1715 & 41.4653 & b08-f10 & 12075 & acs & f475w=27.3 , f814w=26.0 + k527a & 11.1834 & 41.4465 & b08-f10 & 12075 & acs & f475w=27.3 , f814w=26.0 + k574 & 11.2101 & 41.4649 & b08-f10 & 12075 & acs & f475w=27.4 , f814w=26.1 + k594 & 11.2201 & 41.9161 & b15-f10 & 12056 & acs & f475w=27.2 , f814w=26.0 + k856a & 11.4313 & 41.9313 & b16-f05 & 12106 & acs & f475w=27.6 , f814w=26.4 + k891 & 11.5405 & 42.2198 & b21-f04 & 12055 & acs & f475w=27.6 , f814w=26.6 + k908 & 11.6231 & 41.9685 & b18-f03 & 12108 & acs & f475w=28.0 , f814w=27.0 + k934/bw-107 & 11.6467 & 42.2266 & b21-f01 & 12055 & acs & f475w=27.6 , f814w=26.9 + k947/2 - 047 & 11.6689 & 42.1911 & b21-f07 & 12055 & acs & f475w=27.6 , f814w=26.8 + k956a & 11.6792 & 42.2171 & b21-f01 & 12055 & acs & f475w=27.7 , f814w=27.0 + [ tab_snr ] bw-74 & 17@xmath48 & 11@xmath49 & 825 & 4572 & 169 & 0.3 + bw-44 & @xmath50 & @xmath51 & 644 & 2085 & 3 & 1.5 + bw-86 & 9.6@xmath52 & 28@xmath53 & 575 & 4344 & 48 & 0.3 + bw-84 & 8.7@xmath54 & 34@xmath55 & 572 & 5158 & 49 & 0.3 + k527a & 17@xmath56 & 11@xmath57 & 531 & 4925 & 62 & 0.1 + 2 - 049 & 7.6@xmath58 & 47@xmath59 & 517 & 4364 & 60 & 0.6 + bw-81 & 12@xmath60 & 19@xmath61 & 502 & 5212 & 16 & 0.4 + k525a & 20@xmath62 & 9.4@xmath63 & 501 & 4677 & 121 & 0.2 + bw-65 & 8.1@xmath64 & 40@xmath65 & 500 & 5263 & 57 & 0.5 + bw-77 & 7.6@xmath66 & 47@xmath67 & 473 & 4968 & 31 & 0.2 + k934 & @xmath6813 & @xmath6816 & 455 & 1934 & 75 & 0.7 + k908 & 8.6@xmath69 & 36@xmath70 & 418 & 3360 & 47 & 0.3 + bw-66 & 36@xmath71 & 5.5@xmath72 & 411 & 3310 & 63 & 1.6 + bw-71 & 10@xmath73 & 25@xmath74 & 401 & 5281 & 37 & 0.6 + bw-31 & 9.9@xmath75 & 27@xmath76 & 397 & 2288 & 23 & 0.3 + k446 & 9.0@xmath77 & 32@xmath78 & 389 & 4921 & 26 & 0.6 + k594 & 8.4@xmath79 & 38@xmath76 & 379 & 2537 & 26 & 0.5 + bw-106 & 19@xmath80 & 10@xmath81 & 372 & 2642 & 64 & 1.3 + k497 & 10@xmath82 & 26@xmath83 & 360 & 5106 & 26 & 0.2 + k856a & 27@xmath84 & 7@xmath85 & 356 & 2125 & 54 & 0.3 + k376 & 16@xmath86 & 12@xmath86 & 350 & 6874 & 55 & 0.7 + bw-76 & 8.5@xmath87 & 37@xmath88 & 345 & 5013 & 35 & 0.3 + 1 - 008 & 7.6@xmath89 & 46@xmath90 & 336 & 1895 & 61 & 1.7 + bw-61 & 15@xmath91 & 13@xmath92 & 303 & 4835 & 106 & 0.8 + 2 - 050 & 7.6@xmath93 & 47@xmath94 & 289 & 4343 & 11 & 0.5 + k526a & 8.4@xmath79 & 38@xmath76 & 288 & 4842 & 3 & 0.3 + bw-82 & 8.1@xmath95 & 40@xmath96 & 281 & 4017 & 8 & 0.3 + bw-89 & 8.1@xmath97 & 40@xmath98 & 265 & 4296 & 43 & 0.9 + k947 & 13@xmath99 & 16@xmath100 & 245 & 1874 & 37 & 0.9 + bw-60 & 14@xmath60 & 15@xmath101 & 243 & 3277 & 39 & 1.5 + bw-105 & @xmath688 & @xmath10244 & 239 & 1945 & 13 & 0.3 + bw-11 & 10@xmath103 & 26@xmath104 & 235 & 1199 & 105 & 0.9 + 1 - 009 & 8.9@xmath105 & 33@xmath106 & 219 & 739 & 36 & 1.2 + bw-20 & 7.9@xmath75 & 42@xmath107 & 192 & 1339 & 22 & 1.8 + 2 - 044 & 8.4@xmath108 & 37@xmath109 & 186 & 4368 & 13 & 0.2 + 2 - 046 & 11@xmath86 & 21@xmath76 & 185 & 4445 & 16 & 0.4 + 2 - 024 & 16@xmath86 & 12@xmath86 & 164 & 1934 & 5 & 0.5 + bw-39 & @xmath6811 & @xmath10220 & 163 & 844 & 104 & 2.1 + k574 & 8.1@xmath110 & 40@xmath111 & 162 & 4207 & 18 & 0.5 + k180 & 8.8@xmath79 & 33@xmath76 & 162 & 3950 & 23 & 0.6 + 2 - 048 & 7.7@xmath112 & 45@xmath113 & 160 & 3581 & 33 & 0.7 + k891 & @xmath689 & @xmath10236 & 149 & 1441 & 23 & 1.1 + 2 - 025 & 8.2@xmath114 & 39@xmath115 & 140 & 1287 & 13 & 0.3 + bw-18 & 9.9@xmath116 & 27@xmath76 & 117 & 2165 & 25 & 2.0 + k516 & 8.5@xmath117 & 36@xmath118 & 109 & 2210 & 10 & 0.4 + k956a & 8.1@xmath119 & 40@xmath120 & 103 & 1457 & 22 & 1.1 + bw-69 & 24@xmath121 & 7.5@xmath122 & 100 & 1467 & 8 & 1.5 + bw-32 & 9.3@xmath75 & 30@xmath76 & 91 & 562 & 6 & 0.5 + bw-110 & 12@xmath60 & 20@xmath123 & 86 & 1391 & 12 & 0.8 + bw-102 & 8.5@xmath124 & 36@xmath53 & 73 & 1568 & 31 & 1.5 + 2 - 020 & 18@xmath60 & 11@xmath9 & 38 & 1137 & 20 & 1.4 + 1 - 006 & 11@xmath86 & 24@xmath125 & 10 & 601 & 6 & 0.0 + 1 - 010 & 12@xmath86 & 19@xmath9 & 10 & 335 & 9 & 0.2 + + bw-19 & - & - & 291 & 2520 & 0 & 0.5 + bw-36 & - & - & 10 & 4162 & 0 & 1.3 + 2 - 026 & - & - & 191 & 2421 & 0 & 0.1 + 2 - 021 & - & - & 122 & 1268 & 0 & 0.0 + 2 - 028 & - & - & 53 & 374 & 0 & 0.5 + 2 - 016 & - & - & 5 & 466 & 0 & 0.6 + [ tab_results ]

the transport of charged particles in disorganised magnetic fields is an important issue which concerns the propagation of cosmic rays of all energies in a variety of astrophysical environments , such as the interplanetary , interstellar and even extra - galactic media , as well as the efficiency of fermi acceleration processes . we have performed detailed numerical experiments using monte - carlo simulations of particle propagation in stochastic magnetic fields in order to measure the parallel and transverse spatial diffusion coefficients and the pitch angle scattering time as a function of rigidity and strength of the turbulent magnetic component . we confirm the extrapolation to high turbulence levels of the scaling predicted by the quasi - linear approximation for the scattering frequency and parallel diffusion coefficient at low rigidity . we show that the widely used bohm diffusion coefficient does not provide a satisfactory approximation to diffusion even in the extreme case where the mean field vanishes . we find that diffusion also takes place for particles with larmor radii larger than the coherence length of the turbulence . we argue that transverse diffusion is much more effective than predicted by the quasi - linear approximation , and appears compatible with chaotic magnetic diffusion of the field lines . we provide numerical estimates of the kolmogorov length and magnetic line diffusion coefficient as a function of the level of turbulence . finally we comment on applications of our results to astrophysical turbulence and the acceleration of high energy cosmic rays in supernovae remnants , in super - bubbles , and in jets and hot spots of powerful radio - galaxies .

astro-ph0109223

the knowledge of the transport properties of charged particles in turbulent magnetized plasmas is a long standing problem , which bears directly on many astrophysical issues , such as the penetration of low - energy cosmic rays in the heliosphere @xcite , the propagation and escape of galactic cosmic rays in and out of the interstellar magnetic field @xcite , or even the efficiency of fermi acceleration mechanisms , in particular at shocks @xcite . the diffusion coefficient transverse to the mean component of the magnetic field plays a particularly important role in these issues , but to date , there is no satisfactory description of perpendicular transport . some studies have built upon or tried to extend the results of the `` quasi - linear theory '' @xcite , whose validity is limited to very low level turbulence , _ i.e. _ a turbulent component much weaker than the uniform magnetic field , and which calculates the transport coefficients by statistical averages of the displacements perturbed to first order in the inhomogenous field . other studies have appealed to phenomenological approximations such as the bohm estimate for the diffusion coefficient @xmath0 , which corresponds to the assumption that the mean free path for scattering @xmath1 of a particle of velocity @xmath2 , is given by the larmor radius @xmath3 . this approximation originates from laboratory experiments which led d. bohm to the empirical formula @xmath4 for a plasma with temperature @xmath5 . a theoretical derivation of this formula was proposed later by taylor and mcnamara @xcite , and then extended to relativistic particles @xcite , but no theory of bohm diffusion ( relativistic or not ) in magnetic irregularities has been derived _ stricto - sensu _ so far . therefore it appears that important physical and astrophysical issues are yet to be answered : * how do the transport properties change when the level of magnetic turbulence is increased ? what are the transport properties when the mean field vanishes ? notably , what is the relevance of the bohm scaling ? * even for low level turbulence , transverse space diffusion is not well known . it nevertheless plays a crucial role in the confinement of cosmic rays in galaxies or other extragalactic objects ( notably radio - galaxies jets ) . its magnitude is also of direct relevance to the performance of fermi acceleration at perpendicular shocks . * do subdiffusive and more generally anomalous diffusion regimes exist ? if yes , they are also of importance for fermi acceleration . in order to shed light on these issues , we have performed extensive numerical experiments to determine the pitch angle scattering rate , and the parallel and perpendicular spatial diffusion coefficients for a wide range of rigidities and turbulence levels . our experiments are conducted by monte - carlo simulations in which we follow the propagation of a relativistic test particle in a stochastic magnetic field constructed from three - dimensional kolmogorov turbulence , and calculate the diffusion coefficients from the statistical correlations along the trajectory . our study is similar to the recent work of giacalone and jokipii @xcite in which the spatial diffusion coefficients in two and three dimensional magnetostatic turbulence were measured using monte - carlo simulations for various turbulence levels and rigidities . our study is however more extensive than that of ref . in particular we measure the diffusion coefficients in a broader range of rigidities , by studying the diffusion of particles with larmor radii larger than the coherence length , and in a broader range of turbulence levels , by going up to pure turbulence in which there is no uniform component of the magnetic field . in contrast , ref . @xcite studies the case of lower - energy particles and smaller turbulence levels , with a turbulent magnetic field never exceeding the uniform component in strength . we also study in detail the pitch angle scattering rate , which is of central interest in applications to shock acceleration processes , and study in more detail the issue of transverse diffusion and its relation to the chaotic wandering of field lines . finally we will repeatedly compare our results to ref . @xcite where there is overlap , which is important since these numerical experiments are delicate . among our results , we confirm the extrapolation to high turbulence levels of the scalings predicted by the quasi - linear theory for the scattering rate and the parallel diffusion coefficient at low enough rigidity . the perpendicular diffusion coefficient is shown to follow a law which is quite different from the predictions of the quasi - linear theory at low rigidities . we argue that its behavior is compatible with chaotic wandering and diffusion of the magnetic field lines to which particles are `` attached '' . in particular , we demonstrate the chaotic behavior of the magnetic field lines and calculate the associated kolmogorov length and diffusion coefficient in terms of the turbulence level . we also show that the bohm diffusion coefficient only holds in a limited range of rigidities @xmath6 for pure turbulence , and does not exist when the mean field is non - vanishing . in this latter case the bohm value for the coefficient is only obtained at maximum pitch angle scattering , _ i.e. _ for particules with larmor radius of order of the coherence scale . finally , we also found that diffusion operates even for particles whose larmor radii is larger than the coherence length , as far as we have searched in rigidity ( 1.5 decade ) . on these scales , the scattering rate decays as expected , albeit moderately , as the power @xmath7 of the rigidity . our study is conducted with the following main simplifying assumptions : * the magnetic field is composed of a mean homogeneous field @xmath8 and an inhomogeneous component @xmath9 : @xmath10 . * the magnetic disturbances are considered to be static . this assumption is well justified as the waves propagate with velocities of the order of the alfvn velocity @xmath11 , smaller than the velocity of particles @xmath12 ( we consider relativistic particles ) , and the electric force is thus smaller than the magnetic force by a factor @xmath13 . the first correction to the theory is the celebrated second order fermi process which can be described by diffusion in momentum space , with diffusion coefficient @xmath14 , with @xmath15 particle momentum , and @xmath16 , the angular scattering frequency , is an outcome of our study . * the magnetic perturbations are distributed according to isotropic turbulence , whose power spectrum is written in terms of fourier momentum @xmath17 as : @xmath18 for @xmath19 , zero otherwise , and @xmath20 , _ i.e. _ random phases . the exponent @xmath21 characterizes the properties of the turbulence , and we will concentrate on the case @xmath22 in our numerical applications , which describes kolmogorov turbulence . the smallest turbulence wavenumber is related to the maximum scale of the turbulence @xmath23 _ via _ : @xmath24 . this largest scale also corresponds to the correlation length of the magnetic field to within a factor of order unity [ see eq.([eq_correlation ] ) ] . our notations are as follows . the quantities we will be interested in are the scattering rate @xmath25 or scattering time @xmath26 , defined as the correlation time of the pitch angle , the spatial diffusion coefficient along the mean field @xmath27 and the transverse spatial diffusion coefficient @xmath28 . these coefficients are evaluated in terms of turbulence level @xmath29 $ ] , and rigidity @xmath30 . for convenience , the larmor radius @xmath31 is defined with respect to the total magnetic field : @xmath32 for a particle with energy @xmath33 and charge @xmath34 . the larmor pulsation of a particle of energy @xmath33 is defined , for convenience , as @xmath35 , and the larmor time @xmath36 . we define the scattering function as @xmath37 . when useful , we will denote by @xmath38 the larmor pulsation in the mean field . the paper is organised as follows . in section 2 , we recall the relation between spatial diffusion and the scattering off magnetic disturbances and present the numerical method . in section 3 , we present our numerical results and discuss the issue of transverse diffusion and the measurement of magnetic chaos characteristics . a discussion with direct astrophysical consequences of our results is given in section 5 , and conclusions are offered in section 6 . finally , in appendix a , we propose a theoretical interpretation of the regimes of diffusion observed , and in particular of the existence of diffusion for larmor radii larger than the maximum scale of turbulence . high energy particles interact with cosmic matter mostly through scattering on the magnetic field which is more or less frozen in the medium . the interaction is elastic in the frame of a magnetic disturbance and it can be considered as elastic in the plasma rest frame to lowest order in @xmath39 , if the disturbance propagates at small enough velocity @xmath40 . with respect to a given direction , chosen as that of the uniform component of the magnetic field if this latter is non - vanishing , the pitch angle of the particle changes almost randomly if the magnetic field is sufficiently disorganized ( this will be made more precise further on ) . therefore the position of the particle changes according to a random walk on a time scale which is longer than the coherence time of the pitch angle cosine , and pitch angle scattering is thus responsible for the diffusion of particles . however it is generally believed that transverse diffusion may also occur through wandering of the magnetic field lines . in this picture , the transverse velocity of the particle changes through resonant diffusion as before while the guiding center of the approximate helical motion wanders with the magnetic field line to which it is attached , and performs a random walk in the transverse direction . these notions will be quantified in the forthcoming sections . our main objective here is indeed to quantify these various contributions to the process of diffusion . we define the pitch angle @xmath41 with respect to the mean field direction when it exists , otherwise the direction can be arbitrarily chosen . the convenient random function is the pitch angle cosine : @xmath42 , and @xmath41 is a function of time . since we assume a static spectrum of magnetic perturbations , the auto - correlation function of @xmath43 will become stationary in the large time limit . it can then be defined as : @xmath44 where the average can be performed in three different ways . in the original quasi - linear theory , this average is taken over the phases of the magnetic disturbances . in the theory of chaos , the average is performed over the phase space subset of chaotic motions . in practice , and this is what we will use in the numerical experiment , we assume ergodicity and make temporal average . our procedure of calculating averages is explicited further below . the scattering time @xmath45 can then be defined as the coherence time of the pitch angle cosine : @xmath46 in particular , if the auto - correlation function falls off exponentially @xmath47 then @xmath48 . turning to the spatial diffusion coefficient @xmath27 , let @xmath49 be the coordinate of a particle along the mean field direction . then @xmath50 with a constant velocity @xmath2 ( in our case @xmath51 ) , since energy is conserved . consider now a random variation @xmath52 of @xmath49 during the time interval @xmath53 supposed to be larger than the scattering time @xmath45 . one has @xmath54 , and @xmath55 beyond the scattering time @xmath45 , if the stationary random process @xmath43 explores uniformly the interval @xmath56 , the space diffusion coefficient parallel to the mean field stems straightforwardly from its definition , eq . ( [ eq : dxpc ] ) and eq . ( [ eq : tau ] ) : @xmath57 here as well the average can be made according to one of the three ways explained above . the main goal of the computation is then to determine the dependence of @xmath45 , or the scattering function @xmath58 , in terms of the rigidity @xmath59 and the turbulence level @xmath60 . the theoretical result is known in the regime of weak turbulence @xcite , if the correlation time @xmath61 of the force suffered by the particle is much smaller than the scattering time @xmath45 . to make it more precise , particles undergo resonances with the mhd modes such that @xmath62 . the correlation time is related to the width of the resonance in the mode spectrum , such that : @xmath63 where @xmath64 denotes the spectrum width , in the parallel direction . since @xmath65 , @xmath66 is equivalent to @xmath67 . in this case the memory of the initial pitch angle can even be kept and the scattering function @xmath68 . however diffusion coefficients calculated on timescales larger than @xmath45 must be averaged over @xmath69 . due to rotation invariance around the mean field direction , there is a single transverse diffusion coefficient ( when diffusion occurs ) , given by : @xmath70 where @xmath71 denotes the displacement perpendicular to the mean field during the time interval @xmath53 . in weak turbulence theory ( @xmath72 ) , the gyro - phase @xmath73 of the particle is only weakly perturbed by the disorganized component of the field and @xmath74 , where the gyro - pulsation @xmath38 is determined with respect to the mean field . the transverse velocity can be approximated by : @xmath75 , \label{eq : vperp}\ ] ] where @xmath76 denotes the charge of the particle , and we implicitly assumed the mean field to lie along the direction @xmath77 . the pitch angle sine @xmath78 varies on the timescale @xmath45 , which is much longer than the larmor time in the weak turbulence regime . the time correlation function of the pitch angle is obviously the same as that of the cosines since @xmath79 , hence @xmath80 . therefore the transverse diffusion coefficient reads : @xmath81 assuming that the correlation function @xmath82 decays exponentially on the characteristic time @xmath45 , one finally obtains a result similar to the so - called classical diffusion that reads @xmath83 this transverse diffusion based on pitch angle scattering only leads to the ratio @xmath84 where @xmath85 is the mean free path of a particle along the mean magnetic field . this relation can also be obtained by treating the magnetic disturbances as hard sphere scattering centers with weak or strong turbulence . it is also a result of the study of ref . @xcite , which estimate phenomenological diffusion coefficients by using well - motivated assumptions for the velocity auto - correlation functions of the particle orbit . finally , since @xmath86 in the weak turbulence regime , one expects @xmath87 when @xmath88 . however the transverse diffusion may turn out to be larger than predicted by quasi - linear theory , even for moderate turbulence . in particular note that eq . ( [ eq : dpar ] ) for the parallel diffusion coefficient rests on the sole assumption that @xmath82 vanishes on timescales longer than @xmath45 , while the quasi - linear result for the transverse diffusion coefficient , eq . ( [ eq : ddp ] ) , assumes that the particle orbit is only weakly perturbed and the timescale of variation of the pitch angle is much longer than the larmor time , _ i.e. _ , that the level of turbulence @xmath88 . we refer to this result as a prediction of quasi - linear theory ; it neglects the diffusion of the guide center carrying field line and the associated process of chaotic magnetic diffusion which has been analysed by jokipii and parker @xcite , and to which we will come back in the following section . finally in all cases one should obtain @xmath89 when @xmath90 , since the mean field vanishes in this limit and there is no prefered direction anymore . in order to evaluate the transport coefficients , we follow the propagation of particles in stochastic magnetic fields by integrating the standard equation of motion ( lorentz force ) , and measure the statistical quantities of interest to us , namely @xmath91 , @xmath27 and @xmath28 , using the estimators defined respectively in eqs . ( [ eq : tau]),([eq : dpar]),([eq : ddp ] ) . strictly speaking , the averages contained in these expressions should be taken over the phases of the magnetic inhomogeneities . in practice however , one may as well take these averages as follows . for a given @xmath53 [ using the notations of eqs . ( [ eq : dpar]),([eq : ddp ] ) ] , a time @xmath92 is picked at random over the trajectory , and the correlation between the positions at times @xmath92 and @xmath93 is recorded ; this operation is repeated many times and the average is kept . this latter is then further averaged over a population of particles with random initial positions , and then over an ensemble of magnetic field realizations with random phases . in practice , we propagate @xmath94 particles , measure the correlations at @xmath95 different times along the trajectory of each particle , and use a few magnetic realizations . this procedure allows to reach a sufficiently high signal - to - noise ratio in the simulation for a moderate computer time , as indeed setting up the magnetic field and propagating a particle is much more costly than taking averages along the trajectory . in principle one could as well take the average @xmath96 as the variance of the displacement at time @xmath53 over a population of particles originally concentrated at the origin , as in ref . however this method requires to follow the trajectory of a large number of particles @xmath97 in order to achieve a reasonable signal - to - noise ratio . the method we employ , which measures the correlations along the trajectory of each particle , before averaging over a population of particles , is less costly in computer time ( but requires much more memory ) . nevertheless , we also checked ( and found ) that the method which measures the variance of the displacement gave results in agreement with our method within the error attached to the small number of particles propagated . the magnetic field can be constructed in two different ways which both present pros and cons . the first method uses fast - fourier transform ( fft ) algorithms to set up the magnetic field on a discrete grid in configuration space , starting from the magnetic field defined through its power spectrum in fourier space , _ i.e. _ : @xmath98 . \label{eq : b_fft}\ ] ] in this equation , @xmath99 is the tri - dimensional wavenumber vector , with integer coordinates taking values between @xmath100 and @xmath101 , @xmath102 is a unit vector orthogonal to @xmath99 ( this ensures @xmath103 ) , @xmath104 is the amplitude of the field component , and is defined such that @xmath105 and @xmath106 , where the average concerns the phases of the magnetic field . finally , @xmath107 is a numerical prefactor which ensures the correct normalization of the inhomogenous magnetic component with respect to the mean field , by using the following ergodic approximation to @xmath108 : @xmath109 and as before @xmath110 . in practice , the field components are calculated at each vertex @xmath111 of a discrete grid in configuration space beforehand . the boundary conditions are periodic with period @xmath23 , and the fundamental cubic cell size is @xmath112 , where @xmath113 represents the number of wavenumber modes along one direction . one thus has : @xmath114 , where the factor 2 comes from the fact that one must consider both negative and positive @xmath17-modes to respect the hermiticity of @xmath115 . in our simulation , we typically use @xmath116 and in some cases @xmath117 which gives us a dynamic range of two orders of magnitude . during the propagation of particles , it is of course necessary to know the magnetic field at any point @xmath118 for the integration of the equations of motion . our numerical code calculates the value @xmath119 either by tri - linear interpolation between the known values of the field components on the 8 vertices of the cell to which @xmath118 belongs , or by taking the value of @xmath9 at the vertex closest to @xmath118 , which amounts to assuming a constant @xmath9 in cells of size @xmath120 centered on each vertex . while the former method does not respect @xmath121 , the latter implies a discontinuous magnetic field on each cubic cell face . we will show in the following that the results obtained by these methods differ only when scales smaller than the cell size are concerned , as expected . a second algorithm for computing the magnetic field has been proposed by giacalone & jokipii @xcite and calculates the magnetic field as a sum over plane wave modes ( we will refer to this algorithm as gj ) . the expression defining @xmath122 is very similar to eq . ( [ eq : b_fft ] ) above , except that @xmath99 needs not have integral coordinates anymore , as fast - fourier transform methods are not used . indeed , one does not calculate the field on a discrete grid beforehand , but its values are calculated where and when needed during the propagation directly from the sum over plane waves . also the sum is not tri - dimensional , but one - dimensional ; the wavenumbers directions are drawn at random , and the amplitude @xmath123 to account for phase space volume . in practice , it is convenient to have logarithmic spacing of the @xmath124modes between @xmath125 and @xmath126 . one main advantage of the gj method is that there is no restriction in dynamic range due to memory usage , and consequently @xmath127 can be as large as required . however , one is limited in terms of computer usage time since it is expensive to perform the sum over the wavenumber modes at each point of the trajectory if the number of modes @xmath128 becomes significant . in practice , @xmath129 is a strict upper limit for our applications @xcite , and even with @xmath130 the calculation is already much slower than a similar calculation with the above fft algorithm . the number of modes is important as it controls the efficiency of diffusion , since pitch angle scattering proceeds mainly through resonance of the particle momentum on the magnetic field modes . in quasi - linear theory the resonance condition reads @xmath131 , where @xmath132 is the component of the wavenumber along the mean magnetic field direction . the fft and gj share a similar number of resonance modes in this limit @xmath133 . however for each resonant @xmath132 the fft algorithm has @xmath134 transverse components to be compared with one for the gj algorithm . one thus expects that at higher turbulence levels , diffusion should be more effective in the fft algorithm due to the much larger total number of modes than in the gj algorithm . furthermore , in order to preserve a correct spacing of modes in the gj algorithm , one can not indefinitely increase the dynamic range @xmath135 since @xmath128 is fixed for practical reasons , _ i.e. _ computer time . however the fft algorithm suffers from other limitations ( apart from the limitation in dynamic range ) : the interpolation of @xmath9 on scales smaller than the cell size , and the periodicity on the scale @xmath23 . these limitations are not present in the gj algorithm , and imply that the results of the fft method obtained for larmor radii much smaller than the cell size , _ i.e. _ @xmath136 , or much bigger than the periodicity scale , @xmath137 , can not be trusted , since these regimes are likely to be dominated by systematic effects related to the discreteness or to the periodicity . overall both methods appear complementary to each other , and we use them in turn to compare and discuss the robustness of our numerical results with respect to the assumptions made . ( in units of larmor time @xmath138 ) for various rigidity @xmath139 and for @xmath140.,width=377 ] as a function of rigidity @xmath59 . the symbols correspond to the measurements made through our monte - carlo experiments and correspond to various turbulence levels , as indicated . these results have been obtained using the fft numerical method ( see text ) , except for the filled circles which correspond to the gj algorithm . the vertical dashed lines indicate the range of validity of our fft algorithm ( _ i.e. _ all symbols except crosses ) , delimited by @xmath141 , and @xmath142 , which correspond respectively to larmor radii @xmath143 ( @xmath144 cell size ) and @xmath145 . the simulation for @xmath146 shown by crosses has been obtained with a much larger dynamic range than the others , _ i.e. _ @xmath147 . finally , the dotted lines correspond to power law approximations with slopes @xmath148 and @xmath149 . see text for comments . , width=377 ] the first numerical investigation to perform is the self - correlation function of the pitch angle cosine . the behavior of this function is shown in fig . ( [ fig : corr ] ) _ vs _ time interval @xmath150 for various levels of turbulence @xmath60 . two bumps are observed at one and two larmor periods . these bumps are observed as long as the regular magnetic field @xmath151 exists . since the decorrelation times for @xmath152 are larger than one larmor period , the larmor motions are not completely disorganized and contribute to the correlation function with some harmonics generated by nonlinearities . the inflexion of the function indicates that it behaves as @xmath153 as @xmath154 and then decreases exponentially in @xmath155 as @xmath156 . thus the determination of @xmath45 and @xmath25 by a numerical integration of the correlation function is accurate . in fig . ( [ fig : scatt ] ) we show the scattering frequency @xmath157 , which is the main quantity of interest for evaluating the transport coefficients . this figure shows several interesting features which deserve further comments . first of all , one finds that both methods for calculating the magnetic field , _ i.e. _ fft and gj , agree well within the range of validity of the former method , namely for @xmath158 , where @xmath159 , and @xmath160 . these two limiting rigidities correspond to larmor radii of order of the cell size and of the maximum scale of turbulence respectively , and result from the discreteness and periodicity of the magnetic field grid , as explained in section [ subsec.nuco ] . one finds that the scattering function behaves as a power - law with different slopes depending on the rigidity and turbulence level . for @xmath161 , it must be emphasized that the results can not be trusted for the fft results , _ i.e. _ all symbols except filled circles , and the change of slope may be artificial . for @xmath162 and @xmath163 , it appears that @xmath164 , in accordance with the quasi - linear prediction since @xmath165 . for @xmath166 , one finds @xmath167 , an unexpected result , since the resonance conditions can not be satistified at these high rigidities , and the quasi - linear theory thus predicts a sudden drop of the scattering frequency at @xmath166 . in appendix a , we provide a first theoretical explanation of this result by expansion of the particle trajectory in the random displacement and statistical averaging of a non - perturbative resummation of an infinity of graphs of correlations along the trajectory . for @xmath168 and @xmath169 one notes a flattening of the scattering function with recovery of the exponent @xmath148 power law at smaller rigidities . this flattening is definitely present for @xmath146 ( no mean component of the magnetic field ) , and correspond to the phenomenological bohm diffusion regime , as will be seen further below ; however , it only extends over slightly less than a decade in rigidity for @xmath170 , even though for that simulation the dynamic range was very large @xmath171 . at maximum pitch angle scattering , i.e. when @xmath172 and @xmath173 , the scattering function @xmath174 , _ i.e. _ the pitch angle scattering time @xmath45 is of order 2 larmor times @xmath175 . it should be noted that we define the rigidity with respect to the maximum scale of turbulence , which strictly speaking does not coincide with the coherence scale @xmath176 of the turbulent magnetic field . in effect , the spatial correlation function of the turbulent component is defined as @xmath177 with @xmath178 the power spectrum . this integral can not be integrated analytically for a power - law spectrum @xmath179 but one can check numerically that the maximum of the correlation function occurs at scale @xmath180 . turning to the spatial diffusion coefficients , it is interesting to plot the statistical estimators for @xmath181 and @xmath182 given by eqs . ( [ eq : dpar]),([eq : ddp ] ) as a function of time for different turbulence levels , and the result is shown in fig . ( [ fig : diff ] ) . this figure illustrates the transition from the regime in which the particle orbit is weakly perturbed and memory of the initial conditions is kept to the regime in which this memory is lost and the particle diffuse , @xmath183constant . the level of this plateau gives the magnitude of the diffusion coefficient ; fig . ( [ fig : diff ] ) also gives an idea of the uncertainty in our measurement of diffusion coefficients . finally , this figure also confirms the expected results @xmath184 when @xmath185 and @xmath186 as @xmath187 . it should be pointed out that the initial value of the pitch angle cosine was @xmath188 in all simulations ; we have checked that our results are insensitive to this value as long as the turbulence level @xmath189 , as expected . in fig . ( [ fig : dpar ] ) , we show the behavior of the parallel diffusion coefficient as a function of rigidity for various turbulence levels . the dotted lines correspond to the approximation of @xmath181 obtained from the calculation of @xmath45 using eq . ( [ eq : dpar ] ) , and the agreement appears excellent . this study does not confirm the existence of a bohm scaling . more precisely , the bohm diffusion coefficient @xmath190 only applies at @xmath146 in the range @xmath170 , in agreement with the similar conclusion for the scattering function . in all other cases the quasi - linear prediction is verified , _ i.e. _ @xmath191 for @xmath192 . we also found that a diffusion regime exists for rigidities greater than the upper bound of the resonance region , _ i.e. _ @xmath193 , for as far as we have searched , or about 1.5 decade . in this regime @xmath166 , @xmath194 , for all values of @xmath60 . in units of @xmath195 , as a function of the time interval @xmath53 in units of @xmath175 , for various turbulence levels ( @xmath196 ) , and for both the transverse displacement ( lower thin line curves ) and parallel displacement ( upper thick curves ) . one sees the transition from the weakly perturbed propagation regime @xmath197 to the diffusion regime @xmath198 , which appears here as a plateau . the transition duration depends on the turbulence level , and is of order of @xmath45 the scattering time . the diffusion coefficients are given by the levels of the plateaux . obviously , @xmath199 for @xmath152 and the two meet in the limit @xmath200 , as expected.,width=377 ] in units of @xmath195 as a function of rigidity for various turbulence levels . the symbols and vertical dashed lines are as in fig . ( [ fig : scatt ] ) . the dotted lines are obtained from the pitch angle scattering rate , using eq . ( [ eq : dpar ] ) . , width=377 ] in fig . ( [ fig : dperp ] ) , we plot the behavior of the transverse diffusion coefficient as a function of rigidity for various turbulence levels . it is useful to plot also the quantity @xmath201 as shown in fig . ( [ fig : dper2dpar ] ) . indeed , the noise of the simulation is then reduced and this figure allows to compare directly the power law behaviors of @xmath182 and @xmath181 . ) . , width=377 ] as a function of rigidity for various values of @xmath60 . the notations of symbols are as indicated and as in previous figures . the dotted curves overlaid on this figure correspond to the classical scattering result given by eq . ( [ eq : cst ] ) , and correspond from bottom to top to the represented values of @xmath60 in increasing order except for @xmath146 . these models account marginally for the numerical results for high rigidity particles @xmath202 and small turbulence levels @xmath203 , but diverge significantly from the experiment in other regimes.,width=377 ] this figure indeed reveals a clear trend . for all @xmath60 , the ratio @xmath204 is independent of rigidity for @xmath163 , and scales as @xmath205 for @xmath166 . a similar regime has been found by giacalone & jokipii @xcite for @xmath206 , albeit with slightly lower values than ours . this constancy is interpreted in the following as the signature of diffusion due to the chaotic wandering of the guide center carrying field lines . the importance of the guiding center diffusion was pointed out by jokipii @xcite as early as 1966 in order to correct the quasi - linear result ; however this derivation does not apply to high turbulence levels . finally , the ratio @xmath204 converges as expected to @xmath100 for all @xmath59 when @xmath187 . however it is interesting to note that even at @xmath168 , there remains the power law dependence for @xmath166 , @xmath207 . we have found evidence for subdiffusive regimes @xmath208 , with @xmath209 , at low enough rigidities @xmath210 and for @xmath162 . on analytical grounds one expects @xmath211 , corresponding to the so - called process of compound diffusion @xcite , and we have found values of @xmath212 close to this value indeed . however we have not been able to investigate in detail this issue , as it is very consuming in terms of computer time . in effect , this can be studied only using the gj algorithm , since it takes place at low rigidities outside the dynamic range of the fft algorithm . we have thus decided to postpone the study of these anomalous regimes to a subsequent publication . when the magnetic field is a superposition of a mean field and an irregular component depending on all three spatial coordinates , the field line system generically exhibits chaotic solutions . for instance it is sufficient to use a distribution of fourier modes following a power law in wavenumber to obtain a chaotic system . however a two - dimensional field can not have chaotic field lines , and a one - dimensional system can not produce transverse diffusion , as the particles are confined in a flux - tube by conservation of the adiabatic invariant @xcite . an example of this phenomenon is shown in fig . [ fig : nochaos ] in which we show the transverse wandering of a particle in three - dimensional and one - dimensional turbulence . , and for one - dimensional turbulence , the inhomogeneous component is taken to depend on the coordinate @xmath213 , with the homogeneous magnetic field component lying along the @xmath213 axis . note the difference in behavior : in one - dimensional turbulence , the particle is confined to a flux tube and does not diffuse.,width=377 ] in a three - dimensional chaotic system the separation between two initially adjacent field lines first increases exponentially @xmath214 as a function of the abscissa @xmath215 along the field line , with characteristic lyapunov exponent @xmath216 , also called the kolmogorov length . when the separation has become larger than the coherence length of the magnetic field , it behaves diffusively with magnetic diffusion coefficient @xmath217 , where @xmath218 denotes the separation between the two field lines . our numerical computation of the field lines clearly displays this two- step behaviour . in fig . [ fig : div ] , we plotted the separation squared between two field lines as a function of the curvilinear abscissa for @xmath219 . these calculations have been obtained by integrating the equations defining the field lines , namely @xmath220 , instead of integrating the particle equation of motion . figure [ fig : div ] clearly shows this two - step behavior and confirms that the transition from one regime to the other occurs when @xmath221 . .,width=377 ] this calculation allows us to measure the two lengths @xmath216 and @xmath222 with a relatively good accuracy . the results are reported as function of @xmath60 in fig . ( [ fig : kol ] ) . . the two lengths are normalized to the largest scale of turbulence @xmath23.,width=377 ] the effective transverse diffusion of particles in a chaotic magnetic field has been derived by rechester & rosenbluth @xcite . here we extend their argument by assuming that the primary transverse diffusion is anomalous ( sub- or super - diffusive ) . the problem can be stated as follows . after @xmath223 scattering times , parallel diffusion leads to a diffusion in curvilinear abscissa @xmath224 , whereas the transverse primary variation causes transverse displacement such that @xmath225 , with @xmath226 for normal diffusion , and @xmath227 for subdiffusion . because of field line exponential divergence , until the separation is of order the correlation length , say after @xmath228 scatterings , the transverse displacement is amplified exponentially by a factor @xmath229 with @xmath230 . after @xmath231 scatterings ( @xmath232 ) , an effective transverse diffusion coefficient can then be estimated as : @xmath233 because the particles almost follow the field lines , the first factor can be approximated by the magnetic diffusion coefficient @xmath222 , and one gets : @xmath234 the number @xmath231 is obtained by equating the separation distance and the transverse correlation length of the field lines @xmath235 ( in our case @xmath236 ) : @xmath237 which leads to @xmath238 , \label{eq : nc}\ ] ] where @xmath239 ( the scattering length ) . the main result is that magnetic chaos amplifies the transverse diffusion in such a way that it becomes a sizable fraction of the parallel diffusion : @xmath240 } d_{\parallel } \label{eq : dpko}\ ] ] as can be seen , the primary subdiffusion does not refrain the effective diffusion due to chaos . when @xmath241 ( non - anomalous primary diffusion ) , the logarithmic factor reads @xmath242 , with @xmath58 the scattering function as before . finally note that the above regime of diffusion applies at late times after @xmath228 scattering times . the intermediate regime , for @xmath223 scattering times , with @xmath243 , leads to sub - diffusive motion ( compound diffusion ) , with @xmath244 , see for instance ref . we have found evidence for such a regime , but a detailed study of its behavior lies beyond the present work and is defered to a later study . no theory gives the ratio @xmath245 , except for some toy models such as the chirikov - taylor mapping @xcite . however our numerical experiment can provide a fairly accurate estimate of this ratio . in particular we find that the kolmogorov length @xmath246 and that the magnetic diffusion coefficient @xmath247 as long as @xmath248 [ see fig.([fig : kol ] ) ] . beyond this limit , our calculations of @xmath249 and @xmath250 do not provide accurate estimates of these lengths , especially for the kolmogorov length which loses its physical meaning when @xmath60 reaches unity . therefore bearing in mind that the two diffusion coefficients become the same as @xmath251 , we conjecture that the result should be : @xmath252 when @xmath253 . this non perturbative result would be in agreement with the perturbative result obtained by chuvilgin and ptuskin ref . @xcite for small amplitude ( written @xmath254 ) large scale varying fields , the ratio between the two coefficients being proportional to @xmath255 . finally , it is important to note that our numerical results for @xmath204 shown in fig . [ fig : dper2dpar ] have been obtained independently of the above magnetic diffusion law . we find , in agreement with the above relation , that @xmath204 is independent of @xmath59 for @xmath163 , and that @xmath256 provides a good fit to the scaling observed at @xmath152 . however the numerical prefactor in this relation is rather of order @xmath257 for @xmath258 , whereas it should be @xmath259 if the extrapolation could be taken up to @xmath146 . nevertheless , the above provides solid evidence in favor of a dominant contribution of magnetic diffusion to the process of transverse diffusion . in this section , we offer revised estimates of the maximal energy that can be attained by fermi acceleration mechanisms by comparing the acceleration time and the time of escape of cosmic rays outside of the accelerating region using the results obtained in the previous section . we first consider the case of galactic supernovae remnants ( snr ) and so - called superbubbles , and then turn to the case of jets in extragalactic sources . the lagging questions of the production of cosmic rays in supernovae remnants has been recently reviewed in ref . one of the major problems in accounting for the observational data is that the maximum energy achieved by the fermi process in the snr shock is well below the so - called `` knee''-range @xmath260ev . if one uses the bohm approximation to the diffusion coefficient , there is hope to reach the knee energy with sufficient efficiency to expect significant @xmath261-ray emission resulting from @xmath262 decay generated by pp - collisions . the lack of detection of these gamma - rays @xcite or their marginal detection at best , ruined these optimistic assumptions . the fermi acceleration at a shock of velocity @xmath263 is characterized by an acceleration timescale @xmath264 . in most of the shock region , @xmath265 hence @xmath266 , where @xmath267 . the maximum energy is limited by the age of the supernovae remnants , and one thus obtains @xmath268 this result differs from @xcite only by the factor @xmath269 . this factor is close to unity when the bohm scaling applies ; but , as we have found in section [ subsec.nure ] , in fact @xmath270 . strictly speaking this scaling is valid for kolmogorov turbulence , and one expects the turbulent magnetic field downstream to differ from isotropic three - dimensional kolmogorov , but the above scaling serves well for order of magnitude estimates . moreover @xmath271 for larmor radii smaller than turbulence correlation length which could be the case even for the most energetic particles . the bohm approximation thus appears very optimistic . super - bubbles correspond to huge cavities created by @xmath272 snr shock waves built around massive stars associations . the size of these regions can be a sizable fraction of the galaxy disk thickness @xmath273pc ) . in effect , a typical super - bubble radius can be estimated as @xcite @xmath274 where @xmath275 measures the mechanical luminosity of the ob - stars association , @xmath276 the particle density of the surrounding interstellar medium @xmath277@xmath278 , and @xmath279 is the super - bubble lifetime in units of myr . the bubble plasma is more dilute than the interstellar medium by at least two orders of magnitude and thus the alfvn velocity is much greater . this density reads @xcite @xmath280 where @xmath281 is a number of order unity @xcite . the bubble is traversed by many shock fronts propagating with velocity of order or greater than the alfvn velocity ; a second order type of fermi acceleration is thus at work . its acceleration time scale is given by @xmath282 . the maximal energy is limited by escape of the particles , which is governed by the diffusion across the galaxy disk thickness for the most energetic . a strict lower limit to the time of escape @xmath283 can be obtained by using the parallel diffusion coefficient , since @xmath284 with @xmath285 the diffusion coefficient . transverse diffusion would improve the confinement time and thus lead to a higher maximal energy ; however one should then take into account the fact that magnetic lines come out of the galaxy disk and unfold in the halo . let us consider the lower estimate : @xmath286 the maximal energy for acceleration by second order fermi process in super - bubbles then corresponds to @xmath287 , and reads @xmath288 where we used @xmath289 . therefore the second order fermi process in super - bubbles might cover the knee range with slightly optimistic assumptions , since the magnetic field intensity can easily reach @xmath290 g in these super - bubbles , and @xmath291 . moreover , at that maximal energy the rigidity reaches unity and therefore @xmath292 , smaller but close to unity . at this point it is useful to recall that the confinement limiting energy of cosmic rays in the galaxy obtained by equating the larmor radius with the thickness @xmath293 is of order @xmath294ev . extragalactic jets emanating from active galactic nuclei have been considered as possible sources of ultra high energy cosmic rays with @xmath295ev because the confinement limit in a jet of radius @xmath296 and bulk lorentz factor @xmath297 is @xmath298 and for the powerful , strongly collimated and with terminal hot spots fr2 jets , the product @xmath299 is roughly uniform in the jet and is estimated as @xmath300 where @xmath301 is the mass of the central black hole . indeed , asymptotically , the magnetic field at the edge of the jet is dominated by its toroidal component ; therefore the product @xmath299 is governed by the current generated by the central engine along the axis . it slightly decreases along the jet , because the return current progressively establishes through wrapped lines off axis like butterfly wings @xcite . these jets are launched if the magnetic field intensity is close to equipartition with the radiation pressure in the central region , _ i.e. _ , within @xmath302 gravitational radii . this corresponds to @xmath303 within @xmath304 . thus the performance of the jets as uhe cosmic rays accelerators tightly depends on the nature of the central engine . at the base of the jet , @xmath305 mg for @xmath306pc is a reasonable number . in the hot spots of the fr2 jets such as those of cygnus a , @xmath307 g for a region of size @xmath308kpc . thus with @xmath309 , the confinement condition in jets rules out the possibility of generating ultra - high energy cosmic rays of energies larger than @xmath310ev . in the case of fr1 jets , which are less powerful , less collimated , and without hot spots , the limiting energy is even smaller since the product @xmath299 , although not well known , is very likely lower . as usual , the escape of the highest energy cosmic rays is governed by diffusion across the jet and @xmath311 where we have use our previous result that perpendicular diffusion is governed by magnetic diffusion of the field line , and @xmath312 . one thus finds that indeed most high energy cosmic rays escape before reaching the end of the jet , since @xmath313 to be compared with a travel time of @xmath277myr to travel @xmath314kpc , the typical length of extragalactic jets . here as well the maximal energy for fermi acceleration is obtained by equating @xmath315 with the acceleration timescale for acceleration in shocks moving at speed @xmath316 . this gives @xmath317 with the plausible assumption of kolmogorov turbulence , @xmath318 , we finally obtain the maximal energy as a fraction @xmath319 of the confinement energy , the estimate being weakly sensitive to the turbulence level : @xmath320 centaurus a is a well - known example of active galactic nucleus , actually the closest to us ( distance 3.4 mpc ) , which displays fr1 non relativistic jets moving at speeds @xmath321km / s . the jets have both radio and x - ray band synchrotron emission with luminosity @xmath322erg / s extending over several kpc . they have been studied in detail with high resolution interferometry @xcite and recently with the chandra x - ray satellite @xcite . the radio knots and the x - ray knots are identical in the inner jet . the minimum pressure magnetic field is @xmath323 g and the maximum lorentz factor of the electrons @xmath324 . the inner jet has radius @xmath325pc and a constant opening angle of @xmath326 . the product @xmath327g.pc is clearly too low to produce ultra - high energy cosmic rays . in any case it has been shown that even if centaurus a could accelerate cosmic rays to the highest energies observed @xmath328ev , their transport to earth , affected by diffusion according to the rules derived in this paper would lead to strong energy losses by increased travel distance and anisotropy incompatible with present observations @xcite . in the hot spots , the escape is again dominated by diffusion at high energies , but parallel diffusion is more likely unless there is no ordered field . for a hot spot of size @xmath329kpc and magnetic field intensity @xmath330 g as in cygnus a , the confinement limit is @xmath331 and the maximum energy achievable with a non - relativistic shock is @xmath332 the most extreme energy that can be obtained is when the turbulence is high enough that no organised field is set up in the hot spot , and the shock is midly relativistic @xmath333 . but synchrotron emission of hot spots , like those of cygnus a , does not favor this view . indeed the synchrotron emission by relativistic electrons cuts off in the infrared range . since an electron of lorentz factor @xmath261 synchrotron radiates around a frequency @xmath334hz , an observational upper bound on the electron lorentz factor is @xmath335 and the corresponding rigidity is @xmath336 now this maximum electron energy is obtained by the same fermi acceleration process , limited by synchrotron losses . we remind that the characteristic time for a synchrotron radiative process of an electron of energy @xmath337 is @xmath338 where @xmath339 is the thomson cross - section and @xmath340 the magnetic permitivitty of vacuum . by equating this loss time with the first - order fermi acceleration timescale , we obtain the maximum lorentz factor that can be achieved @xmath341 and the scattering function is calculated for the maximum rigidity @xmath342 . therefore , assuming again kolmogorov turbulence with @xmath343 , we obtain the approximate value of the turbulence level by equating the two expressions for @xmath344 : @xmath345 since @xmath346 is very likely , the required turbulence level is rather low . this , in turn , reduces drastically the maximum energy of cosmic ray acceleration , using eq . ( [ eq : emhs ] ) , eq . ( [ eq : eclhs ] ) and knowing that @xmath347 : @xmath348 where we explicited the scaling with @xmath21 the exponent of the power spectrum of magnetic fluctuations ; however note that the estimate of @xmath60 must be changed with @xmath21 . for kolmogorov turbulence , @xmath22 and using the upper bound on @xmath60 eq . ( [ eq : turl ] ) above , the prefactor is of order @xmath349 , and acceleration is not sufficient to account for the highest energy cosmic rays by several orders of magnitude . this limit can not be circumvented easily , since it is severely constrained by the cut - off frequency of the electrons synchrotron emission . the only parameter that could be modified without affecting this cut - off frequency is the turbulence index @xmath21 . if one considers kraichnan turbulence @xmath350 instead of kolmogorov turbulence , the prefactor @xmath351 is changed into @xmath352 , but @xmath60 itself is lowered by a factor @xmath302 due to the modified dependence of @xmath58 on @xmath342 . furthermore if hot spots were to radiate synchrotron emission in x - rays , this would increase @xmath353 by a factor @xmath302 only , and would not affect drastically our conclusions . finally the numbers considered above are consistent with recent observations of the cygnus a hot spots by chandra @xcite , which give an accurate measurement of the magnetic field intensity @xmath354 g to within a few tens of percents , as obtained by the ratio of the synchrotron - self compton luminosity over the synchrotron luminosity , a value which is furthermore close to the equipartition value if there are no protons ! these measurements also confirm model dependent estimates proposed in 1986 @xcite . finally , as an aside , the same reasoning allows to estimate the level of turbulence required to get the x - ray emission in centaurus a : with @xmath355 , @xmath356 , and @xmath357 . let us first summarize the results we have obtained . the scattering function @xmath58 has been found to follow the scaling predicted by quasi - linear theory in the inertial range @xmath358 for weak to strong turbulence . however we found that scattering still operates for @xmath359 , contrary to the predicted sudden drop of the scattering function ; this facilitates the injection of particles in fermi processes . for larmor radii larger than the correlation length @xmath166 , scattering decreases as a power - law in rigidity unlike the predicted sudden drop of @xmath58 . therefore high rigidity particles still diffuse . one should also mention that the lack of scattering encountered in weak turbulence theory for particles having pitch angle close to @xmath360 is cured in strong enough turbulence . the perpendicular diffusion turns out to be very different from the prediction of the quasi - linear theory . our investigation of the chaos of magnetic field lines characterised by a kolmogorov length and a diffusion coefficient with space increment indicates that this process of magnetic diffusion governs the transverse diffusion of particles . our numerical experiment shows that the phenomenological bohm approximation , characterized here by @xmath361 and @xmath362 with @xmath363 , only applies in a limited range of rigidities @xmath364 , and only in the case of pure turbulence @xmath146 . many estimates in astroparticle physics , that rely on the bohm conjecture , must be reconsidered . the slow decrease @xmath365 of scattering for cosmic rays with larmor radius larger than the correlation length of the magnetic field , which implies @xmath366 , is of potential importance to the transport of high energy cosmic rays in our galaxy as well as ultra - high energy cosmic rays in the intergalactic medium . the accurate knowledge of the transport coefficients allows to be more conclusive than before on the performances of fermi acceleration in some astronomical sources of high energy cosmic rays such as supernovae remnants , super - bubbles and extragalactic jets . using new chandra data , the turbulence level and the maximum energy for electrons and for cosmic rays can be determined . we confirm the difficulty to obtain energies larger than @xmath367ev in supernovae remnants and shows that the `` knee '' range of the cosmic ray spectrum could be accounted for by second order fermi acceleration in super bubbles . we also confirm that fr1 jets , such as centaurus a , although radiating synchrotron in x - rays , can not produce uhe cosmic rays . on the contrary , fr2 jets can produce cosmic rays up to @xmath368ev , but presumably not more , owing to a fairly good confinement ; however most high energy cosmic rays escape before reaching the end of the jet . hot spots of powerful radio - galaxies have always been considered as a promising source , but we have found that , because of the low turbulence level implied by the synchrotron cut off frequency , cosmic rays escape rapidly along the mean field lines by fast parallel diffusion and acceleration is not effective above @xmath369ev for a shock velocity @xmath370 . our paper left opened several important issues that we are currently investigating . in particular it seems crucial to investigate in more detail the existence of subdiffusive regimes at low ridigities for which we have found evidence . these regimes play a crucial role in the acceleration processes at perpendicular shocks @xcite . furthermore , we have described magnetic turbulence as an ensemble of magneto - static modes distributed according to a power law spectrum . this approximation is justified by the small alfvn velocity when compared to the velocity of the particles . however it would be interesting to investigate the effect of temporal and spacial intermittency on the transport properties . finally we are currently investigating the transport properties of particles in non - isotropic turbulence as may be encountered in the vicinity of a shock wave , in particular in the downstream medium . the consequences on fermi acceleration will be presented in a forthcoming paper . the diffusion resulting from the random variations of the momentum due to the irregular magnetic field can be formalized as follows . energy conservation , and thus @xmath15 conservation , allows to treat the problem as random rotations of the unit vector @xmath371 such that @xmath372 : @xmath373 . assuming that the correlation functions of the components of @xmath371 are integrable over a caracteristic time @xmath45 , the diffusion coefficients are given by @xmath374 the correlation matrix is derived by making the appropriate average , after solving the stochastic equation : @xmath375 where the gyro - matrix @xmath376 , @xmath377 being the reduced components of the magnetic field experienced by the wandering particle and @xmath378 is a @xmath379 matrix , with components @xmath380 , where @xmath381 is the fully anti - symmetric levi - civita tensor , and @xmath382 . note that the @xmath383 are generators of a lie algebra such that @xmath384 where @xmath385 represents the identity matrix , @xmath386 and @xmath387 , where @xmath388 is the orthonormal basis , @xmath389 the orthogonal projector over the plane transverse to the direction @xmath41 . we have @xmath390 , @xmath391 and @xmath392 , with @xmath393 and @xmath394 . moreover @xmath395 $ ] , with @xmath396 . the time variable is measured in larmor time units , the space variables are reduced to @xmath23 and wave numbers are accordingly dimensionless and varies from @xmath100 to @xmath397 , where @xmath398 . we make the three following assumptions : * the random process becomes stationary beyond the correlation time @xmath45 . * the random process can be approximated by a markoff process beyond the integration time @xmath45 . * the random process @xmath399 is supposed to be specified . for instance , it is gaussian with a known correlation function @xmath400 . assumptions i ) and ii ) allow to calculate the correlation function with an average matrix that describes the relaxation of the correlations : @xmath401 where @xmath402 . assumption iii ) is not exact , of course ; however the numerical experiments provide correlation functions that allow to get a good `` guess '' . then the theoretical method allows to calculate the solution through iterations , starting with a gaussian approximation , and then estimating , if necessary , non - gaussian corrections , given a skewness factor . the formal solution reads : @xmath403\right\rangle \ , \label{eq : a4a}\ ] ] where the symbol @xmath5 represents the `` time ordering operator '' that organises the expansion of the exponential operator in products of non commutative operators that are in chronological order . the result can be factorized as the product of the unperturbed rotation in the mean field times some relaxation operator : @xmath404\right\rangle \ , \label{eq : a5}\ ] ] with @xmath405 ( see @xcite for technical details ) . for @xmath60 small and for a broad magnetic spectrum insuring a short correlation time of the random force compared to the scattering time , the quasi linear theory applies @xcite . this allows to make two approximations . first , the relaxation operator can be calculated to the lowest order , the so - called `` bouret approximation '' @xcite , which corresponds to a summation of all the `` unconnected diagrams '' : @xmath406 with @xmath407 the simplest way to derive this result is to linearize the evolution equation for @xmath408 : @xmath409 which one then solves to lowest order for @xmath410 as a function of @xmath411 and inserts the result in the evolution equation for @xmath412 . then we use the isotropy of the spectrum to obtain @xmath413 the sum of operators can be simplified to : @xmath414 where @xmath415 is the reduced larmor pulsation in the mean field , thus @xmath416 . second , the correlation function of the magnetic irregularities experienced by the particles is calculated with unperturbed trajectories . @xmath417 where @xmath418 , and @xmath419 , as the notation indicates is the three - dimensional power spectrum in fourier space . these two calculations , thanks to commutation properties , lead to a matrix of the form @xmath420 \label{eq : ab}\ ] ] the factors @xmath58 are small numbers of order @xmath60 that contrains the usual resonances of the quasi linear theory in @xmath421 . these resonances come from the cosine and sine factors in eq . ( [ eq : aa8 ] ) and from the expansion into a fourier sequence in @xmath422 of the exponential involved in @xmath423 , see eq . ( [ eq : a9 ] ) , which introduces bessel functions of all orders . however , in pratice , only the main resonances for @xmath424 are retained because the higher resonances involve shorter and shorter wavelengths which contain less and less energy for usual power law spectra . the contribution in @xmath425 modifies the gyro - pulsation in the rotation matrix @xmath426 and therefore is unimportant . we finally retain the following result : @xmath427 where @xmath428 is the product of the rotation and the transverse projector . we thus obtain the unexpected result that the transverse relaxation is longer than the parallel relaxation since @xmath429 . then the correlation functions are obtained by averaging over @xmath430 and there comes a major problem of quasi - linear theory because the functions @xmath58 are proportional to @xmath431 for @xmath432 and vanish for @xmath433 because of the lack of resonance . this introduce long tail contributions to the correlation functions . this is the symptom of the `` sticky '' regime for pitch angles close to @xmath360 that tends to dominate the diffusion coefficients ; which requires to take into account mirroring effects and/or overlapping of the resonances for @xmath434 close to @xmath435 and those for @xmath436 close to @xmath437 , as suggested in @xcite . this difficulty disappears in strong turbulence and for large enough larmor radii . the fully deductive theory of this regime is quite difficult . however some attempt can be proposed in the case where @xmath438 , i.e. @xmath439 , when the correlation time [ decay time of @xmath440 is shorter than the scattering time . thus , for a time longer than the correlation time , we can keep part of the quasi - linear theory , namely the expression of the relaxation operator involving the integral over @xmath441 . technically , this corresponds to the summation of the unconnected diagrams of the expansion of @xmath442 in eq . ( [ eq : a4a ] ) , the other diagrams ( `` nested '' and `` crossed '' ) being of smaller orders . therefore the correlation functions @xmath443 asymptotically decay like @xmath444 and @xmath445 now the main difference comes from the estimation of the correlation function of the field experienced by the particles : @xmath446 we propose the following heuristic estimate . because the particles follow the field lines when their larmor radius is smaller than the wavelength of the modes , we consider only the modes such that @xmath447 . the dominant contribution in the averaged exponential is then for short time , @xmath448 . because of the random distribution of @xmath430 over the unitary sphere , we get @xmath449 a similar result is obtained with a gaussian evaluation of the average @xmath450 \ . \label{eq : a14}\ ] ] inserted into the integral over the spectrum ( restricted to @xmath451 ) , it leads to : @xmath452 \ . \label{eq : a15}\ ] ] in the integral , @xmath453 can be approximated by @xmath150 for @xmath454 ( @xmath455 in reduced units ) and by @xmath456 for @xmath457 . therefore @xmath458 \ , \label{eq : a16}\ ] ] where @xmath459 . when @xmath58 is small , because @xmath460 , we get a simple result close to the previous one : @xmath461 this gaussian evaluation indicates the error made by the previous assumption . thus , for small @xmath59 , we obtain the extension of the quasi - linear result , namely @xmath462 , and for @xmath463 to @xmath464 . these two approximations are in agreement with the numerical experiments , except that the measured drop is in @xmath465 instead of @xmath466 here . now the range of @xmath59 values where @xmath58 is on the order of unity corresponds to the `` bohm estimate '' , which is , in fact , the maximum value of @xmath58 achieved for @xmath467 only . we thus propose the following final estimate for the scattering function : @xmath468 the error on the coefficient is of order ten percent .

the massive clump g10.6 - 0.4 is an ob cluster forming region , in which multiple uc hii regions have been identified . in the present study , we report arcsecond resolution observations of the cs ( 10 ) transition , the nh ( 3,3 ) main hyperfine inversion transition , the choh j=5 transitions , and the centimeter free free continuum emissions in this region . the comparisons of the molecular line emissions with the free free continuum emissions reveal a 0.5 pc scale massive molecular envelope which is being partially dispersed by the dynamically expanding bipolar ionized cavity . the massive envelope is rotationally flattened and has an enhanced molecular density in the mid plane . in the center of this massive clump lies a compact (.1 pc ) hot ( k ) toroid , in which a cluster of o type stars has formed . this overall geometry is analogous to the standard core collapse picture in the low mass star forming region , with a central ( proto)stellar object , a disk , an envelope , and a bipolar outflow and outflow cavity . however , g10.6 - 0.4 has a much larger physical size scale (.1 pc for typical low mass star forming core ) . based on the observations , we propose a schematic picture of the ob cluster forming region , which incorporates the various physical mechanisms . this model will be tested with the observations of other embedded ob clusters , and with numerical simulations .

1101.3459

a m@xmath210 m@xmath4 massive star has a short kelvin helmholtz ( cooling ) timescale , and starts the nuclear burning while it is still deeply embedded in the dense molecular clump . during this earliest evolutionary stage , the massive stars can continue to increase their stellar mass via accretion of more dense molecular gas ( kahn 1974 ; garay and lizano 1999 ) . meanwhile , these massive stars produce large amounts of strong ionizing photons , and create the * ultracompact ( uc ) * * * h**ii * regions*. the uv radiation ( krumholz , mckee , klein 2005 ; peters et al . 2010 ) , the pressure of the ionized gas ( keto 2002b ; krumholz et al . 2009 ; peters et al . 2010 ) , and the strong stellar wind directly interact with the surrounding molecular gas . in addition , as massive stars typically form in clusters , their environments may be also influenced by their interactions both radiatively and dynamically . these interactions potentially can disperse the accretion flow , unless the system is confined or trapped by the high density molecular gas with the right physical profiles ( example solutions see keto 2002b , 2003 , 2007 ; also see galvn - madrid et al . 2008 for an observational evidence of accretion ) . this is the greatest contrast to the formation of the low mass stars , where accretion only occurs before the pre main sequence phase when the surrounding gas is neutral . to understand how massive stars finally attain a high stellar mass ( e.g. @xmath520 m@xmath4 ) , it is necessary to understand the accretion after the formation of an uc hii . _ one key factor in determining the relative importance between these feedback mechanisms as well as gravity , and hence the subsequent evolution of the dense molecular core / envelope , is the overall geometry / morphology of the entire system . _ for example , the density distribution of the molecular gas around massive stars determines how the molecular structures are self shielded from the ionizing photons ; and the overall geometry of the system determines how the radiation and the ionized photons can leak out of the system . owing to their typical large distances ( a few kpc ) , the molecular structures around the uc hii regions , are not well resolved in previous observations . how the accretion flows around individual stars and in the entire cluster is unknown . hence how the ob stars gain mass via accretion after nuclear burning begins , is also unknown . we perform new interferometric observations toward a well studied o type cluster forming region g10.60.4 , in order to resolve the molecular structures . the high column density and high brightness temperature of this source allow us to follow the detailed structures of the molecular gas over an extended region , at arcsecond resolutions . here we emphasize that the subject under consideration is the molecular and the ionized gas flow around a cluster of ob stars , rather than the accretion around a single massive ( proto)stellar object . this is due to the still insufficient spatial resolution . the uc hii region complex g10.6 - 0.4 is at a distance of 6 kpc ( caswell et al . 1975 ; downes et al . 1980 ) , and is still deeply embedded in a dense molecular core / envelope . the brightest uc hii region is extremely luminous ( integrated flux is 2.6 jy at 23 ghz ) ( ho & haschick 1981 ; sollins et al . 2005 ; sollins & ho 2005 ) , and contains an ob cluster within a 0.05 pc region . detections of water and oh masers ( genzel & downes 1977 ; ho & haschick 1981 ; ho et al . 1983 ; hofner & churchwell 1996 ; fish et al . 2005 ) indicate that the formation of massive stars is still ongoing . the early nh@xmath0 , cs and c@xmath6o line observations ( ho & haschick 1986 ; keto , ho , & haschick 1987 ; keto , ho , & haschick 1988 ; omodaka et al . 1992 ; ho , terebey , & turner 1994 ) unveiled a 0.5 pc scale molecular envelope . these observations also suggested that the dynamics of the majority of the molecular gas have not yet been strongly disturbed by the ysos or stellar activities . at the 0.5 pc scale , the general motion of the molecular gas appears to be dominated by gravity , showing rotation along a flattened geometry ( liu et al . 2010 ) . the inner part at the 0.1 pc scale appears to rotate faster while infalling towards the uc hii region ( keto , ho , & haschick 1987 ; keto , ho , & haschick 1988 ; liu et al . 2010 ; beltrn et al . recombination line studies further suggest that the accretion flow continues across the hii boundary , while evidence of outflow can be seen in the motions and structures of the hii region at the sub arcsecond angular scale ( keto 2002 ; keto & wood 2006 ) . structures found in projection against the uc hii continuum emission from nh@xmath0 optical depth studies , suggest accretion via a very clumpy and geometrically thick rotating core at the 0.1 pc scale ( sollins & ho 2005 ) . in this paper , we present the new arcsecond resolution ( 1@xmath7 corresponds to 0.03 pc at the distance of g10.6 - 0.4 ) molecular line ( ch@xmath0oh , nh@xmath0 , cs ) data , together with very sensitive and 0@xmath7.5 resolution 3.6 centimeter continuum observations . the linear scale of the synthesized beams is smaller than the natural scale for self - gravity ( jeans length @xmath80.1 pc for temperature @xmath82030 k and h@xmath9 number density @xmath810@xmath10 @xmath11 ) . we see an unprecedented detailed morphology of the massive accretion flow which clarifies the geometrical relation between the hot core and the envelope . with the aid of the new results and the discussions on this target source in previous publications , a more thorough understanding is achieved . in massive star forming regions , the nh@xmath0 inversion transitions are usually regarded as reliable molecular tracers for 20100 k gas . the main hyperfine components of these nh@xmath0 lines are optically thick and may mask the detailed density structures . the satellite hyperfine components are more optically thin ( see ho & townes 1983 for a review ) . however , in most cases , the satellite hyperfine components of the nh@xmath0 ( j , k ) = ( 1,1 ) inversion transitions are blended with the main hyperfine component ; the satellite hyperfine components of the higher ( j , k ) level transitions are much weaker and are difficult to be observed with high resolutions . with a factor of @xmath8100 higher einstein a coefficient , the selected ch@xmath0oh transitions potentially allow us to probe the denser regions . we simultaneously observed multiple ch@xmath0oh transitions which have comparable einstein a coefficients ( and therefore critical densities ) , but trace a broad range of upper level energy ( e@xmath12 ) . these observations allow us to distinguish the hottest molecular gas directly associated with the uc hii region from the cooler envelope . the low e@xmath12 and the modest critical density of the cs ( 1 - 0 ) transition provide sufficient optical depth , which allows us to trace the molecular gas to a much more extended region . however , in the case of g10.6 - 0.4 , significant local structures can still be identified without being severely affected by the sidelobes of the dirty beam and the missing flux . the high optical depth of cs ( 1 - 0 ) also provides better signal to noise ratio to the geometrically thinner expansional signatures , and provides information for the on going feedback processes . the observational results are presented in section [ chap_result ] . in section [ chap_discussion ] , we discuss the physical implication of the overall geometrical picture of the region . in addition , we estimate the feedback from various physical processes ( radiative pressure , ionized gas pressure , stellar wind ) and compare with our observational measurements . a brief summary of the results are provided in section [ chap_summary ] . we observed the ch@xmath0oh j=5 transitions in the 230 ghz band using the submillimeter array ( sma ) ; we observed the nh@xmath0 ( 3,3 ) hyperfine inversion transitions in the k - band using the nrao ( expanded ) very large array ( vla / evla ) in the c configuration ; and we observed the cs ( 1 - 0 ) transition using the vla / evla in the dnc configuration . for more on the sma and its specifications , see ho , moran & lo ( 2004 ) . the size of the synthesized beams in these observations are comparable . continuum emissions are averaged from the line - free channels and then subtracted from the line data . the properties of the selected molecular transitions are listed in table 1 ; the instrumental parameters are summarized in table 2 ; the observational settings are provided in table 3 . the ch@xmath0oh 5(-2,4)4(-2,3 ) e and ch@xmath0oh 5 ( 2,3)4(2,2 ) e lines are closely blended . the frequency separation of these two lines is only marginally resolved . considering that these two lines trace similar physical conditions , and the line - widths are broad in the observed region , we regard the blended lines as one single line in this research , and denote it henceforth as ch@xmath0oh 5(2,3)4(2,2 ) . basic calibrations are carried out in the ` mir ` package for the sma data ; and are carried out in the ` aips ` package for the the vla / evla data . all data are additionally self calibrated in the ` aips ` package . imaging is also carried out in the ` aips ` package . we observed the x band continuum emission toward g10.6 - 0.4 in the vla a configuration including the vlba pie town antenna on 2005 january 2 , and in the vla / evla c configuration on 2009 july 27 . we observed 1331 + 305 , 0319 + 415 , and 1820 - 254 as absolute flux , passband , and gain calibrators . the basic calibrations , self - calibration , and imaging of these data are carried out in the ` aips ` package . we combine the a array+pie town visibility data with the c - array visibility data , yielding a 0@xmath7.5@xmath130@xmath7.4 synthesized beam with a position angle of -43.2@xmath14 . the observed rms noise of the 3.6 cm continuum image is about 0.24 mjy / beam ( @xmath821 k in brightness temperature ) , and the peak brightness temperature is about 8700 k. we note that the 3.6 cm continuum emission is free free emission based on spectral index measurements . hence its high brightness temperature indicates that the emission is moderately optically thick . the observing results are separately presented in two sections . in section [ chap_geometry ] , we focus on discussions of the warm and dense molecular gas , mainly traced by the nh@xmath0 and the ch@xmath0oh emissions . in section [ chap_dynamics ] , we examine the dynamics of the more extended components , based on the observations of the cs ( 10 ) transition . cross comparisons of the molecular lines have been omitted , when there are no additional significant physical implications . figure [ fig_mnt0 ] shows the velocity integrated maps of the ch@xmath0oh 5(0,5)4(0,4 ) a+ transition , the blended ch@xmath0oh 5(-2,4)4(-2,3 ) e and ch@xmath0oh 5(2,3)4(2,2 ) e transitions , and the ch@xmath0oh 5(-3,3)4(-3,2 ) e transition . in figure [ fig_nh3 ] , we present these three images in rgb colors , and an overlay with the velocity integrated map of the nh@xmath0 ( 3,3 ) main hyperfine line . in the projected area of the uc hii region , the nh@xmath0 ( 3,3 ) main hyperfine line is detected in strong absorption ( see keto , ho & haschick 1987 ; keto , ho , & haschick 1988 ; sollins et al . 2005 ; liu et al . 2010 for more descriptions ) and is presented with much larger contour intervals . we see that : * the highest excitation ch@xmath0oh transition ( blue ) exclusively traces a compact ( @xmath80.1 pc ) flattened hot toroid in the center ( @xmath8300k , klaassen et al . 2009 ; @xmath887 k , beltrn et al . the lower excitation ch@xmath0oh transitions trace the warm ( @xmath850 k ; keto , ho & haschick 1987 ) and dense molecular gas to an extended ( @xmath80.5 pc ) region . the ratio of the velocity integrated flux of the ch@xmath0oh 5(0,5)4(0,4 ) a+ transition to that of the ch@xmath0oh 5(-3,3)4(-3,2 ) e transitions in the hot toroid region is about 0.56 . in the optically thin limit , this line ratio reflects an averaged excitation temperature of @xmath8500 k. if the lower excitation line is saturated owing to high optical depth , the excitation temperature would be overestimated . * we see the complicated structures from the extended ch@xmath0oh emissions . the majority of the ch@xmath0oh emission has a southeast northwest distribution . two low column density cavities are clearly seen northeast and southwest of the hot toroid . we note that the size scale of the region where we detect significant emissions is small as compared with the angular size scale of the sma primary beam ( @xmath81@xmath15 ) and that of the vla primary beam ( @xmath82@xmath15 ) . since the detected structures are well centered within the primary beams , they are not affected by the primary beam attenuation . we therefore do not perform primary beam corrections to the maps . the shortest baselines of the ch@xmath0oh and the nh@xmath0 data sets ( table [ table_molecule_list2 ] ) correspond to the maximum angular size scale of @xmath835@xmath7 and @xmath890@xmath7 , respectively . hence the absence of structures with angular size scales @xmath520@xmath7 suggests that the compact appearance of the emissions is physical in nature , and is not caused by sampling defects . from figure [ fig_nh3 ] , we see that most structures traced by the ch@xmath0oh emissions have their nh@xmath0 counterparts ; however , the nh@xmath0 emission shows more significantly detected structures than the ch@xmath0oh emissions . a similar effect appears in the velocity domain . figure [ fig_ch3ohpv ] shows the position - velocity ( pv ) diagrams of the nh@xmath0 ( 3,3 ) main hyperfine line and all selected ch@xmath0oh j=5 transitions . the pv cut is centered on the brightest free - free continuum source at the coordinates of r.a . = 18@xmath1610@xmath1728.64@xmath18 and decl = -19@xmath1455@xmath1549.22@xmath7 with a position angle pa = 140@xmath14 ( the positive angle is defined in the usual counterclockwise fashion ) . from these pv diagrams , we see that : * the ch@xmath0oh 5(-3,3)4(-3,2 ) e transition has the highest upper level energy e@xmath12 of 96.9 k. the pv diagram of this transition reveals the fast rotating hot toroid . * the ch@xmath0oh transitions with lower upper level energies trace the molecular gas in a more extended region . a significant velocity gradient is also seen in the extended emission . * from the left most panel of figure [ fig_ch3ohpv ] , we see that the dominant emission of the nh@xmath0 ( 3,3 ) main hyperfine inversion line shows a good agreement with the ch@xmath0oh emissions . the nh@xmath0 ( 3,3 ) absorption line consistently traces the fast rotating hot toroid as seen in the pv diagram of the ch@xmath0oh 5(-3,3)4(-3,2 ) e transition . additionally , the nh@xmath0 ( 3,3 ) emission shows significant detections of broad line emission features ( e.g. at 7@xmath7 and 11@xmath7 ) and some diffuse broad line emission ( e.g. around -3@xmath7 -11@xmath7 ) . the nh@xmath0 molecule has a stable abundance over a large range of densities and temperature without depletion onto grains . hence its emission lines can be used to trace the dynamics in the bulk of the dense molecular accretion flow . the detected velocity gradients can be explained by the marginally centrifugally supported motion of the molecular gas , in response to the gravity of the enclosed molecular and the stellar mass ( see liu et al . 2010 for more discussions ) . the projected area and the rotational motions traced by the ch@xmath0oh are highly consistent with those which are traced by the nh@xmath0 emissions . this suggests that these two molecules are tracing dense gas in the same region . nevertheless , there are differences in their detailed brightness distributions which suggest that in this particular source , the ch@xmath0oh emissions are tracing different excitation / chemical conditions ( e.g. temperature ) , in the same flow traced by nh@xmath0 ( see also the discussions for the ch@xmath0oh emissions in liu , ho & zhang 2010 ) . in addition , the abundance of the ch@xmath0oh molecule may also be affected by outflow . the nh@xmath0 ( 3,3 ) main hyperfine inversion transition has comparable upper level energy ( e@xmath12/k = 124.5 k ) with the highest excitation ch@xmath0oh 5(-3,3)4(-3,2 ) e transition ( e@xmath12/k = 96.9 k ) , but in contrast it shows an extended distribution . one explanation is that the nh@xmath0 molecule has a higher abundance ( [ nh@xmath0]/[h@xmath9 ] = 10@xmath21 ; keto , ho & haschick 1987 ) than the ch@xmath0oh molecule ( [ ch@xmath0oh]/[h@xmath9 ] = 2@xmath2210@xmath23 ; takakuwa , ohashi , & hirano 2003 ) , and thus becomes optically thick and shows strong emission . the excellent match between the nh@xmath0 emission and the lowest excitation ch@xmath0oh emission suggests that the emissions of both molecules trace the real physical structures . we know that for the dense gas with temperature around 50 k , the statistical weight compensates the boltzmann factor , and therefore the population of nh@xmath0 at the ( 3,3 ) level can be comparable to the population at the lowest excitation ( 1,1 ) level . the ch@xmath0oh emissions and the nh@xmath0 ( 3,3 ) main hyperfine emission may both trace a population of molecular gas which is significantly warmer and denser than the ambient molecular gas . _ we call this warm ( @xmath2430 k ) and dense ( @xmath510@xmath10 @xmath11 ) region the massive envelope hereafter . _ the previous observations on c@xmath6o ( 21 ) unveiled a parsec scale filamentary structure in the south of the massive envelope ( ho , terebey , & turner 1994 ) . the distribution of the molecular mass on an even larger scale has to be further examined by sensitive single dish maps . the selected ch@xmath0oh transitions have higher einstein a coefficients than the nh@xmath0 ( 3,3 ) main hyperfine inversion transition , and potentially only trace the denser components . the observations in the clearest example of photon - dominated regions ( pdr ) , the orion bar , further suggest that the ch@xmath0oh emissions trace the clumpy materials , which have an h@xmath9 volume density higher than 10@xmath25 @xmath11 , and are self - shielded from photo ionization ( leurini et al . 2010 ) . the abundance of the ch@xmath0oh molecule can be greatly enhanced if the molecular gas is shocked by high velocity outflows with velocities greater than @xmath820 kms@xmath26 ( see discussions in takakuwa , ohashi , & hirano 2003 ) . we therefore can not rule out the possibilities that zones with locally enhanced ch@xmath0oh abundance / emission are distributed over the entire region . part of those extended nh@xmath0 emissions can also be attributed to the externally heated lower density surfaces of the extended structures . other possibilities will have to be investigated by higher resolution observations which define the detailed temperature and density profiles of individual local structures . in figure [ fig_ionout ] , we compare the distribution of the molecular gas traced by the ch@xmath0oh transitions with the distribution of the ionized gas traced by the 3.6 cm free free continuum emission . by comparing the left panel of figure [ fig_nh3 ] with this figure , we find that the 3.6 cm free - free continuum emissions peak at the hot toroid . the distribution of the extended free - free continuum emission fits into the projected area of the low column density cavities seen in the ch@xmath0oh rgb map . these observed features suggest an overall geometry of a core envelope cavity system . this overall geometry resembles the standard disk envelope system in the low mass ( @xmath31 m@xmath4 ) star forming region with bipolar outflow and outflow cavity , except on a larger and more massive scale . this overall geometry remains even after the embedded ob cluster starts the nuclear burning , and other physical processes start to affect the dynamical evolution ( section [ chap_introduction ] ) . we emphasize that the subject under consideration is the formation and accretion of a cluster of ob star . by comparing the flux of the ionizing photons and the bolometric luminosity , ho and haschick ( 1981 ) suggested that a few massive stars ( from 06.5 to b0 ) have already been formed and are embedded in the central 0.1 pc region ( see also the discussions in keto 2002 , sollins & ho 2005 , and keto & wood 2006 ) . this cluster of ob stars contains @xmath8200 m@xmath4 of stellar mass in total . as pointed out by sollins and ho ( 2005 ) , the source g10.6 - 0.4 stands distinct from the cases where a single massive star dominates at the center ( e.g. iras 20126 , iras 18089 , g192.16 , etc . zhang et al . 1998 ; beuther et al . 2004 ; shepherd et al . 2001 ) , which might have formed in a process similar to low mass stars ( keto & zhang 2010 ) . in g10.6 - 0.4 , a parsec scale massive envelope ( @xmath242500 m@xmath4 over the central 30@xmath7 region ; ho , terebey & turner 1994 ) appears to be undergoing a coherent motion in response to the gravity of the enclosed mass and the specific angular momentum . in this case , the mass of the embedded ob stars is a less significant fraction ( @xmath110% ) as compared to the molecular mass in the envelope . the fact that a concentration of ob stars formed in the small central 0.1 pc region suggests that the global contraction of the massive envelope may have been efficient . we note that in another ob cluster forming region g20.08 - 0.14n ( l@xmath86.6@xmath2210@xmath10 l@xmath4 , d@xmath812.3 kpc ) , the emissions of the nh@xmath0 molecules and various hot core tracers also show a similar geometry and kinematic profile with those in g10.6 - 0.4 ( galvn - madrid et al . 2009 ) . figures [ fig_cs ] and [ fig_csx ] show the velocity integrated ( moment 0 ) map of the cs ( 10 ) emission , the 1.3 mm continuum image , and the 3.6 cm continuum image . in the 3.6 cm continuum image ( figure [ fig_csx ] ) , we see three isolated uc hii regions ( from east to west , uc hii region a , b , c hereafter ) , of which we also detect their free - free continuum emission at 1.3 mm ( figure [ fig_cs ] , right panel ) . the visibilities of the cs ( 10 ) data cover a range of uv distances of 4245 @xmath27 , which correspond to @xmath852@xmath70@xmath7.8 in angular scale . the 1@xmath28 rms in the cs velocity integrated map is @xmath877 mjy / beam@xmath22kms@xmath26 ( 24 k@xmath22kms@xmath26 ) . the achieved brightness sensitivity is comparable to the theoretical rms noise ( also achieved in the channel maps ) . with the presence of the bright uc hii regions , the detection of cs ( 10 ) is locally biased by the continuum subtraction . structures distributed in front of the uc hii regions will be detected in emission line only if they emit brighter cs ( 1 - 0 ) emission than the local free - free continuum background ; otherwise , they will be detected in absorption line . we omit plotting the negative flux due to absorption to enhance the contrast of emission features in the gray scale . structures behind the uc hii region are not affected by the continuum subtraction and are always detected in emission line , unless the 49 ghz continuum emission is optically thick . in regions where the brightness temperature of the free - free continuum emission is comparable with the excitation temperature of the molecular gas , the absolute values of the flux density of the emission line and the absorption line are comparable . smoothed by the finite synthesized beam , the emission line and the absorption line signatures may cancel each other , and reduce the signal to noise ratio . from figures [ fig_cs ] and [ fig_csx ] we see : * the brightest regions of the cs ( 10 ) emission agrees well with the 1.3 mm continuum emission . ( the significant deficit of cs ( 10 ) emission in the middle of the maps is due to the subtraction of the bright free - free continuum in the uc hii region . ) * we see an extended dense flattened structure with a size scale of @xmath80.5 pc . the orientation is indicated by the red dashed line , which has a position angle of 135@xmath14 , and is consistent with the orientation of the central 0.1 pc hot toroid ( liu et al . * northeastern to the dense flattened structure , the extended cs emission shows filamentary structures . the size scale of those filamentary structures is a few times of 0.1 pc . southwestern to the dense flattened structure , we see an upside down * v * shaped cavity wall signature . the 1.3 mm continuum emission arises predominantly from the free - free continuum emission in the central 2@xmath73@xmath7 uc hii region ; in the extended region , the 1.3 mm continuum emission traces the thermal dust emission ( liu et al . 2010 ; liu , ho & zhang 2010 ) . the thermal dust emission is usually regarded as a reliable tracer of the molecular hydrogen . the general agreement between the cs ( 10 ) image and the 1.3 mm continuum image indicates that cs ( 10 ) also traces the molecular hydrogen . however , in the optically thick case , the velocity integrated flux of cs ( 10 ) can be biased by saturation and/or foreground absorption , and also by the non uniform excitation temperature . these explain some inconsistency between the cs ( 10 ) velocity integrated image and the 1.3 mm continuum image . by inspecting the position velocity diagrams ( section [ chap_pv ] ) , we do not see a gap of emission line , and therefore we argue that the self absorption does not severely affect the overall picture of geometry . the excitation temperature only changes the relative strength of emissions , and should not greatly alter the overall picture either , as long as the excitation does not change by a very large factor . the distribution of the cs ( 10 ) emissions is consistent with the overall geometry suggested by our observations of the ch@xmath0oh j=5 transitions . however , instead of seeing a clean biconical feature as in the ch@xmath0oh maps , we detect significant cs ( 1 - 0 ) emission in the projected area of the biconical cavity . this can be due to the fact that the cs ( 10 ) transition has a much lower upper level energy ( e@xmath12 = 2.35 k ) , and a factor of @xmath850 lower einstein a coefficient . these two factors make cs easily excited in the outer layer of the cavity wall with lower temperature ( and molecular gas density ) . this property allows us to robustly detect the dynamics of the cavity wall , and explicitly measure the expansional velocity . our cs ( 10 ) maps additionally unveil an unprecedentedly detailed morphology in the massive cluster forming clump . the velocity integrated map apparently shows abundant structures with a size scale of a few arcseconds ( 0.03 pc ) . in figures [ fig_chm78 ] , we present an example channel image at @xmath29@xmath30= -7.8 kms@xmath26 . in this channel , we do not detect compact ( 2@xmath73@xmath7 ) absorption features towards the brightest uc hii region a , which alleviates the potential confusion from the sidelobes of a strong source . we see that the arcsecond scale structures have a high contrast with respect to the background diffuse emission . we note that it is unlikely that our images can achieve the theoretical rms noise level if the extended diffuse emission has dominant contributions to the total flux . the small scale emissions may come from compact dense molecular cores , or from local structures with high temperature or high cs abundance owing to the interactions with the ( proto)stellar activities and shocks . the compact dense molecular cores are potentially the future or current sites of star formation . in figure [ fig_chm78 ] , we see that two significant local maxima of cs ( 10 ) emission are spatially associated with water maser sources ( water maser e , and the cluster of water maser sources around the uc hii region a ; hofner & churchwell 1996 ) . figure [ fig_pvwater ] shows the pv diagrams of three cuts through the water maser sources n , e , and w. from this figure we see detections of bright components near ( @xmath31@xmath7 ) the water maser sources e and w. the water maser source e may be associated with the local maxima of cs ( 10 ) emission at -7.8 kms@xmath26 , or -5.4 kms@xmath26 . we do not find bright local cs ( 10 ) emission peak around the water maser source n , which may be due to the fact that its parent molecular core is already dispersed by the outflows . recalling the small high opacity clumps detected in the nh@xmath0 ( 3,3 ) hyperfine inversion line absorption ( sollins and ho 2005 ) , it is not surprising that more of them are detected in emission line . however , our current data are insufficient to robustly distinguish the dense cores from the local shock and outflow signatures , which leads to the difficulties in a quantitative analysis of the core mass function . in the other paper on high velocity molecular outflows in this region , we will discuss the feedback from those local sites of star formation and the implications of clustering of stars ( liu , ho & zhang 2010 ) . we show the mean velocity map ( moment one ) and the velocity dispersion map ( moment two ) of the cs ( 10 ) emission in figure [ fig_mnt12 ] . from figure [ fig_mnt12 ] , we see that the entire system has a bimodal velocity profile . in addition to the general rotation with velocity gradient from the southeast to the northwest , we see blueshifted gas in the northeast and redshifted gas in the southwest . these additional blueshifted and redshifted gas cover a larger velocity range than the general rotational motion , and the orientation resembles a bipolar expansion with a certain inclination . geometrically , the upside down * v * shaped cavity wall signature in the southwest can also be explained by the bipolar expansion . we note that our 3@xmath28 detection limit is about 18 k , which is only sensitive to structures which are sufficiently warm and have high optical depth . the moment one and moment two maps thus can be biased due to the cutoff of the fainter high velocity line wings , and over interpretations should be omitted . to understand the dynamics of the cavity wall , we make two pv cuts through the southern part of the biconical cavity ( centered at * cut1 : * r.a . = 18@xmath1610@xmath1728.51@xmath18 , decl = -19@xmath1455@xmath1554.3@xmath7 , and * cut2 : * r.a . = 18@xmath1610@xmath1728.48@xmath18 , decl = -19@xmath1456@xmath151.0@xmath7 ) with a position angle 140@xmath14 ( consistent with the rotational plane of the central compact hot core pa=140@xmath14@xmath315@xmath14 ; liu et al . 2010 ) ; and we make one pv cut through the uc hii regions b , c , and the northern part of the biconical cavity ( centered at * cutn : * r.a . = 18@xmath1610@xmath1727.45@xmath18 , decl = -19@xmath1455@xmath1548.8@xmath7 , with position angle 70@xmath14 ) . features we identify from these three pv cuts are described as follows . owing to the complexity of the observed region and the limited signal to noise ratio of the observations , there could be other dynamical features not revealed in the current maps . [ [ cutn ] ] cutn + + + + from the pv diagram of the cutn ( figure [ fig_pv2 ] ) , we identify the expansional signature in the northern part of the biconical cavity . around the uc hii region b , we marginally see the arc shaped signatures which have broad velocity width , and may also be explained as expansional signatures around the uc hii region b. the expansional velocities in these two regions are about 3 kms@xmath26 . in these two regions , the expansional signatures are confused with the global rotation / contraction of the dense gas , and therefore the size and the expansional velocities are not robustly constrained . the detection of the expansional signature in the uc hii region c is marginal , with a suggested expansional velocity of 3.8 kms@xmath26 . [ [ cut1-and - cut2 ] ] cut1 and cut2 + + + + + + + + + + + + + both the cut1 and cut2 are located south to the dense flattened structure ; the center of cut2 is south of the center of cut1 . we argue that pv cuts in the position angle of 140@xmath14 is less confused by the dynamics of the fast rotating ( @xmath834 kms@xmath26 rotational velocity ) dense gas in the mid - plane of the envelope , and provides the clean information of the dynamics of the molecular cavity wall . the top and bottom panels of figure [ fig_pv ] show the pv diagram of cut1 and cut2 , respectively . the cs ( 10 ) emission in both panels clearly show arc shaped expansional signatures centered at v@xmath30@xmath8 - 3 kms@xmath26 , with an averaged brightness temperature of 2030 k , and the peak brightness temperature of 45 k. supposedly cs ( 10 ) is optically thick at the peak , then the brightness temperature represents the excitation temperature of the molecular gas . the maximum excitation temperature of 45 k in the projected area of the biconical cavity explains why the ch@xmath0oh transitions are difficult to be excited and have low optical depth . the projected linear scale of the expansional signature is about 10@xmath7 ( 0.3 pc ) , which is consistent with the scale of the visually identified biconical features in the cs ( 1 - 0 ) maps ( figures [ fig_cs ] , [ fig_csx ] ) , and the ch@xmath0oh maps ( figures [ fig_nh3 ] , [ fig_ionout ] ) . the arc shaped signatures in the pv diagrams have a different physical size scale and geometry from the typical outflow signatures ( liu , ho , & zhang 2010 ) . the arc shaped feature shown by the cut1 may be explained as the redshifted expansional motion of the cavity wall . since the center of cut1 is closer to the free - free continuum peak , its blueshifted counter part may be affected by the free - free continuum subtraction , and therefore we can not obtain significant detections ; or , it can be explained by the non - uniformity of excitation temperature and density distribution ( the 1@xmath28 detection limit is 5.9 k ) . the arc shaped feature shown by cut2 can be explained as the blueshifted expansional motion . the lack of the redshifted counter part of the expansional motion can either be explained by the asymmetry of molecular gas density , or asymmetrical stellar ionization . the line of sight expansional velocity of both features is about 2.55 kms@xmath26 . this interpretation can be constrained by observing the lower density tracers ( e.g. @xmath32co ( 2 - 1 ) or c@xmath6o ( 2 - 1 ) ) , to sample the lower density / temperature part of the cavity wall . the emissions of those lower density tracers are extended , and the single dish observations will be needed in order to robustly interpret the data . we summarize our series of high resolution observations on g10.60.4 by a schematic model , which is shown in figure [ fig_shematic ] . the observational results enable the discussions of the importance of the radiation and the pressure force of the ionized gas . in sections [ chap_radiation ] and [ chap_timescale ] , we provide the order of magnitude estimates of the radiative pressure and the pressure forces of the ionized gas and the stellar wind . we discuss their implications for the dynamical evolution of the system in section [ chap_dp ] . we are looking forward to comparing our results with numerical hydrodynamical simulations in the future . the dominant molecular structure in this model is a 0.5 pc scale massive envelope . the massive envelope rotates and contracts globally . a flattened overdensity formed in the mid plane of the massive envelope ( this is supported by the cs ( 10 ) observations , and also the @xmath32cs ( 54 ) observations in liu et al . meanwhile , the molecular gas also contracts locally , and forms intermediate mass or b type massive ( proto)stars , and some low mass ( proto)stars . the protostellar objects eject molecular outflows with high momentum and high energy into the ambient environment , which may affect the global contraction of the massive envelope ( li & nakamura 2006 ; nakamura & li 2007 ; carroll et al . 2009 ; wang et al . 2010 ) . those molecular outflows can be revealed as highly blueshifted and redshifted @xmath33co emissions ( @xmath34@xmath2420 kms@xmath26 ) if the protostellar objects are not located in the mid plane ( liu , ho , & zhang 2010 ) . in the mid plane , the outflows are impeded and decelerated by the high density ambient gas to have lower terminal velocities , which have to be diagnosed from the molecular outflow / shock tracers ( liu , ho , & zhang 2010 ) . the regulated global contraction may lead to the formation of the o type cluster in the center ( li & nakamura 2006 ) , which creates an uc hii region . the ionized gas and the ionizing photons leak out from the hii region mainly in the bipolar direction , either because of a configuration where the rotationally flattened system has a low gas density in the bipolar region , or because of a strong protostellar mhd wind which created the biconical cavities in that region . the outflowing molecular / ionized gas may flush through the ambient molecular gas on a timescale comparable to or much shorter than the global dynamical timescale10@xmath10 years . ] , and leave behind the elongated relics of interactions . this may explain the detected large filamentary structures by cs ( 10 ) in the northeast ( section [ chap_csdis ] ) . due to the short expansional timescale , the filamentary structures were not yet relaxed by the global rotational / infall motion . the pressure of the ionized gas can also push the cavity wall to a quick dynamical expansion ( section [ chap_pv ] ) . however , in the rotationally flattened system , most of the molecular mass are concentrated in the mid plane . owing to the self gravitational or the rayleigh taylor instabilities , the molecular gas becomes clumpy ( sollins & ho 2005 ) , and additionally reduces the effective solid angle as presented to the central stars . in such cases , the momentum impulse exerted by the ionized gas would be much smaller because of this smaller solid angle . the inertia of the molecular accretion flow in the mid plane can easily overcome the impulse exerted by the stellar radiative pressure ( also see section [ chap_discussion ] ) and the pressure of the ionized gas . this is consistent with the lack of a clear expansional motion in the pv diagram of the molecular lines ( figure [ fig_ch3ohpv ] ) . when the molecular gas gets sufficiently close to the central ob cluster , it is heated by the stellar radiation to a significantly higher temperature , and is further ionized to a temperature of 10@xmath35 k ( also see discussions in keto 2002 ; keto & wood 2006 ) . this overall scenario allows the o type stars to continue accreting after the nuclear burning begins . the biconical cavity provides a low opacity channel for the photons to leak out of the envelope . this mechanism can significantly reduce the radiative pressure on the molecular gas , and can therefore enhance the molecular accretion flow . to quantitatively examine this effect , we first compare the relative importance of radiative pressure and the gravitational force in a system without the biconical cavity . following the derivations in jijina & adams ( 1996 ) , we assume the radiation field is nearly isotropic ; and the stellar mass predominantly contributes to the gravitational force in the relevant scale . we assume the temperature distribution follows the power law @xmath36 ( keto , ho , & haschick 1987 ) , and quote the resulting effective potential from jijina & adams ( 1996 ) : @xmath37 where @xmath38 is the mass of the embedded stellar cluster . the parameter @xmath39 is defined by @xmath40 where @xmath41 is the bolometric luminosity of the embedded stellar cluster , @xmath42 is the planck mean opacity at the fiducial temperature @xmath43 , and @xmath44 is the radius where @xmath45 . the first term in equation [ eq1 ] represents the radiative pressure , and the second term represents the gravitational force . from equation [ eq1 ] , we see that when @xmath46 , the radiative pressure is comparably important with the gravity . the radiative pressure becomes dominant at smaller radii . in g10.6 - 0.4 , the molecular hot toroid with temperature about 300 k is detected inward of the 0.05 pc radius . to match the measured molecular gas temperature , we adopt , which is the dust sublimation temperature . if the temperature is higher than @xmath47 , the dust is sublimated and the radiative pressure is reduced . the temperature power law of @xmath36 potentially overestimates the temperature at the 0.3 pc radius by a factor of 2 since the temperature decay faster with radius in regions with lower opacity , and has to be fitted with a more negative power law index . however , our discussions about the radiative pressure focus on the inner region which has high opacity . ] @xmath48 2300 k ( @xmath49@xmath830 @xmath50g@xmath26 ) and @xmath51 8.5@xmath2210@xmath52 pc . given the g10.6 - 0.4 bolometric luminosity @xmath5310@xmath25 l@xmath4 and stellar mass @xmath54200 m@xmath4 , we obtain @xmath550.07 pc ( ho & haschick 1981 ; keto 2002 ; sollins & ho 2005 ; keto & wood 2006 ) . in such a case , the radiative pressure can significantly affect the dynamics at the scale of the hot toroid , or even reverse an inflow . however , based on the numerical radiation transfer calculations of a set of systems with different geometries , krumholz , mckee and klein ( 2005 ) suggested that the presence of the biconical outflow cavity allows the photons to leak out , and therefore potentially reduces the radiative pressure force by a factor of @xmath810 . this effectively reduce @xmath41 by a factor of 10 , and therefore reduce @xmath56 by a factor of 100 . assuming this conclusion can be generally applied , with the presence of the biconical cavity , the value of @xmath57 in the case of g10.60.4 might be effectively reduced to 7@xmath2210@xmath52 pc ( 140 au ) , which is much smaller than the radius of the uc hii region in g10.60.4 ( @xmath80.03 pc ) . qualitatively this estimate is consistent with the case in krumholz , mckee & klein ( 2005 ) , suggesting that without considering the ionization , the radiative pressure will only be important on the scale of a few hundred au . we therefore suggest that in g10.60.4 , the radiative pressure is unimportant for the dynamics of the molecular accretion flow . we note that we may overestimate the reduction factor of the radiative pressure owing to the non - self - consistent treatment of the temperature distribution . however , since @xmath57 scales as the square of the reduction factor , it can easily become smaller than the radius of the hii region with a small reduction factor . the presence of the biconical cavity can also lead to a temperature gradient steeper than our assumption , and may makes @xmath57 shrink to an even smaller radius . the dusty disk around the stars may redirect the radiation ( yorke & sonnhalter 2002 ; kuiper et al . 2010 ) , and potentially play the role of reducing the radiative pressure on the accretion flow . however , it is still uncertain whether the disks can stably exist in uc hii regions , where multiple o type stars are embedded . this has to be examined in future observations . the large molecular mass in the contracting massive envelope contains enormous inertia . from the pv diagrams ( figure [ fig_ch3ohpv ] ) , we find that the rotational velocity at the 0.25 pc radius is about 45 kms@xmath26 . this rotational motion is gravitationally bound by an enclosed mass of 10001600 m@xmath4 . while the embedded ob stars contribute @xmath8200 m@xmath4 , the rest of the binding mass is dominantly contributed by the molecular gas . for a spherical geometry , this molecular mass corresponds to a mean molecular density @xmath58 of ( 35)@xmath2210@xmath10 @xmath11 . the momentum flux in the molecular gas flow can be estimated by @xmath59@xmath22@xmath58@xmath22@xmath60 , where @xmath29 is the infall velocity of the molecular gas , which can be larger than 1 kms@xmath26 ( ho & haschick 1986 ; keto , ho & haschick 1987 , 1988 ; keto 1990 ; klaassen & wilson 2008 ) , and @xmath59 is the mean molecular weight . in the embedded uc hii region , the momentum flux in the ionized gas can be estimated by @xmath61@xmath22@xmath62@xmath22@xmath63 , where @xmath61 is the ion mass , @xmath62 is the ion density , and @xmath63 is the thermal velocity of the ions , which has the order of magnitude of 10 kms@xmath26 . we assume that the ion density @xmath62 is equal to the electron density @xmath64 , which is constrained to be 10@xmath6510@xmath35 in previous observations ( ho and haschick 1981 ; for a more sophisticated model fitting see keto , zhang , & kurtz 2008 ) . assuming @xmath59@xmath82@xmath61 , we find that the momentum flux in the molecular gas flow ( 610@xmath2210@xmath10@xmath22@xmath61 m@xmath4kms@xmath26@xmath66s@xmath26 ) and the momentum flux in the ionized gas ( 110@xmath2210@xmath10@xmath22@xmath61 m@xmath4kms@xmath26@xmath66s@xmath26 ) have the same order of magnitudes . this suggests that the two components are in an approximate dynamical equilibrium . in the real case , the mass is anisotropically distributed . the dynamics of the accretion flow in the denser region will not be significantly affected by the pressure of the ionized gas , while the lower density region can undergo a dynamical expansion powered by the pressure of the ionized gas . we estimate the momentum budget of the ionized gas and the stellar wind in this section . in the next section , we will suggest a simple geometric model for the density distribution , and estimate the velocity of the dynamical expansion , to compare with the observations . the embedded massive stars in uc hii regions b and c have similar spectral type ( ho & haschick 1981 ) with the embedded ob stars in uc hii region a. some of the energetic relations estimated in uc hii region a can also be applied in uc hii region b and c. [ [ expansional - time - scale ] ] expansional time scale + + + + + + + + + + + + + + + + + + + + + + the expansional signatures of the biconical cavity have a characteristic radius @xmath67 of about 5@xmath7 . assuming the fastest expanding front of the cavity wall uniformly expands with the terminal velocity @xmath68 of 5 kms@xmath26 , the total expanding time @xmath69 is about 3@xmath2210@xmath35 years , which is much shorter than the global dynamical timescale ( @xmath510@xmath10 years ) . the timescale estimation constrains the total momentum budget from the stellar wind and the ionized gas . [ [ the - stellar - wind - feedback ] ] the stellar wind feedback + + + + + + + + + + + + + + + + + + + + + + + + + the uc hii region a contains a few o6o9 stars with a total stellar mass m@xmath70 of @xmath8200 m@xmath4 . the stellar wind from each of these massive stars has a mass loss rate of the order of magnitude of 10@xmath71 m@xmath4yr@xmath26 , and terminal velocity of @xmath82000 kms@xmath26 ( tout et al . 1996 ) . given the number of the embedded o stars @xmath72 ( @xmath84 ; ho & haschick 1981 ) , the wind momentum feedback rate of the massive cluster @xmath73 is @xmath72@xmath222@xmath2210@xmath74 m@xmath4kms@xmath26yr@xmath26 . assuming the embedded stellar cluster starts to feed back the stellar wind by the beginning of the expansion of the biconical cavity , the total wind momentum budget @xmath75 is @xmath72@xmath2260 m@xmath4kms@xmath26 isotropically spread in the entire 4@xmath76 solid angle . the exerted pressure force has a @xmath77 radial dependence . [ [ the - ionized - gas - pressure ] ] the ionized gas pressure + + + + + + + + + + + + + + + + + + + + + + + + we adopt the characteristic radius @xmath78 of 0.15 pc , and adopt the electron density @xmath64 = 10@xmath6510@xmath35 @xmath11 according to the measurements in ho & haschick ( 1981 ) , to estimate the thermal pressure force of the ionized gas . assuming the ion density equals to the electron density , and has the thermal velocity of @xmath810 kms@xmath26 , the momentum feedback rate @xmath79 can be estimated by @xmath80 , where @xmath81 is the characteristic solid angle . as an order of magnitude estimation , we adopt @xmath81=4@xmath76 , which leads to @xmath79@xmath86@xmath2210@xmath526@xmath2210@xmath74 m@xmath4kms@xmath26yr@xmath26 . the total ionized gas momentum feedback @xmath82 depends on the expansion history , and has the order of magnitude of 10@xmath8310@xmath84 m@xmath4kms@xmath26 . the ionized gas feedback differs from the stellar wind feedback in the sense that it is a pressure effect ( see also figure [ fig_shematic ] ) , which does not exert force in the radial direction , but in the direction perpendicular to the cavity wall . in addition , the exerted pressure force depends only on the temperature and density of the ion , which is determined by the ionization , the recombination , and the pressure balance . from the observations of the 1.3 mm continuum emission , and the measurements of the rotational velocity of the molecular gas , we constrained the molecular mass inward of the 0.15 pc radius around the uc hii region a to have the order of magnitude of 400 m@xmath4 . we assume a simple geometry in this region , and perform order of magnitude estimates of the energetic relation . assume the molecular gas has an initial mass density distribution @xmath85 before the creation and the dynamical expansion of the biconical cavity . defining @xmath86 to be the angular separation from the plane of rotation , if the system is approximately axisymmetric , @xmath85 can be represented by @xmath87 . if @xmath88 can be represented as a polynomial of @xmath89 , for a rotationally flattened system , we expect the leading order dependence to be @xmath89 . this suggests that about 70% of molecular mass ( 280 m@xmath4 ) are concentrated in the region with @xmath9045@xmath14 . after the creation and the dynamical expansion of the biconical cavity , the molecular mass initially distributed in the region @xmath9145@xmath14 ( 120 m@xmath4 ) are accumulated on the geometrically thin cavity wall , which has an opening angle of 90@xmath14 . the assumption of the mass accumulation on the cavity wall is consistent with the fast expansion ( see section [ chap_timescale ] for the estimate of timescale ) . the molecular gas initially distributed in the region @xmath9045@xmath14 forms a dense flattened structure with density distribution @xmath92 . we argue that the opening angle of 90@xmath14 is a reasonable value while comparing with the geometrical picture revealed by the ch@xmath0oh images ( figure [ fig_nh3 ] , [ fig_ionout ] ) . the cavity wall has two components , w@xmath93 and w@xmath94 , of which the surface area are perpendicular and parallel to the radial direction , respectively . given the opening angle , the ratio of the surface area of these two components is approximately 1 ; we assume the ratio of the mass accumulated on these two components is also 1 . [ [ w_parallel ] ] w@xmath94 + + + + + + + + + + + + + + + + + + + + + + + + + + + this component has a cone shape , with the total mass of @xmath860 m@xmath4 . for it to expand at 2.55 kms@xmath26 , it requires a momentum of 150300 m@xmath4kms@xmath26 . from the estimations in the previous section , we see that this required momentum has the same order of magnitude of the feedback from the ionized gas pressure . the gravitational force acts in the radial direction and does not retard the expansion in the @xmath95 direction . we expect the stellar wind to have weak effects on the dynamics of w@xmath94 since the wind is parallel to the cavity wall . [ [ w_perp ] ] w@xmath93 + + + + + + + + + + + + + + + + + + + + + + + this part is directly driven by the stellar radiation and the stellar wind , which are competing with gravity in the radial direction . at the radius of 0.15 pc , the stellar wind and the ionized gas pressure can have comparable effect on w@xmath93 . assuming the mass is 60 m@xmath4 , the gravitational force has the order of magnitude of 2@xmath2210@xmath74 m@xmath4kms@xmath26yr@xmath26 , which is also comparable to the force of the stellar wind and the ionized gas . observationally this part of cavity wall is not as prominently detected as w@xmath94 . since the embedded stellar cluster in the uc hii region a has a high ionizing photon emission rate ( @xmath96 = 10@xmath97 s@xmath26 ; ho & haschick 1981 ) , which is able to ionize on the order of 10@xmath65 m@xmath4 of gas in the biconical region . even if the recombination rate marginally balances the ionization rate , the stellar ionization can still explain why the w@xmath93 is not clearly detected . some local high density clumps which have higher recombination rate may be self - shielded from the stellar ionization and have long lifetime . those high density clumps occupy small solid angles , and are less accelerated by the stellar wind and the pressure of the ionized gas . some of those clumps can still fall toward the ob cluster owing to the gravitational attraction . the clumpy 3.6 cm free - free continuum emissions in the uc hii region a , especially , northeastern to the emission peak , may be explained by those clumps externally ionized by the central ob cluster . we note that we may overestimate the dynamical expansion timescale of the biconical cavity since the cavity can start expansion from a finite size owing to the initial ionization , which leads to the overestimation of the momentum feedback . however , those ionized gases are not gravitationally bound and are free to leak out . therefore , the accumulated mass on the cavity walls w@xmath94 and w@xmath93 is reduced if the initial ionization is efficient , and requires smaller momentum feedback to reach the final state velocity . similar arguments are valid if the biconical cavity is initially created by the massive bipolar outflow ( an example see g240.31 + 0.07 : qiu et al . 2009 ) and the molecular gas is evacuated . [ [ the - dense - flattened - structure ] ] the dense flattened structure + + + + + + + + + + + + + + + + + + + + + + + + + + + + + this structure has the mass of @xmath8280 m@xmath4 , and continues from the outer of @xmath80.15 pc radius to the inner @xmath80.03 pc ( liu et al . 2010 ) . the total gravitational force on this structure depends on the embedded mass . a lower limit of 9.6@xmath2210@xmath74 m@xmath4kms@xmath26yr@xmath26 can be given assuming that the majority of mass is distributed at the 0.15 pc radius . this lower limit is already on the same order of magnitude as the feedback from the stellar wind and the ionized gas pressure ; the actual value of the gravitational force should be much larger . depending on the effective solid angle it occupies , the stellar wind and the ionized gas pressure may have negligible effect on the radial motion of the flattened structure . with the biconical cavity structure which leads to the photon leakage , the radiative pressure is also negligible . we therefore suggest that the dynamics in this region is dominated by gravity and rotation . [ [ uc - hii - region - b - and - c ] ] uc hii region b and c + + + + + + + + + + + + + + + + + + + + + the stellar wind and ionized gas pressure feedback in uc hii region b and c have the same order of magnitude as those in the uc hii region a. however , these two regions have lower initial gas density , which leads to a smaller accumulated mass on the expansional shell . the observed expansional velocities in these two regions are comparable to but smaller than the thermal velocity of the ionized gas , suggesting that the ionized gas pressure is an important driving source of the dynamical expansion . from the estimations in the previous section , the stellar wind feedback can be comparably important , depending on the detailed geometry of these system . we present high resolution observations of molecular lines and free - free continuum for the uc hii region g10.60.4 . the resolved projected distributions of the molecular gas and the ionized gas suggest an overall picture consisting of an extended ( @xmath80.5 pc ) envelope , a single compact ( @xmath80.1 pc ) hot rotating toroid , and a biconical molecular cavity filled with ionized gas . this overall geometry resembles the standard envelope disk and protostellar outflow model for the low mass star forming region . with the presence of the biconical cavity , we suggest that the radiative pressure can be significantly reduced.in the plane of rotation , at the scale larger than 1@xmath7 ( 0.03 pc ) , we see that the rotational motion of the dense gas is not severely disturbed , which consistently suggests that the radiation is not yet important for the molecular accretion flow . the stellar radiation may increase and play a more important role as the stellar mass is increased . from the observations of cs ( 10 ) , we suggest that the biconical cavity around the uc hii region a is undergoing an expansional motion , with velocity of 2.55 kms@xmath26 . we perform simple order of magnitude estimates , and suggest that the feedback from the ionized gas pressure can account for the required momentum of this dynamical expansion . the expansional signatures are also detected in the uc hii region b and c , which are driven by the ionized gas pressure and the stellar wind . the expansional motions of the uc hii regions significantly disturb the local dynamics of the molecular gas , inject the energy , and may induce the non - uniformity in the molecular accretion flow , which is an important feedback mechanism in the massive molecular clump . oh j=5 transitions ( * r : * ch@xmath0oh 5(0,5)4(0,4 ) a+ ; * g : * the blended ch@xmath0oh 5(-2,4)4(-2,3 ) e and ch@xmath0oh 5 ( 2,3)4(2,2 ) e ; * b : * ch@xmath0oh 5(-3,3)4(-3,2 ) e ) . * right : * the rgb image of the ch@xmath0oh j=5 transitions overlaid with the nh@xmath0 ( 3,3 ) main hyperfine inversion emission ( contour ) . solid contours start from 10% of the emission peak with 10% intervals ; dashed contours start from -200% of the emission peak with -200% intervals . note the primary beams of the sma ( @xmath81@xmath15 ) and the vla ( @xmath82@xmath15 ) observations are much larger than the angular size scale of the detected structures . ] @xmath131.2@xmath7 , with position angle 77@xmath14 . * contour : * the 1.3 mm continuum image . this data is taken by sma subcompact+compact+very extended array with 0@xmath7.79@xmath130@xmath7.58 resolution . contours show the levels of [ 1 , 2 , 4 , 8 , 16 , 32 , 64 , 96]% of the peak brightness temperature 49 k. a proposed plane of rotation is indicated by a red dashed line . the red dashed line passes through two free - free continuum peaks , and has a position angle of 135@xmath14 , which is consistent with the position angle of the plane of rotation of the central 0.1 pc scale hot core ( 140@xmath14@xmath315@xmath14 , liu et al . 2010 ) . ] @xmath131.2@xmath7 , with position angle 77@xmath14 . * contour : * the 3.6 cm free - free continuum image ; resolution 0@xmath7.5@xmath130@xmath7.4 . contours show the levels of [ 1 , 2 , 4 , 8 , 16 , 32 , 64 , 96]% of the peak intensity of 0.1 jy / beam . ]

we investigate the phase structure of gauge theory in four and five dimensions with one compact dimension by using perturbative one - loop and pnjl - model - based effective potentials , with emphasis on spontaneous gauge symmetry breaking . when adjoint matter with the periodic boundary condition is introduced , we have rich phase structure in the quark - mass and compact - size space with gauge - symmetry - broken phases , called the split and the re - confined phases . our result is qualitatively consistent with the recent lattice calculations . when fundamental quarks are introduced in addition to adjoint quarks , the split phase becomes more dominant and larger as a result of explicit center symmetry breaking . we also show that another phase ( pseudo - reconfined phase ) with negative vacuum expectation value of polyakov loop exists in this case . we study chiral properties in these theories and show that chiral condensate gradually decreases and chiral symmetry is slowly restored as the size of the compact dimension is decreased .

1302.2196

now that the higgs - like particle has been discovered in large hadron collider ( lhc ) @xcite , one of primary interests of particle physics is to understand mechanism of dynamical electroweak symmetry breaking . to settle down the unsolved issues including the hierarchy problem , there have been proposed several promising courses including supersymmetric , composite - higgs and extra - dimensional models . among them , the gauge - higgs unification with extra dimensions @xcite produces brilliant dynamics of gauge symmetry breaking , called the hosotani mechanism @xcite , with the higgs particle as an extra - dimensional component of the gauge field : when adjoint fermions are introduced with periodic boundary condition ( pbc ) in gauge theory with a compact dimension , the compact - space component of the gauge field can develop a non - zero vacuum expectation value ( vev ) , which breaks the gauge symmetry to its subgroup spontaneously . this phenomenon originates in the non - abelian aharonov - bohm effect . there have been proposed a number of bsm models directly and indirectly based on this mechanism @xcite recently , the same phenomenon has been observed in a different context . as well - known , the imaginary time dimension is compacted in the finite - temperature ( quantum chromodynamics ) qcd . when the adjoint fermion is introduced with pbc in the theory , it was shown that exotic phases appear @xcite , where the color trace of polyakov line takes non - trivial vev leading to the dynamical gauge symmetry breaking . this can be interpreted as @xmath3 realization of the hosotani mechanism . further study is now required to understand detailed properties and phase diagram for this case . as for the five - dimensional gauge theory on @xmath4 , lattice simulation is still at the stage to understand the nature of 5d pure yang - mills theory . first attempt was done in ref . @xcite and recent progress was shown in ref . @xcite . at the present , we have no detailed investigation on the phase structure of the five - dimensional gauge theory with dynamical quarks . the purpose of our work is to understand properties of gauge theories with quarks on @xmath3 and @xmath4 by using effective theories and show a guideline for further lattice simulations . we focus on the phase structure in @xmath0 gauge theory by mainly using the perturbative one - loop effective potential . when we look into chiral properties , the polyakov - loop - extended nambu jona - lasinio ( pnjl ) model @xcite is introduced . we obtain the phase diagram in the quark - mass and compacted - size space for several different choices of fermion representations and boundary conditions . in the theory with adjoint fermions with pbc , we find rich phase diagram with unusual phases where @xmath0 gauge symmetry is spontaneously broken to @xmath1 ( split phase ) or @xmath2 ( re - confined phase ) . this result is consistent with the recent lattice simulations @xcite and the perturbative calculations for massless quarks @xcite . we show that chiral condensate is gradually decreased and the chiral symmetry is slowly restored with the compacted size being decreased although small chiral transitions coincide with the deconfined / split and split / reconfined transitions . we also study phase diagram for the case with both fundamental and adjoint quarks , and show that the split phase becomes more dominant as a result of the explicit breaking of @xmath5 center symmetry . this result indicates that the @xmath6 phase , which is of significance in terms of phenomenology , can be controlled by adjusting the number of fundamental flavors . in this case we point out that another unusual ( @xmath2 ) phase with negative vev of color - traced polyakov loop emerges , which we call pseudo - reconfined phase " . we note similar investigation on the phase diagram of compact - space gauge theory with pbc fermions were shown in @xcite , where the authors focus on the volume independence of the vacuum structure from the viewpoint of the large n reduction @xcite . in order to reproduce the confined phase at low temperature , they used the gluon effective potentials with the mass - dimension parameter , which leads to explicit gauge symmetry breaking . since these effective potentials do not suit our purpose of studying spontaneous gauge symmetry breaking due to the hosotani mechanism , we mainly use setups with manifest gauge symmetry . it is also notable that gauge symmetry breaking for all clasical lie groups with pbc matters are classified from topological viewpoints in @xcite . this paper is organized as following . in sec . [ sec : ep ] , we show our setups including the one - loop potential and the pnjl model . [ sec : ps4 ] shows our numerical results for four dimensional cases . [ sec : ps5 ] shows the results for five - dimensional cases . in sec . [ sec : ob ] we discuss how our predictions can be checked in the lattice simulations . [ sec : sum ] is devoted to summary . in this section we discuss our setups for su(n ) gauge theory . the one - loop effective potential is composed of two parts , gluonic and quark contributions . since our goal is to study the phase diagram in terms of wilson - line phases ( polyakov - line phases ) , @xmath7 @xmath8 , we will write the effective potential as a function of these variables . besides the one - loop potential , we consider the contribution from the chiral part , which originates in the njl - type four - point interactions . the total effective potential is then regarded as that of the pnjl model @xcite as a function of the @xmath7 and the chiral condensate @xmath9 . depending on our purposes , we in some cases use only the perturbative one - loop effective potential , and in other cases use the pnjl effective potential . we also comment on possible deformation reflecting non - perturbative effects in the gluonic contribution . we begin with one - loop effective potential for @xmath10 gauge theory . in four dimensions , the finite - temperature one - loop effective potential for gauge bosons and fermions is well studied in ref . what we do in the present study is just to regard the compacted imaginary - time direction as one of the spatial direction , and to replace @xmath11 by @xmath12 where @xmath13 is a size of the compacted dimension . firstly , we rewrite the @xmath10 gauge boson field as @xmath14 where @xmath15 stands for a compact direction and @xmath16 is a vacuum expectation value ( vev ) while @xmath17 is fluctuation from it . the vev can be replaced by @xmath18 where @xmath19 s color structure is @xmath20 with @xmath21 . we note that eigenvalues of @xmath7 are invariant under all gauge transformations preserving boundary conditions . then we can easily observe spontaneous gauge symmetry breaking from values of @xmath7 in the vacuum . for detailed argument on gauge transformation for this topic , see @xcite for example . the gluon one - loop effective potential @xmath22 is expressed as @xmath23 where @xmath24 and @xmath25 is the number of color degrees of freedom . the contribution from massless fundamental quarks @xmath26 is given by @xmath27}{n^4},\end{aligned}\ ] ] where @xmath28 is the number of fundamental flavors . depending on boundary conditions , we should replace @xmath29 by @xmath30 . for example , the choice of @xmath31 describes quarks with periodic boundary conditions . from here , we denote @xmath32 as effective potential of the fundamental fermion with boundary angle @xmath33 . the contribution from massive fundamental quarks is expressed by using the second kind of the modified bessel function @xmath34 as @xmath35,\end{aligned}\ ] ] where @xmath36 is the fundamental fermion mass . ( we assume the same mass for all flavors . ) here the second kind of the modified bessel function @xmath37 is defined as @xmath38 where @xmath39 is the gamma function . the adjoint quark contribution @xmath40 can be easily obtained by the following replacement , @xmath41 , \label{re}\end{aligned}\ ] ] where @xmath42 and @xmath43 are the number of flavors and the mass for adjoint fermions . for the gauge theory with @xmath44 fundamental and @xmath45 adjoint fermions with arbitrary boundary conditions , the total one - loop effective potential is given by @xmath46 this total one - loop effective potential contains eight parameters including the compactification scale @xmath13 , the number of colors @xmath25 , the fermion masses @xmath36 , @xmath43 , the number of flavors @xmath44 , @xmath45 , and the boundary conditions @xmath33 for two kinds of matter fields . all through the present study we keep @xmath47 , then obtain phase diagram in @xmath12-@xmath43 space with @xmath36 , @xmath44 , @xmath45 and @xmath33 fixed to several values . the reason we change @xmath43 while fixing @xmath36 is that gauge symmetry phase diagram is more sensitive to the former than the latter . up to this point , we work in first - principle perturbative calculations and have no model - parameter - fixing process dictated by physical results or numerical calculations . we here comment on ability and limitation of the perturbative one - loop effective potential . this weak - coupling - limit potential at least works to investigate weak - coupling physics . in this study a small compactification scale @xmath13 ( or high temperature @xmath48 ) corresponds to weak coupling , and the @xmath12-@xmath43 phase diagram obtained from the effective potential is valid at smaller @xmath13 while the strong - coupling physics near larger @xmath13 or small temperature can not be reproduced . we note that gauge symmetry breaking due to hosotani mechanism takes place even at weak - coupling due to the compact dimension topology @xcite . indeed , the exotic phases in the recent lattice simulations @xcite have been shown to be at weak - coupling regime . our potential thus work to reveal properties of these exotic phases . now we consider contribution from the chiral sector . we introduce the four - point interaction at the action level as @xmath49 , \label{gs}\ ] ] where @xmath50 is a two - flavor fermion field and @xmath51 is the effective coupling constant with the mass dimension minus two . ( we here concentrate on the two - flavor case . ) the introduction of the above term ( [ gs ] ) leads to addition of the following zero - temperature contribution to the one - loop effective potential ( [ eq_ep_pert ] ) , @xmath52,\ ] ] where @xmath53 is a cutoff scale of the effective theory and @xmath54 stands for constituent quark mass @xmath55 with @xmath9 being chiral condensate @xcite . depending on fundamental or adjoint representations , @xmath56 takes @xmath25 or @xmath57 respectively . in the case with both fundamental and adjoint quarks , we need two different sets of chiral sectors . the total effective potential combined with the chiral contribution is given by @xmath58 we note that this total effective potential is identical to that of the pnjl model @xcite adopting the one - loop potential as the gluon contribution , @xmath59 + { \cal v}_{g},\ ] ] with @xmath60 . we thus call the total effective potential ( [ pnjl - like ] ) pnjl or pnjl - based effective potential . in the present study we will use this pnjl - based potential to study gauge theory with adjoint or fundamental quarks with periodic boundary conditions(pbc ) . in this model , we have two model parameters , the cutoff scale @xmath53 and the effective - coupling @xmath51 in addition to the eight parameters in the one - loop perturbative potential shown below ( [ eq_ep_pert ] ) . these two parameters are the very model parameters and should be fixed to reproduce physical results or first - principle lattice calculations . on the other hand , the lattice qcd simulation on the compactified space has been done as finite - temperature lattice qcd , but for only anti - periodic boundary conditions ( apbc ) for quarks . thus , we choose the following criterion for fixing the two parameters for pbc quarks ; we first fix them so as to reproduce the zero - temperature constituent mass and the chiral critical temperature in the lattice finite - temperature qcd with apbc matters @xcite , then we use the same parameter set for the pbc case . according to @xcite , the zero - temperature constituent mass for apbc adj . qcd is given by @xmath61 and the chiral phase transition takes place at @xmath62 where @xmath63 and @xmath64 are critical values for chiral transition . we can fix @xmath53 and @xmath51 by using these lattice results . this method is also applied to the fundamental quarks . apart from the chiral contribution , the pnjl potential we obtained is based on perturbative calculations , and chiral properties we will obtain from it is again valid for a small @xmath13 region . in addition to this limitation , we have the cutoff scale @xmath53 in this model . it means that we can not apply the model beyond this cutoff . the model is thus valid for intermediate compactification scale as @xmath65 . in the next section we will find specific chiral properties within this limitation . we here comment on non - perturbative modification of the pnjl model . in the standard use of the pnjl model , the gluonic contribution @xmath66 is replaced by the non - perturbatively "- deformed ones : in order to mimic the confinement / deconfinement phase transition in the study on qcd phase diagram , , the one - loop gluon potential should be replaced by some nonperturbative versions . we have several schemes including the simple scale introduction in ref . @xcite , center - stabilized potential in @xcite , the one - loop _ ansatz _ in ref . @xcite and the strong - coupling lattice potential in ref . the first one in @xcite is given by @xmath67 where the second term is the modification incorporating non - perturbative effect and the mass scale @xmath68 works as a scale for confinement / deconfinement transtion . the center - stabilized potential in @xcite has a similar form given as @xmath69}_{n=1 } \sum_{i , j=1}^n a_{n } \bigl ( 1 - \frac{1}{n } \delta_{ij } \bigr)\cos ( 2 n \pi q^{ij } ) , \label{g_np2}\end{aligned}\ ] ] where @xmath70 in the deformation term is a mass dimension-3 model parameter , which works as the phase transition scale in this case . the one - loop ansatz potential is based on simple introduction of the scale @xmath68 to the original gluon potential as @xmath71 where some of compactification scale @xmath13 is just replaced by @xmath68 in the original potential . the potential from the lattice strong - coupling expansion is non - perturbative deformation adopted in the original pnjl model @xcite , which is given by @xmath72 , \label{g_np4}\end{aligned}\ ] ] where @xmath73 stands for the cutoff scale and @xmath9 for the string tension for confinement . @xmath74\sim{\rm diag}[e^{2\pi i q_{1}}, ... ,e^{2\pi i q_{n}}]$ ] is a compactified dimension polyakov ( wilson ) loop . these two model parameters work as the scale for the phase transition . all the modifications able to reproduce the phase transition , however , it is notable that the @xmath10 gauge symmetry is explicitly broken due to the mass - dimensionful model parameters in them . it means that they do not suit our purpose of classifying the gauge - broken phases . we also emphasize that , in the first place , the exotic phases including gauge - broken phases are likely to emerge at weak - coupling region @xcite and we do not need the above deformations for the strong - coupling region for our purpose . thus , we will mainly adopt one - loop effective potential as the gluonic contribution in our pnjl model and just discuss how the phase diagram is changed by the deformation in the next section . in the five - dimensional case , most of the setup is parallel to the four - dimensional case except for difference of mass dimensions of fields and parameters . we here show the one - loop effective potential of gluon and quarks below . the five - dimensional one - loop effective potential in the gluon sector is given by @xmath75 which has @xmath76 dimension . the effective potential for massless fermions in this case is given by @xmath77 \nonumber\\ & = \frac{3}{l^5 \pi^2 } \sum_{n=1}^\infty \frac{\cos [ 2 \pi n ( q^{ij}+1/2 ) ] } { n^5}. \label{integral}\end{aligned}\ ] ] the effective potential of the massless fundamental and adjoint fermions with arbitrary boundary condition is expressed as @xmath78}{n^5 } , \\ { \cal v}_a^\phi & = \frac{3 n_a}{\pi^2l^5 } \sum_{i , j=1}^n \sum_{n=1}^\infty \bigl ( 1 - \frac{1}{n } \delta_{ij } \bigr ) \frac{\cos [ 2 \pi n ( q_{ij } + \phi ) ] } { n^5}.\end{aligned}\ ] ] for the massive fermion , we should replace @xmath79 by @xmath80 in eq . ( [ integral ] ) . to evaluate the integration , we should expand the logarithm in the same way as the four dimesional case . individual integration is done by using the bessel function as @xmath81 \nonumber\\ & = \frac{e^{\pm 2i \pi nq^{ij}}}{n } \frac { 2^{3/2 } m}{\sqrt{\pi } } \bigl ( \frac{m}{n l } \bigr)^{3/2 } k_{5/2}(n l m).\end{aligned}\ ] ] then , the potential contributions from massive fermions are obtained by using the @xmath82 as @xmath83 , \\ { \cal v}_a^\phi ( n_{a } , m_{a } ) & = \frac { \sqrt{2 } n_a ( m_{a}/l ) ^{5/2 } } { \pi^{5/2 } } \sum_{i , j=1}^n \sum_{n=1}^\infty \bigl ( 1 - \frac{1}{n } \delta_{ij } \bigr ) \frac{k_{5/2 } ( n m_{a } l)}{n^{5/2 } } \cos [ 2 \pi n ( q_{ij } + \phi ) ] .\end{aligned}\ ] ] this total one - loop effective potential in five dimensions again contains eight parameters including the compactification scale @xmath13 , the number of colors @xmath25 , the fermion masses @xmath36 , @xmath43 , the number of flavors @xmath44 , @xmath45 , and the boundary conditions @xmath33 for two kinds of matter fields . as with the four - dimensional case we keep @xmath47 , then obtain phase diagram in @xmath12-@xmath43 space with @xmath36 , @xmath44 , @xmath45 and @xmath33 fixed to several values . the @xmath12-@xmath43 phase diagram obtained from this effective potential is valid at smaller @xmath13 or the weak coupling regime , and thus works to reveal properties of the exotic phases at weak coupling . we here comment on treatment of parity pairs . when we need a parity - even mass term in odd dimensions ( e.g. @xmath84 orbifolded theory ) , we need a set of parity pairs @xmath85 in the action @xcite . although our study does not take much care about such a case , we note that the factor two appears in front of the fermion potentials if the parity pair is introduced . it is thus the same situation as doubled flavors within this setup . in sec . [ sec : ps5 ] , we will see that the phase diagram changes depending on the choice of the parity pair in five dimensions . in this section , we investigate the phase structure in the compacted - size @xmath86 and quark - mass @xmath87 space for @xmath0 gauge theories on @xmath3 . because of the relation @xmath88 , the vacuum structure is discussed just from the @xmath89 potential . in the following we begin with investigation on vacuum and phase structures based on the one - loop potential ( [ eq_ep_pert ] ) for several choices of matters and boundary conditions . by far , lattice qcd simulations in four dimensions with one compacted dimension have been done for apbc fundamental , pbc fundamental , apbc adjoint @xcite and pbc adjoint @xcite matters . there are no lattice study on systems including both fundamental and adjoint matters . we also have no lattice corresponding results for five dimensional cases . in the present study we mainly focus on the case with pbc adjoint , and with pbc adjoint and fundamental matters , and make comparison to the lattice result if exists . as shown in the above section , we make use of the other lattice results to fix model parameters in pnjl models . we note that we will sometimes consider one - flavor cases as @xmath90 or @xmath91 . since the anomaly is not implemented and difference of one- and multiple - flavors just appears as an overall factor in the effective potential , our four - dimensional results are qualitatively unchanged even for the two - flavor cases as @xmath92 or @xmath93 . thus , these one - flavor examples are sufficient for our purpose of studying phase diagram qualitatively . ( things are different in five dimensions as shown in sec . [ sec : ps5 ] . ) in the appendix . [ sec : vsf ] , we summarize the potential minima for gauge - symmetry - preserved cases , which are not our main results . we show the contour plots of the effective potentials for @xmath94 in fig . [ fig_p_g_3d ] , @xmath95 with the anti - periodic boundary condition ( apbc ) in fig . [ fig_p_gfa_3d ] , @xmath95 with the periodic boundary condition ( pbc ) in fig . [ fig_p_gfp_3d ] , and @xmath90 with apbc in fig . [ fig_p_gaa_3d ] . we set @xmath96 in these cases . in the pure - gauge and gauge+adjoint theories , the three degenerate vacua clearly reflects the @xmath5 center symmetry while the introduction of fundamental fermions lift of the degeneracy , which means breaking of the center symmetry . we note that the global minima are given by any of @xmath97 , and we have @xmath98 ( mod 1 ) in the vacuum . it obviously shows that the @xmath0 symmetry remains in all these cases . gauge theory on @xmath99 with one adjoint fermion with pbc @xmath100 l^4 $ ] . ( right ) the contour plot as a function of @xmath101 and @xmath102 . thicker region stands for deeper region of the potential . ( left ) the effective potential as a function of @xmath101 with @xmath103 . the global minima are located at @xmath104 . , title="fig:",scaledwidth=40.0% ] gauge theory on @xmath99 with one adjoint fermion with pbc @xmath100 l^4 $ ] . ( right ) the contour plot as a function of @xmath101 and @xmath102 . thicker region stands for deeper region of the potential . ( left ) the effective potential as a function of @xmath101 with @xmath103 . the global minima are located at @xmath104 . , title="fig:",scaledwidth=40.0% ] now , we shall look into the case with spontaneous gauge symmetry breaking . we consider @xmath90 with pbc . we note that this theory has exact center symmetry , and all the phases , even the gauge - broken phase , should reflect this symmetry . figure [ fig_p_gap_3d ] shows the effective potential @xmath105 l^4 $ ] . the left contour plot is obviously different from the gauge - symmetric cases . careful search shows that the minima are located at @xmath106 , @xmath107 , @xmath108 , @xmath109 , @xmath110 , @xmath111 . it means the vacua are given by permutations of @xmath112 , and @xmath0 gauge symmetry is broken into @xmath2 . this is the famous result , known as the hosotani mechanism , where the aharonov - bohm effect in the compacted dimension nontrivially breaks gauge symmetry @xcite . we note that this situation is sometimes called re - confined phase " @xcite since the color fundamental trace of the polyakov loop @xmath113 becomes zero . to study the phase diagram , we introduce nonzero quark mass . figure [ fig_p_gapm_2d ] shows the effective potential @xmath114l^4 $ ] as a function of @xmath115 with @xmath116 for @xmath117 , @xmath118 , @xmath119 and @xmath120 from left to right panels ( @xmath121 ) . it is clearly seen that there is the first - order phase transition in the vicinity of @xmath122 . this is a transition between the re - confined phase and the other gauge - broken phase , which we call the split phase " @xcite . the contour plots for @xmath123 and @xmath124 are shown in fig . [ fig_p_gapm_3d ] . the @xmath124 case corresponds to the split phase . the global minima in the split phase are given by * @xmath125 : @xmath126 , @xmath127 , @xmath128 , * @xmath129 : @xmath130 , @xmath131 , @xmath132 , * @xmath133 : @xmath134 , @xmath135 , @xmath136 . these results indicate that the vacuum in the first set is given by permutations of @xmath137 , and @xmath0 gauge symmetry is broken into @xmath1 . the vacua for other two sets are derived by the @xmath138 transformation of @xmath137 . from the viewpoint of phenomenological symmetry breaking , this phase and similar phases for large color numbers are the most preferable . in fig . [ 4d_phase_p ] we depict the @xmath86-@xmath87 phase diagram with one pbc adjoint quark @xmath139 based on the one - loop effective potential . we note that , as @xmath87 appears as @xmath140 in this one - loop effective potential , we have scaling in the phase diagram and we can choose arbitrary mass - dimension unit . since we drop the non - perturbative effect in the gluon potential , we have no confined phase at small @xmath86 ( at low temperature ) . the order of the three phases in fig . [ 4d_phase_p ] ( deconfined @xmath0 @xmath141 split @xmath1 @xmath141 reconfined @xmath2 from small to large @xmath86 . ) is consistent with that of the lattice simulation except that we have no confined phase . we note that all the critical lines in the figure are first - order . in fig . [ fig_pl_distribution ] we depict a schematic distribution plot of @xmath142 in the complex plane . it is obvious that each phase reflects @xmath5 symmetry . in the split phase , @xmath142 takes nonzero values but in a different manner from the deconfined phase . in the re - confined phase , we exactly have @xmath143 with the vacuum breaking the gauge symmetry . gauge theory on @xmath99 with one pbc adjoint quark as a function of @xmath101 with @xmath103 @xmath144 l^4 $ ] , for @xmath117 ( reconfined ) , @xmath118 ( reconfined@xmath145split ) , @xmath119 ( split@xmath145deconfined ) and @xmath120 ( deconfined ) . , title="fig:",scaledwidth=23.0% ] gauge theory on @xmath99 with one pbc adjoint quark as a function of @xmath101 with @xmath103 @xmath144 l^4 $ ] , for @xmath117 ( reconfined ) , @xmath118 ( reconfined@xmath145split ) , @xmath119 ( split@xmath145deconfined ) and @xmath120 ( deconfined ) . , title="fig:",scaledwidth=23.0% ] gauge theory on @xmath99 with one pbc adjoint quark as a function of @xmath101 with @xmath103 @xmath144 l^4 $ ] , for @xmath117 ( reconfined ) , @xmath118 ( reconfined@xmath145split ) , @xmath119 ( split@xmath145deconfined ) and @xmath120 ( deconfined ) . , title="fig:",scaledwidth=23.0% ] gauge theory on @xmath99 with one pbc adjoint quark as a function of @xmath101 with @xmath103 @xmath144 l^4 $ ] , for @xmath117 ( reconfined ) , @xmath118 ( reconfined@xmath145split ) , @xmath119 ( split@xmath145deconfined ) and @xmath120 ( deconfined ) . , title="fig:",scaledwidth=23.0% ] gauge theory on @xmath99 with one pbc adjoint quark @xmath146 l^4 $ ] , for @xmath122 and @xmath147 ( @xmath1 split phase ) as a function of @xmath101 and @xmath102 . thicker region indicates deeper region of the potential.,title="fig:",scaledwidth=40.0% ] gauge theory on @xmath99 with one pbc adjoint quark @xmath146 l^4 $ ] , for @xmath122 and @xmath147 ( @xmath1 split phase ) as a function of @xmath101 and @xmath102 . thicker region indicates deeper region of the potential.,title="fig:",scaledwidth=40.0% ] -@xmath87 phase diagram for @xmath0 gauge theory on @xmath99 with one pbc adjoint quark based on one - loop effective potential . d stands for deconfined ( @xmath0 ) " , s for split ( @xmath1 ) " and r for re - confined ( @xmath2 ) " phases . phase transitions are first - order.,scaledwidth=40.0% ] as a function of @xmath148 and @xmath149 for @xmath0 gauge theory on @xmath99 with one pbc adjoint quark . , scaledwidth=40.0% ] we next turn on the chiral sector and consider the pnjl effective potential ( [ pnjl - like ] ) . we investigate chiral properties of @xmath92 @xmath0 gauge theory on @xmath3 .. corresponds to the conformal window . however , since we introduce nonzero quark mass in our study , the conformality , even if exists , is broken . ] before proceeding to the main topic , we discuss validity of the effective model ( [ pnjl - like ] ) for the purpose of studying chiral properties at weak - coupling region . although we have no confined phase nor confined / deconfined phase transition in our model , the chiral restoration associated with the phase transition is correctly reproduced in this model for the known cases : @xmath150 with apbc , @xmath150 with pbc and @xmath92 with apbc . we depict behavior of the constituent mass for these cases in fig . [ fig_chiral ] , where the chiral phase transition takes place at some point for the three cases . the parameters are chosen so as to have the correct critical temperatures in finite - temperature @xmath0 gauge theory with apbc quarks @xcite , @xmath151 gev and @xmath152 for fundamental quarks and @xmath153 gev and @xmath154 for adjoint quarks @xcite . we are now convinced that the model can work to study chiral properties , and we go on to the main topic , @xmath92 with pbc . we calculate the pnjl effective potential in eq . ( [ pnjl - like ] ) and search for the vacua for @xmath115 , @xmath155 and @xmath9 . we depict the phase diagram for this case in fig . [ fig_chi_phase ] . due to nonzero constituent mass , the whole phase diagram is shifted , and we have phase transitions even for a @xmath156 massless case . we also note that the critical lines are curved due to the dimensionful parameters introduced . although our analysis is valid only for the weak - coupling regime , the phase diagram is qualitatively consistent with that of the lattice simulations @xcite except that ours have no confinement / deconfinement phase transition . as a function of @xmath86 for @xmath150 with apbc , @xmath150 with pbc , @xmath92 with apbc and @xmath150 with pbc with the bare mass fixed as @xmath157 gev . we choose the parameter set as @xmath151 gev and @xmath152 for fundamental cases and @xmath153 gev and @xmath154 for adjoint cases . , scaledwidth=40.0% ] to look into chiral properties , we simultaneously depict the real part of vev of polyakov loop @xmath142 and the constituent mass @xmath54 as a function of @xmath86 with the bare mass fixed as @xmath157 gev and @xmath158 gev in fig . [ fig_chi_p ] . we here normalize the constituent mass as @xmath159 . it is notable that chiral condensate , or equivalently , constituent quark mass does not undergo a clear transition even for large values of @xmath86 , and it gradually decreases . we thus have no clear chiral restoration transition while chiral symmetry is gradually restored in this theory . this result and the standard - pnjl result in @xcite are consistent with those of the lattice simulation @xcite , which argues that the chiral restoration at weak coupling should occur at a quite small value of the compacted size @xmath13 . the other notable point is that the chiral condensate undergoes quite small transitions coinciding with the deconfined / split and split / re - confined phase transitions . ( it can be seen better in fig . [ fig_chiral ] or the right panel for @xmath158 gev in fig . [ fig_chi_p ] . ) this kind of the transition propagation is well studied in @xcite , but they may be too small to be observed in the lattice simulations . -@xmath87 phase diagram for @xmath0 gauge theory on @xmath99 with two - flavor pbc adjoint quarks @xmath92 based on the pnjl - type effective potential . d stands for deconfined " , s for split " and r for re - confined " phases . critical lines are first - order.,scaledwidth=40.0% ] ( blue dashed ) with @xmath103 and the constituent mass @xmath54 ( red solid ) as a function of @xmath86 with the bare mass fixed as @xmath157 gev ( left ) and @xmath158 gev ( right ) . the constituent mass is normalized as @xmath160 . the result indicates that the chiral symmetry is gradually restored without the clear phase transition . , title="fig:",scaledwidth=40.0% ] ( blue dashed ) with @xmath103 and the constituent mass @xmath54 ( red solid ) as a function of @xmath86 with the bare mass fixed as @xmath157 gev ( left ) and @xmath158 gev ( right ) . the constituent mass is normalized as @xmath160 . the result indicates that the chiral symmetry is gradually restored without the clear phase transition . , title="fig:",scaledwidth=40.0% ] let us briefly comment on non - perturbative deformation of the gluon potential discussed in sec . [ sec : ep ] . as a result , in appendix . [ sec : vsf]2 we show a result for the deformation ( [ g_np ] ) . indeed , with introducing the dimension parameter , for example @xmath161 mev , the phase diagram in fig . [ 4d_phase_p ] is modified as fig . [ 4d_phase_np ] . here the confined phase emerges at small @xmath86 region , but it is connected with the re - confined phase through the small mass region . this result clearly shows that the gauge symmetry is broken as @xmath162 even at zero - temperature or infinite-@xmath13 . we note that the similar result with the same deformation in the gluon potential is shown in ref . @xcite , where it is argued that the unified confined phase implies the volume - independence of the confined phase structure . for the other deformations , we have the same situation with explicit breaking of gauge symmetry . for our purpose of clarifying and classifying phases of gauge symmetry , the deformations are not appropriate although they may work as more phenomenological means . we next consider @xmath91 with pbc . we concentrate on the case with a massless fundamental quark , and the potential is given by @xmath163 . in this case , since the fundamental quark breaks the @xmath5 center symmetry , the minima at @xmath164 become true vacua in the deconfined and split phases as shown in fig . [ fig_pl_distribution2 ] . ( fundamental matter with pbc moves the vacua to @xmath164 direction . ) in addition , we have no exact re - confiend phase since @xmath143 vacuum can not be chosen because of the center symmetry breaking in fig . [ fig_pl_distribution2 ] . we term this unusual phase as pseudo - reconfined phase " , where the @xmath0 is broken to @xmath2 and @xmath142 takes a nonzero and negative value . the existence of this phase is consistent with the research on flavor - number dependence of gauge - symmetry - broken manners in ref . @xcite . in fig . [ 4d_phase_fp ] , we depict the phase diagram for the same case . we emphasize that the split phase gets larger by introducing pbc fundamental quarks , compared to fig . [ 4d_phase_p ] . it is because the center symmetry breaking chooses @xmath165 and its permutations as true vacua among all the possible minima fig . [ fig_pl_distribution2 ] , and makes it more stable than the center - symmetric case in fig . [ fig_pl_distribution ] . for @xmath0 gauge theory on @xmath99 with one pbc adjoint and one pbc massless fundamental quarks @xmath91 . points painted over stand for vacua in this case . @xmath5 symmetry is broken , and some of the three minima are chosen as true vacua in deconfined and split phase . the polyakov loop @xmath142 in the pseudo - reconfined phase ( pc ) takes a nonzero and negative value.,scaledwidth=40.0% ] @xmath0 gauge theory with one pbc adjoint and one massless pbc fundamental quarks @xmath91 based on the one - loop effective potential . @xmath87 is the adjoint quark mass @xmath166 and @xmath86 is the inverse of the compacted size . pc stands for pseudo - reconfined " phase.,scaledwidth=40.0% ] to look into the details of pseudo - reconfined phase and the phase transition to the split phase , we depict the expanded effective potential near the global minima as a function of @xmath101 with @xmath167 in fig . [ skew - rec4d ] . the left panel shows the result for the massless adjoint quark ( @xmath168 ) , which corresponds to the pseudo - confined phase . the minimum is not located at @xmath169(split case ) nor at @xmath170(re - confined case ) . in the pseudo - confined phase we totally have six minima for @xmath101 and @xmath102 as @xmath171 , @xmath172 , @xmath173 , @xmath174 , @xmath175 , @xmath176 , which means that the vacua are given by the permutation of @xmath177 . the right panel shows the first - order phase transition between pseudo - confined and split phases . since the potential barrier at the phase transition is quite low , the fluctuation could break down clear phase transition . for the cases with @xmath178 and @xmath179 where the flavor of adjoint quarks is larger than that of fundamental quarks , the minima of the potential in the pseudo - reconfined phase becomes deeper , and we can observe the first - order phase transition more distinctly . @xmath0 gauge theory with one pbc adjoint and one massless pbc fundamental quarks as a function of @xmath101 with @xmath103 . left one shows the case with the pseudo - reconfined phase ( @xmath180 ) , where we have the minimum at @xmath181 . right one shows the first - order phase transition between the pseudo - reconfined and split phase at @xmath182.,title="fig:",scaledwidth=40.0% ] @xmath0 gauge theory with one pbc adjoint and one massless pbc fundamental quarks as a function of @xmath101 with @xmath103 . left one shows the case with the pseudo - reconfined phase ( @xmath180 ) , where we have the minimum at @xmath181 . right one shows the first - order phase transition between the pseudo - reconfined and split phase at @xmath182.,title="fig:",scaledwidth=40.0% ] the chiral sector can be introduced by extending to @xmath183 . in this case we have two chiral sectors for fundamental and adjoint quarks , and we have arbitrariness how to implement four - point interactions and choose relative parameters . for example , we may consider the following forms of the four - point and eight - point interactions . @xmath184 + ( g_{s})_{a}[(\bar{\psi}_{a}\psi_{a})^{2 } + ( \bar{\psi}_{a}i\gamma_{5}{\vec \tau}\psi_{a})^{2 } ] \nonumber\\ \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\ , & + ( g_{s})_{fa}[\{(\bar{\psi}_{f}\psi_{f})^2 + ( \bar{\psi}_{f}i\gamma_{5}{\vec \tau}\psi_{f})^2 \}^2 \{(\bar{\psi}_{a } \psi_{a})^2 + ( \bar{\psi}_{a}i\gamma_{5}{\vec \tau}\psi_{a})^2 \}^2],\end{aligned}\ ] ] where @xmath185 and @xmath186 stand for fundamental and adjoint quark fields , and @xmath187 stand for fundamental , adjoint and mixing effective coupling . even if we fix a form of the four - point interactions , we still have no criterion on how to set the parameters since there is no lattice study on this case either for apbc or pbc . thus we just show results of the chiral properties for two representative sets of the parameters . in either set , @xmath188 and @xmath53 are set the same value used in fig . [ fig_chiral ] and the mixing term is set as @xmath189 . as the first case we consider @xmath190 , where @xmath191 is the coefficient obtained from the fierz transformation . we call it scenario a " , where the fundamental chiral condensate becomes zero for all the region of @xmath12 while the adjoint chiral condensate has the qualitatively similar behavior to @xmath192 case as shown in left - panel of fig . [ fig_chi_p1 ] . as the second case we consider @xmath193 , which we call scenario b. in this case the fundamental chiral condensate has nonzero value at @xmath194 as shown in the right - panel of fig . [ fig_chi_p1 ] . in this case , unlike the adjoint chiral condensate , the fundamental chiral condensate never reacts to the gauge symmetry phase transitions and the decreasing behavior becomes relatively gentle at large @xmath12 ( @xmath195 gev ) . we consider that either of the scenarios of chiral properties for @xmath183 will be detected in the on - going lattice simulation . ( black dotted ) , the adjoint fermion constituent mass @xmath43 ( red solid ) and the fundamental fermion constituent mass @xmath36 ( blue dashed ) as a function of @xmath86 for the scenario a ( left ) and b ( right ) . the constituent masses are normalized as by @xmath196 gev , title="fig:",scaledwidth=40.0% ] ( black dotted ) , the adjoint fermion constituent mass @xmath43 ( red solid ) and the fundamental fermion constituent mass @xmath36 ( blue dashed ) as a function of @xmath86 for the scenario a ( left ) and b ( right ) . the constituent masses are normalized as by @xmath196 gev , title="fig:",scaledwidth=40.0% ] besides the cases we have shown above , we also have other interesting choices of matters . as shown in @xcite , the @xmath0 gauge theory with three fundamental flavors with flavored twisted boundary conditions @xmath197 can accidentally keep the @xmath5 center symmetry . for example , in the case of @xmath198 with the flavored twisted boundary conditions @xmath199 for fundamental fermions and pbc for the adjoint fermion , the distribution plot of @xmath142 becomes symmetric as fig . [ fig_pl_distribution ] and we have exact re - confined phase although the theory contains fundamental quarks . moreover , in @xcite , the present authors and collaborators discuss the @xmath0 gauge theory with @xmath200 with the flavored twisted boundary conditions @xmath199 can lead to the spontaneous gauge symmetry breaking without adjoint quarks . this case is also fascinating as a future research topic . in this section , we discuss the vacuum and phase structure in @xmath0 gauge theories on @xmath4 . since the five - dimensional gauge theory is not renormalizable , we can not discuss its non - perturbative aspects in a parallel way to the four - dimensional case . thus we will concentrate on the one - loop part of the effective theory , and will not consider the contribution from the chiral sector . it is sufficient for our purpose of investigating the phase structure at weak - coupling . we consider the five - dimensional one - loop effective potential @xmath201 l^5 $ ] . we first look at the case with @xmath90 with pbc . the qualitative properties are common with the four - dimensional case . figure [ fig_p_gapm_d5_2d ] shows the effective potential @xmath202l^5 $ ] as a function of @xmath115 with @xmath103 fixed . the adjoint quark mass is set as @xmath203 , @xmath204 and @xmath119 from left to right panels . the three cases correspond to re - confined ( left ) , split ( center ) , and deconfined ( right ) phase . the contour plots at @xmath203(reconfined ) and @xmath205(split ) are depicted in fig . [ fig_p_gfapm_d5_3d ] . the manners of @xmath0 symmetry breaking and the distribution of polyakov line @xmath142 in each phase are the same as the four - dimensional case ; @xmath206 in the re - confined phase and @xmath6 in the split phase . we depict the phase diagram in @xmath86-@xmath87 for this case in fig . [ fig_pd_gap ] . the qualitative configuration of the three phases in the phase diagram is indifferent from the four - dimensional case although the split phase is smaller in five dimensions . gauge theory on @xmath207 with one pbc adjoint quark as a function of @xmath101 with @xmath103 @xmath208 . depicted for @xmath203 ( re - confined ) , @xmath204 ( split ) and @xmath119 ( deconfined ) . , title="fig:",scaledwidth=30.0% ] gauge theory on @xmath207 with one pbc adjoint quark as a function of @xmath101 with @xmath103 @xmath208 . depicted for @xmath203 ( re - confined ) , @xmath204 ( split ) and @xmath119 ( deconfined ) . , title="fig:",scaledwidth=30.0% ] gauge theory on @xmath207 with one pbc adjoint quark as a function of @xmath101 with @xmath103 @xmath208 . depicted for @xmath203 ( re - confined ) , @xmath204 ( split ) and @xmath119 ( deconfined ) . , title="fig:",scaledwidth=30.0% ] gauge theory on @xmath207 with one pbc adjoint quark @xmath209 for @xmath210 and @xmath204 as a function of @xmath101 and @xmath102 . , title="fig:",scaledwidth=40.0% ] gauge theory on @xmath207 with one pbc adjoint quark @xmath209 for @xmath210 and @xmath204 as a function of @xmath101 and @xmath102 . , title="fig:",scaledwidth=40.0% ] -@xmath43 phase diagram for @xmath0 gauge theory on @xmath207 with one pbc adjoint quark based on the one - loop effective potential . , scaledwidth=40.0% ] we next consider the case for @xmath91 with pbc . we again concentrate on @xmath211 . due to explicit breaking of @xmath5 center symmetry , @xmath212 minima are chosen as vacua in the deconfined and split phases in the same way as the four dimensional case ( fig . [ fig_pl_distribution2 ] ) . it is notable that the split phase is again enlarged by introducing fundamental fermions , but more effectively in five dimensions . enhancement of the split phase due to the fundamental matter in five dimensions is sensitive to choice of parity pairs , or equivalently , choice of the number of flavors . for @xmath91 without the parity pair , the pseudo - confined phase disappears and the split phase becomes the unique gauge - broken phase as shown in fig . [ fig_pd_gapfp ] . this result is consistent with that of the massless case ( @xmath156 ) in ref . @xcite . on the other hand , when we introduce parity pairs for @xmath213 or equivalently consider @xmath93 , the split phase becomes wider than that in fig . [ fig_pd_gap ] but the pseudo - reconfined phase still survives as shown in fig . [ fig_pd_gapfp2 ] . in the five - dimensional pseudo - confined phase we again have six minima for @xmath101 and @xmath102 as @xmath171 , @xmath172 , @xmath173 , @xmath174 , @xmath175 , @xmath176 , which indicates that the minima are given by the permutation of @xmath177 . we depict the expanded effective potential in fig . [ skew - rec5d ] . the left panel shows the massless case ( @xmath214 ) , which corresponds to the pseudo - reconfined phase . the right panel shows the first - order phase transition between the pseudo - reconfined and split phases ( @xmath215 ) . in the cases with @xmath216 or @xmath217 , the potential minima in the pseudo - reconfined phase becomes deeper and the phase transition gets more distinct . -@xmath43 phase diagram for @xmath0 gauge theory on @xmath4 with one adjoint and one massless fundamental quarks ( @xmath91 , @xmath211 , pbc ) based on the one - loop effective potential . , scaledwidth=40.0% ] -@xmath43 phase diagram for @xmath0 gauge theory on @xmath4 with a set of parity pairs of @xmath91 , or equivalently @xmath93 , based on the one - loop effective potential . , scaledwidth=40.0% ] gauge theory on @xmath4 with a set of parity pairs of @xmath91 or @xmath93 as a function of @xmath101 with @xmath103 . left one shows the pseudo - confined phase ( @xmath180 ) , where we have the minimum at @xmath181 . right one shows the first - order phase transition between the pseudo - reconfined and split phase at @xmath215.,title="fig:",scaledwidth=40.0% ] gauge theory on @xmath4 with a set of parity pairs of @xmath91 or @xmath93 as a function of @xmath101 with @xmath103 . left one shows the pseudo - confined phase ( @xmath180 ) , where we have the minimum at @xmath181 . right one shows the first - order phase transition between the pseudo - reconfined and split phase at @xmath215.,title="fig:",scaledwidth=45.0% ] in the end of this section we comment on the other aspect of gauge theory with a compacted dimension . if we regard the compacted direction as time direction , the boundary condition for the polyakov - loop phases can be seen as imaginary chemical potential . the periodic and anti - periodic boundary conditions correspond to different roberge - weiss transition points on the qcd phase diagram . from this viewpoint , it is clear that fundamental fermions with pbc works to move the vacua to @xmath218 direction as shown in fig . [ fig_pl_distribution2 ] while those with apbc move it to @xmath219 direction as shown in fig . [ fig_pl_distribution3 ] for su(3 ) gauge theory with one pbc adjoint and one apbc fundamental quarks . points painted over stand for vacua in this case . @xmath5 symmetry is broken in the different manner from fig . [ fig_pl_distribution2].,scaledwidth=40.0% ] in this section we discuss observables quantitatively in our study , which can be compared to existing and on - going lattice simulations . _ mass spectrum _ : we fist consider the mass spectrum in the gauge - broken phases , @xmath1 and @xmath2 phases . as shown in ref . @xcite , the kalza - klein spectrum for gauge bosons is given by @xmath220 where @xmath221 stands for kk index . we here focus on @xmath222 modes and the case with zero quark mass . as long as @xmath223 @xmath224 , these modes are massless . on the other hand , when @xmath225 is realized at the vacuum , some or all of @xmath222 modes become massive . this phenomenon is consistent with the higgs mechanism with the gauge boson obtaining mass due to gauge symmetry breaking . in our study , we find the two gauge - broken phase @xmath1 and @xmath2 for @xmath0 gauge theory on @xmath226 . in the @xmath1 phase of @xmath0 gauge theory , where we originally have 8 massless gauge bosons , we have 4 massive modes , whose mass is given by @xmath227 where we substitute @xmath228 to @xmath7 . this result of the gauge boson mass is common for the adj . case and the adj .- fund . irrespective of the matter field , @xmath1 phase has the five gauge bosons with the mass ( [ su2u1 m ] ) . things change in @xmath2 phase . in @xmath2 phase ( re - confined phase ) for @xmath139 , we have 6 massive modes , whose masses are given by @xmath229 for @xmath213 , the @xmath2 phase ( pseudo - confined phase ) again has 6 massive modes , but the masses are different from the above as @xmath230 thus , if we look into gauge mass as a function of the compactification scale , there should be clear difference between different choices of the matter field even in the same symmetry phase . it is good indication not only for the gauge symmetry breaking but also for specifying the phases . these results are the case with both four - dimensional and five - dimensional cases . we note that our results of mass spectrum are valid for small @xmath13 ( weak - coupling ) , and the lattice simulation for relatively weak - coupling can reproduce our results . _ polyakov - loop _ : as we have discussed , the trace of compact - dimensional polyakov loop is also a good indication of the exotic phases @xcite . we first show how the exotic phases seen in the lattice qcd with pbc adjoint fermions @xcite is interpreted from the gauge symmetry breaking phases . [ pod ] is the distribution plot of the polyakov loop in the lattice simulations . each point corresponds to each of the results for different gauge configurations . we also depict corresponding results in our analytical calculation of fig . [ fig_pl_distribution ] on the figure . by comparing them , we find that each of the cases can be understood as one of @xmath0 deconfined , @xmath1 split or @xmath2 re - confined phases . ( the strong - coupling confined phase can not be reproduced in our weak - coupling study . ) the lattice artifacts makes @xmath5 symmetry no - exact , and one of @xmath5 minima is selected depending on @xmath231 . this result means that we can predict distribution of polyakov - loop for other choices of matter fields in the on - going lattice simulations . for example , @xmath91 case should have the distribution of polyakov - loop depending on @xmath231 as shown in fig . [ pod2 ] . the explicit @xmath5 symmetry breaking shifts the true vacua to the @xmath232 side in the complex space . as a result , the vacuum with @xmath232 should be chosen in @xmath1 phase while the @xmath2 phase also has @xmath232 . this behavior of the polyakov - loop distribution will be observed in the on - going simulation for the case with both adj . and fund . quarks . as shown in fig . [ fig_pl_distribution3 ] , when we take apbc for fundamental quarks instead of pbc , the behavior is changed as the vacua at @xmath233 are chosen . these predictions are valid for both four - dimensional and five - dimensional theories . we note that our results catch physical properties at least at small @xmath13 ( weak - coupling ) , and the lattice simulation for relatively weak - coupling can reproduce our results . on the lattice @xcite and that of the one - loop effective potential for @xmath0 gauge theory on @xmath99 with pbc adjoint quarks . ( watermarks stand for distribution of plot points on the lattice @xcite . ) apart from the strong - coupling confined phase , all of the specific behavior can be interpreted as the phases we found in our analytical calculations . , scaledwidth=60.0% ] based on the one - loop effective potential for @xmath0 gauge theory on @xmath234 with pbc adjoint and fundamental quarks.,scaledwidth=80.0% ] _ chiral condensate and chiral susceptibility _ : the special behavior of constituent mass ( chiral condensate ) as a function of @xmath12 , which we calculated in the pnjl model , is also peculiar to the theories with spontaneous gauge symmetry breaking . and it can be detected on the lattice . even if it is not easy to look into exact value and details of constituent mass on the lattice , the chiral susceptibility works to investigate the subtle behavior . chiral susceptibility in our model is defined as @xmath235 with @xmath236 where @xmath87 stands for the adjoint or fundamental quark mass . for example , the chiral behavior in the @xmath0 gauge theory with pbc adjoint matter in fig . [ fig_chi_p ] indicates that the chiral condensate reacts to the gauge - symmetry phase transitions slightly and it slowly decreases in the re - confined phase with @xmath12 getting large . chiral susceptibility in this case should have two discrete changes at small @xmath12 . the chiral properties in the @xmath0 gauge theory with pbc adjoint and fundamental matters in fig . [ fig_chi_p1 ] can have more special behavior . whichever of the scenarios a and b are realized on the lattice , peculiar behaviors of chiral condensate can be detected on the lattice . we note that our results from the pnjl model are valid for @xmath237 , which roughly means the region for @xmath1 phase and @xmath2 phases at @xmath238 gev . we expect the lattice results for these two phases reproduce our results . in this paper we have studied the phase diagram for @xmath0 gauge theories with a compact spatial dimension by using the effective models , with emphasis on the dynamical gauge symmetry breaking . we show that introduction of adjoint matter with periodic boundary condition leads to the unusual phases with spontaneous gauge symmetry breaking , whose ranges are controlled by introducing fundamental matter . we also study chiral properties in these theories and show that the chiral condensate remains nonzero even at a small compacted size . in sec . [ sec : ep ] , we developed our setup based on one - loop effective potential and four - point fermion interactions . the effective potential is composed of the gluon , quark and chiral sectors . the total effective potential corresponds to that of the pnjl model with one - loop gluon potential . this setup is effective enough to investigate vacuum structure and chiral properties at weak - coupling or small size of the compact dimension . in sec . [ sec : ps4 ] , we elucidated the vacuum and phase structure in @xmath0 gauge theory on @xmath3 . the theory with pbc adjoint quarks has three different phases in the @xmath86-@xmath87 space , including the deconfined phase ( @xmath0 , nonzero @xmath142 ) , the split phase ( @xmath1 , nonzero @xmath142 ) and the re - confined phase ( @xmath2 , zero @xmath142 with nontrivial global minima ) . the configuration of these phases in the phase diagram is consistent with that of the lattice calculations although we have no confined phase . by using the pnjl effective potential , we argued that chiral symmetry is slowly restored without clear phase transition when the size of the compact dimension is decreased . in this section we also studied vacuum and phase structure for the case with both fundamental and adjoint matters , and showed that the split phase is generically widened by adding pbc fundamental quarks . we consider that it is because one of the three possible minima for the split phase becomes more stable due to the center symmetry breaking . we showed that another @xmath2 phase with a negative value of @xmath142 emerges in this case ( pseudo - reconfined phase ) . in sec . [ sec : ps5 ] , we studied the vacuum and phase structure in @xmath0 gauge theory on @xmath4 . in the five - dimensional case we concentrate on the one - loop part of the effective potential the theory with pbc adjoint quarks again has the split ( @xmath1 ) and re - confined ( @xmath2 ) phases . introduction of fundamental quarks works to enlarge the split phase more effectively than the four - dimensional case . especially in the case with one adjoint and one fundamental quarks without parity pairs , the split phase overcomes the pseudo - reconfined phase and becomes a unique gauge - symmetry - broken phase . in sec . [ sec : ob ] , we discuss observables comparable to the lattice simulations . we list up observables including particle mass spectrum , in particular gauge boson mass ( lowest kk spectrum ) , polyakov - loop in the compact dimension , constituent mass . they can be good indications of exotic phase and properties both in the existing and on - going simulations . all through this paper , we treat pbc and apbc cases in a parallel manner . we note that the reference @xcite argues that gauge theory with pbc fermions has no thermal fluctuation , and thermal interpretation is inappropriate in such a case . this means that all the results here should be understood as topological phenomena . in the end of the paper , we discuss whether the lattice simulation can check our predictions . one of our main results is that enlargement of the split phase in the presence of fundamental fermions . this property would be observed in the on - going lattice simulations on four - dimesional gauge theory with a compact direction @xcite . the pseudo - reconfined phase can be also observed where the vev of polyakov loop becomes negative but different from that of the split phase . on the other hand , it seems difficult to show the first - order phase transition between the pseudo - reconfined and split phases . the lattice calculation can check whether chiral condensate remains finite in the re - confined phase even at a very small compacted size . the small chiral transitions coinciding with the gauge - symmetry phase transitions are subtle . if the lattice simulation succeeds to measure chiral susceptibility with high precision , this phenomena may be able to be observed . during preparing this draft , the collaboration @xcite kindly informed us that it also performed perturbative calculations on the phase diagram in gauge theory on @xmath3 and @xmath4 . it could be valuable for readers to compare the results from the two independent groups . the authors are grateful to e. itou for giving them chances to have interest in the present topics and reading the draft carefully . they are thankful to y. hosotani , h. hatanaka and h. kouno for their careful reading of this manuscript and valuable comments . they also thank g. cossu and j. noaki for the kind co - operation . k. k. thanks h. nishimura for useful comments . is supported by riken special postdoctoral researchers program . is supported by grand - in - aid for the japan society for promotion of science ( jsps ) postdoctoral fellows for research abroad(no.24 - 8 ) . in this appendix , we discuss the vacuum structure for the several gauge - symmetric cases in @xmath3 @xmath0 gauge theory . we concentrate on massless cases as @xmath239 . the potential contour plot for @xmath240 ( pure gauge theory ) is shown in fig . [ fig_p_g_3d ] . the effective potential has the minima at @xmath241 , @xmath242 , @xmath243 . since it means that the three polyakov - line phases are equivalent in the vacuum @xmath244 , the @xmath0 gauge symmetry is intact . the case with one fundamental fermion with anti - periodic boundary condition is shown in fig . [ fig_p_gfa_3d ] . the fundamental quark breaks @xmath138 symmetry explicitly and thus , two of the three minima become the meta - stable . if the boundary condition of fermion is changed to the periodic one , the global minima move to @xmath245 as shown in fig . [ fig_p_gfp_3d ] . on the other hand , the case for an adjoint fermion with anti - periodic boundary condition is similar to the pure gauge case as shown in fig . [ fig_p_gaa_3d ] since it keeps the @xmath5 center symmetry . in fig . [ 4d_phase_np ] , we depict the phase diagram for @xmath3 @xmath0 gauge theory with @xmath139 with pbc based on the nonperturbatively deformed gluonic potential . as an example , we consider the following modification from the perturbative potential in @xcite : @xmath251 where @xmath68 is the mass - dimension 1 parameter . we set the scale parameter as @xmath161 mev . the confined phase and the first - order phase transition show up , but it is merged into the gauge - broken re - confined phase . we note that the @xmath0 gauge symmetry is explicitly broken in the confined ( re - confined ) phase . su(3 ) theory with one pbc adjoint quark based on the effective potential with non - perturbative deformation ( @xmath161 mev ) . c stands for confined " , d stands for deconfined " and s for split " phases . compared to the perturbative one - loop case , the confined phase emerges but merges into the re - confined phase.,scaledwidth=40.0% ]

small perturbations in spherical and thin disk stellar clusters surrounding massive a black hole are studied . due to the black hole , stars with sufficiently low angular momentum escape from the system through the loss cone . we show that stability properties of spherical clusters crucially depend on whether the distribution of stars is monotonic or non - monotonic in angular momentum . it turns out that only non - monotonic distributions can be unstable . at the same time the instability in disk clusters is possible for both types of distributions . [ firstpage ] galaxy : centre , galaxies : kinematics and dynamics .

0710.1193

the study of the gravitational loss - cone instability , a far analog of the plasma cone instability , has begun with the work of v. polyachenko ( 1991 ) , in which a simplest analytical model of thin disk stellar cluster has been treated . the interest to the problem of stability of stellar clusters has been revived recently by detailed investigation by tremaine ( 2005 ) and polyachenko , polyachenko , shukhman , ( 2007 ; henceforth , paper i ) of low mass clusters around massive black holes . the both papers have considered stability of small amplitude perturbations of stellar clusters of disk - like and spherical geometry . tremaine ( 2005 ) has shown using goodman s ( 1988 ) criterion that thin disks with symmetric dfs over angular momentum and empty loss cone are generally unstable . by contrast , analyzing perturbations with spherical numbers @xmath0 and @xmath1 , he deduced that spherical clusters with monotonically increasing df of angular momentum should be generally stable . later we demonstrated ( see paper i ) that spherical systems with non - monotonic distributions may be unstable for sufficiently small - scale perturbations @xmath2 , while the harmonics @xmath3 are always stable . for the sake of convenience , we have used two assumptions . the first one is that the keplerian potential of the massive black hole dominates over a self - gravitating potential of the stellar cluster ( which does not mean that one can neglect the latter ) . then the characteristic time of system evolution is of the order of the orbit precessing time , which is slow , compared to typical dynamical ( free fall ) time . since a star makes many revolutions in its almost unaltered orbit , we can regard it as to be `` smeared out '' along the orbit in accordance with passing time , and study evolution of systems made of these extended objects . the second assumption is a so called _ spoke approximation _ , in which a system consists of near - radial orbits only . this approximation was earlier suggested by one of the authors ( polyachenko 1989 , 1991 ) . the spoke approximation reduces the problem to a study of rather simple analytical characteristic equations controlling small perturbations of stellar clusters . there are two questions that naturally arise in this context . first : does the instability remain when abandoning the assumption of strong radial elongation of orbits ? second : does the instability occur in spheres with monotonically increasing distributions in angular momentum if one consider smaller - scale perturbations with @xmath4 ? the aim of the paper is to provide answers to these questions . to achieve the task we use semi - analytical approach based on analysis of integral equations for slow modes elaborated recently in polyachenko ( 2004 , 2005 ) for thin disks , and in paper i for spherical geometry . following paper i , we shall restrict ourselves to studying monoenergetic models with dfs in the form @xmath5 the models specified by function @xmath6 are suitable for studying the effects of angular momentum distribution on gravitational loss - cone instability . on the other hand , the dirac @xmath7-function permits one to reduce the integral equations for slow modes to one - dimensional integral equations , and to advance substantially in analytical calculations . several arguments can be brought in favour of our simplified approach . first of all , the lynden - bell derivative ( see paper i , eq . 4.7 ) of the df with respect to angular momentum @xmath8 , keeping @xmath9 constant ( here @xmath10 is the radial action ) in the limit where the slow mode approximation is applicable , can be replaced by a derivative , keeping energy @xmath11 constant : @xmath12 because @xmath13 is small . thus , the derivative over energy is not included into the slow integral equation , and one can loosely say , that dependence on energy is only parametric . another argument is that the results of independent study by tremaine ( 2005 ) , who used a non - monoenergetic df , are in agreement with our conclusions . section 2 is devoted to spheres , section 3 to thin disks with symmetric dfs . the sections are organized alike . in the beginning we derive integral equations for initial distribution functions in the form ( [ eq:1.1 ] ) . then follow analytical and numerical investigations of these equations . we demonstrate that by contrast to the case of near - keplerian sphere , the loss - cone instability in disks takes place even for the monotonic df , @xmath14 , provided the precession is retrograde and the loss cone is empty : @xmath15 . 2 is complimented by stability analysis of models with circular orbits , which of course does nt belong to the class of monoenergetic models of ( [ eq:1.1 ] ) type . in the last , section 4 , we discuss the results and some perspectives of further studies . for the near - keplerian systems , the slow integral equation , which has been derived in our paper i ( see there eq . ( 4.8 ) ) , is neatly suited . in contrast to paper i , we shall not assume here strong elongation of orbits , i.e. we shall go beyond the spoke approximation . since energies of all stars are identical , the unperturbed df depends on one variable only . it is convenient to use a dimensionless angular momentum @xmath16 , where @xmath17 is the angular momentum on circular orbits : @xmath18 , @xmath19 is the central point mass , @xmath20 is the gravitational constant . the frequency of stellar radial oscillations @xmath21 , and the radius of the system @xmath22 are independent of the angular momentum . for shorthand notations , we shall omit the argument @xmath23 . the normalization constant @xmath24 is taken so that a mass of the spherical system surrounding the central mass is equal to @xmath25 ( we assume the ratio @xmath26 to be small : @xmath27 ) : @xmath28 if one defines the normalization of the dimensionless df over angular momentum @xmath29 ( see ( 1.1 ) ) as @xmath30 , then normalization factor @xmath24 in ( 1.1 ) is @xmath31 it allows to represent the kernel of the integral equation ( formula ( 4.8 ) in the paper i ) in the form @xmath32 where @xmath33 is the index of the spherical harmonic , @xmath34 ^ 2 $ ] and @xmath35 is the bessel function . the coefficient @xmath36 can be calculated analytically : @xmath37 . numerical calculations show decreasing @xmath38 with increasing the mode number @xmath33 : @xmath39 , @xmath40 , @xmath41 , @xmath42 , and so on . ] the functions @xmath43 satisfy to the condition @xmath44 ; their explicit form is given later . then substitution of the df in the form ( [ eq:1.1 ] ) leads to the following integral equation : @xmath45 where @xmath46 is the fourier harmonics of the radial part of the perturbed potential ( for more detail , see paper i ) , @xmath47 is the orbital precession rate , @xmath48 for odd @xmath33 , and @xmath49 for even @xmath33 . the coefficients @xmath50 are calculated by the formula @xmath51 ^ 2 } , & |l - s| \ { \rm even},\\ \\ 0 & |l - s|\ \ { \rm odd}. { \end{array}}\right . \end{aligned}\ ] ] recall that eq . ( [ eq:1s.1 ] ) is written in a noninertial reference frame centered on the mass @xmath19 . then , additional indirect potential arising from the acceleration of the frame should be considered ( see , e.g. , tremain 2005 ) @xmath52 where @xmath53 is the perturbation to the background df . tremain ( 2005 ) argued that for the secular perturbations , this indirect potential must be omitted . indeed , in studying secular evolution one should consider perturbations @xmath53 averaged over keplerian orbits . in this case the perturbed density is a superposition of contributions of individual orbits , averaged over their periods . a special feature of a keplerian orbit is that the average force from this orbit acting to the material point located in a focus of the ellipse is equal to zero . one must be careful however , since the perturbation is not well defined for orbits with low angular momenta . below we shall consider systems with `` small amount '' of stars with low angular momenta only ( see also discussion in sec . 2.2.1 ) . by changing the unknown function @xmath54\,\varphi_{s}(\alpha ) = \phi_{s}(\alpha)\ ] ] eq . ( [ eq:1s.1 ] ) can be reduced to the linear eigenvalue problem @xmath55 \varphi_{s}(\alpha ) = 2\,\omega_1\,\epsilon\,c_l \!\sum\limits_{s'=s_{\rm min}}^l s'^{\,2}\,d^{s'}_{l } \times\\ \times\int\limits_0 ^ 1 \omega_{\rm pr}(\alpha')\,\alpha ' \frac{d\,f(\alpha')}{d\alpha'}\ , { \cal k}_{\,s\,s'}^{(l)}(\alpha,\alpha')\,\varphi_{\,s'}(\alpha')\,d\alpha'.\end{gathered}\ ] ] for almost radial orbits , when @xmath56 or eccentricity @xmath57 , the precession rate is @xmath58.\end{aligned}\ ] ] for orbits with smaller eccentricity , the numerical coefficient preceding the small parameter @xmath59 is somewhat greater than @xmath60 . since one suggests that the characteristic frequencies of the problem under consideration are of the order of typical precession velocities , @xmath61 , it is convenient to change to the dimensionless frequencies , measured in the natural `` slow '' frequency : @xmath62 for the spherical systems , the precession is retrograde ( see tremaine 2005 , or paper i ) , so @xmath63 . then the dimensionless integral equation becomes @xmath64\,\varphi_{s}(\alpha ) = -2\,c_l \sum\limits_{s'=s_{\rm min}}^l s'^{\,2}\,d^{s'}_{l } \times\\ \times\int\limits_0 ^ 1 \nu(\alpha')\,\alpha'\ , \frac{d\,f(\alpha')}{d\alpha'}\ , { \cal k}_{\,s,\,s'}^{(l)}(\alpha,\alpha')\,\varphi_{\,s'}(\alpha')\,d\alpha',\end{gathered}\ ] ] to obtain the eigenfrequency spectrum for a model it is necessary to compute preliminarily the kernels @xmath65 ( universal for all models ) , and the precession rate profile @xmath66 for the given model . the integration over keplerian orbits is most conveniently expressed using the variable @xmath67 , which is connected with the current radius @xmath68 and the true anomaly @xmath69 of a star as follows : @xmath70 then after some transformations , the kernel @xmath71 can be reduced to the form @xmath72 where @xmath73 and @xmath74 specify the position of a star on the orbit with the eccentricity @xmath75 corresponding to the variable @xmath76 , and the notation @xmath77 is used . the expression for the precession rate can be obtained by transformation of expression ( 4.2 ) of tremaine ( 2005 ) ( see also paper i ) : @xmath78 and @xmath79 , where the density @xmath80 is defined by ( [ eq:1s.10 ] ) . for the monoenergetic models , the minimal and maximal radii are @xmath81 where @xmath82 , and @xmath83 is the minimal dimensionless angular momentum corresponding to the boundary of the loss cone . the function @xmath84 is a ratio of the mass of a spherical system inside the sphere of radius @xmath68 to the total mass @xmath25 : @xmath85 @xmath86 , and the density is calculated by the formula @xmath87 where @xmath88 . from here on we shall assume @xmath89 . using ( [ eq:1s.6 ] ) and ( [ eq:1s.8])([eq:1s.10 ] ) one can transform the expression for the scaled precession rate @xmath66 to a more universal form : @xmath90 where the kernel @xmath91 does nt depend on a df and equals to @xmath92 with @xmath93 here @xmath94 and @xmath95 for the near radial orbits @xmath96 , so that one obtains the above mentioned result ( [ eq:2.5 ] ) : @xmath97 . tremaine ( 2005 ) has noted that for an arbitrary distribution @xmath98 with empty loss cone , @xmath99 , a zero frequency lopsided mode @xmath0 must exist . the mode corresponds to a non - trivial perturbation arising under shift of the spherical system as a whole relative to the central mass . the perturbed potential in such a mode is @xmath100 , where @xmath101 is the displacement . in terms of the function @xmath102 , this perturbation has a form @xmath103 or in terms of the function @xmath104 from ( [ eq:1s.3 ] ) , @xmath105 one can check that ( [ eqv:2.15 ] ) and @xmath106 provided the condition @xmath107 is a solution of ( [ eq:1s.1 ] ) or ( [ eq:1s.3 ] ) for @xmath0 , taking into account the expressions ( [ eqv:2.12 ] ) and ( [ eqv:2.13 ] ) , written in the form @xmath108 and also the expression for the kernel @xmath109 @xmath110 the lopsided solution with zero frequency is specific for spherical systems . at the first glance , it defies common sense to argue that the stationary mode in which the center of mass of a spherical system does not coincide with the bh is physical . indeed , it seems that movement ( oscillations ) of the stellar cluster and the bh relative to the common center of mass must occur . however , it does not occur . it is , by all means , clear in the case of the empty loss cone of finite size , @xmath111 ( here @xmath83 is a minimal value of dimensionless angular momentum @xmath112 , for which @xmath113 ) . indeed , let us consider the spherically symmetric cluster . since the loss cone is finite , there is a spherical empty cavity of finite radius in the centre of sphere . according to the newton s first theorem ( binney & tremaine 1987 ) , in this cavity the bh does not experience a net gravitational force from the cluster . thus , if the bh is initially deposited at some arbitrary point within the cavity , it would remain at this position ( and hence , acceleration of stellar cluster due to non - coincidence of centers of mass does not occur ) . in the case when @xmath114 the situation is not so obvious , but the net force acting to the bh from the shifted spherical system can be zero as well . in order to assure this , one should write down the indirect potential taking into account the expression for perturbed density in zero lopsided mode @xmath115 : @xmath116 @xmath117 hence , the condition for omitting of the indirect potential is @xmath118 . in what follows we suppose this condition to be fulfilled . the conditon is not equivalent to the condition @xmath119 , imposed to the df for the existence of such a solution of our governing integral equation , but it is equivalent to the stronger condition : @xmath120 . indeed , it is easy to show that if @xmath121 for small @xmath112 then @xmath122 for small @xmath68 . so , the condition @xmath123 must be fulfilled . by contrast , in the disk systems the analogous @xmath124 zero mode does not exist , because there is no analog of the newton s first theorem . the very existence of zero modes is crucial for stability analysis of spherical clusters with monotonic distributions . indeed , the role of destabilizing contribution of the second term in the right side of ( [ eq:1s.3 ] ) falls off with increasing the number @xmath33 . so , it is expected that the most `` dangerous '' modes correspond to the lowest values of @xmath33 . but it turns out that @xmath0 mode is neutrally stable , and the next dangerous mode @xmath1 is stable . note , however , that such a reasoning is not valid for systems with non - monotonic distributions . by analyzing ( [ eq:1s.3 ] ) , it is easy to find one more analytical solution with the frequency @xmath125 at arbitrary values of @xmath33 , for the models with highly elongated orbits . first of all we note that the frequency of this stable mode corresponds to the resonance on the _ tail _ of a narrow distribution , and so it decays exponentially slowly . in this way the mode differs from the unstable modes of interest which have a resonance in a region where the distribution is localized , i.e. at @xmath126 ; so they have characteristic frequencies and growth rates of the order of @xmath127 . after setting @xmath128 in ( [ eq:1s.3 ] ) , omitting the second summand in l.h.s . , turning to the spoke approximation , and taking into account the equality @xmath129 , one finds @xmath130 it is essential that this _ high - frequency _ mode is independent of details of the df . note also that in the systems with prograde precession , this mode describes the well - known radial orbit instability ( instead of the neutral oscillations ) . using ( [ eq:1s.3 ] ) , one can prove two important statements : 1 . for spherical system models with monotonic distributions @xmath131 , the eigenfrequency squared , @xmath132 , must be a real number . this means that either the instability is absent at all , or aperiodic instability with @xmath133 occurs . 2 . rotating ( or oscillating ) unstable modes may appear only in models with non - monotonic distributions . let us write eq . ( [ eq:1s.3 ] ) in the form @xmath134 where @xmath135 we multiply both parts of eq . ( [ eq:4.1 ] ) by @xmath136 , sum the result over @xmath137 ( asterisk means the complex conjugation ) , and integrate over @xmath112 with the weight @xmath138 . then we obtain @xmath139 \,[g(\alpha')\,\varphi_{s'}(\alpha')].\end{gathered}\ ] ] the reality of the coefficients of @xmath140 in the l.h.s . of ( [ eq:4.2 ] ) and the first term in the r.h.s . is evident . with the help of ( [ eq:1s.7 ] ) , one can show that the kernel in eq . ( [ eq:4.2 ] ) has the following property of symmetry : @xmath141 so , one can readily see that the second term in the r.h.s is real also . consequently , taking the imaginary part of eq . ( [ eq:4.2 ] ) , one obtains @xmath142 > from the last equality , the statements formulated above follow immediately . if the function @xmath138 ( or , equivalently , @xmath143 ) is constant - sign , then the integral should be non - zero , and so @xmath144 . in contrast , when @xmath145 , the integral must be equal to zero . consequently , the function @xmath138 should change its sign , i.e. df @xmath131 is non - monotonic . let us explain the term _ variational principle _ used in the title of this subsection . consider a dynamic equation in the form @xmath146 . provided that `` elasticity operator '' @xmath147 is hermitian , the dynamic equation may be obtained from the conditions @xmath148 and @xmath149 . here @xmath150 and @xmath151 are considered formally as independent variations while the functional @xmath152 is @xmath153 ( @xmath154 is a nonnegative weight function ) . in such a case it is used to speak about the variational ( or energy ) principle ( see , e.g. , review by kadomtsev 1966 on mhd - stability of plasma ) . from the other hand it is easy to see that if @xmath155 for any nontrivial @xmath101 then @xmath152 is real . thus usually ( as is the case in mhd - stability theory of plasma where @xmath147 is hermitian and @xmath156 ) the notions `` variational principle '' and reality of @xmath152 are identical . however , in our case the condition latexmath:[$\int @xmath101 unless the df is monotonic . assuming that this condition is fulfilled , following the tradition that originates from plasma physics we speak that the variational principle takes place . only in this case the dynamical equation can be interpreted mechanically , in terms of elastic forces . evidently , the condition ( [ eq:4.4 ] ) is a serious obstacle to obtain unstable rotating modes . so , one might want to get round this obstacle . for instance , if we slightly change the initial monotonically increasing df in a narrow region near @xmath158 , to vanish quickly but smoothly , then a modified system would be practically indistinguishable from the initial one . but then the variational principle breaks down . the question can be formulated as follows : may the discontinuous vanishing of @xmath131 at @xmath158 be considered as the violation of monotony for the df ? importance of this question is known since stability study of stellar systems with isotropic dfs , @xmath159 ( antonov , 1960 , 1962 ) . the variational principle there required a df to be decreasing function of energy @xmath11 , @xmath160 , everywhere . the systems with @xmath161 need separate examinations , that was carried out in some cases ( see , e.g. , antonov , 1971 , kalnajs , 1972 , polyachenko and shukhman , 1972 , 1973 , fridman and polyachenko , 1984 ) . an essential difference of the second type of dfs is in jumps to zero at the phase space boundary @xmath162 . in fact , we have in this case an interval degenerated into the single point @xmath162 where @xmath163 is decreasing . we checked numerically a possibility of the instability development connected with the maximum on the edge of the distribution function s domain . for this purpose , number of models smoothed near @xmath158 were computed . the computations showed no sign of instability , in contrast to isotropic distributions , @xmath159 . the reason for the tolerance of our present models is in fact that the kernels @xmath164 of integral operators in ( [ eq:4.1 ] ) vanish for the circular orbit @xmath158 , so details of the initial distribution near circular orbits can not affect much solutions of the integral equation ( [ eq:4.1 ] ) . roles of different terms in eq . ( [ eq:4.1 ] ) can be easily understood . when @xmath165 , the first term of the right side in eq . ( [ eq:4.1 ] ) favors stabilization , while the second term gives destabilization ( taking into account that the operator involved into this construction is self - adjoint and positively defined ) . in principle , this destabilizing effect could lead to instability . however , this is not the case because the stabilizing contribution exceeds destabilizing one in all cases considered both by tremaine ( 2005 ) , and in the present paper ( see the following sections ) . instability boundaries in terms of the angular momentum dispersion @xmath166 found in paper i for the monoenergetic df with @xmath167 ( @xmath168 is the normalization constant , @xmath169 is the dimensionless angular momentum dispersion , @xmath170 is the real number ) have a qualitative character only : formally , these boundaries lie outside the validity of the spoke approximation , since @xmath171 . obtaining such critical dispersions means only that the spoke models , in which @xmath172 by definition , are certainly unstable . so the quantitative determination of these boundaries with help of the exact integral equation is required . the _ power exp _ model ( [ eq:2.1 ] ) is studied in sec . 2.3.1 . in sec . 2.3.2 we study a simplest _ heaviside _ model consisting of two steps ( at @xmath173 and at @xmath174 ) ( both in the spoke approximation framework and using exact integral equation ) : @xmath175,\ \ h_1<h_2<1\end{aligned}\ ] ] ( @xmath176 denotes the heaviside function ) . finally , in sec . 2.3.3 we consider the _ log exp _ model with df @xmath177 for @xmath178 , and @xmath179 for @xmath180 , i.e. , with the empty loss cone ( @xmath168 is the normalization constant ) . following paper i , here we consider the stability of models ( [ eq:2.1 ] ) with @xmath181 and @xmath182 relative to the spherical harmonic @xmath183 . we remind that at the limit @xmath172 , both these models were unstable ( the stability boundaries obtained using spoke approximation were @xmath184 for @xmath181 , and @xmath185 for @xmath182 ) . for distribution ( [ eq:2.1 ] ) one finds @xmath186 particularly , in the case @xmath172 , the normalization constant is @xmath187 . from ( [ eq:1s.10 ] ) , we obtain @xmath188 further calculations of the density ( [ eq:1s.10 ] ) and precession rate ( [ eqv:2.12 ] ) profiles should be evaluated numerically . solutions of the integral equation ( [ eq:1s.3 ] ) for @xmath181 and @xmath182 show similar behavior . at small values of @xmath169 , each model has one unstable mode . with increasing the dimensionless angular momentum dispersion @xmath169 , the growth rate of the instability decreases , and then vanishes at some critical value @xmath189 : for the model @xmath181 , @xmath190 , for the model @xmath182 , @xmath191 ( see fig.[fig:0 ] ) . we conclude that the spoke approximation in this case is qualitatively correct , but quantitatively poor . the instability becomes saturated at some critical value @xmath189 , while the discrepancy between exact and approximate values of @xmath189 for both models are not small . apart from the unstable mode , the spectrum of each model has a discrete mode , the growth rate of which is equal to zero at small @xmath169 , and becomes negative with increasing @xmath169 . this is just that weakly decaying mode with the frequency @xmath192 ( at @xmath193 ) which was mentioned in sec . the dependence of the frequency on @xmath33 for this mode was one of the tests for numerical code of the integral equation solver . another test was detecting the zero lopsided mode @xmath0 mentioned in sec . the third test was the evaluation of @xmath194 dependence in the spoke approximation limit . assuming that @xmath195 , and using @xmath196 , @xmath197 , and @xmath198 in ( [ eq:1s.1 ] ) , one can obtain the equation for the @xmath183 mode @xmath199 by setting the r.h.s . to zero , one obtains an unstable mode for each @xmath170 : @xmath200 for @xmath181 , and @xmath201 for @xmath182 . the same values obtained from solution of the exact integral equation ( [ eq:1s.3 ] ) for @xmath202 are @xmath203 and @xmath204 , correspondingly . vs. dimensionless angular momentum dispersion @xmath169 of the mode @xmath183 for models @xmath181 ( diamonds ) and @xmath182 ( circles ) . dashed lines show the asymptotic behavior obtained using spoke approximation equation ( [ eq:2.power_sa ] ) : @xmath205 and 0.532 for @xmath181 and 3 respectively ( exact solution for @xmath202 gives 0.185 and 0.529).,width=385 ] the simplest non - monotonic model that allows to progress rather far by analytical methods is the model with a piecewise constant distribution over momentum ( [ eqv:2.2 ] ) . in other words , we assume the df to be non - zero only within the interval @xmath206 , where it is taken constant . when studying stability of discontinuous distributions such as ( [ eqv:2.2 ] ) , it is more convenient to start with the integral equation in the form ( [ eq:1s.1 ] ) . substituting ( [ eqv:2.2 ] ) into eq . ( [ eq:1s.1 ] ) , one obtains @xmath207.\end{gathered}\ ] ] let us turn again to the natural slow scale of frequencies according ( [ eq:1s.2s ] ) and then substitute in ( [ eq:2.15 ] ) particular values @xmath173 and @xmath174 . for brevity sake , the following designations are used : @xmath208 , @xmath209 . we have @xmath210,\end{gathered}\ ] ] @xmath211.\end{gathered}\ ] ] this set of equations relative to @xmath212 and @xmath213 , ( @xmath214 $ ] ) can be reduced to the standard linear set . introducing new unknown functions @xmath215 one obtains @xmath216,\end{gathered}\ ] ] @xmath217.\end{gathered}\ ] ] the precession rates in these equations can be expressed through the complete elliptical integrals @xmath218 and @xmath219 : @xmath220 where @xmath221 and the function @xmath222 is @xmath223.\ ] ] in the limit @xmath224 , the frequencies @xmath225 and @xmath226 are coincident : @xmath227 , where @xmath228 , and @xmath229 . note that a determinant of the set of equations ( [ eq:2.20 ] ) and ( [ eq:2.21 ] ) has a rank @xmath230 $ ] . particularly , for the mode @xmath0 , the rank is equal to 2 . roots of the determinant are calculated numerically . the difference @xmath231 has a meaning of dispersion , i.e. , it is analogous to the parameter @xmath169 in our models with smooth distributions . a simple analytical task is to ascertain that @xmath232 is the eigenvalue of the determinant for @xmath0 . we have for @xmath233 @xmath234 where @xmath235 . in particular , @xmath236 @xmath237 setting @xmath232 in the determinant of the set ( [ eq:2.20 ] ) and ( [ eq:2.21 ] ) , and using the expressions for the elements of the kernel ( [ eq:2.29 ] ) and ( [ eq:2.30 ] ) , we can show that it is equal to zero identically . this just means the occurrence of a zero mode in the spectrum . another root @xmath238 for @xmath0 is positive for any values of @xmath239 and @xmath240 , which agrees with our previous conclusion ( paper i ) that the instability is absent for the mode @xmath0 . on the plane @xmath241 for the heaviside model ( left : exact calculations , right : spoke approximation calculations ) : _ a _ and _ b _ for the mode @xmath183 ; _ c _ and _ d _ for the mode @xmath242 , _ e _ and _ f _ for mode @xmath243 . the spoke approximation is reliable in the lower left corner of the domain . it is also seen that for @xmath183 , the growth rate sharply decreases when the ratio of the difference @xmath244 ( the analog of the dispersion @xmath169 for models with smooth dfs ) to the size of the loss cone , @xmath239 , becomes greater than 2.2.,width=385 ] on the plane @xmath241 for the heaviside model ( left : exact calculations , right : spoke approximation calculations ) : _ a _ and _ b _ for the mode @xmath183 ; _ c _ and _ d _ for the mode @xmath242 , _ e _ and _ f _ for mode @xmath243 . the spoke approximation is reliable in the lower left corner of the domain . it is also seen that for @xmath183 , the growth rate sharply decreases when the ratio of the difference @xmath244 ( the analog of the dispersion @xmath169 for models with smooth dfs ) to the size of the loss cone , @xmath239 , becomes greater than 2.2.,width=385 ] on the plane @xmath241 for the heaviside model ( left : exact calculations , right : spoke approximation calculations ) : _ a _ and _ b _ for the mode @xmath183 ; _ c _ and _ d _ for the mode @xmath242 , _ e _ and _ f _ for mode @xmath243 . the spoke approximation is reliable in the lower left corner of the domain . it is also seen that for @xmath183 , the growth rate sharply decreases when the ratio of the difference @xmath244 ( the analog of the dispersion @xmath169 for models with smooth dfs ) to the size of the loss cone , @xmath239 , becomes greater than 2.2.,width=385 ] it is useful to derive equations in ( [ eq:2.20 ] ) and ( [ eq:2.21 ] ) in the spoke limit , when the distribution is localized in a region of small @xmath112 . this means that we suggest @xmath245 , @xmath246 , @xmath247 , set in ( [ eq:2.15 ] ) @xmath248 , @xmath249 , and finally obtain @xmath250.\end{aligned}\ ] ] for the precession frequencies @xmath225 and @xmath226 , one has in this limit @xmath251 , @xmath252 introducing @xmath253 , let us write down , e.g. , the characteristic equation for the mode @xmath183 . in this case @xmath254 , @xmath255 , hence @xmath256.\end{gathered}\ ] ] due to the denominator @xmath257 , the role of `` self - gravity '' may be made sufficiently large in spite of small parameter @xmath258 . this may give the oscillating instability under certain conditions for @xmath239 and @xmath240 . the limiting solutions serves a test for the model with arbitrary @xmath239 and @xmath240 . the results for the modes @xmath183 , @xmath242 and @xmath243 are presented in fig.[fig:1]_a f_. they show the boundaries of instability domains on the plane @xmath241 . left panels show the results of computations from the exact set of equations ( [ eq:2.20 ] ) and ( [ eq:2.21 ] ) ; right panels the results from the spoke equation for this model ( [ eq:2.31 ] ) . it is seen that in the region @xmath245 , @xmath259 , the results obtained from the spoke equation and those from the exact equations do coincide . location of the growth rate maxima at figs.[fig:1]_a _ , 1_c _ , 1_e _ , as well as the values of growth rates are practically the same . are explained by more complicated structure of the characteristic equation for this mode , compared to the modes @xmath183 and @xmath242 . ] we conclude that for the non - monotonic df the instability is insensitive to a number @xmath33 of the mode . this is a characteristic feature for the loss - cone instability . recall that in models with monotonic distributions , the destabilizing term quickly decreases with increasing spherical number @xmath33 of the mode . in some numerical models ( see , e.g. , cohn , kulsrud , 1978 , berczik et . 2005 , spurzem et . al . 2005 ) the initially isotropic distribution transforms under action of a massive black hole into one monotonically increasing with angular momentum , @xmath260 , where @xmath83 defines the minimum angular momentum of a star which is not absorbed by the black hole . in this section , we consider the stability of df ( [ eq:2.3 ] ) . for it , one finds @xmath261,\end{gathered}\ ] ] where @xmath262 is the exponential integral . density profile is obtained from ( [ eq:1s.10 ] ) : @xmath263 much as in the power exp model considered above , the calculations of the precession rate should be performed numerically . the qualitative pattern of the spectrum for this model is similar to that of the power exp model : when the dispersion is not too large , two discrete modes occur , one of which being unstable . in fig.[fig:2 ] ( left panel ) , the isolines @xmath264 on the plane of parameters @xmath265 for the mode @xmath183 are presented . comparison of the figures [ fig:1 ] and [ fig:2 ] shows qualitative coincidence of growth rates behavior on the dispersion of df and the size of the loss cone in this and heaviside models . the right panel shows the ratio of imaginary part to real part of @xmath266 vs. dimensionless angular momentum dispersion @xmath169 for several values of loss cone size parameter @xmath83 . for the log exp model . left : isolines @xmath267 on the plane @xmath268 . right : the ratio of the growth rate @xmath269 to the real part of the frequency @xmath270 , for values @xmath271.,title="fig:",width=170 ] for the log exp model . left : isolines @xmath267 on the plane @xmath268 . right : the ratio of the growth rate @xmath269 to the real part of the frequency @xmath270 , for values @xmath271.,title="fig:",width=170 ] instability of spherical clusters around massive black holes was first studied by tremaine ( 2005 ) . he considered distributions of the form @xmath272 in the domain @xmath273 , @xmath274 ( and zero outside this domain ) ; @xmath275 , @xmath276 are the action variables , @xmath277 ; @xmath278 and @xmath83 are the real parameters . in the distribution , the loss cone is empty for dimensionless angular momentum @xmath180 . tremaine studied the most large - scale perturbations with the spherical indices @xmath0 and @xmath1 . in this section we consider two monoenergetic models . the first one is @xmath279 where @xmath280 characterize size of the loss cone , @xmath168 is the normalization constant . dependence of distributions ( [ eq:3.1 ] ) and ( [ eq:1.1 ] ) with @xmath6 from ( [ eq:3.2 ] ) on the angular momentum is identical . just this dependence is crucial for stability or instability of each specific distribution . stability of distribution ( [ eq:3.2 ] ) for arbitrary values of @xmath33 is proved in sec . another distribution ( sec . 2.4.2 ) is the simplest _ monotonic heaviside _ model in a form of the step - like df , @xmath281 the factor @xmath282 is added to reflect that the df domain is bounded by circular orbits . in sec . 2.4.3 , we prove the stability of spherical systems ( in the field of a central massive body ) , all orbits of which are circular . distribution in the form ( [ eq:3.2 ] ) allows to calculate the density @xmath80 explicitly . using ( [ eq:1s.10 ] ) one obtains : @xmath283 where the normalization constant satisfies the relation : @xmath284.\ ] ] from the condition @xmath285 , it follows that @xmath286 in presence of the loss cone , @xmath111 , the radius of the system @xmath287 is less than the apocenter radius for a radial orbit , @xmath288 , since stars with low angular momentum , @xmath289 , are absorbed by the black hole . integration ( [ eq:3.5 ] ) gives for the density @xmath290 . \ ] ] as is seen the density vanishes smoothly at the boundaries of the spherical layer @xmath291 and @xmath292 . the expressions for the precession velocity are defined by the formulas ( [ eq:1s.8])([eq:1s.9 ] ) . in fig.[fig:3 ] , the frequency spectrum , @xmath293 , of the spherical harmonic @xmath0 for the _ log_-model is presented for different values of the parameter @xmath83 . all calculations detect zero modes . for @xmath0 . in all cases the eigenfrequencies are neutral and consist of one zero mode ( circled point ) and continuous part of spectrum ( points run into a line ) . to save room the spectra are shown in a single plot , but vertically separated from one another ( @xmath83 grows from bottom to top).,width=377 ] for other values of @xmath33 ( we considered @xmath294 ) , the discrete modes are absent in the frequency spectra for all @xmath83 . the spectra are continuous and lie at the region of real and positive values of @xmath238 . we conclude that the log models are turned out to be stable . following the procedure described in sec . 2.3.2 for the unstable heaviside model , one can derive the following equation for the distribution function ( [ eq:3.4 ] ) : @xmath295 where @xmath296 , @xmath297 $ ] . it is easy to see that for @xmath0 there is a zero mode only , since @xmath298 . introducing new variables @xmath299 one can reduce ( [ eqv:5.6 ] ) to a standard linear set @xmath300 where the matrix @xmath301 is hermitian . so , the eigenfrequencies @xmath302 are real . here again we have a competition of opposite factors , expressed by the first and the second terms in the r.h.s of ( [ eqv:5.7 ] ) . to conclude whether the instability occurs , we must find numerically zeros of the determinant @xmath303 as a function of @xmath302 for @xmath304 . a rank of the determinant is equal to @xmath305 $ ] . , for @xmath306.,width=340 ] ) eigenfrequencies squared , @xmath307 , for @xmath308 and 7.,width=340 ] the results are represented in figs . [ fig:4 ] and [ fig:5 ] . this figures show the dependence of @xmath309 , @xmath310 $ ] as a function of the loss cone size @xmath83 . the instability is clearly absent . each eigenvalue @xmath311 has only a weak dependence on @xmath83 and approximately equals to @xmath312 . the least stable mode is @xmath183 ( see fig . [ fig:5 ] ) , but it is still far from instability . let us consider the simplest monotonic model , in which all orbits are circular . in this section we do not assume a distribution to be monoenergetic , otherwise the density distribution would be degenerated to a thin spherical layer . let us assume that the df is @xmath313 $ ] , where @xmath314 is normalization factor and @xmath11 changes in some range @xmath315 . in terms of radial and transverse velocities @xmath316 and @xmath317 , the df is : @xmath318 , \ \ v_0(r)=r\omega(r),\ ] ] where @xmath319 is the angular velocity of a star on the circular orbits . this velocity is determined by a balance of the centrifugal force and a sum of the gravitational forces from the central body and from the spherical cluster : @xmath320 here we also suggest that @xmath321 in this approximation , the orbits are near - keplerian , and the following relations are valid : @xmath322 for the precession rate we have ( see also tremaine , 2001 ) @xmath323,$ ] or , taking into account ( [ eqv:5.8 ] ) , @xmath324 since @xmath325 , one has the following scalings : @xmath326 , @xmath327 , and for slow modes @xmath328 . we start from the equation derived by palchik et . al . ( 1970 ) , for the models with circular orbits ( the equation can also be found in the monograph by fridman and polyachenko ( 1984 ) . it has a form : @xmath329 where @xmath330 is a radial part of the potential perturbation @xmath331 , coefficients @xmath332 and @xmath333 are @xmath334[\omega\!-\!(s\omega\!+\!\kappa)]},\end{aligned}\ ] ] @xmath335[\omega- ( s\omega+\kappa)]}\right]+ \right.\\ + \omega_0 ^ 2\,\left[\frac{4s\omega(\omega - s\omega)+ s^2\bigl[(\omega - s\omega)^2 + 4\omega^2-\kappa^2\bigr ] } { ( \omega - s\omega)^2\ , [ \omega-(s\omega-\kappa)][\omega- ( s\omega+\kappa)]}+\right.\\ \left.\left . \frac{(l+s+1)(l - s)}{(\omega - s\omega)\,(\omega - s\omega-2\omega)}\right]\right\}.\end{gathered}\ ] ] now we need to distinguish between even and odd values of @xmath33 , since @xmath33 and @xmath137 should be of the same parity ( i.e. both even or both odd ) . for _ even _ @xmath33 , the dominating contributions is expected from @xmath336 and @xmath337 . however , one can see that for even @xmath33 , the contributions from @xmath336 and @xmath337 cancel each other . indeed , setting @xmath338 one has : @xmath339 after taking into account the relation @xmath340\bigr\},\ ] ] one obtains : @xmath341 . so the equation ( [ eqv:5.9 ] ) is reduced to the trivial relation @xmath342 , which means the absence of the slow density perturbations . for _ odd _ @xmath33 the terms @xmath343 give the main contribution to the sum , while other terms ( @xmath344 ) are beyond the accuracy of the slow mode equation . thus , one has : @xmath345 to study this case we transform the differential equation ( [ eqv:5.9 ] ) with @xmath346 and @xmath347 from ( [ eqv:5.12 ] ) to an integral equation . eq.([eqv:5.9 ] ) can be represented in the form of the poisson equation @xmath348 with the perturbed density @xmath349\biggr.\nonumber\\ \biggl . & + \frac{d}{dr } \bigl[\,\frac{2}{r}\,s(\omega^2,r)\bigr]\,\chi_l\biggr\},\end{aligned}\ ] ] where @xmath350 . the solution of eq . ( [ eqv:5.13 ] ) in the integral form is : @xmath351 where the kernel @xmath352 or , substituting ( [ eqv:5.14 ] ) into ( [ eqv:5.15 ] ) , and integrating by parts , @xmath353\,\times\\ \times\left[\frac{d\,{\cal f}_l(r , r')}{dr'}+\frac{2}{r'}\,{\cal f}_l(r , r')\right].\end{gathered}\ ] ] applying the operator @xmath354 to both parts of eq . ( [ eqv:5.17 ] ) and denoting @xmath355 we obtain an integral equation ) can also be derived from the general `` slow '' integral equation , by considering of circular orbit limit . however , that derivation is much more cumbersome than one given here . ] @xmath356 with the new symmetrical kernel @xmath357 introducing the new function @xmath358 we obtain the required integral equation @xmath359\,z_l(r)=\\ -\frac{2d_l^1}{2l+1 } \int dr'\,\omega_{\rm pr}(r)\,\omega_{\rm pr}(r')\,{\cal k } _ l(r , r')\,z_l(r'),\end{gathered}\ ] ] with the kernel @xmath360 . since the kernel defines a self - adjoint integral operator , all eigenfrequencies @xmath152 should be real . to determine whether negative values of @xmath152 are possible , let us write out the kernel @xmath361 explicitly : @xmath362 it contains two contributions : the first is negative , the second is positive . substituting ( [ eqv:5.20 ] ) into ( [ eqv:5.19 ] ) , one finds @xmath363 for @xmath0 one can see that @xmath364 satisfies this equation since @xmath365 . however , the most interesting fact consists in stability of all higher modes , @xmath366 . indeed , since @xmath367 for @xmath366 , and the integral operator in the r.h.s . is positively defined , we conclude that all eigenvalues , @xmath152 , are positive . consequently , the instability is absent in the limit of circular orbits . the result is universal and does not depend on a particular choice of the model @xmath368 . in this section we shall consider the monoenergetic distributions of ( [ eq:1.1 ] ) type assuming the function @xmath131 to be even , @xmath369 . the function @xmath98 is normalized as follows : @xmath370 which gives the normalization constant @xmath371 provided that @xmath372 . the integral equation for slow modes ( paper i ) can be represented in the form : @xmath373 or , using the evenness of @xmath374 which stems from evenness of @xmath131 , oddness of @xmath375 , and symmetry properties of the kernel , @xmath376 , @xmath377 where @xmath266 and @xmath66 are the dimensionless pattern speed and the dimensionless precession rate : @xmath378 changing the unknown function , integral equation ( [ eq : disk.4a ] ) takes the form of linear eigenvalue problem : @xmath379\,\psi(\alpha)= \frac{2\,c_m}{\pi^3 } \int\limits_{0}^1 \nu(\alpha')\,\frac{df}{d\alpha'}\ , { \cal k}_m(\alpha,\alpha')\,{\psi}(\alpha')\,d\alpha ' . \end{aligned}\ ] ] the kernel functions for thin disks @xmath380 can be transformed from the corresponding expression in paper i to a suitable form as follows : @xmath381 where dependence of @xmath68 and anomaly @xmath69 on @xmath67 and @xmath382 are the same as in the spherical case ( [ eq:1s.6 ] ) , but the function @xmath383 is @xmath384 as before , the kernel @xmath380 is normalized to unity : @xmath385 , which means @xmath386 equal to @xmath387 the latter formula immediately follows from ( [ eq : disk.5 ] ) if one reminds that for radial orbits @xmath388 , @xmath389 @xmath390^{-1/2}$ ] . for the lowest azimuthal numbers , functions @xmath383 can be expressed through elliptical integrals of the first and the second kind @xmath391 and @xmath392 : @xmath393 @xmath394,\end{aligned}\ ] ] where @xmath395 , @xmath396 , @xmath397 . using ( [ eq : disk.8 ] ) and ( [ eq : disk.9 ] ) one can obtain numerically @xmath398 , @xmath399 . for the surface density we have : @xmath400 where @xmath401 and @xmath402 as for spheres . the relation between the precession rate and the potential @xmath403 is the same as in the spherical systems , ( see tremaine 2005 and paper i ) @xmath404 but the relation between the potential and the surface density is much more complicated ( see , e.g. , tremaine , 2001 ) @xmath405 . \end{aligned}\ ] ] using ( [ eq : disk.10])([eq : disk.3.16 ] ) one obtains a suitable expression for the scaled precession rate @xmath66 ( [ eq : disk.4s2 ] ) in the integral form @xmath406 where @xmath91 is a universal function ( i.e. does not depend on form of the distribution ) : @xmath407 @xmath408 . \end{aligned}\ ] ] here @xmath409 . the integral of the first term is understood in the principle value sense . using the same trick as in sec . 2 , one can change to new integrating variables @xmath67 and @xmath76 , where @xmath410 and @xmath411 . then , for @xmath412 one obtains @xmath413 . \end{gathered}\ ] ] as we see from sec . 2 , for spherical systems with the monotonic df , the variational principle takes place . besides , for @xmath0 and the empty loss cone , a zero frequency solution exists which stands for a sphere displacement from the massive center ; all other eigenmodes being stable . a thin disk is completely different . the displacement is no longer an eigenmode . moreover , models with analogous distributions are turn out to be unstable . let us prove the instability of lopsided @xmath124 mode provided that @xmath414 note that spherical models with the analogous condition are stable . disks with even dfs satisfying condition ( [ eq : disk 3.20 ] ) obey the variational principle , which means the eigenfrequencies squared are real . so one can formulate a _ sufficient _ condition of instability for @xmath124 azimuthal perturbations as follows : if the loss cone is empty ( @xmath15 ) , the df is monotonically increasing @xmath415 , and the precession is retrograde for all values of angular momentum ( @xmath416 ) , then @xmath124 perturbations are unstable . to prove the statement we shall use integral equation ( [ eq : disk.4a ] ) , in which @xmath417 with @xmath418 is assumed : @xmath419 where operator @xmath420 is @xmath421 now we consider another eigenvalue problem @xmath422 eigenvalues @xmath423 of the problem ( [ eq : disk 3.21 ] ) correspond to eigenvalues @xmath424 of the problem ( [ eq : disk 3.23 ] ) . let us define an inner product as @xmath425 where the weight function @xmath426>0.$ ] operator @xmath427 has the following properties . @xmath427 is hermitian , i.e. @xmath428 \2 . @xmath427 is continuous , when @xmath429 . one might think that the first term @xmath430 in the r.h.s . of ( [ eq : disk 3.22 ] ) breaks down the continuity , which in turn means that system of proper functions is incomplete . however , this is not the case , since @xmath430 can be absorbed by introducing new eigenvalue @xmath431 in ( [ eq : disk 3.23 ] ) . since @xmath131 is even , for smooth df we have @xmath432 , @xmath433 . this condition guarantee the weight function @xmath434 to be finite even for @xmath435 , despite @xmath436 at @xmath437 . @xmath427 is positive definite at sufficiently large @xmath423 . this is evident , since @xmath438 , and the second term in ( [ eq : disk 3.22 ] ) becomes small at large @xmath423 . > from the first two properties it follows that for fixed @xmath439 eigenvalues @xmath440 , @xmath441 of @xmath420 are real , and the system of proper functions is complete . the third property means that at large @xmath423 all eigenvalues @xmath440 are positive . if we find a test function @xmath442 , for which the scalar product @xmath443 is negative , it mean @xmath444 is not positive definite for the given @xmath445 . so , at least one eigenvalue must be negative : @xmath446 . this minimal eigenvalue @xmath447 increases with @xmath423 and becomes positive , as all other @xmath440 . we conclude that there must be a value of @xmath423 , @xmath448 for which @xmath449 . this value is an eigenvalue for ( [ eq : disk 3.22 ] ) , which means the existence of the eigenmode describing aperiodic instability with growth rate @xmath423 . for the test function @xmath442 one can take a displacement of the disk from the center , which is similar to the sphere displacement ( [ eqv:2.15 ] ) and correspond to the lopsided perturbation @xmath124 : @xmath450 , and @xmath451 . one can show that @xmath452 let the l.h.s . be @xmath453 , or , explicitly @xmath454 ^ 2\ , \nu(\alpha)\ , + \frac{2 c_1}{\pi^3}\int\limits_0 ^ 1 d\alpha \,\frac{e(\alpha)}{\alpha}\times\\ \times\frac{df(\alpha)}{d\alpha}\,\int\limits_0 ^ 1 d\alpha ' \ , \frac{e(\alpha')}{\alpha'}\,\frac{df(\alpha')}{d\alpha'}\,{\cal k}_1(\alpha,\alpha ' ) . \end{gathered}\ ] ] after some lengthy manipulations using ( [ eq : disk.5 ] ) , ( [ eq : disk.8 ] ) , ( [ eq : disk 3.19 ] ) , and condition @xmath15 , one can show that @xmath455 is positive , so inequality ( [ eq : disk 3.24 ] ) is fulfilled . tremaine ( 2005 ) has also obtained a sufficient condition for a lopsided mode in the symmetrical disk using goodman s ( 1988 ) criterion . his condition , however , differs from ours . namely : if the loss cone is empty , @xmath99 , and @xmath456 throughout the radial range containing most of the disk mass , then disk is unstable with respect to @xmath124 perturbations . this formulation does not use the requirement for precession to be retrograde and the df to be monotonically increasing , although monotonic increase of @xmath457_{l=0}=0 $ ] , @xmath458_{l=0}>0 $ ] is implied , at least for small angular momentum . thus , the comparison between spherical and disk case can hardly be made , unless the conditions of stability is formulated in similar terms . to perform the comparison , we give our own criterion that follows directly from the integral equation . it needs to be emphasized that the sufficient condition by tremaine ( 2005 ) is different . lack of the condition for the sign of precession possibly means that his criterion include two type of instability simultaneously the radial orbit instability arising in disks with prograde precession , and the loss cone instability which require retrograde precession . for disks composed of near - radial orbits , tremaine s condition gives the result obtained in paper i : a disk with symmetrical distribution @xmath131 obeying the conditions @xmath432 , @xmath433 is unstable if the precession is retrograde . in turn , precession of near - radial orbits is retrograde if @xmath456 . to support the mathematical rationale given above and to provide a basis for possible simulations , it is useful to obtain eigenfrequencies of unstable modes for particular models . here we consider the _ power exp _ model with symmetrical distribution @xmath459 where the normalization constant is @xmath460 for @xmath172 the constant @xmath461 . distributions become monotonic in the interval @xmath462 $ ] when @xmath463 . note , that when @xmath464 df is simply @xmath465 in the interval @xmath466 $ ] and does nt depend on @xmath169 . evaluation of integral equation ( [ eq : disk.4s ] ) requires preliminary calculations of the kernel function @xmath467 using ( [ eq : disk.5 ] ) , and the scaled precession rate @xmath66 using ( [ eq : disk 3.17 ] ) and ( [ eq : disk 3.19 ] ) . for brevity , we skip the details here , just noting that calculation of function @xmath468 is turn out to be rather difficult numerical task . the calculations show that for the model ( [ eq : disk_4.1 ] ) the precession rate @xmath66 is retrograde for all @xmath112 , i.e. @xmath416 . the results for values of azimuthal number @xmath469 are collected in fig.[fig:6 ] . since the initial distribution is symmetric , real parts of the eigenvalues @xmath293 are equal to zero . hence in fig.[fig:6 ] we show the imaginary parts @xmath470 , which are the growth rates of the unstable modes divided by azimuthal number @xmath471 , vs. dimensionless angular momentum dispersion @xmath169 . one can see that instability exists for all @xmath169 and never becomes saturated . moreover , it is easy to obtain the asymptotic values @xmath472 for different @xmath471 at @xmath473 : 0.289 , 0.108 , and 0.026 for @xmath469 correspondingly . for small angular momentum , growth rates increase linearly with @xmath169 , such as @xmath474 are equal to 0.454 , 0.463 , and 0.481 for @xmath469 correspondingly . ( growth rate divided by azimuthal number @xmath471 ) vs. dimensionless angular momentum dispersion @xmath169 of the initial df for azimuthal numbers @xmath124 ( crosses ) , @xmath475 ( circles ) , and @xmath476 ( squares).,width=359 ] we have studied the stability of the spherically - symmetric and thin disk stellar clusters around a massive black hole . we conclude that stability properties of spherical clusters depend crucially on monotonity of initial distribution functions , while thin disk clusters are almost always unstable . if the initial distribution of the spherical cluster is monotonic , the cluster is most likely to be stable . this conclusion was first made in tremaine ( 2005 ) , where stability of @xmath0 mode was generally proved , and @xmath1 was tested numerically . we confirm this conclusion by considering a number of monotonic distributions for modes with arbitrary @xmath33 . besides , we have checked distributions obtained from monotonic ones by making them vanish quickly but smoothly at circular orbits . these models were also stable . however , a general proof of stability for any monotonic distributions was not yet found . spherical clusters with the non - monotonic dfs should be generally affected by the gravitational loss - cone instability . the instability was first found in our paper i using a simplification of systems with near - radial orbits . in the sec . 2 we show that this instability is due to just non - monotony of distributions over angular momentum , the orbits may not necessary be near - radial . in our opinion , both monotonic and non - monotonic distributions are important for possible applications to real stellar clusters around black holes . the dfs monotonically increasing from the loss cone radius up to circular orbits are formed naturally due to two - body collisions of stars . it follows from numerical experiments ( see , e.g. , cohn and kulsrud , 1978 ) , which predict establishment of such distributions after a characteristic time for collisional relaxation . these distributions may be approximated by the formula @xmath477 . such a slowly increasing function is , in fact , predetermined by the boundary conditions imposed in the cited numerical study and some other investigations . indeed , the vanishing condition at @xmath478 , and the matching condition to isotropic ( maxwellian ) distribution , @xmath159 , at the boundary @xmath479 of the phase space @xmath480 ( boundary separates stars which is gravitationally coupled to the black hole from the others ) is required . the last condition means the asymptotic ( when @xmath481 ) independence of the function @xmath98 on the momentum @xmath8 . so monotonic , or logarithmic , dependence of type of ( [ eq:3.2 ] ) is quite reasonable . the non - monotonic distributions are also real . if the cluster , is formed , for example , as a result of the collisionless collapse ( several free fall times ) , then it remains collisionless for a long timescale of collisional relaxation ( see , e.g. , merritt & wang , 2005 ) . in principle , the system can have almost arbitrary df both in the energy and in the angular momentum . during the collapse , a typical non - monotonic distribution of stars over the angular momentum , with empty loss cone and maximum at some value @xmath482 , is formed . in paper i we argued that stability properties of such a distribution is effectively analogous to one of typical plasma distributions of the `` beam - like '' type . but they can readily become unstable , as it is well - known in plasma physics ( and also confirmed by direct stability study of corresponding stellar systems in paper i ) . it is possible ( as it is often so in plasma ) that for the time of collisionless behavior , df can undergo a dramatic change from its initial form . in particular , the collective flux of stars into the loss cone caused by the instability could , in principle , lead to the formation of a considerable part of the black hole . checking of such possibilities is the most urgent task for future studies of unstable _ non - monotonic _ models . since spherically - symmetric models with the _ monotonic _ df are apparently stable , but analogous disk systems are unstable ( see tremaine 2005 and sec . 3 ) , a critical flatness of ellipsoid models at which the instability begins is expected . study of such systems , as well as systems with more complex triaxial ellipsoids can be performed using numerical simulations . we are grateful to v. a. mazur for stimulating interest in our work . the careful review of referee helped in improving the presentation . the work was supported in part by russian science support foundation , rfbr grants no . 05 - 02 - 17874 , 08 - 02 - 00928 and 07 - 02 - 00931 , `` leading scientific schools '' grants no . 7629.2006.2 , 900.2008.2 and `` young doctorate '' grant no . 2010.2007.2 provided by the ministry of industry , science , and technology of russian federation , and the `` extensive objects in the universe '' grant provided by the russian academy of sciences , and also by programs of presidium of russian academy of sciences no 16 and ofn ras no 16 .

bisch and jones suggested the skein theoretic classification of planar algebras and investigated the ones generated by 2-boxes with the second author . in this paper , we consider 3-box generators and classify subfactor planar algebras generated by a non - trivial 3-box satisfying a relation proposed by thurston . the subfactor planar algebras in the classification are either or the ones from representations of quantum . we introduce a new method to determine positivity of planar algebras and new techniques to reduce the complexity of computations . = 1 [ section ] [ theorem]lemma [ theorem]assumption [ theorem]proposition [ theorem]corollary [ theorem]definition [ theorem]definition [ theorem]notation [ theorem]example

1606.00779

jones intiated the modern theory of subfactors to study quantum symmetry @xcite . a deep theorem of popa says the standard invariants completely classify strongly amenable subfactors @xcite . in particular , the @xmath2 , @xmath3 , @xmath4 classification of subfactors up to index 4 is a quantum analogy of mackey correspondence . moreover , popa introduced standard @xmath5-lattices as an axiomatization of the standard invariant @xcite , which completes ocneanu s axiomatization for finite depth subfactors @xcite . jones introduced ( subfactor ) planar algebras as an axiomatization of the standard invariant of subfactors , which capture its ulterior topological properties . he suggested studying planar algebras by skein theory , a presentation theory which allows one to completely describe the entire planar algebra in terms of generators and relations , both algebraic and topological . an @xmath6-box generator for a planar algebra is usually represented by a @xmath7-valent labelled vertex . a planar algebra is a representation of labelled planar diagrams , namely _ fully labelled planar tangles _ , on vector spaces in the flavor of tqft @xcite : the representation is well defined up to isotopy ; the target vector space only depends on the boundary condition of diagrams ; in particular , diagrams without boundary are mapped to the ground field , called the partition function . a planar algebra is called a subfactor planar algebra if its partition function is positive definite . in this case , the vector spaces become hilbert spaces . a skein theory is a presentation of a planar algebra by generators and relations , such that the partition function can be evaluated by the relations . the planar algebra is determined by these generators and relations once an evaluation algorithm is obtained . therefore , one can ask for a skein theoretic classification of planar algebras , as suggested by bisch and jones . one can find interesting skein theories for planar diagrams in graph theory , such as the exchange relation @xcite motivated by biprojections from intermediate subfactors @xcite , and the relations coming from the discharge method @xcite . other types of skein theory were introduced , including thurston s relation @xcite , the jellyfish relation @xcite and the yang - baxter relation @xcite , motivated by knot theory and suggested by bisch and jones @xcite . the simplest planar algebra is the temperley - lieb - jones planar algebra which has neither generators nor relations . planar algebras generated by 1-boxes were completely analyzed by jones @xcite . bisch and jones initiated the classification of planar algebras generated by a single 2-box @xcite . based on the subsequent work of bisch , jones and the second author @xcite , a classification of singly generated yang - baxter relation planar algebra was achieved in @xcite , where a new family of planar algebras was constructed . planar algebras with multiple 2-box generators were discussed in @xcite . in this paper , we study planar algebras generated by a single 3-box . there is a known two parameter family of examples @xmath8 related to quantum @xmath1 , see example 2.5 in @xcite . we classify all @xmath9 for which @xmath8 has a positive partition function , ( see theorem [ thm : positivity ] ) . its semisimple quotient is a subfactor planar algebra . the corresponding subfactors are known as jones - wenzl subfactors @xcite . the subfactor planar algebras were constructed by xu @xcite . the skein theory of @xmath8 is inherited from the homfly - pt skein relations @xcite . thurston provided an intrinsic skein theory which was designed for 6-valent planar graphs @xcite , which we call here thurston s relation . in this paper , we classify subfactor planar algebras with thurston s relation generated by a non - trivial 3-box , namely singly generated thurston - relation planar algebras : [ thm : main ] any singly generated thurston - relation planar algebra is either @xmath10 or ( the semisimple quotient of ) @xmath8 . moreover , @xmath11 for some @xmath12 , @xmath13 , and @xmath14 for some @xmath15 , @xmath16 , or @xmath17 . we remark that it would be interesting to find a skein theory for a single 3-box which gives a family of subfactor planar algebras involving @xmath10 . the paper is organized as follows . in [ sec : pre ] , we recall planar algebras , homfly - pt skein theory , and thurston s skein theory . the @xmath6-box space of a singly generated thurston - relation planar algebra @xmath18 generically has dimension @xmath19 . in [ sec : generic ] , we classify the generic case , namely @xmath20 . we set up five formal variables for thurston s relation using 3-box relations . we prove that only two variables survive after considering 4-box relations ( theorems [ thm1 ] , [ thm2 ] ) . then we identify the two - parameter family with @xmath8 ( theorem [ thm : main1 ] ) . technically we simplify the computation by working on the reduced planar algebra with respect to the second jones - wenzl idempotent @xmath21 . the reduced planar algebra has smaller @xmath6-box spaces and their dimensions are @xmath22 . in [ sec : reduced ] , we classify the reduced case , namely @xmath23 . in [ section : pos ] , we classify all @xmath9 for which @xmath24 has a positive markov trace . in this case , @xmath8 has a unique involution @xmath25 , so that its semi - simple quotient is a subfactor planar algebra . then we complete our classification , theorem [ thm : main ] . we refer the readers to @xcite for definition , properties , examples and skein theory of planar algebras . the homfly - pt polynomial is a link invariant given by a braid @xmath26{upb+.pdf}}}$ ] satisfying reidemeister moves i , ii , iii and the hecke relation . @xmath27{upb+.pdf}}}-{\raisebox{-.4cm}{\includegraphics[height=1cm]{upb-.pdf}}}&=(q - q^{-1}){\raisebox{-.4cm}{\includegraphics[height=1cm]{upb2id.pdf } } } , & & \\ & \text{reidemeister moves i : } & { \raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly11.pdf}}}&=r{\raisebox{-.4cm}{\includegraphics[height=1cm]{upb1id.pdf } } } ; & { \raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly12.pdf}}}&=r^{-1}{\raisebox{-.4cm}{\includegraphics[height=1cm]{upb1id.pdf } } } ; \\ & & { \raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly13.pdf}}}&=r { \raisebox{-.4cm}{\includegraphics[height=1cm]{upb1id-.pdf } } } ; & { \raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly14.pdf}}}&=r^{-1}{\raisebox{-.4cm}{\includegraphics[height=1cm]{upb1id-.pdf}}},\\ & \text{reidemeister moves ii : } & { \raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly21.pdf}}}&={\raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly22.pdf } } } ; & { \raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly25.pdf}}}&={\raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly26.pdf } } } ; \\ & & { \raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly23.pdf}}}&={\raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly24.pdf } } } ; & { \raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly27.pdf}}}&={\raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly28.pdf}}},\\ & \text{reidemeister moves iii : } & { \raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly31.pdf}}}&={\raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly32.pdf } } } ; & { \raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly33.pdf}}}&={\raisebox{-.4cm}{\includegraphics[height=1cm]{uphomfly34.pdf}}},\\ & \text{circle parameter : } & { \raisebox{-.4cm}{\includegraphics[height=1cm]{upr0+.pdf}}}&={\raisebox{-.4cm}{\includegraphics[height=1cm]{upr0-.pdf}}}=\delta,\label{circpara } \\ & & r - r^{-1}&=\delta(q - q^{-1})\end{aligned}\ ] ] when @xmath28 , we have @xmath29 . then the skein relation is determined by the circle parameter @xmath30 . when @xmath31 , @xmath32 . let @xmath33 , @xmath34 , be the diagram obtained by adding @xmath35 oriented ( from bottom to top ) through - strings on the left of @xmath26{upb+.pdf}}}$ ] . the hecke algebra of type @xmath2 is a ( unital ) filtered algebra @xmath36 . the algebra @xmath37 is generated by @xmath33 , @xmath38 and @xmath37 is identified as a subalgebra of @xmath39 by adding an oriented through string on the right . over the field @xmath40 , the equivalence classes of minimal idempotents of @xmath37 are indexed by young - diagrams with @xmath41 cells . the trace formula is given by [ traceformula ] let @xmath5 be a young diagram and @xmath42 the minimal idempotent corresponding to @xmath5 , then @xmath43 where @xmath44 is the content of the cell @xmath45 in @xmath5 and @xmath46 is its hook length . jones studied the planar algebras @xmath47 associated with homfly - pt skein relation @xcite . its @xmath41-box space consists of homfly - pt diagrams which have @xmath48 boundary points and alternating orientation on the boundary as follows : @xmath49{uphomflypa.pdf}}}.\ ] ] moreover , he proved that this planar algebra is generated by a 3-box : [ thm : jones3-box ] the planar algebra @xmath47 is generated by @xmath50{uphomfly.pdf}}}$ ] . when @xmath11 for some @xmath12 and @xmath14 for some @xmath15 or @xmath17 , @xmath47 admits an involution * such that the markov trace is positive semidefinite . therefore , the semisimple quotient of @xmath47 is a subfactor planar algebra , which can be constructed from the representation theory of quantum su(n ) @xcite . ( when @xmath51 , we have @xmath52 and @xmath53 . ) we prove that these are the only possibilities such that @xmath47 has positivity in theorem [ thm : positivity ] . first , we identify the isomorphism classes of @xmath47 : [ lemsym ] the four planar algebras @xmath47,@xmath54,@xmath55,@xmath56 are isomorphic . note that @xmath57 and @xmath58 define the same planar algebra . the isomorphism between @xmath47 and @xmath55 is induced by sending @xmath26{uphomflyg.pdf}}}$ ] to @xmath59{uphomflyg.pdf}}}$ ] . the isomorphism between @xmath47 and @xmath56 is induced by sending @xmath26{uphomflyg.pdf}}}$ ] to @xmath26{uphomflyg-.pdf}}}$ ] . as emphasized in the introduction , skein theory provides an important perspective from which to understand a planar algebra for many reasons . skein theories are important starting points for the construction and classification of planar algebras . in this paper we will study subfactor planar algebras generated by a 3-box . recall that @xmath47 is generated by a 3-box . although it has a skein theory derived from the homfly - pt skein relation , thurston provides a skein theory intrinsic to the 3-box generator of @xmath47 @xcite : we say a @xmath60-box @xmath61 satisfies the thurston s relation if @xmath62{upuncappable1.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upuncappable2.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upuncappable3.pdf}}},\ \dots\ , { \raisebox{-.6cm}{\includegraphics[height=1.5cm]{upuncappable4.pdf}}}=(lower\ terms ) ; \label{equ : h - i0}\\ \text{unshaded } & 2\leftrightarrow2 \text { move : } & & { \raisebox{-.6cm}{\includegraphics[height=1.5cm]{upi.pdf}}}=a\ { \raisebox{-.6cm}{\includegraphics[height=1.5cm]{uph.pdf}}}+(lower~terms ) , ~a\neq0 \label{equ : h - i1 } \;;\\ \text{shaded } & 2\leftrightarrow2 \text { move : } & & { \raisebox{-.6cm}{\includegraphics[height=1.5cm]{upsi.pdf}}}=b{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upsh.pdf}}}+(lower~terms ) , ~b\neq0 \label{equ : h - i2 } \;,\end{aligned}\ ] ] where lower terms are a linear combination of diagrams with less generators . one can generalize thurston s relation for multiple 3-box generators . he proved that the above relations are evaluable , in the sense that there exists an algorithm for evaluating every closed planar diagram to a scalar . moreover , the standard forms in the sense of thurston @xcite give a basis of the @xmath41-box space when their dimension achieves the maximum of @xmath63 . suppose @xmath64 admits thurston s relation . then @xmath65 is spanned by the standard forms and @xmath66 . [ cor:24-basis ] in the generic case , namely @xmath67 , we have the basis of the 4-box space given by the standard form as follows : moreover , one can replace them by the other diagrams with two generators using thurston s relation , . thurston s relation look similar to the 6j symbol in a monoidal category , but it provides a much better evaluation algorithm . the 6j symbols in a monoidal category are the coefficients of the change of basis matrix in the hom spaces , and a monoidal category is determined by the 6j symbols up to monoidal equivalence , however it seems hopeless to determine the 6j symbols in general . actually for the example @xmath47 appearing in our classification , it is difficult to compute the 6j symbols . thurston s relation only requires partial 6j symbols for two objects , together with the data for the dual planar algebra . this combination , rather than considering only one side , appears to be powerful , and determines the planar algebra completely . the main purpose of this paper is to classify all subfactor planar algebras generated by a 3-box @xmath61 satisfying thurston s relation . we give the classification for the generic case in [ sec : generic ] and for the reduced case in [ sec : reduced ] . since the subfactor with index at most @xmath68 is classified , we only need to consider the case @xmath69 . actually , @xmath10 and extended @xmath10 subfactor planar algebras are generated by a 3-box with thurston s relation . the dimension of their 4-box spaces are 21 and 22 respectively . in this section , we classify subfactor planar algebras @xmath18 generated by a 3-box @xmath61 with thurston s relation , such that @xmath67 . in this case , we have @xmath70 and @xmath71 . most results in this section also work for case @xmath72 . suppose @xmath18 is a spherical planar algebra . we use the following notations : @xmath73 is the @xmath74 jones projection ; @xmath75 is the @xmath74 jones - wenzl idempotent ; and @xmath76 is the basic construction ideal in @xmath65 . since @xmath77 , there exists two minimal idempotents @xmath78 and @xmath79 in @xmath80 and @xmath81 . since @xmath82 , we assume that @xmath83 . we take @xmath84 , where @xmath85 , as the generator for @xmath64 . then @xmath61 has the following relations : 1 . @xmath61 is totally uncappable , i.e , @xmath86{upuncappable1.pdf}}}={\raisebox{-.8cm}{\includegraphics[height=2cm]{upuncappable2.pdf}}}=\cdots={\raisebox{-.8cm}{\includegraphics[height=2cm]{upuncappable3.pdf}}}={\raisebox{-.8cm}{\includegraphics[height=2cm]{upuncappable4.pdf}}}.\end{aligned}\ ] ] 2 . @xmath61 is an eigenvector of the ( 2-click ) rotation @xmath87 . i.e @xmath88 3 . @xmath61 satisfies a quadratic relation : @xmath89 note that property ( 1 ) is @xmath90 move in thurston s relation . now we focus on the @xmath21-cutdown of @xmath91 instead of the entire subfactor planar algebra . technically , this reduces the dimension of the @xmath41-box space and simplify the computation . elements in the @xmath41-box cutdown space will be elements @xmath92 of the form @xmath49{upcut.pdf } } } \;.\ ] ] then the @xmath21-cutdown of @xmath93 has a basis @xmath94{uptltl.pdf}}}\}$ ] . we can consider the elements in the @xmath21-cutdown as morphisms in the @xmath95 bimodule category ( the even part of the subfactor planar algebra ) . in terms of @xmath95 bimodule maps , we rewrite @xmath21 as a single string labelled by @xmath21 , and we ignore the label if there is no confusion . in this setting , we express @xmath61 and @xmath50{uptltl.pdf}}}$ ] as @xmath95 bimodule maps as follows : @xmath96{uptl.pdf}}}&:={\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptltl.pdf } } } \;,\\ { \raisebox{-.4cm}{\includegraphics[height=1cm]{upws.pdf}}}&:={\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upscutdown.pdf}}}.\end{aligned}\ ] ] where the position of @xmath61 indicates the position of the @xmath97 sign . now let us set up and simplify the formal variables for the relations of and in @xmath93 [ skein ] we have the following skein relations in the @xmath98 cut - down in terms of @xmath99 and one new parameter @xmath100 : + * @xmath101{ups.pdf}}}=\omega{\raisebox{-.4cm}{\includegraphics[height=1cm]{upws.pdf}}};$ ] + * @xmath102{upbox1s.pdf}}}={\raisebox{-.4cm}{\includegraphics[height=1cm]{upbox1tl.pdf}}}=0;$ ] + * @xmath103{upbox2stl.pdf}}}=0;$ ] + * @xmath102{upbox2tltl.pdf}}}=\frac{\delta^2 - 2}{\delta}{\raisebox{-.4cm}{\includegraphics[height=1cm]{upvi.pdf}}};$ ] + * @xmath101{upbox2ss.pdf}}}=\gamma\frac{\delta^3 - 2\delta}{\delta^2 - 1}{\raisebox{-.4cm}{\includegraphics[height=1cm]{upvi.pdf}}};$ ] + * @xmath104{upbox3tltltl.pdf}}}=\frac{\delta^2 - 3}{\delta}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptltriv.pdf}}};$ ] + * @xmath105{upbox3stltl.pdf}}}=-\frac{1}{\delta}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upsv.pdf}}};$ ] + * @xmath50{upbox3tlss.pdf}}}=(\gamma-1){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upsv.pdf}}}-\gamma\frac{\delta}{\delta^2 - 1}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptltriv.pdf}}};$ ] + * @xmath104{upbox3sss.pdf}}}=\delta_{\omega,1}\varepsilon{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upsv.pdf}}}+\gamma(\gamma-1)\frac{\delta^2}{\delta^2 - 1}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptltriv.pdf}}}$ ] , where @xmath106 is an undetermined variable . the relations follow from relations [ equ : s1][equ : s3 ] . since and form an orthogonal basis of the @xmath21-cutdown of @xmath93 , one obtains the coefficients in ( viii ) and ( ix ) by computing the inner product . we proceed to discuss the relations in the @xmath21-cutdown of @xmath107 . we define @xmath108{uptlc.pdf}}}:={\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upc.pdf } } } \;.\ ] ] [ lem : generic - cutdown-4-basis ] the set @xmath109 defined as @xmath110{upbox4tlstls.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upiss.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphtls.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphstl.pdf } } } , { \raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upitls.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlid.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptle.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlc.pdf}}}\right\}\ ] ] forms a basis of the @xmath21-cutdown of @xmath111 . note that the @xmath21-cutdown of the 24 diagrams in the basis of @xmath107 in corollary [ cor:24-basis ] is in the linear span of @xmath109 , thus @xmath21-cutdown of @xmath111 is spanned by @xmath109 . suppose @xmath112{upbox4tlstls.pdf}}}+b{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upiss.pdf}}}+c{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphtls.pdf}}}+d{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphstl.pdf}}}+e{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}}\\ + f{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upitls.pdf}}}+g{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlid.pdf}}}+h{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptle.pdf}}}+i{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlc.pdf}}}=0 , \end{gathered}\ ] ] for some @xmath113 . recall that each string in equation represents in @xmath18 . we consider it as an equation in @xmath107 and rewrite it in terms of the basis of @xmath107 in corollary [ cor:24-basis ] . the coefficients of @xmath26{upsh.pdf}}}$ ] and @xmath26{upi.pdf}}}$ ] are @xmath114 and @xmath115 respectively , thus @xmath116 . furthermore , the coefficients of @xmath117{upphtls.pdf}}}}$ ] , @xmath117{upphstl.pdf}}}}$ ] , @xmath26{uppistl.pdf}}}$ ] and @xmath26{uppitls.pdf}}}$ ] are @xmath118 , @xmath119 , @xmath120 and @xmath121 respectively , thus @xmath122 . finally , the coefficients of @xmath26{upptlid.pdf}}}$ ] , @xmath26{upptle.pdf}}}$ ] and @xmath26{uppc.pdf}}}$ ] are @xmath123 , @xmath124 , and @xmath125 respectively , thus @xmath126 . therefore @xmath109 is linearly independent , and @xmath109 is a basis . since @xmath109 is a basis , we have the unshaded @xmath127 move for @xmath61 in terms of formal variables : @xmath128{uphss.pdf}}}&=a{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upiss.pdf}}}+b_1{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphstl.pdf}}}+b_2{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}}+ \nonumber\\ & + b_3{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphtls.pdf}}}+b_4{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upitls.pdf}}}+c_1{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlid.pdf}}}+c_2{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptle.pdf}}}+d{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlc.pdf } } } \;. \label{equ : relation}\end{aligned}\ ] ] using the rotational symmetry , we can immediately simplify the formal variables . [ lem : a ] either @xmath129 , and @xmath128{uphss.pdf}}}&={\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upiss.pdf}}}+b\left({\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphstl.pdf}}}-{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upitls.pdf}}}+{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphtls.pdf}}}-{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}}\right)+ \nonumber \\ & + c\left({\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlid.pdf}}}-{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptle.pdf}}}\right ) \;. \label{equ : a=1}\end{aligned}\ ] ] or @xmath130 , and @xmath128{uphss.pdf}}}&=-{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upiss.pdf}}}+b\left({\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphstl.pdf}}}+{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upitls.pdf}}}+{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphtls.pdf}}}+{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}}\right)+ \nonumber\\ & + c\left({\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlid.pdf}}}+{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptle.pdf}}}\right)+d{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlc.pdf}}}. \label{equ : a=-1}\end{aligned}\ ] ] if @xmath130 , then we apply rotations by @xmath131 to equation and obtain @xmath128{upiss.pdf}}}&=-{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphss.pdf}}}+b_4{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphstl.pdf}}}+b_1{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}}+ \nonumber\\ & + b_2{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphtls.pdf}}}+b_3{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upitls.pdf}}}+c_2{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlid.pdf}}}+c_1{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptle.pdf}}}+d{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlc.pdf } } } \;. \label{equ : relation1}\end{aligned}\ ] ] subtracting equations from yields @xmath132{uphstl.pdf}}}+(b_1-b_2){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}}+ \nonumber\\ & + ( b_2-b_3){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphtls.pdf}}}+(b_3-b_4){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upitls.pdf}}}+(c_1-c_2)({\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlid.pdf}}}+{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptle.pdf } } } ) \;. \label{equ : relation2}\end{aligned}\ ] ] thus @xmath133 and @xmath134 and we obtain equation . if @xmath135 , then we apply rotation by @xmath136 to equation and take the alternating sum of the four resultant equations . we have @xmath137{uphss.pdf}}}&=2(1+a){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upiss.pdf}}}+\nonumber\\ & + ( b_1-b_2+b_3-b_4)\left({\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphstl.pdf}}}-{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upitls.pdf}}}+{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphtls.pdf}}}-{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}}\right)+ \nonumber\\ & + 2(c_1-c_2)\left({\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlid.pdf}}}-{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptle.pdf}}}\right).\end{aligned}\ ] ] taking the quotient by @xmath138 , we obtain equation . furthermore , we can determine the coefficients @xmath139 in lemma [ lem : a ] in terms of @xmath140 . [ thurston ] if @xmath129 , then @xmath141 if @xmath130 , then @xmath142 when @xmath129 , applying to the bottom on both sides of equation , we obtain @xmath143{ups.pdf}}}-\gamma\frac{\delta}{\delta^2 - 1}{\raisebox{-.4cm}{\includegraphics[height=1cm]{uptl.pdf}}}=b\left((\omega+\omega^{-1})(-\frac{1}{\delta}){\raisebox{-.4cm}{\includegraphics[height=1cm]{ups.pdf}}}-\frac{\delta^2 - 2}{\delta}{\raisebox{-.4cm}{\includegraphics[height=1cm]{ups.pdf}}}\right)+c{\raisebox{-.4cm}{\includegraphics[height=1cm]{uptl.pdf}}}.\ ] ] by comparing the coefficient in and , we solve @xmath144 as in eqution . when @xmath130 , we apply @xmath26{uptl.pdf}}}$ ] to the bottom on both sides of equation . by a similar computation , we solve @xmath139 as in equation . we remark that the variable @xmath100 in lemma @xmath145 does not appear in the solutions [ solution : a=1 ] , [ solution : a=-1 ] . we can solve for @xmath100 in terms of @xmath140 by applying to the bottom of both sides of equations , . we will determine @xmath100 later . [ cor1 ] suppose @xmath64 is a planar algebra generated by a non - trivial 3-box satisfying thustron relations , then @xmath64 is determined by @xmath146 , where @xmath147 is the index , @xmath148 the ratio of the trace of the two orthogonal minimal idempotents , @xmath149 is the rotation eigenvalue and @xmath150 are the signs in the unshaded and shaded @xmath151 moves . suppose @xmath18 and @xmath152 are two planar algebras and there exists orthogonal minimal idempotents @xmath153 and @xmath154 . then we can construct uncappable rotation eigenvectors @xmath61 and @xmath155 as in section @xmath60 satisfying the same quadratic relation . since the planar algebra is generated by a 3-box , we define a map @xmath156 by sending @xmath61 to @xmath155 . since all the coefficients in the skein theory are determined @xmath157 , @xmath158 extends to a planar algebra isomorphism . with the structure of the thurston - relation coefficients determined , we are ready to turn to the classification of thurston - relation planar algebras . in this section , we prove our classification result ( theorem [ thm : main ] ) for the generic case . in theorem [ cor1 ] , we show that @xmath18 is determined by @xmath159 and @xmath160 for the unshaded and shaded @xmath161 moves . first we prove that @xmath162 and @xmath163 in the @xmath161 move . thus @xmath18 is determined by @xmath164 . then we identify @xmath18 with @xmath47 . [ thm1 ] if @xmath18 is a planar algebra generated by a non - trivial 3-box satisfying thurston s relation with parameters @xmath146 , @xmath70 , then the rotation eigenvalue @xmath162 . a direct computation using lemma [ skein ] shows that in general , @xmath108{upbox4stltls.pdf}}}=(\gamma-1){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphstl.pdf}}}+\gamma{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlid.pdf}}}-\frac{\gamma\delta}{\delta^2 - 1}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlc.pdf}}}-\frac{1}{\delta}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upiss.pdf}}}\ ] ] suppose @xmath129 , notice that @xmath50{upbox4sstls.pdf}}}=\omega\omega^{-1}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{up3481.pdf}}}$ ] , + and apply the ( @xmath127 ) move in the dotted circle , we can rewrite with respect to the basis in corollary [ cor:24-basis ] . the coefficient of in the linear representation of @xmath50{upbox4sstls.pdf}}}$ ] with respect to this basis is @xmath165 . + similarly @xmath50{upbox4sstls.pdf}}}=\omega\omega^{-1}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{up3491.pdf}}}$ ] , + so applying the ( @xmath127 ) move in the dotted circle , we see that the coefficient of is @xmath166 . equating these coefficients from the two different computations yields @xmath167 since @xmath168 , @xmath169 , and so @xmath170 hence @xmath171 . + note that from thurston s relation , we obtain @xmath108{upbox3sss.pdf}}}=(\frac{\gamma\delta}{\delta^2 - 1}(\delta^2 - 3)-2(\gamma-1)^2\frac{\delta}{\delta^2 - 3}){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upsv.pdf}}}+\gamma(\gamma-1)\frac{\delta}{\delta^2 - 1}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptltriv.pdf}}}\ ] ] since the coefficient before is 0 with @xmath172 , we obtain @xmath173 , which contradicts the fact that @xmath70 . + now suppose @xmath130 . again , we can evaluate as a sum of basis diagrams in two different ways as above , which will again give us that @xmath171 . + similarly by checking @xmath50{upbox3sss.pdf}}}$ ] , we obtain @xmath174 which yields the equation @xmath175 this has no real solutions with @xmath70 . + this concludes the proof of the theorem . [ thm2 ] if @xmath18 is a planar algebra generated by a non - trivial 3-box satisfying thurston s relation with parameters @xmath176 , @xmath70 , then @xmath177 . suppose @xmath130 . first note that by [ thm1 ] , we have @xmath162 . applying ( @xmath127 ) move in the dotted circle , we obtain the relation @xmath108{up2241.pdf}}}=b{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upbox4tlstls.pdf}}}+(-(\gamma-1)+b(-\frac{2}{\delta})){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upiss.pdf}}}+\frac{\gamma\delta}{\delta^2 - 1}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}}\ ] ] @xmath178{uphstl.pdf}}}+(b(\gamma-1)+d(-\frac{1}{\delta})){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upitls.pdf}}}\ ] ] @xmath179{uptlid.pdf}}}+\frac{b\gamma}{\delta^2 - 1}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptle.pdf}}}+2b(-\gamma\frac{\delta}{\delta^2 - 1}){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlc.pdf}}}\ ] ] applying ( @xmath127 ) move in the dotted circle , we obtain @xmath108{up2242.pdf}}}=b{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upbox4tlstls.pdf}}}+(-(\gamma-1)+b(-\frac{2}{\delta})){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphss.pdf}}}+\frac{\gamma\delta}{\delta^2 - 1}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphstl.pdf}}}\ ] ] @xmath178{upistl.pdf}}}+(b(\gamma-1)+d(-\frac{1}{\delta})){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphtls.pdf}}}\ ] ] @xmath179{uptle.pdf}}}+\frac{b\gamma}{\delta^2 - 1}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlid.pdf}}}+2b(-\gamma\frac{\delta}{\delta^2 - 1}){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlc.pdf}}}.\ ] ] let @xmath5 be the coefficient of in the linear expansion of with respect to basis . ( note that if @xmath180 , then @xmath181 . we do not use this assumption , so that the proof also work for the case @xmath182 . ) equating the coefficients of @xmath50{upiss.pdf}}}$ ] in the two expressions yields @xmath183 thus @xmath184 note that @xmath185 , when @xmath162 , so @xmath186 hence @xmath187 @xmath70 implies that @xmath171 and thus @xmath188 . therefore , @xmath108{upbox4sstls.pdf}}}=\frac{\delta}{\delta^2 - 1}{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}}+c{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphstl.pdf}}}+d(-\frac{1}{\delta}){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upitls.pdf}}}\ ] ] @xmath189{uphstl.pdf}}}+c{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}}+d(-\frac{1}{\delta}){\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphtls.pdf}}}\ ] ] hence , @xmath190 . note that this contradicts @xmath171 . these two theorems show that a singly genrated thurston - relation planar algebra with parameters @xmath146 satisfies @xmath191 . thus the planar algebra is parameterized by @xmath192 . with the following lemma and theorem , we will identify any thurston - relation planar algebra as @xmath47 for some @xmath57 . [ lemma : idem ] there exist two idempotents @xmath193 with @xmath81 and @xmath194 consider the basis @xmath195{uptl3c2.pdf}}},{\raisebox{-.4cm}{\includegraphics[height=1cm]{uptl3e.pdf}}},{\raisebox{-.4cm}{\includegraphics[height=1cm]{uptl3f.pdf}}}\right\}$ ] of @xmath196 . + first set @xmath197{uptl3b.pdf}}}+c\left({\raisebox{-.4cm}{\includegraphics[height=1cm]{uptl3c1.pdf}}}+{\raisebox{-.4cm}{\includegraphics[height=1cm]{uptl3c2.pdf}}}\right)+e{\raisebox{-.4cm}{\includegraphics[height=1cm]{uptl3e.pdf}}}+f{\raisebox{-.4cm}{\includegraphics[height=1cm]{uptl3f.pdf}}}$ ] . since we want @xmath198 , comparing the coefficients with respect to @xmath199 of @xmath200 for @xmath201 we obtain the following equations : @xmath202 which have solutions in terms of @xmath203 and @xmath204 : @xmath205 we want @xmath79 to be an idempotent , so we set up equations by comparing the coefficients with respect @xmath206 of @xmath207 and @xmath79 and we find two possible solutions : @xmath208 taking ( 1 ) , we have @xmath209 applying a similar computation , we see that @xmath210 is an idempotent with @xmath211 we see that taking the second case ( 2 ) simply switches the role of @xmath78 and @xmath79 . another strategy to show the above trace formula is to use the trace formula for @xmath204 a root of unity , and proving that the trace formula for idempotents are rational functions on @xmath9 . we are now ready for the following theorem : [ thm : main1 ] suppose @xmath64 is a singly generated thurston - relation planar algebra and @xmath212 . then it is isomorphic to the semisimple quotient of @xmath47 for some @xmath57 . suppose @xmath64 is a singly generated thurston - relation planar algebra with parameter @xmath213 . we find @xmath57 such that @xmath47 has the same parameter @xmath192 . case 1 : @xmath214 . in this case , we show that that @xmath215 with a circle parameter @xmath30 is a solution . first note that this planar algebra has the desired @xmath30 . therefore , we only need to show that the ratio of two traces of the two minimal idempotents in @xmath216 equals to @xmath148 . by lemma [ lemma : idem ] , we know that @xmath216 with circle parameter @xmath30 has two minimal idempotents @xmath217 with @xmath218 the ratio equals to @xmath219 . therefore , @xmath215 with circle parameter @xmath30 gives a solution . case 2 : @xmath220 . let @xmath57 be the solution of @xmath221 ( note that the assumption @xmath220implies that @xmath31 . ) from equation , we see that @xmath222 from equation , we obtain @xmath223 then we have the following equations : @xmath224 recall that @xmath31 , we let @xmath225 to be defined by the formulas in lemma [ lemma : idem ] @xmath226 then we have @xmath227 combined with the equation , and equation , we see that either @xmath228 or @xmath229 in both cases , the singly generated thurston - relation planar algebra with @xmath192 and @xmath47 have the same skein theory . therefore , the planar algebras are isomorphic . combined with equation and equation , one can check that the following equations holds @xmath230 therefore with the above solution @xmath57 , @xmath47 is the homfly - pt planar algebra with desired condition . suppose @xmath64 is a singly generated thurston - relation planar algebra with parameter @xmath192 . we want to find @xmath57 such that @xmath231 has same @xmath192 , i.e , @xmath232 equation implies that @xmath233 subtracting from , we obtain that @xmath234 + similarly , + shows that @xmath235 + therefore we have @xmath236 using @xmath237 , we get @xmath238 note that @xmath239 , so @xmath240 therefore , @xmath241 thus we can solve @xmath204 in terms of @xmath192 . it might be the case that the solution @xmath51 for equation . in this case , we determine that @xmath192 must satisfy the relation from equation , @xmath242 there are 8 solutions @xmath57 in the above theorem . the 8 corresponding planar algebras are isomorphic in proposition [ lemsym ] . in this section , we will classify subfactor planar algebras @xmath18 generated by a 3-box with thurston - relation for the reduced case , namely @xmath243 . [ smallthm ] if @xmath64 is @xmath244-supertransitive , then @xmath245 if and only if @xmath246 . in this case , @xmath18 is either the @xmath247 or @xmath248 subfactor planar algebra . thus the main theorem [ thm : main ] holds for the reduced case @xmath246 . if @xmath249 , then @xmath18 is either the @xmath247 or @xmath248 by the classification of subfactors up to index 4 @xcite(popa ) . moreover , @xmath246 . if @xmath246 , then @xmath245 by ocneanu s triple point obstruction @xcite(ocn88 ) . in this section , we classify subfactor planar algebras @xmath18 generated by a 3-box @xmath61 with thurston - relation , such that @xmath250 . in this case , we have @xmath70 and @xmath251 . by the result of jones @xcite , the 14 temperley - lieb diagrams and the 8 diagrams in the annular consequence are linearly independent . then one of the diagram with two generators @xmath26{upi.pdf}}}$ ] , @xmath26{uph.pdf}}}$ ] , @xmath26{upsi.pdf}}}$ ] and @xmath26{upsh.pdf}}}$ ] is linearly independent with these 22 diagrams . otherwise @xmath252 . up to rotation and the duality of the shading , we can assume that @xmath26{upi.pdf}}}$ ] and the other 22 diagrams form a basis . similarly to lemma [ lem : generic - cutdown-4-basis ] , the basis for the @xmath21-cutdown of @xmath107 is given by @xmath253{upiss.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphtls.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uphstl.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upistl.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{upitls.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlid.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptle.pdf}}},{\raisebox{-.6cm}{\includegraphics[height=1.5cm]{uptlc.pdf}}}\right\}.\ ] ] note that all results in [ sec : generic ] work for the case @xmath72 , except theorem [ thm1 ] . there we used the fact that @xmath50{upbox4tlstls.pdf}}}$ ] is linearly independent with @xmath254 , which is no longer true for the reduced case . now we give a different proof of @xmath162 for the reduced case . consequently the main theorem [ thm : main ] holds for the reduced case @xmath255 . [ thm : omega23 ] if @xmath64 is a singly genrated thurston - relation subfactor planar algebra with parameters @xmath146 , then @xmath162 . one can deduce that the principal graph starts with the graph @xmath49{uppgr1.pdf}}}\ ] ] where p and q correspond to 2 minimal projections in @xmath80 described above . now , we can apply penney s triple point obstruction [ pen13 ] : @xmath256}=-(1+\gamma)\frac{[4]}{[3]}\left(coef_{\in\cap_{n+1}\bar{p^{'}}}(s)\right)\ ] ] where @xmath257 and @xmath258=\frac{q^{n}-q^{-n}}{q - q^{-1}}$ ] is the @xmath259 quantum number ( where @xmath204 is defined so that @xmath260=q+q^{-1}$ ] and @xmath261 ) , @xmath262 is the projection labelled in our picture , and @xmath263 is the dual projection . under the assumption @xmath264 or @xmath265 , @xmath266 . this implies that @xmath267}=-(1+\gamma)\frac{[4]}{[3]}\left(coef_{\in\cap_{n+1}\bar{p^{'}}}(s)\right)\ ] ] we now consider @xmath244 subcases : suppose @xmath268 is self - dual , i.e @xmath26{uppcap.pdf}}}=p$ ] . in this case , we have @xmath269tr(p)-[3]}{tr(p)}\frac{1}{\gamma+1}=([2]-\frac{[3]}{[4]}(\gamma+1))\frac{1}{\gamma+1}\ ] ] therefore , @xmath270}=-\frac{[4][2]}{[3]}+1+\gamma,\ ] ] which implies @xmath271 hence @xmath272 , contradicting our assumption that @xmath70 . now suppose that @xmath273 is not self - dual , i.e @xmath26{uppcap.pdf}}}=q$ ] . in this case , we have @xmath274tr(p)-[3]}{tr(q)}\frac{1}{\gamma+1}=-([2]\frac{1}{r}-\frac{[3]}{[4]}(\frac{1}{\gamma}+1))\frac{1}{\gamma+1}\ ] ] therefore , @xmath270}=-\frac{[4]}{[3]}([2]\frac{1}{\gamma}-\frac{[3]}{[4]}(1+\frac{1}{\gamma})),\ ] ] hence @xmath275\gamma^2\pm\gamma-[5]=0 $ ] . now suppose @xmath129 , then by considering evaluating @xmath50{upbox3sss.pdf}}}$ ] , we have the following equations , @xmath276\gamma^2\pm\gamma-[5]&=0\;. \label{ee2 } \end{aligned}\ ] ] one can solove equation for @xmath148 in terms of @xmath30 and plug it in back to . we use mathematica to solve numerical solutions for @xmath30 . they are far below 2 hence there is no solution when @xmath70 . if @xmath130 , we obtain two equations following simialr arguments , @xmath277\gamma^2\pm\gamma-[5]&=0\ ; , \end{aligned}\ ] ] where @xmath144 are given in equation . one can show there is no solution with @xmath70 in a similar way . we conclude that @xmath162 for the dimension 23 case . [ thm : main2 ] suppose @xmath64 is a singly generated thurston - relation planar algebra and @xmath278 , then it is isomorphic to the semisimple quotient of @xmath47 for some @xmath57 . it follows from theorem [ thm : omega23 ] , [ thm2 ] and the proof of theorem [ thm : main1 ] . in this section , we will determine the positivity of @xmath47 . recall that when @xmath11 for some @xmath12 and @xmath14 for some @xmath15 or @xmath17 , @xmath47 admits an involution @xmath25 such that the markov trace is semipositive - definite . we show that they are the only values of @xmath57 and involution such that positivity holds . if @xmath47 has positivity , then the idempotents @xmath78 and @xmath79 in lemma [ lemma : idem ] are projections , therefore the generator @xmath61 is self - adjoint . since @xmath47 is generated by @xmath61 , the involution @xmath25 is uniquely determined . in theorem [ thm : main1 ] and [ thm : main2 ] , we show that a singly generated thurston - relation planar algebra with parameters @xmath192 and @xmath70 is isomorphic to @xmath47 for some @xmath57 . by lemma [ lemsym ] , we can assume that @xmath289 , @xmath290 . note that @xmath70 and @xmath291 , then by equation , we have @xmath292 ( one can check the term in the square root is positive . ) we have @xmath293 with @xmath294 or @xmath17 . case 1 : @xmath261 . by lemma [ lemsym ] , we can assume that @xmath297 . by @xmath298 , we have that @xmath299 . if @xmath11 , for some @xmath12 , then we know that @xmath47 has positivity . otherwise , @xmath300 . then the idempotent @xmath301}$ ] is well - defined and @xmath302})<0 $ ] . by proposition [ embed ] , @xmath47 does not have positivity . therefore we obtain our classification result , theorem [ thm : main ] . the ones with positivity all come from representations of the quantum groups @xmath316 or @xmath10 . note that @xmath317 really comes from @xmath318 a. ocneanu , _ quantized groups , string algebras and galois theory for algebras _ , operator algebras and applications , vol . 2 , london math . lecture note ser . , 136 , cambridge univ . press , cambridge , 1988 , pp . 119172 .

we study rare exclusive decays based on the quark - level transition , where or , in the context of supersymmetric theories with minimal flavour violation . we present analytic expressions for various mixing matrices in the presence of new-violating phases , and examine their impact on observables involving and decays . an estimate is obtained for-violating asymmetries in and decays for the dilepton invariant mass region . as a typical result , we find a-violating partial width asymmetry of about ( ) in the case of ( ) in effective supersymmetry with phases of , taking into account the measurement of the inclusive branching fraction . on the other hand , asymmetries of less than are predicted in the case of . we argue that it is not sufficient to have additional phases of to observe large-violating effects in exclusive decays . # 1#1 # 1#2#3#1|#2| # 3 # 1#1% # 1eq . ( [ # 1 ] ) # 1#2eqs . ( [ # 1])([#2 ] ) # 1fig . [ # 1 ] # 1#2figs . [ # 1][#2 ] # 1 # 1 # 1sec . [ # 1 ] # 1*#1 * # 1^_#1 # 1^#1 # 1 ^ 0_#1 # 1_#1 # 1_#1 # 1_#1 # 1_#1 # 1_#1 # 1_#1 # 1#2r^#1_#2 # 1#2#3acta phys . pol . b * # 1 * , # 3 ( 19#2 ) # 1#2#3annu . rev . nucl . part . sci . * # 1 * , # 3 ( 19#2 ) # 1#2#3eur . phys . j. c * # 1 * , # 3 ( 19#2 ) # 1#2#3int . j. mod . phys . a * # 1 * , # 3 ( 19#2 ) # 1#2#3_ibid . _ * # 1 * , # 3 ( 19#2 ) # 1#2#3_ibid . _ * # 1 * , # 3 ( 20#2 ) # 1#2#3*#1 * , # 3 ( 19#2 ) # 1#2#3mod . phys . lett . a * # 1 * , # 3 ( 19#2 ) # 1#2#3nuovo cimento a * # 1 * , # 3 ( 19#2 ) # 1#2#3nucl . phys . * b#1 * , # 3 ( 19#2 ) # 1#2#3nucl . phys . * b#1 * , # 3 ( 20#2 ) # 1#2#3nucl . phys . b ( proc . suppl . ) * # 1 * , # 3 ( 19#2 ) # 1#2#3phys . lett . b * # 1 * , # 3 ( 19#2 ) # 1#2#3phys . rev . * # 1 * , # 3 ( 19#2 ) # 1#2#3phys . rev . d * # 1 * , # 3 ( 19#2 ) # 1#2#3phys . rev . d * # 1 * , # 3 ( 20#2 ) # 1#2#3phys . rev . lett . * # 1 * , # 3 ( 19#2 ) # 1#2#3phys . rev . lett . * # 1 * , # 3 ( 20#2 ) # 1#2#3phys . rep . * # 1 * , # 3 ( 19#2 ) # 1#2#3prog . theor . phys . * # 1 * , # 3 ( 19#2 ) # 1#2#3rev . mod . phys . * # 1 * , # 3 ( 19#2 ) # 1#2#3z . phys . c * # 1 * , # 3 ( 19#2 )

hep-ph0002089

within the standard model ( sm ) , @xmath4 violation is caused by a non - zero complex phase in the cabibbo - kobayashi - maskawa ( ckm ) quark mixing matrix @xcite . while the experimentally observed indirect @xmath4 violation in the neutral kaon system , @xmath18 , can be accommodated in the sm , it is still an open question whether the sm description of @xmath4 violation is consistent with the new experimental result on direct @xmath4 violation , @xmath19 , since the theoretical prediction of its precise value suffers from large hadronic uncertainties @xcite . on the other hand , if the baryon asymmetry of the universe has been generated via baryogenesis at the electroweak phase transition , the ckm mechanism of @xmath4 violation can not account for the observed amount of baryon asymmetry . this feature could be a hint of the existence of @xmath4-violating sources outside the ckm matrix @xcite . important tests of the sm are provided by flavour - changing neutral current ( fcnc ) reactions involving @xmath0 decays @xcite , thus offering an opportunity to search for supersymmetric extensions of the sm @xcite . there are at present only a few fcnc processes which have been observed experimentally , but the situation will change considerably after the completion of @xmath0 factories in the near future . in this work , we analyse the exclusive decays @xmath20 and @xmath21 in the context of supersymmetry ( susy ) with minimal particle content and r - parity conservation @xcite . the inclusive reaction @xmath22 within supersymmetric models has been extensively studied in and , more recently , in . new physics effects in the exclusive channels have been investigated in . we place particular emphasis on @xmath4-violating effects associated with the partial rate asymmetry between @xmath0 and @xmath5 decays as well as the forward - backward asymmetry of the @xmath23 . within the sm these effects turn out to be unobservably small ( @xmath24 ) in the decays @xmath6 @xcite , and amount to only a few per cent in @xmath7 @xcite . however , in models with new @xmath4-violating phases in addition to the single phase of the ckm matrix , larger effects may occur due to the interference of amplitudes with different phases . the purpose of the present analysis is to explore @xmath4-violating observables in the aforementioned fcnc reactions that could provide evidence of a non - standard source of @xmath4 violation , and hence may be useful in analysing supersymmetry in future collider experiments . the paper is organized as follows . in , we exhibit the various mixing matrices of the minimal supersymmetric standard model ( mssm ) in the presence of additional @xmath4-violating phases . within such a framework we discuss different scenarios for the susy parameters . in , we are primarily concerned with the short - distance matrix element and wilson coefficients governing @xmath1 in the mssm . we also briefly describe an approximate procedure to incorporate quark antiquark resonant intermediate states namely @xmath25 , and the @xmath26 family which enter through the decay chain @xmath27 . section [ decays : exc ] is devoted to the exclusive decay modes @xmath20 and @xmath28 , where formulae are given to calculate @xmath4 asymmetries which can be determined experimentally by measuring the difference of @xmath0 and @xmath5 events . in , we present our numerical results for @xmath4-violating observables in the non - resonant domain @xmath8 , taking into account experimental bounds on rare @xmath0 decays such as @xmath14 . we summarize and conclude in . the analytic formulae describing the short - distance effects in the presence of susy as well as the explicit expressions for the form factors are relegated to the appendices . in the mssm , there are new sources of @xmath4 violation . in general , a large number of @xmath4-violating phases appear in the mass matrices as well as the couplings . after an appropriate redefinition of fields one ends up with at least two new @xmath4-violating phases , besides the phase of the ckm matrix and the qcd vacuum angle , which can not be rotated away . for instance , in the mssm with universal boundary conditions at some high scale only two new physical phases arise ; namely @xmath29 associated with the higgsino mass parameter @xmath3 in the superpotential and @xmath30 connected with the soft susy - breaking trilinear mass terms . in order to fulfil the severe constraints on the electric dipole moments ( edm s ) of electron and neutron , one generally assumes that the new phases are less than @xmath31 . since there is no underlying symmetry which would force the phases to be small , this requires fine - tuning . of course , one can relax the tight constraint on these phases by having masses of the superpartners in the @xmath32 region ; this heavy susy spectrum may , however , lead to an unacceptably large contribution to the cosmological relic density . it has recently been pointed out by several authors that it is possible to evade the edm constraints so that phases of @xmath13 still remain consistent with the current experimental upper limits . methods that have been advocated to suppress the edm s include cancellations among different susy contributions @xcite , and nearly degenerate heavy sfermions for the first two generations while being consistent with naturalness bounds . the latter can be realized within the context of so - called ` effective susy ' models @xcite , thereby solving the susy fcnc and @xmath4 problems . to get an idea how supersymmetry affects @xmath4 observables in rare @xmath0 decays , we will consider as illustrative examples the following types of susy models : mssm coupled to @xmath33 supergravity with a universal susy - breaking sector at the grand unification scale . effective susy with near degeneracy of the heavy first and second generation sfermions . in the present analysis , we restrict ourselves to the discussion of flavour - diagonal sfermion mass matrices that is , we assume the ckm matrix to be the only source of flavour mixing . this subsection concerns the mass and mixing matrices relevant to our analysis . in what follows , we will adopt the conventions of . the mass - squared matrix of the charged higgs bosons reads [ massmatrix : higgs ] m_^2= ( ccb+ m_w^2 ^2 + t_1/v_1 & b+m_w^2 + b+ m_w^2 & b+ m_w^2 ^2 + t_2/v_2 ) , with [ bmu - term ] b=2(m_h_1 ^ 2 + m_h_2 ^ 2 + 2||^2 ) , ||^2= -m_z^2 + . here @xmath0 and @xmath3 refer to the complex soft susy - breaking and higgsino mass parameters respectively , @xmath34 are the soft susy breaking higgs - boson masses at the electroweak scale , and @xmath35 stand for the renormalized tadpoles . the mixing angle @xmath36 is defined as usual by @xmath37 , with @xmath38 denoting the tree level vacuum expectation values ( vev s ) of the two neutral higgs fields . in we have adjusted the phase of the @xmath3 parameter in such a way that @xmath39 is real at tree level , thereby ensuring that the vev s of the two higgs fields are real . consequently , the mass matrix becomes real and can be reduced to a diagonal form through a biorthogonal transformation @xmath40 . at the tree level , i.e. @xmath41 in , we have o= ( cc - & + & ) . before proceeding , we should mention that radiative corrections to the higgs potential induce complex vev s . as a matter of fact , @xmath4 violation in the higgs sector leads to an additional phase which , in the presence of chargino and neutralino contributions , can not be rotated away by reparametrization of fields @xcite . as a result , the radiatively induced phase modifies the squark , chargino , and neutralino mass matrices . in the present analysis , we set this phase equal to zero . we now turn to the @xmath42 squark mass - squared matrix which can be written as [ squark : massmatrix ] m^2_= ( cc m^2_&m^2_e^-i _ + m^2 _ e^i _ & m^2 _ ) , = , , in the @xmath43 basis , and can be diagonalized by a unitary matrix @xmath44 such that ( m_^)^2=r_^m_^2 r_^. for subsequent discussion it is useful to define the @xmath45 matrices [ def : stt ] ( ^q_l)_ai=(r_)_ai , ( ^q_r)_ai=(r_)_a , i+3 , q = u , d , with @xmath46 and @xmath47 denoting up- and down - type quarks respectively . working in the so - called ` super - ckm ' basis @xcite in which the @xmath48 quark mass matrices @xmath49 and @xmath50 are real and diagonal , the submatrices in take the form [ squark : up ] m^2__ll=(m_^2)_ll + m_u^2 + m_z^22(3 - 4 ^ 2_w ) , m^2__lr= m_u ( a_u-^ * ) , [ squark : down ] [ squark : down : relation ] m^2__ll= ^(m_^2)_ll+ m_d^2 - m_z^22(3 - 2 ^ 2_w ) , m^2__lr= m_d |a_d-^*| , m^2__rr=(m_^2)_rr + m_d^2 - m_z^22 ^ 2_w , _ = ( a_d-^ * ) . here @xmath51 denotes the weinberg angle , @xmath52 represents a @xmath48 unit matrix , @xmath53 and @xmath54 are hermitian scalar soft mass matrices , and @xmath55 is the usual ckm matrix . in deriving , we have used the relation @xmath56 , which is due to su(2 ) gauge invariance . since we ignore flavour - mixing effects among squarks , @xmath53 and @xmath54 in eqs . ( [ squark : up ] ) and ( [ squark : down ] ) are diagonal and hence real whereas the @xmath57 s are given by [ aterms : diag ] a_u=(a_u , a_c , a_t ) , a_d=(a_d , a_s , a_b ) , a_i |a_i|e^i_a_i . consequently , the squark mass - squared matrix , , in the up - squark sector decomposes into a series of @xmath58 matrices . as far as the scalar top quark is concerned , we have m^2_= ( cc m__l^2 + m_t^2 + m_z^22(3 - 4 ^ 2_w ) & m_t|a_t-^*|e^-i _ + m_t|a_t-^*|e^i _ & m__r^2 + m_t^2 + m_z^22 ^ 2_w ) , where @xmath59 and @xmath60 are diagonal elements of @xmath61 and @xmath62 respectively , while @xmath63 can be readily inferred from . diagonalization of the stop mass - squared matrix then leads to the physical mass eigenstates @xmath64 and @xmath65 , namely ( c _ 1 + _ 2 ) = ( cc _ & _ e^-i _ + -_e^i _ & _ ) ( c _ l + _ r ) ( cc ^u_l_33&^u_r_33 + ^u_l_63&^u_r_63 ) ( c _ l + _ r ) , where the mixing angle @xmath66 is given by the expression ( @xmath67 ) 2_=. the chargino mass matrix can be written as [ chargino : mass - matrix ] m_= ( cc m_2 & m_w + m_w & ||e^i _ ) , where we have adopted a phase convention in which the mass term of the w - ino field , @xmath68 , is real and positive . note that without loss of generality , we can always perform a global rotation to remove one of the three phases from the gaugino mass parameters @xmath69 ( @xmath70 ) . the mass matrix can be cast in diagonal form by means of a biunitary transformation , namely m_^= u^ * m_v^ , where @xmath71 is diagonal with positive eigenvalues , and @xmath46 , @xmath72 are unitary matrices . solving the eigenvalue problem [ chargino : sq ] ( m_^)^2= u^ * m_^m_^u^t= v m_^m_^v^ , we find u= ( cc _ u & _ u e^-i_u + -_u e^i_u&_u ) , v= ( cc _ ve^-i_1 & _ v e^-i(_v+_1 ) + -_v e^i(_v-_2)&_v e^-i_2 ) , with the mixing angles 2_u= , 2_v= , _ v=- , _ 1= , _ we have chosen @xmath73 , @xmath74 , where @xmath75 , and the chargino mass eigenvalues read [ chargino : mass ] + & & \{(m_2 ^ 2-||^2)^2 + 4m_w^4 ^ 2 2 + 4m_w^2 [ m_2 ^ 2+||^2 + 2 ||m_2 2_]}^1/2 ] . we present here the susy lagrangian relevant to the fcnc processes of interest which will also serve as a means of fixing our notation . the interactions of charged higgs bosons , charginos , neutralinos , and gluinos in the presence of new @xmath4 phases can be written as @xcite & = & [ ( |um_u p_l d ) + ( |um_d p_r d)]h^+ + & + & _ j=1 ^ 2[^(x_j^u_lp_l+x_j^u_rp_r ) d + ^(x_j^l_lp_l+x_j^l_rp_r)l ] + & + & _ k=1 ^ 4 [ ^(z_k^d_lp_l+z_k^d_rp_r ) d + ^(z_k^l_lp_l+z_k^l_rp_r)l ] + & - & g_s_b=1 ^ 8 ^(g^d_lp_l - g^d_rp_r)t^b d + , where generation indices have been suppressed , and @xmath76 . the mixing matrices in the super - ckm basis are given by [ feyn : x ] x^u_l_j = g , x^u_r_j= g u_j2^u_l , x^l_l_j =- g v_j1^ * r _ , x^l_r_j= g u_j2r _ , z_k^d_l = - , z_k^d_r = - , z_k^l_l = , z_k^l_r = - , g^d_l = e^-i_3/2^d_l , g^d_r = e^i_3/2^d_r , @xmath77 being the phase of the gluino mass term @xmath78 . ( for scalar lepton as well as neutralino mass and mixing matrices , we refer the reader to . ) in the remainder of this section , we briefly discuss two susy models with quite distinct scenarios for the @xmath4-violating phases . in order to solve the fcnc problem in the mssm and to further reduce the number of unknown parameters , the mssm is generally embedded in a grand unified theory ( gut ) . this leads to the minimal supergravity ( msugra ) inspired model , commonly referred to as the constrained mssm @xcite . in this model one assumes universality of the soft terms at some high scale , which we take to be the scale of gauge coupling unification , @xmath79 , implying that ( i ) all gaugino mass parameters are equal to a common mass @xmath80 ; ( ii ) all the scalar mass parameters share a common value @xmath81 ; and ( iii ) all the soft trilinear couplings are equal to @xmath82 . as a result , the msugra model has only two new independent phases which are associated with the @xmath83 and @xmath82 parameters . after all , we have at the gut scale [ set : modeli : gut ] , m_1/2 , m_0 , |a_0| , |_0| , _ a_0 , _ _ 0 , with @xmath80 and @xmath81 being real . the parameters at the electroweak scale are then obtained by solving the renormalization group equations ( rge s ) . a few remarks are in order here . first , the phases of the gaugino mass terms @xmath69 are not affected by the renormalization group evolution , and therefore the low energy gaugino mass parameters are real . second , the phase that appears together with the @xmath3 parameter does not run at one - loop level so that @xmath84 . moreover , to satisfy the constraints on the edm s of electron and neutron , @xmath85 has to be of @xmath31 unless strong cancellations between different contributions occur . third , solving the rge s for the evolution of the @xmath4-violating phase of the @xmath86-term yields a small value for @xmath87 , and thus is not constrained by the edm s @xcite . lastly , off - diagonal entries occur in the squark mass matrices due to renormalization group evolution of the parameters even in the absence of flavour mixing at the gut scale . however , these effects are found to be small and therefore the squark mass matrix is essentially flavour diagonal at the electroweak scale ( see also ) . as an example of susy models with large cp phases , we consider the effective supersymmetry picture @xcite without assuming universality of sfermion masses at a high scale . within such a framework , the first and second generation sfermions are almost degenerate and have masses above the @xmath88 scale , while third generation sfermions can be light enough to be accessible at future hadron colliders . consequently , fcnc reactions as well as one - loop contributions to the edm s of electron and neutron are well below the current experimental bounds . however , it should be noted that the edm s also receive contributions at two - loop level involving scalar bottom and top quarks that may become important for phases of order unity in the large @xmath89 regime @xcite . in our numerical work , @xmath90 is assumed to be in the interval @xmath91 . let us start with the qcd - corrected matrix element describing the short - distance interactions in @xmath92 within the sm . it is given by [ mael : short ] + & & - ] |l^ p_l l + ( c_10-c_10 ) |l^ p_r l } , where @xmath93 is the four - momentum of the lepton pair , and @xmath94 ( @xmath95 ) in the case of @xmath96 ( @xmath97 ) . in the sm , the wilson coefficients @xmath98 and @xmath99 are real with values of @xmath100 and @xmath101 respectively , and the leading term in @xmath102 has the form @xcite [ wilson : c9app ] = c_9 + ( 3 c_1 + c _ 2)\{g(m_c , q^2 ) + _u}+ , where @xmath103 . the wilson coefficients will be discussed in detail in appendix [ app : wilson ] . in the above expression [ lambdau ] _u \ { l - ^2(-i ) f = s , + - f = d , . with @xmath104 and @xmath105 being the wolfenstein parameters @xcite , where the latter reflects the presence of @xmath4 violation in the sm . for definiteness , we will assume @xmath106 and @xmath107 @xcite . finally , the one - loop function @xmath108 at the scale @xmath109 is given by parameter of the superpotential , we use the notation @xmath110 for the renormalization scale . ] [ loopfunc ] + & & \ { ( 1-y_i)+ ( y_i-1 ) 2 } , where @xmath111 . observe that @xmath112 and @xmath113 loops provide absorptive parts that are mandatory , as we show below , for a non - zero partial width asymmetry besides the presence of a @xmath4-violating phase . throughout this paper , we will assume that in the presence of new physics there are no new operators beyond those that correspond to the wilson coefficients appearing in . ( for a discussion of the implications of new operators for rare @xmath0 decays , see , e.g. , . ) thus , the effect of new physics is simply to modify the matching conditions of the wilson coefficients , i.e. their absolute values and phases at the electroweak scale . as a result , we are left with additional susy contributions at one - loop level to the wilson coefficients @xmath98 , @xmath102 , and @xmath99 in . in fact , they arise from penguin and box diagrams with ( i ) charged higgs boson up - type quark loops ; ( ii ) chargino up - type squark loops ; ( iii ) neutralino down - type squark loops ; and ( iv ) gluino down - type squark loops . thus , the short - distance coefficients can be conveniently written as [ wilson : komplett ] c_i(m_w ) = c_i^(m_w ) + c_i^(m_w ) + c_i^(m_w ) + c_i^(m_w)+ c_i^(m_w ) ( i=7 , , 10 ) . the explicit expressions for the various wilson coefficients are given in appendix [ app : wilson ] . since we limit our attention to flavour - conserving effects in the squark sector , the neutralino and gluino exchange contributions in will be omitted in our numerical calculations . for future reference , we parametrize the new physics contributions as follows : [ ratio : np - sm ] r_i = |r_i| e^i _ i , = , where @xmath114 is real to a good approximation within the models under study . we have considered so far only the short - distance interactions . a more complete analysis , however , has also to take into account resonance contributions due to @xmath113 , @xmath115 , and @xmath112 intermediate states , i.e. @xmath116 , and so forth . a detailed discussion of the various theoretical suggestions of how to describe these effects is given in . we employ here the approach proposed in which makes use of the renormalized photon vacuum polarization , @xmath117 , related to cross - section data ( s ) , with @xmath118 . in fact , the absorptive part of the vacuum polarization is given by ^_(s ) = ( s ) , whereas the dispersive part may be obtained via a once - subtracted dispersion relation @xcite ^_(s)= p _ 4m_^2^d s , with @xmath119 denoting the principal value . for example , in the case of the @xmath26 family ( i.e. @xmath120 ) the imaginary part of the one - loop function @xmath121 , , can be expressed as [ im : gmc ] g(m_c , s)=^j/(s ) , ^j/(s)(s)+ ( s ) , where the subscripts ` cont ' and ` res ' stand for continuum and resonance contributions respectively , while the real part is given by [ re : gmc ] g(m_c , s)= -(m_c / m_b)-+ p _ 4m_^2^d s. the contributions from the continuum can be determined by means of experimental data given in , whereas the narrow resonances are well described by a relativistic breit - wigner distribution . however , in order to reproduce correctly the branching ratio for direct @xmath26 production via the relation @xmath122 ) [ rel : factorize ] ( bh v_c|ch l^+l^-)= ( bh v_c|c)(v_c|cl^+l^- ) , where @xmath123 stands for pseudoscalar and scalar mesons , one has to multiply @xmath124 in eqs . ( [ im : gmc ] ) and ( [ re : gmc ] ) by a phenomenological factor @xmath125 , regardless of which method one uses for the description of the resonances @xcite . in the approximation of which has to be corrected to @xmath126 , taking into account non - factorizable contributions in two - body @xmath0 decays ( see , e.g. , ) . ] using the form factors of ( see next section ) together with experimental data on @xmath127 , @xmath128 , and @xmath129 @xcite , we find a magnitude for @xmath125 of @xmath130 to @xmath131 . at this point two remarks are in order . first , the branching ratio for direct @xmath26 and @xmath132 production is enhanced by a factor @xmath133 , while it is essentially independent of @xmath125 outside the resonance region . second , the numerical results for average @xmath4 asymmetries in the non - resonant continuum @xmath134 are not affected by the uncertainty in @xmath125 . similar considerations also hold for @xmath113 and @xmath115 systems except that the @xmath135 resonance is described through ( s ) = ( 1-)^3/2|f_(s)|^2 , where the pion form factor is given by a modified gounaris - sakurai formula @xcite . the hadronic matrix elements in exclusive @xmath0 decays can be written in terms of @xmath138-dependent form factors , where @xmath138 is the invariant mass of the lepton pair . in the work described here , we employ heavy - to - light @xmath16 and @xmath139 form factors determined by melikhov and nikitin @xcite within a relativistic quark model . to get an estimate of the theoretical uncertainty that is inherent to any model for the form factors , we also utilize the parametrization of colangelo @xcite , which makes use of qcd sum rule predictions . for simplicity of presentation , we do not display corrections due to a non - zero lepton mass , which can be found in . ( the same remark applies to the light quark masses @xmath140 . ) henceforth we shall denote pseudoscalar and vector mesons by @xmath141 and @xmath142 respectively . the hadronic matrix elements for the decays @xmath143 can be parametrized in terms of three lorentz - invariant form factors ( see appendix [ formfactors ] for details ) , namely = [ ( 2p - q ) _ f_+(q^2)+ q _ f_-(q^2 ) ] , = -[(2p - q)_q^2 - ( m_b^2 - m_p^2)q_]s(q^2 ) , with @xmath144 . here we assume that the form factors are real , in the absence of final - state interactions . note that the terms proportional to @xmath145 may be dropped in the case of massless leptons . introducing the shorthand notation [ def : triangle ] (a , b , c ) = a^2 + b^2 + c^2 - 2 ( a b + b c + a c ) , [ def : x ] x_i = ^1/2(m_b^2 , m_i^2 , s ) , and recalling @xmath118 , the differential decay rate can be written as ( neglecting @xmath146 and @xmath147 ) [ diff : btop ] = |v_tb^v_tf^|^2 x_p^3 ^2_l . here @xmath148 is the angle between @xmath23 and the outgoing hadron in the dilepton centre - of - mass system , and the wilson coefficients are collected in appendix [ app : wilson ] . defining the forward - backward ( fb ) asymmetry as [ fb ] a_(s)= , which is equivalent to the energy asymmetry discussed in , it follows directly from the distribution in that @xmath149 vanishes in the case of @xmath150 transitions . we note in passing that , given an extended operator basis ( e.g. in models with neutral higgs - boson exchange ) , new dirac structures @xmath151 and @xmath152 may occur in , giving rise to a non - zero fb asymmetry in @xmath150 . the hadronic matrix elements describing the decays @xmath153 are characterized by seven independent form factors , which we present in appendix [ formfactors ] , defined through ( @xmath154 ) [ def : formfactors : btov ] + & & = i _ ^p^q^ g(q^2 ) - \ { _ ^ * f(q^2 ) + ( ^*q)[(2p -q)_a_+(q^2)+ q_a_-(q^2 ) ] } , + & & ( ^*q)\{(2p - q)_+ q _ } , @xmath155 being the polarization vector of the final - state meson , and @xmath144 . the differential decay rate for @xmath156 in the case of massless leptons and light quarks takes the form ( @xmath96 or @xmath157 ) [ diff : btov ] = |v_tb^v_tf^|^2 x_v , with @xmath158 as in . the quantities @xmath159 , @xmath0 , and @xmath160 are [ double : a ] a(s ) & = & , [ double : b ] b(s ) = 8 x_v \{c_10^*[s a_x a_y - m_b ( a_x b_y + a_y b_x ) ] } , [ double : c ] c(s ) = , where @xmath161 , and [ definition : alphai ] _ 1(s ) = ( ||^2 + |c_10|^2 ) a_x^2 + b_x^2 -_x b_x , _ 2(s ) = _ 1(s ) _ xy , _ 3 ( s ) = _ 1(s ) _ xz , _ 4(s ) = ( ||^2 + |c_10|^2 ) a_y a_z + _ yb_z -(a_y b_z+a_z b_y ) , in which the @xmath162 s and @xmath163 s are defined as a_x = g(s),a_y= f(s),a_z= a_+(s ) , b_x= g_+(s ) , b_y= g_+(s)(m_b^2-m_v^2)+ s g_-(s ) , b_z= - . the complex wilson coefficients @xmath98 , @xmath102 , and @xmath99 are given in appendix [ app : wilson ] . finally , using eqs . ( [ fb ] ) and ( [ diff : btov ] ) , we derive the forward - backward asymmetry [ fb : btov ] a_(s)= 12 x_v . to discuss @xmath4-violating asymmetries , let us first recall the necessary ingredients . suppose the decay amplitude for @xmath164 has the general form [ matrixelement : itof ] a(f)=e^i_1a_1 e^i_1 + e^i_2a_2 e^i_2 , where @xmath165 and @xmath166 denote strong phases ( @xmath4-conserving ) and weak phases ( @xmath4-violating ) respectively ( @xmath167 and @xmath168 being real ) . together with the decay amplitude for the conjugate process ( b|f ) = e^-i_1a_1 e^i_1 + e^-i_2a_2 e^i_2 , which can be obtained from by means of @xmath169 invariance , we may define the @xmath4 asymmetry as [ def : cpasym : gen ] a_= , with @xmath170 , @xmath171 , and @xmath172 . as can be easily seen from , a non - zero partial rate asymmetry requires the simultaneous presence of a @xmath4-violating phase @xmath173 as well as a @xmath4-conserving dynamical phase @xmath174 , the latter being provided by the one - loop functions @xmath121 and @xmath175 present in the wilson coefficient @xmath102 [ ] . notice that in the limit in which the charm quark mass equals the up quark mass there is no @xmath4 violation in the sm . given the differential decay distribution in the variables @xmath176 and @xmath177 , we can construct the following @xmath4-violating observables : [ def : cpasym ] [ def : cpasymsd ] a_^d , s(s)= , _ d , s_0 ^ 1_-1 ^ 0 , where we have introduced _ = ( |bh l^+l^-)- ( b |hl^+l^- ) , _ = ( |bh l^+l^-)+ ( b |hl^+l^- ) , with @xmath178 . it should be noted that the asymmetry @xmath179 represents a @xmath4-violating effect in the angular distribution of @xmath23 in @xmath0 and @xmath5 decays while @xmath180 is the asymmetry in the partial widths of these decays . as can be seen from eqs . ( [ diff : btop ] ) , ( [ diff : btov ] ) , and ( [ fb : btov ] ) , the latter involves the phases of @xmath98 and @xmath102 while the former is also sensitive to the phase of the wilson coefficient @xmath99 . given the susy contributions presented in the preceding sections , we now proceed to study the implications of supersymmetry for exclusive @xmath0 decays . in our numerical analysis , we scan the susy parameter space as given in and take as input the parameters displayed in appendix [ app : numerics ] . in addition , we take into account the following experimental constraints : from the measurement of the inclusive branching ratio @xmath181 , which probes @xmath182 , one can derive upper and lower limits @xcite : [ range : btosg ] 2.010 ^ -4 < ( x_s ) < 4.510 ^ -4 ( ) . this is specially useful to constrain extensions of the sm . indeed , following the model - independent analysis performed in , and taking the wilson coefficient @xmath98 in leading - log approximation [ see of the appendix ] , we obtain [ btosg : theo ] ( x_s ) & & [ 0.801 + 0.444 |r_7|^2 + 0.002 |r_8|^2 + 1.192r_7 + & & + 0.083r_8 + 0.061 ( r_7 r_8^*)]10 ^ -4 , where @xmath183 and @xmath184 are as in . from we infer that the present @xmath185 measurement already excludes many solutions for @xmath183 . a cdf search for the exclusive decays of interest yields the upper limits @xmath186 and @xmath187 at the @xmath188 @xcite . note that the @xmath189 upper limit is close to the sm prediction @xcite . as for the modes @xmath190 and @xmath191 , we are not aware of any such limits . non - observation of any susy signals at lep 2 and the tevatron imposes the following lower bounds @xcite : m _ > 86 , m _ > 260 , m_h^ > 90 . as mentioned earlier , we investigate @xmath4 asymmetries in the low dilepton invariant mass region , i.e. @xmath192 , which is of particular interest because the low-@xmath176 region is sensitive to the wilson coefficient @xmath98 ( in the case of @xmath193 ) . in fact , it can receive large susy contributions and be complex , as well as change sign , while being consistent with the experimental measurement of @xmath185 . on the other hand , the new - physics effects are known to alter the remaining wilson coefficients @xmath102 and @xmath99 only slightly within msugra and effective susy with no additional flavour structure beyond the usual ckm mechanism @xcite . moreover , @xmath4 asymmetries above the @xmath26 resonance are dominated by @xmath112 resonant intermediate states , whereas below @xmath194 the @xmath135 resonance has a strong influence on the asymmetry . this can be seen from , where we show the @xmath4 asymmetry @xmath180 [ ] between @xmath195 and @xmath196 as a function of the dilepton invariant mass within the sm and in the presence of susy contributions with new @xmath4-violating phases . it is evident that the predictions for @xmath4 asymmetries suffer from large theoretical uncertainties in the neighbourhood of the @xmath104 resonance and above the @xmath26 , as discussed in . using eqs . ( [ diff : btop ] ) and ( [ diff : btov ] ) together with the definition for @xmath4-violating asymmetries , eqs . ( [ def : cpasym ] ) , we can summarize our main findings as follows : the @xmath4-violating asymmetry in the @xmath23 spectra of @xmath0 and @xmath5 decays , @xmath179 , vanishes in the case of @xmath143 transitions . within the framework of the constrained mssm with phases of @xmath31 numerical values for the average asymmetry @xmath198 in the low-@xmath176 region are comparable to the sm predictions with asymmetries of @xmath199 and @xmath10 for @xmath200 and @xmath201 transitions respectively . our results for @xmath180 between the decays @xmath202 and @xmath203 in the context of effective susy with a light stop @xmath204 and phases of @xmath205 are shown in figs . [ acpseff - kdecay ] and [ acpseff - pidecay ] for low and large @xmath90 solutions which correspond to @xmath206 and @xmath207 respectively . observe that the @xmath4-violating asymmetry @xmath180 in @xmath143 depends only weakly on the sign and phase of @xmath98 . this is due to the fact that @xmath98 , which is constrained by the @xmath185 measurement and not enhanced by a factor of @xmath208 in the low-@xmath176 region , is nearly one order of magnitude smaller than the leading term in @xmath102 [ cf . ] . the @xmath4 asymmetry in the partial widths of @xmath209 and @xmath210 changes sign for large values of @xmath211 , while @xmath212 ( see ) . however , non - standard contributions to @xmath211 are found to be small and hence @xmath213 . on the other hand , average asymmetries of @xmath214@xmath215@xmath216 are predicted for @xmath198 in the case of @xmath11 , even for values of @xmath211 as small as @xmath217 ( see ) . given a typical branching ratio of @xmath218 and a nominal asymmetry of @xmath219 , a measurement at @xmath220 level requires @xmath221 @xmath222 pairs . ( this rather challenging task might be feasible at lhc and the tevatron . ) the small magnitude of the @xmath4 asymmetry is also due to a suppression factor multiplying the indispensable absorptive part in @xmath102 , which is only slightly affected by new - physics contributions . indeed , it follows from eqs . ( [ wilson : c9app ] ) and ( [ def : cpasym : gen ] ) that a_r~~10 ^ -2 , where the weak and strong phases @xmath173 and @xmath174 can be of order unity . numerical estimates for average @xmath4 asymmetries are mildly affected by the parametrization of form factors ( see also ) . the contribution of the wilson coefficient @xmath98 ( or equivalently @xmath183 ) to the decay rate in @xmath153 modes is enhanced by a factor of @xmath208 in the low-@xmath176 region . as seen from , in the case of @xmath12 , the @xmath4-violating asymmetry @xmath180 can change sign for @xmath223 ( i.e. @xmath224 ) , while for large @xmath90 it is always negative . for small values of @xmath211 an average @xmath4 asymmetry of about @xmath10 is predicted for both @xmath223 and @xmath225 solutions . since the distributions of @xmath180 for @xmath226 are very similar to the ones obtained for @xmath227 , we refrain from showing the corresponding plots . as we have already mentioned , the @xmath4-violating asymmetry in the angular distribution of @xmath23 in @xmath0 and @xmath5 decays can , in principle , probe the phase of the wilson coefficient @xmath99 . this is shown in , where we have plotted the @xmath4 asymmetry as a function of @xmath211 and @xmath228 for large @xmath90 ( i.e. @xmath207 ) . unfortunately , the msugra and effective susy predictions for the average asymmetry @xmath229 turn out to be unobservably small . in this paper , we have studied the consequences of new @xmath4-violating phases for exclusive @xmath0 decays within the framework of supersymmetric extensions of the sm , ignoring intergenerational mixing in the squark sector . we have examined @xmath4-violating asymmetries in the partial widths as well as angular distribution of @xmath23 between the exclusive channels @xmath230 and @xmath231 in the invariant mass region @xmath232 . the essential conclusion of our analysis is that it is not sufficient to have additional @xmath4 phases of @xmath13 in order to obtain large @xmath4-violating effects . within the constrained mssm and effective susy with a complex ckm matrix and additional @xmath4 phases , we obtain values for the average asymmetry @xmath198 of about @xmath9 ( @xmath10 ) in the decays @xmath233 and @xmath234 , taking into account experimental constraints on edm s of electron and neutron , as well as rare @xmath0 decays such as @xmath14 . as for the asymmetry in the angular distribution , @xmath229 , it probes the phase of the wilson coefficient @xmath99 , but will be unobservable at future colliders . numerical estimates of the @xmath4 asymmetries in the decays @xmath6 and @xmath235 turn out to be small ( less than @xmath15 ) and are comparable to the sm result . our analysis shows that the smallness in @xmath4 asymmetries is mainly due to the coefficient @xmath236 which multiplies the requisite absorptive part in @xmath102 [ ] , and which is only slightly affected by the new - physics contributions discussed in . therefore , any sizable @xmath4-violating effect in the low-@xmath176 region requires large non - standard contributions to the short - distance coefficient @xmath102 and/or @xmath99 , as well as additional @xmath4 phases of @xmath13 . by the same token , any large @xmath4-violating effect would provide a clue to physics beyond the sm . a detailed discussion of this point will be given elsewhere . one could argue , however , that the inclusion of flavour off - diagonal contributions ( i.e. gluino and neutralino diagrams ) to the wilson coefficients might lead to higher @xmath4 asymmetries . in fact , it has been pointed out in that even in the presence of large supersymmetric @xmath4 phases , a non - trivial flavour structure in the soft - breaking terms is necessary in order to obtain sizable contributions to @xmath4 violation in the @xmath237 system and to @xmath4 asymmetries in two - body neutral @xmath0 decays ( see also ) . using the mass insertion approximation , such effects have recently been studied in which predicts a partial width asymmetry for @xmath238 of a few per cent in the low-@xmath176 domain . finally , we would like to recall that , nevertheless , large @xmath4 asymmetries may occur in rare @xmath0 decays like the observed @xmath14 modes , where @xmath239 can be substantial ( up to @xmath240 ) in some part of the parameter space @xcite . f.k . would like to thank gudrun hiller for useful discussions . this research has been supported by the tmr network of the ec under contract erbfmrx - ct96 - 0090 . unless otherwise specified , we use the experimental values as compiled by the particle data group @xcite and the parameters displayed in . [ eqs : input ] l m_t=175 , m_b=4.8 , m_c=1.4 , m_s=170 , m_d=10 , + m_u=5 , = 1/129 , for the sake of convenience , we provide in this appendix formulae for the wilson coefficients @xmath98 , @xmath102 , and @xmath99 in the presence of susy , using the results derived in . since we study the case of massless leptons , we retain only those contributions that do not vanish in the limit @xmath241 . as for @xmath242 leptons in the final state , there are further charged and neutral higgs - boson contributions [ see also eqs . ( [ feyn : x ] ) ] . introducing the shorthand notation [ notation ] _ s= , r_w= , r_= , = , the wilson coefficient @xmath98 evaluated at @xmath109 has the form ( in leading - log approximation ) [ wilson : c7eff ] = _ s^16/23c_7(m_w)+(_s^14/23-_s^16/23)c_8(m_w)+_i=1 ^ 8 h_i_s^a_i , with the coefficients @xmath243 , @xmath244 tabulated in . recalling eqs . ( [ feyn : x ] ) and ( [ wilson : komplett ] ) , and using the one - loop functions @xmath245 listed in appendix [ app : auxfuncs ] , the various contributions to @xmath246 can be written as follows : standard model : [ c7:sm ] c_7^(m_w)= r_w f_1(r_w ) . charged higgs boson : c_7^(m_w)=[r _ f_1(r_)^2 + 2 f_2(r _ ) ] . chargino : . however , using the explicit form for the squark mixing matrices [ eqs . ( [ feyn : x ] ) ] , the constant term vanishes reflecting the unitarity of the mixing matrices . ] [ c7mw : chargino ] + & & , where we have defined n=\ { l1f = d , + 2f = s . . for completeness , we also give the expressions for the neutralino and gluino contributions which vanish in the limit of flavour - diagonal squark mass matrices . neutralino : [ c7:neutralino ] + & = & -_a=1 ^ 6_k=1 ^ 4 . gluino : c_7^(m_w)= -_a=1 ^ 6 . the corresponding expressions @xmath247 are obtained changing @xmath248 in , with @xmath249 collected in appendix [ app : auxfuncs ] , while the gluino contribution reads c_8^(m_w)= -_a=1 ^ 6 . the wilson coefficient @xmath102 at @xmath109 in next - to - leading approximation is given by [ wilson : c9 ] & = & c_9 + g(m_c , s ) ( 3 c_1 + c _ 2 + 3c_3 + c_4 + 3c_5 + c_6 ) + & + & _u(3 c_1 + c _ 2 ) - g(m_f , s ) ( c_3 + 3 c_4 ) + & - & g(m_b , s)(4c_3 + 4c_4 + 3c_5+c_6)+ ( 3c_3 + c_4 + 3c_5+c_6 ) , where @xmath250 and @xmath251 are defined in eqs . ( [ lambdau ] ) and ( [ loopfunc ] ) respectively , with @xmath118 . as far as the wilson coefficients @xmath252@xmath253 are concerned , we have numerically c_1=-0.249 , c_2=1.108 , c_3=0.011 , c_4=-0.026 , c_5=0.007 , c_6=-0.031 , using the values given in appendix [ app : numerics ] . further , [ c9 ] c_9= c_9(m_w)-+ p_0 + p_e _ i e^i , with @xmath254 , and [ c9mw ] c_9(m_w)= _ i(-4z^i)+ , where the analytic expressions for @xmath255 , @xmath256 , and @xmath257 are given in . since @xmath258 , we shall keep only the sm contribution in the last term of . moreover , as discussed in , the order @xmath259 correction in due to one - gluon exchange may be regarded as a contribution to the form factors , and hence we set @xmath260 in . turning to the wilson coefficient @xmath99 , it has the simple form c_10(m_w ) = -_i . note that the corresponding operator does not renormalize and thus @xmath261 . the expressions for the various contributions to @xmath262 and @xmath263 read as follows : standard model : y^= f_9(r_w ) , z^= f_10(r_w ) . charged higgs : y^y^_z , z^z^_+z^_z , z^_=-f_6(r_)^2 , y^_z= z^_z =- r_w f_5(r_)^2 . chargino : y^y^_z+ y^ _ , z^z^_+z^_z , z^_= _ a=1 ^ 6 _ j=1 ^ 2 [ ( x_j^u_l)_na(x_j^u_l)_a3f_7 ( ) ] , + & -&c_2(m^2_,m^2_,m^2 _ ) _ abv^*_i1v_j1^+m_m _ c_0(m^2 _ , m^2 _ , m^2_)_ab u_i1^u^*_j1 ] } , y^_= _ a=1 ^ 6 _ i , j=1 ^ 2[(x_i^u_l)_na(x_j^u_l)_a3 d_2(m^2 _ , m^2 _ , m^2 _ , m^2_)v_i1^*v_j1 ] , with @xmath264 in the case of @xmath265 in the final state . neutralino : y^y^_z+ y^ _ , z^z^_+z^_z+ z^ _ , z^_=- _ a= 1 ^ 6_k= 1 ^ 4 [ ( z_k^d_l)_na(z_k^d_l)_a3 f_8 ( ) ] , + & & } , + & & } , + & & } , with @xmath266 for @xmath265 in the final state . gluino : y^y^_z , z^z^_+z^_z , z^_= -_a= 1 ^ 6 [ ( g^d_l)_na(g^d_l)_a3 f_8 ( ) ] , y^_z = z^_z= _ a , b= 1 ^ 6[(g^d_l)_na(g^d_l)_b3 c_2(m^2 _ , m^2 _ , m^2 _ ) ( g^d_rg^d_r)_ab ] . the functions @xmath245 , @xmath267 , and @xmath268 are given in below . here we list the functions @xmath245 , @xmath249 , @xmath267 , and @xmath268 introduced in the previous section . [ fis ] f_1(x)= - x , f_2(x)= + x , f_3(x)= + x , f_4(x)= + x , f_5(x)= + x , f_6(x)= + x , f_7(x)= + x , f_8(x)= + x , f_9(x)=+x , f_10(x)= -x , [ gis ] g_1(x)=-3f_3(x),g_2(x)=3[f_4(x)-1/2 ] , g_3(x)=-3f_3(x ) , g_4(x)=-3f_4(x ) , g_5(x)= , g_6(x)= , [ cis ] c_0(m_1 ^ 2 , m_2 ^ 2 , m_3 ^ 2 ) = - , [ cis : c2 ] c_2(m_1 ^ 2 , m_2 ^ 2 , m_3 ^ 2 ) = - , [ dis ] + & = & - , + & = & - . in tables [ table : ff : mel ] and [ table : ff : col ] we summarize the two different sets of form factors discussed in , which are related via [ formfactors : relation ] f_1(q^2)=f_+(q^2 ) , f_0(q^2)=f_+(q^2 ) + f_-(q^2 ) , f_t(q^2)= - ( m_b + m_p)s(q^2 ) , v(q^2 ) = ( m_b+m_v ) g(q^2 ) , a_1(q^2)= , a_2(q^2)= -(m_b+m_v ) a_+(q^2 ) , a_0(q^2)= , t_1(q^2)= - g_+(q^2 ) , t_2(q^2)= - , t_3(q^2)= . n. cabibbo , ; m. kobayashi and k. maskawa , . for recent reviews , see s. bertolini , j. o. eeg , and m. fabbrichesi , hep - ph/0002234 , and references therein ; a. j. buras , hep - ph/9908395 , talk given at _ 1999 chicago conference on kaon physics ( kaon 99 ) _ , chicago , il , 1999 . a. g. cohen , d. b. kaplan , and a. e. nelson , ; a. d. dolgov , hep - ph/9707419 ; v. a. rubakov and m. e. shaposhnikov , usp . nauk * 166 * , 493 ( 1996 ) [ phys . usp . * 39 * , 461 ( 1996 ) ] . a. ali , . j. l. hewett and j. d. wells , . m. misiak , s. pokorski , and j. rosiek , in _ heavy flavours ii _ , edited by a. j. buras and m. lindner ( world scientific , singapore , 1998 ) , p. 795 , hep - ph/9703442 . h. p. nilles , ; h. e. haber and g. l. kane , ; j. f. gunion and h. e. haber , ; ( e ) ; j. f. gunion , h. e. haber , g. l. kane , and s. dawson , _ the higgs hunter s guide _ ( addison - wesley , reading , ma , 1990 ) . for recent reviews , see m. kuroda , hep - ph/9902340 ; j. c. romo , hep - ph/9811454 . s. bertolini , f. borzumati , a. masiero , and g. ridolfi , . a. ali , g. f. giudice , and t. mannel , . p. cho , m. misiak , and d. wyler , . t. goto , y. okada , y. shimizu , and m. tanaka , ; t. goto , y. okada , and y. shimizu , . t. goto , y .- y . keum , t. nihei , y. okada , and y. shimizu , . y. g. kim , p. ko , and j. s. lee , . s. baek and p. ko , ; , and references therein . huang , w. liao , and q .- s . yan , ; c .- s . huang and q .- s . e. lunghi , a. masiero , i. scimemi , and l. silvestrini , . see , for example , c. greub , a. ioannissian , and d. wyler , ; g. burdman , ; ; d. melikhov , n. nikitin , and s. simula , ; t. m. aliev and m. savci , . a. ali , p. ball , l. t. handoko , and g. hiller , , and references therein . e. lunghi and i. scimemi , . f. krger , l. m. sehgal , n. sinha , and r. sinha , . f. krger and l. m. sehgal , ; ( e ) . t. ibrahim and p. nath , ; ( e ) ; ; m. brhlik , g. j. good , and g. l. kane , ; a. bartl , t. gajdosik , w. porod , p. stockinger , and h. stremnitzer , ; s. pokorski , j. rosiek , and c. a. savoy , . s. a. abel and j. m. frre , . a. g. cohen , d. b. kaplan , and a. e. nelson , ; a. pomarol and d. tommasini , ; a. g. cohen , d. b. kaplan , f. lepeintre , and a. e. nelson , . see the second paper in . a. pilaftsis , ; ; m. brhlik and g. l. kane , ; d. a. demir , ; ; a. pilaftsis and c. e. m. wagner , . t. falk and k. a. olive , ; e. accomando , r. arnowitt , and b. dutta , phys . rev . d ( to be published ) , hep - ph/9907446 . d. a. demir , a. masiero , and o. vives , , and references therein . d. chang , w .- y . keung , and a. pilaftsis , ; ( e ) . a. j. buras and m. mnz , ; m. misiak , ; ( e ) . for a review , see g. buchalla , a. j. buras , and m. e. lautenbacher , . l. wolfenstein , . f. parodi , p. rodeau , and a. stocchi , ; a. ali and d. london , 999687 . d. guetta and e. nardi , ; t. g. rizzo , ; s. fukae , c. s. kim , t. morozumi , and t. yoshikawa , . a. ali and g. hiller , . f. krger and l. m. sehgal , ; . n. cabibbo and r. gatto , . h. burkhardt and b. pietrzyk , . a. ali , t. mannel , and t. morozumi , . z. ligeti and m. b. wise , ; see also z. ligeti , i. w. stewart , and m. b. wise , . f. m. al - shamali and a. n. kamal , , and references therein . d. melikhov and n. nikitin , hep - ph/9609503 . particle data group , c. caso , 3981 . t. kinoshita , b. nii , and y. okamata , ; f. jegerlehner , . p. colangelo , f. de fazio , p. santorelli , and e. scrimieri , ; ( e ) . d. melikhov , n. nikitin , and s. simula , . aleph collaboration , r. barate , ; cleo collaboration , s. ahmed , hep - ex/9908022 ; see also g. eigen , hep - ex/9901005 , talk given at _ 4th international symposium on radiative corrections ( radcor 98 ) _ , barcelona , 1998 . a. l. kagan and m. neubert , 7995 . cdf collaboration , t. affolder , . a. ali , . d. melikhov , n. nikitin , and s. simula , . m. brhlik , l. everett , g. l. kane , s. f. king , and o. lebedev , ; d. a. demir , a. masiero , and o. vives , hep - ph/9911337 . a. l. kagan and m. neubert , . m. ciuchini , g. degrassi , p. gambino , and g. f. giudice , ; ; c. bobeth , m. misiak , and j. urban , .

we invoke star formation triggered by cloud - cloud collisions to explain global star formation rates of disk galaxies and circumnuclear starbursts . previous theories based on the growth rate of gravitational perturbations ignore the dynamically important presence of magnetic fields . theories based on triggering by spiral density waves fail to explain star formation in systems without such waves . furthermore , observations suggest gas and stellar disk instabilities are decoupled . following gammie , ostriker & jog ( 1991 ) , the cloud collision rate is set by the shear velocity of encounters with initial impact parameters of a few tidal radii , due to differential rotation in the disk . this , together with the effective confinement of cloud orbits to a two dimensional plane , enhances the collision rate above that for particles in a three dimensional box . we predict . for constant circular velocity ( ) , this is in agreement with recent observations ( kennicutt 1998 ) . we predict a b - band tully - fisher relation : , also consistent with observations . as additional tests , we predict enhanced star formation in regions with relatively high shear rates , and lower star formation efficiencies in clouds of higher mass . subject headings : galaxies : starburst and spiral ism : clouds stars : formation

astro-ph9906355

understanding how the global star formation rates ( sfrs ) of galaxies and starbursts depend on their physical properties is essential for an understanding of galaxy evolution . furthermore , such knowledge can also reveal much about the star formation process itself . empirically , in disk galaxies ( kennicutt 1989 & 1998 , hereafter k89 & k98 ) and the circumnuclear disks of starbursts ( kenney 1997 ; downes & solomon 1998 , hereafter ds98 ) , star formation occurs in regions where the gas disk is unstable to gravitational perturbation growth . this can be expressed as a condition on the surface density of gas : @xmath3 ( toomre 1964 , quirk 1972 ) , where @xmath4 is the gas velocity dispersion ; @xmath5 is a dimensionless constant near unity , to account for deviations of real disks from the idealized toomre thin disk , single fluid model ; @xmath6 is a dimensionless parameter and @xmath7 is the epicyclic frequency : @xmath8 @xmath9 is the circular velocity at a particular galactocentric radius @xmath10 , and @xmath11 , which is @xmath12 for a flat rotation curve . from the outermost galactic star forming regions , k89 finds @xmath13 , assuming @xmath14 . @xmath15 is expected , because of the destabilizing influence of a stellar disk ( jog & solomon 1984 , jog 1996 ) . where @xmath16 , for a certain range of scales , the gas disk is gravitationally unstable , and fragments into bound clouds . when stars form , the energy they release raises @xmath4 and star formation is hypothesized ( e.g. silk 1997 ) and observed ( k89 , ds98 ) to self - regulate , so that @xmath17 . all star formation is observed to occur in molecular clouds , and the majority in giant molecular clouds ( gmcs ) , with masses @xmath18 ( see blitz & williams 1999 and mckee 1999 for reviews ) . however , k89 reported the surprising result that the correlation of sfr with the surface density of molecular gas was much weaker than with the total ( atomic + molecular ) . uncertainties in co to @xmath19 conversion may account for some of the poor correlation , however , the data suggest the immediate supply of gas controlling the sfr is both atomic and molecular . this implies the atomic to molecular conversion timescale , @xmath20 , is short compared to the timescale on which star formation is regulated . spitzer ( 1978 ) finds the rate constant for molecule formation on dust grains to be approximately @xmath21 , for typical galactic interstellar medium ( ism ) metallicities . ignoring destruction processes , a naive estimate of the time to convert a region with @xmath22 , perhaps created from the collision of two atomic clouds , to h@xmath23 , gives @xmath24 , which is a relatively short timescale . where @xmath25 , the sfr is observed to be correlated with gas density . schmidt ( 1959 ) introduced the parameterization of the volume densities @xmath26 , with @xmath27 . by looking at about one hundred different galactic and circumnuclear starburst disk systems , k98 found a similar relation for disk averaged surface densities of gas and star formation , valid over five orders of magnitude in @xmath28 , @xmath29 with @xmath30 ( figure [ fig : ken1 ] ) ( however , see taniguchi & ohyama 1998 ) . k98 finds the sfr is also correlated with the orbital angular frequency , @xmath31 , via @xmath32 ( figure [ fig : ken2 ] ) . @xmath31 is measured at the outer radius of the star forming region ( see [ diskav ] ) . previous theories for explaining these relations fall into two broad categories , based either on the growth rate of gravitational perturbations in a disk or on the triggering of star formation in gas passing through spiral or bar density waves . in this paper we present a third paradigm , in which cloud collisions determine the sfr . the timescale for perturbation growth can be expressed as @xmath33 ( e.g. larson 1988 , 1992 ; elmegreen 1994 ; wang & silk 1994 ) , and so @xmath34 . assuming a constant disk scaleheight , we obtain equation ( [ kenlaw ] ) with @xmath35 for local surface densities . however , disk averaged quantities will depend on the radial gas distribution . we can also express @xmath36 . perturbation growth via swing amplification in a differentially rotating disk grows in a similar manner ( e.g. larson 1988 ) . by assuming star formation self - regulates and keeps @xmath6 constant , larson ( 1988 ) and wang & silk ( 1994 ) predict @xmath37 , since @xmath38 , for disks with flat rotation curves . however , these theories neglect the effects of magnetic fields and the viscosity of the ism . gammie ( 1996 ) finds these significantly reduce the growth rate of non - axisymmetric perturbations for typical galactic conditions . furthermore , most galactic disk stars form in localized , highly clustered regions ( lada , strom & myers 1993 ) in gmcs , and most of the gas in the disk , including most of the bound gas , is not directly involved in the star formation . in gmcs static magnetic fields play a dynamically important role ( e.g. mckee 1999 , heiles et al 1993 ) . their presence sets a critical mass , @xmath39 ( bertoldi & mckee 1992 ) for spherical clouds , where @xmath40 and @xmath41 . for the diffuse ism ( elmegreen 1985 ; mestel 1985 ) with @xmath42 and @xmath43 , we have @xmath44 , which is typical for a gmc . below this mass gravitational collapse is impossible without ambipolar diffusion , and even for greater masses , there will still be a significant effect . collapse may also be impeded by turbulent magnetic pressure , generated from energy injected by the first stars to form in a cloud . mckee ( 1999 ) has modeled these higher mass clouds as being in approximate hydrostatic equilibrium with low mass star formation providing support . the data supporting the empirical schmidt laws ( equations [ kenlaw ] and [ larson ] ) are sensitive only to high mass stars . since collapse mediated by ambipolar diffusion occurs on a timescale , @xmath45 , much greater than the free - fall time , @xmath46 , we conclude it is inaccurate to use the rate of purely gravitational perturbation growth of the disk averaged ism and ignoring magnetic fields , to predict global sfrs . the spatial correlation of star formation with large scale spiral structure in some disk galaxies motivates theories for the triggering of star formation during the passage of gas through density waves . wyse ( 1986 ) and wyse & silk ( 1989 ) propose a sfr law of the form @xmath47 where @xmath48 is the pattern frequency of the spiral density wave . in the limit of small @xmath48 and for @xmath49 we recover equation ( [ larson ] ) . the outer radius of star formation is predicted to be the co - rotation radius , in approximate agreement with observations . increased cloud collision rates and increased perturbation growth rates in the arms , where @xmath6 is locally lowered , have been suggested as the star formation triggering mechanism . these will be further discussed in [ sec : spiral ] . ho , filippenko & sargent ( 1997 ) investigate the influence of bar density waves on star formation in the nuclear regions of disk galaxies , finding enhancements in the sfrs of early type spirals . they argue this is due to the bar channeling gas to the central region , however the presence of a bar is neither a necessary nor sufficient condition for nuclear star formation . the wide range in the strength of nuclear hii regions , whether a bar is present or not , suggests that it is not the passage of gas through a density wave which mediates the star formation rate within a particular starburst . one prediction of these theories is a correlation of sfr with the density wave amplitude . however this is not observed ( elmegreen & elmegreen 1986 ; mccall & schmidt 1986 , k89 ) . furthermore such theories have difficulty explaining star formation in galaxies where there is a lack of organized star formation features , as in flocculent spirals ( block et al 1994 ; sakamoto 1996 ; thornley & mundy 1997a and 1997b ; grosbol & patsis 1998 ) , some of which even contain moderate amplitude density waves as revealed in the near infra - red . gmcs are present and sfrs are similar to those systems where star formation is organized into spiral patterns . this suggests stellar disk instabilities , which create spiral density waves , and gas instabilities , which lead to gmcs and large - scale star formation , are decoupled in the sense that one does not cause the other ( k89 ; seiden & schulman 1990 ) . this decoupling highlights the need for a theory which physically motivates the schmidt law of equation ( [ larson ] ) without the need for coherent density waves . in this paper we outline such a theory . the sfr , dominated by stars forming in clustered regions , with high mass stars present , is controlled by collisions between gravitationally bound gas clouds , which can be atomic , molecular or both . we find the collision time is a fraction of the orbital period . collisions create localized , over - dense regions where high mass star formation occurs . the pre - collision clouds are formed relatively quickly by the action of gravitational , thermal or parker instabilities , growing in regions where @xmath25 . however , their collapse is halted by static and turbulent magnetic pressure support . the latter may be produced by low mass star formation regulated by ambipolar diffusion ( e.g. mckee 1989 ) , which does not dominate the global galactic sfr . thus the rate limiting step for star formation is not the formation of bound clouds , but the compression of these , or parts of these , in cloud - cloud collisions . therefore at any particular time , most of the bound gas is not undergoing collision induced star formation . there is no specific need for large scale , coherent density waves . there is some evidence for collision induced star formation in the galaxy . scoville , sanders & clemens ( 1986 ) noted the efficiency per unit mass of h@xmath23 for ob star formation decreases significantly with increasing cloud mass over the range @xmath50 to @xmath51 ( see also [ eff ] ) and concluded the principal trigger for star formation is not an internal mechanism , such as the growth rate of gravitational instability or sequential star formation . scoville et al suggested the approximately quadratic dependence of the galactic h ii region distribution on the local h@xmath23 density ( averaged on scales @xmath52 pc ) was evidence for cloud collisions causing massive star formation . detailed observations of individual star forming regions also suggest cloud collisions are an important triggering mechanism ( scoville et al 1986 ; maddalena et al 1986 ; hasegawa et al 1994 ; greaves & white 1991 ; womack , ziurys , & sage 1993 ) . clouds with embedded clusters of star formation have broader distributions of optical polarization angles than clouds without star formation ( myers & goodman 1991 ) . this may indicate accumulation of gas at super - alfvenic speeds in these star forming regions from the collision of two clouds with distinct magnetic field alignments . note , the results of myers & goodman suggest the enhanced dispersion in polarization angles is more closely associated with the presence of dense gas than with the young stars which subsequently form . computer simulations of collisions between inhomogeneous clouds ( klein & woods 1998 ) reveal the formation of high density clumps embedded in filamentary structures , via a bending mode instability . such structures are abundant in omc-1 ( wiseman & ho 1994 , 1996 ) and taurus ( ungerechts & thaddeus 1987 ) . in modeling starbursting systems ic 1908 and ngc 6872 , mihos , bothun & richstone ( 1993 ) were unable to reproduce observations of enhanced sfrs in regions of high velocity dispersion and circular velocity gradients , where cloud collision rates are high , with traditional , non - collisional prescriptions of star formation . while there is much evidence for cloud collisions playing an important role in inducing star formation , it is not yet clear if the majority of star formation is triggered by this process . the data supporting equations ( [ kenlaw ] ) and ( [ larson ] ) are sensitive only to high mass star formation , although the bulk of stars are expected to form in these regions ( lada et al 1993 ) . the theory of collision induced star formation outlined below , requires the initial trigger for most star formation to be a cloud collision . however , subsequent triggering by other processes , such as self - propagating star formation , from the initial site , is not excluded . the rest of this paper is set out as follows . in [ setup ] we set out our preliminary assumptions , and derive results for gas disks , independent of the hypothesis of collision induced star formation . in [ sec : sfr ] we derive a sfr law of the form @xmath53 in the case of uniform circular velocity , based on star formation from cloud collisions . in [ testing ] we compare this law to observations and make predictions of radial sfr profiles , sfr fluctuations due to different shear velocities , dependencies of disk averaged sfr with gas density ( schmidt law ) and circular velocity ( b - band tully - fisher relation ) and how the efficiency of star formation depends on cloud mass . such tests are required to distinguish this theory from those involving the growth rate of gravitational perturbations or triggering by density waves . in [ sec : spiral ] we examine how large scale spiral density waves affect the collisional theory . finally , in [ sec : sb ] we consider the theory s application to the circumnuclear disks of starbursts , which make up most of the dynamic range in the data supporting a global schmidt law ( k98 ) . the star forming regions under consideration are thin , self - gravitating disks . self - regulated star formation ( e.g. silk 1997 ) enforces the condition @xmath17 . the circumnuclear disks of starbursts ( ds98 ) and the star - forming regions of disk galaxies ( k89 ) satisfy these conditions . for @xmath25 , an overdensity on scales in the critical range leads to a bound object and we assume instabilities drive most of the gas mass into bound clouds . gravitational , parker and thermal instabilities have been considered ( e.g. wada & norman 1999 ; burkert & lin 1999 ; elmegreen 1991 ) . the effect of cloud growth via collisional coagulation has also been examined ( e.g. oort 1954 ; field & saslaw 1965 ; kwan & valdes 1987 ; das & jog 1996 ) . gas in the bound clouds can be either atomic or molecular . in the milky way , extended hi envelopes are commonly observed around molecular clouds ( e.g. moriarty - schieven , andersson & wannier 1997 ; williams & maddalena 1996 ; elmegreen 1993 ; blitz & williams 1999 , 2.3 ) . although andersson & wannier ( 1993 ) conclude the hi around low mass ( @xmath54 ) clouds is not gravitationally bound , for larger gmcs ( @xmath55 ) , the situation is probably reversed ( blitz , private communication ) . in the solar neighborhood the mass in atomic envelopes is similar to the mass in gmcs ( blitz 1990 ) . thus , out to about 8 or 9 kpc , most of the gas in the galaxy is in self - gravitating clouds . as gas densities and pressures increase towards the centers of galaxies , the molecular fraction of the gas is expected to increase , until it is almost completely molecular in the circumnuclear disks of starbursts ( e.g. liszt & burton 1996 , ds98 ) . for simplicity we describe the cloud population with a single , typical mass , @xmath56 . in galactic disks , this approximation is justified by the observed mass spectrum of gmcs in the milky way , @xmath57 , with @xmath58 , and with an exponential cutoff above @xmath59 ( e.g. solomon et al 1987 ) . note , these are only the molecular masses . most of the gas mass is in the large clouds . observations suggest circumnuclear disks are clumpy and the typical mass is much larger ( ds98 ) . however , there is little evidence for the form of the mass function . for galactic disks we set @xmath60 , while for circumnuclear disks we consider clouds with @xmath61 . the properties and timescales associated with these clouds are shown in tables ( [ tab : prop ] ) and ( [ tab : times ] ) . the cloud radius , @xmath62 , is smaller than its tidal radius , @xmath63 , defined as the radial distance from the cloud s center , at which the shear velocity , @xmath64 , due to differential galactic rotation , is equal to the escape velocity from the cloud at that distance . the shear velocity of two orbits separated by a radial distance , @xmath65 , is @xmath66 and so for @xmath67 @xmath68 @xmath69 is the galactic mass interior to @xmath10 , assuming a spherical distribution . this approximation is valid at larger @xmath10 , when the dark matter halo begins to dominate over the disk mass . for smaller @xmath10 , and particularly for the circumnuclear disks of starbursts , this is not the case . however , for simplicity , we keep this formalism , where @xmath69 is understood to be the `` equivalent interior galactic mass '' , if the distribution was spherical instead of disk - like . equation ( [ r_tgeneral ] ) implies @xmath70 for solid body rotation , when @xmath71 and thus @xmath72 . for most of the star - forming circumnuclear and galactic disks rotation curves are flat ( ds98 ; k89 ) , @xmath73 and we have @xmath74 @xmath75 is of order 100 pc for the fiducial values of @xmath76 in circumnuclear and galactic disks . the relation between @xmath77 and @xmath75 will be discussed in more detail in [ sec : sfr ] . the dimensions of the clouds are comparable to the scaleheight of the gas disk ( e.g. solomon et al 1987 ; ds98 ) , and so we describe the cloud distribution with a thin , two dimensional disk . we also assume an approximately axisymmetric distribution , so there is a single value of @xmath6 at any particular @xmath10 . galaxies with strong spiral arms and thus non - axisymmetric gas distributions will be discussed in [ sec : spiral ] . we assume the cloud velocity dispersion , @xmath4 , results from a balance of heating via gravitational torquing from non - collisional encounters and cooling via dissipative collisions . gammie et al ( 1991 ) numerically integrated orbits for two - body encounters to obtain @xmath78 valid for @xmath79 and in approximate agreement with galactic observations ( e.g. stark & brand 1989 ; knapp , stark & wilson 1985 ; clemens 1985 ) . for a flat rotation curve , this is approximately the shear velocity of an encounter of impact parameter , @xmath67 . the surface densities of real disks , set by @xmath17 , are such that the effects of many - body interactions may be important . n - body simulations are required to probe these effects . substituting for @xmath4 in equation ( [ q ] ) , we derive the radial distribution of gas , @xmath80 note k89 , assumed @xmath4 was independent of @xmath10 , which leads to an underestimation of @xmath6 , by factors of a few , in the central galactic regions compared to the case where equation ( [ gammie ] ) is used instead . this may explain the slight trend of @xmath6 decreasing by factors of a few as one moves towards the centers of galaxies , ( k89 , figure 11 ) , rather than remaining constant . however , better statistics are required before a proper comparison can be made . when @xmath81 , the assumption that most of the gas mass is organized in bound clouds breaks down , together with our use of equation ( [ gammie ] ) . the presence of a large scale stellar bar , channeling gas radially inwards , will deplete the gas from certain regions , thus raising @xmath6 . here we expect little or no star formation . this may be the situation in the inner few kpc of the milky way ( e.g. binney et al 1991 ) . our principal hypothesis is that cloud collisions , by compressing parts of the clouds , induce the majority of star formation in galactic and circumnuclear disks . however , collisions can also be disruptive . a simple theoretical condition for colliding clouds to remain bound has been given by larson ( 1988 ) . neglecting post - shock cooling , the clouds stay bound if the ram pressure , @xmath82 , is less than the binding pressure , @xmath83 , where @xmath84 and @xmath85 are the cloud volume and surface densities respectively . for typical galactic disk cloud properties in table [ tab : prop ] , this implies @xmath86 , which is higher than the mean value for gmcs ( @xmath87 ) by a factor of about five . realistic clouds are probably more extended , with a gradually decreasing density profile . interactions in these outer layers reduce the actual relative velocity from the value quoted in table [ tab : prop ] . radiative post - shock cooling reduces the disrupting pressure , and thus also relaxes the above condition . therefore we expect some collisions to lead to an increase in mass , density and the gravitational potential energy of the clouds involved , and hence the likelihood of faster sfrs . the outcome of cloud collisions has also been investigated numerically ( e.g. lattanzio et al 1985 ) , but the simulations have usually been unable to resolve the jeans length , thus violating the numerical jeans condition ( truelove et al 1997 ) . furthermore , there has been no systematic attempt to probe the parameter space of cloud collisions , as defined by the angle of collision , impact parameter , mach number and mass ratio . magnetic fields have yet to be included . for the idealized cases considered , the results of a collision depend sensitively on the collision parameters ( klein , private communication ; lattanzio et al 1985 ) . we make the simplifying assumption that the fraction of a cloud converted into stars in a typical collision , averaging over the parameter space of possible collisions , is constant . we consider the thin disk of self - gravitating clouds described in [ setup ] , where @xmath17 . we hypothesize @xmath88 is , on average , inversely proportional to the collision time , @xmath89 , of these clouds . a fraction , @xmath90 , of each gas cloud is converted into stars in each burst of collision induced star formation . the time between bursts is @xmath91 , where @xmath92 is the fraction of collisions which lead to star formation . thus , @xmath93 where @xmath94 is the surface number density of gravitationally bound clouds per unit area of the disk . by numerically solving the equations of motion , gammie et al ( 1991 , figure 8) , found cloud - cloud collisions result from encounters caused by differential rotation , primarily with initial impact parameters of about @xmath95 , and with a spread in values of order @xmath75 ] . for typical gmc parameters in the galaxy , the associated shear velocity is @xmath97 . this sets the collision rate , together with the cloud surface density , @xmath98 , and the probability of collision , @xmath99 , of these encounters . note , the random velocity dispersion of the cloud population ( @xmath100 e.g. stark & brand 1989 ) sets the clouds moving on epicycles , but is not the velocity directly influencing the collision rate . the effect of these random motions has been accounted for in the calculations of gammie et al , since they consider the collision of clouds which are already moving on epicycles . increasing the random motions increases the initial impact parameters at which most cloud collisions occur , raising the shear velocity and thus the collision rate . we express @xmath89 as @xmath101 where the first factor of @xmath102 accounts for clouds either catching up with others at larger @xmath10 or being caught up with by clouds at smaller @xmath10 . @xmath103 is the mean free path of a cloud to catch up , or be caught up to , by another . @xmath104 is the shear velocity of an encounter with impact parameter @xmath105 , due to differential rotation . we evaluate the factor @xmath106 via @xmath107 as in equation ( [ radialgas ] ) , we have used @xmath108 and assumed the velocity dispersion of the gas clouds results from gravitational torquing ( gammie et al 1991 ) so that @xmath109 , with @xmath110 . so @xmath111 and is constant where @xmath6 is constant . thus every area element , @xmath112 , of the disk approximately contains the mass of gas , @xmath76 , required to set @xmath63 . thus , from equation ( [ taucoll ] ) , @xmath113 from gammie et al ( 1991 ) we set @xmath114 . we expect it to scale as @xmath115 . we consider cloud boundaries to be set by pressure confinement from the general ism pressure , @xmath116 . following elmegreen ( 1989 ) we have @xmath117 where @xmath118 and @xmath119 are the stellar surface density and velocity dispersion respectively . the boundary pressure of the self - gravitating clouds is a few times less than the interior cloud pressure , @xmath120 , where @xmath121 . since @xmath122 implies @xmath123 , and with @xmath124 , we have @xmath125 observationally , @xmath126 and @xmath118 have approximately similar spatial distributions , and so from equation ( [ rcrt ] ) we see that @xmath127 , and thus @xmath99 , varies only very slowly with @xmath10 . from here on we take it to be a constant . substituting equation ( [ colltime ] ) into equation ( [ sfr1 ] ) , we obtain @xmath128 this is a new `` modified''-schmidt law , to be tested against observations ( [ testing ] ) . for our fiducial location in the galactic disk ( @xmath129 kpc ) we have @xmath130 disk averaged sfrs , with the appropriate gas distribution , are estimated in [ diskav ] . with high resolution data for @xmath88 , @xmath126 , including atomic and molecular components , and @xmath9 , equation ( [ sfr3 ] ) can be directly tested . this is practical for the milky way and nearby galaxies , but difficult for circumnuclear disks of starbursts because of their small size . star formation from cloud collisions is a stochastic process and so statistically significant data sets are required . properly identifying bound clouds requires atomic and molecular observations , so the masses of both components can be accounted for . the assumption that the cloud velocity dispersion is caused by gravitational torquing ( gammie et al 1991 ) , also leads to the prediction of @xmath131 ( equation [ radialgas ] ) . combining this with equation ( [ sfr3 ] ) leads to @xmath132 which is proportional to @xmath133 for constant @xmath9 . if observations of @xmath126 are lacking , then the theory can still be tested using equation ( [ sfrr ] ) and sfr and circular velocity data , for an assumed constant @xmath6 . note , @xmath134 is , in general , difficult to determine . however surveys of galactic co ( e.g. sanders et al 1986 ) find no strong evidence for systematic variation ( solomon et al 1987 ; scoville et al 1987 ) . furthermore any variation is weakened by being raised to the @xmath135 power in equation ( [ sfrr ] ) . if galactic stellar disks have been built up primarily through self - regulated star formation , where @xmath17 , then we also have @xmath136 as an additional prediction . several authors have presented radial profiles of @xmath126 and @xmath88 for individual galaxies ( e.g. tacconni & young 1986 ; kuno et al 1995 ) . however , problems of accounting for the varying extinction of the tracers of star formation , such as h@xmath5 and br@xmath137 make direct comparison difficult . similarly , where fir emission is used as a sfr estimator , the heating contributions from young stars , old stars and possible agn activity must be disentangled . a follow - up paper to k98 , ( martin & kennicutt 2000 ) , will present radial data for many galaxies , accounting for these effects . one distinct prediction of this theory results from the extra dependence of the sfr on variations in the circular velocity . statistically , we expect negative velocity gradients in the rotation curve to increase the sfr and positive gradients to decrease it . regions of solid body rotation will be free of collisions resulting from shearing motions . thus we expect star formation here to have a different triggering mechanism . these regions provide a good control environment for testing the collisional theory . in general , we expect a positive correlation between the sfr and the velocity dispersion of the clouds . with higher random velocities they move on larger epicycles , and encounters with greater initial impact parameters and greater shear velocities occur , leading to increased collision rates . no such prediction is made by the theory of star formation triggered by the rate of gas passage through spiral arms . in a future paper we plan to investigate this question with high resolution bima and vla data . we also test equation ( [ sfr3 ] ) by examining the disk averaged properties of galaxies and starbursts . we note , however , that such tests , while good consistency checks , do not discriminate well between the different theories of how star formation is triggered . we take the area - weighted mean of equation ( [ sfr3 ] ) over the whole region of a disk , where @xmath138 and @xmath9 is constant , to obtain @xmath139 current observations do not have the spatial resolution to estimate @xmath140 , except for the nearest galaxies . however , k98 presents data revealing a correlation between @xmath141 and @xmath142 , where @xmath143 is the angular rotation frequency at the outer radius , @xmath144 , of the star forming region ( figure [ fig : ken2 ] ) . since we are considering the flat rotation curve case , we rewrite equation ( [ sfr5 ] ) as @xmath145 for @xmath146 ( equation [ radialgas ] ) we obtain @xmath147 where @xmath148 , @xmath149 being the inner radius where the rotation curve is flat . if @xmath150 is uncorrelated with @xmath151 and @xmath28 , then we predict k98 s observed correlation . applying equation ( [ sfr7 ] ) to the inner 8.5 kpc of the milky way , assuming @xmath152 and @xmath153 ( see [ eff ] ) , gives @xmath154 this is consistent with estimates based on observations of thermal radio emission from hii regions , which give @xmath155 ( gsten & mezger 1982 ; mckee 1989 ) . scaling to a typical starbursting circumnuclear disk , with @xmath156 , @xmath157 and @xmath158 , gives @xmath159 . one of the most uncertain factors in these estimates is @xmath92 . numerical studies , ( klein , private communication ) , although not yet incorporating magnetic fields , can constrain this number . we also predict the form of the b band tully - fisher relation , @xmath160 . assuming @xmath161 we have @xmath162 where we have evaluated @xmath28 using equation ( [ radialgas ] ) assuming @xmath163 . this result compares favorably with the observed b band exponent of @xmath164 ( burstein et al 1995 ; strauss & willick 1995 ) . the tully - fisher relation at longer wavelengths becomes more and more contaminated by light from older stellar populations . collision induced star formation predicts a variation in the star formation efficiency , @xmath90 , of gmcs , dependent on their mass . variations in @xmath90 , resulting from this and other mechanisms , have been considered by a number of authors ( e.g. elmegreen & clemens 1985 ; scoville et al ( 1986 ) ; pandey , paliwal & mahra 1990 ; franco , shore & tenorio - tagle 1994 ; ikuta & sofue 1997 ; williams & mckee 1997 ) . scoville et al ( 1986 ) found the ratios of lyman - continuum luminosity to @xmath56 and number of high luminosity hii regions to @xmath56 , decreased with increasing @xmath76 . they argued this was evidence for collision induced star formation , since , if the collision rate scaled as the cloud surface area ( @xmath165 ) , then the efficiency of star formation per unit cloud mass , @xmath90 , would scale as @xmath166 . however , if a more appropriate mass - size relationship ( @xmath167 ; larson 1981 ) is applied with their reasoning , then no scaling of @xmath90 with @xmath76 is predicted . ikuta & sofue ( 1997 ) considered the ( radio ) luminosity of hii region , rather than simply the number , associated ( @xmath168 pc away in the plane of the sky ) with molecular clouds . they found @xmath169 . however , they did not allow for higher mass clouds being larger than smaller ones . the centers of large clouds may be much further than 10 pc away from hii regions associated with their periphery . there is also no comment on the completeness of the data . in particular the hii region sample , being flux limited at @xmath170 , is incomplete below luminosities of @xmath171 for sources in the inner galaxy , where @xmath172 is the ionizing luminosity ( @xmath173 ) in units of @xmath174 . for example , orion a has @xmath175 and @xmath176 . note , by definition an object with a radio flux of 1 jy at a distance of 1 kpc has a luminosity of @xmath177 , i.e. there is no factor of @xmath178 . ] . we now revisit the question of how the efficiency , @xmath179 , of star formation depends on cloud mass . observationally , we measure @xmath90 by summing the radio luminosity , @xmath180 , of hii regions associated with a cloud . this luminosity is converted into a stellar mass using @xmath181 ( mckee & williams 1997 ) . dividing by @xmath76 then gives @xmath90 . we use the hii region data of downes et al ( 1980 ) and the molecular cloud data of solomon et al ( 1987 ) . for each of the 106 hii regions , with resolved distance ambiguity , in the same region as the cloud survey , molecular clouds within @xmath182 pc ( i.e. 2 @xmath77 ) on the sky , within the correct radial distance bounds and with relative velocities of @xmath183 are identified . the hii region is associated with the closest cloud , if more than one is identified . with these criteria , 83 hii regions are associated with 39 clouds . this data is shown in figure [ fig : eff]a , with the typical errors shown by the cross . however , there are also two issues of completeness for the sample . firstly , as already mentioned , the hii data are incomplete for @xmath184 . we adopt the more conservative limit of @xmath185 , which is shown by the diagonal dashed line in figure [ fig : eff]a . the data below this line are incomplete . secondly , the molecular cloud data are incomplete for @xmath186 ( williams & mckee 1997 ) . this is shown by the vertical dashed line . assuming , for a particular @xmath76 , the probability of detection of a cloud is uncorrelated with @xmath180 , we ignore the effect of cloud incompleteness in the following analysis . the total @xmath180 for each cloud is recalculated , now only summing individual hii regions with @xmath187 . this data is shown in figure [ fig : eff]b . the best fit straight line is shown by the long dashed line . we now compare these observations to theory . we model a cloud of mass @xmath76 , with hii regions , as resulting from a binary collision of two smaller clouds of mass @xmath188 and @xmath189 , with @xmath190 and @xmath191 . the assumption of binary collisions is justified by the paucity of high mass star formation in the total cloud population . given our lack of understanding of the detailed results of cloud collisions , the assumption that several hii regions , rather than just one , may result from a single collision event is reasonable . @xmath188 ( and @xmath189 ) are chosen from the mass spectrum of clouds : @xmath57 , with @xmath58 . following our original hypothesis applied to equal mass clouds , that , on average , in a collision a constant fraction , @xmath90 , of the total cloud mass forms stars , we assume for unequal mass collisions , the mass of stars formed is @xmath192 . thus @xmath193 , and the maximum star formation efficiency , @xmath194 , is statistically , achieved for collisions of equal mass clouds ( @xmath195 ) , and is thus independent of @xmath76 . from the data in figure [ fig : eff]b , we set @xmath196 . the observationally determined minimum luminosity , above which the hii region sample is complete , defines a minimum value , @xmath197 , for @xmath188 . thus @xmath198 . these two constraints meet when @xmath199 , and thus the star formation efficiency data for a complete sample of hii regions associated with clouds will be contained within a triangular region in a diagram of @xmath200 vs. @xmath201 ( figure [ fig : eff]b ) . the typical value of @xmath188 , @xmath202 , is an average of @xmath188 weighted by the collision rate . this scales with @xmath76 via @xmath203 in the region where @xmath204 , since from equation ( [ taucoll ] ) , we have @xmath205 and since @xmath206 is approximately independent of @xmath188 , and @xmath207 . thus @xmath208 and so @xmath209 performing an averaging of the distribution of @xmath188 ( and thus @xmath90 ) given by equation [ m1dist ] , in logarithmic space , leads to line a in figure [ fig : eff]b . we also consider three other theoretical models for comparison . models b and c ( stochastic star formation , williams & mckee 1997 ) assume a cloud s star formation rate is linearly proportional to its mass , provided @xmath194 is not exceeded . model b imposes a mass dependence , at high masses , on @xmath194 by assuming there is a physical limit to the maximum luminosity of ob associations . model c assumes @xmath194 independent of @xmath76 , and the reason for the observed maximum association luminosity is simply due to sparse sampling of the distribution . by normalizing to the observed galactic ob association luminosity and molecular cloud mass functions , a probability distribution of association luminosities expected in a cloud of mass @xmath76 , approximately proportional to @xmath210 is predicted . after summing the expected number of associations in a cloud , and then averaging the resulting distribution of @xmath90 at each @xmath76 in logarithmic space , we obtain lines b and c. finally we consider an unphysical model ( d ) with a uniform distribution of @xmath211 for clouds of a particular mass . the four model predictions are shown in figures [ fig : eff]b and [ fig : eff]c . in the latter , we also show the 95% confidence limits on the linear best fit from the existing sample ( 1 ) , a hypothetical sample ten times larger with the same distribution ( 2 ) , and as for ( 2 ) but with individual measurement errors reduced by a factor of two ( 3 ) . from the existing data , model d is marginally excluded ( because of its slope ) , while a , b , c are all consistent . although the confidence limits on fitting a non - linear function will differ in detail , figure [ fig : eff]c qualitatively shows that with ten times more data and with modest improvements in measurement accuracy , we can hope to discriminate between these models . such a sample can be achieved by a survey of radio hii regions in the entire inner galaxy to fluxes of order @xmath212 . improving completeness of co observations of smaller mass gmcs will also improve the statistics . uncertainties in @xmath76 result from the assumption of virialization and the co/@xmath213 ratio . errors in @xmath180 result in part from distance uncertainties . direct infra red observations of massive stars may improve estimates of association luminosities . as a final point of caution , we note that evolutionary effects will cause a cloud s measured @xmath90 to change . at the onset of star formation @xmath90 will be small , while @xmath90 will increase as the cloud is destroyed by energy injection from its high mass stars . large enough samples are required to average over this effect . in summary , collision induced star formation predicts a decrease in the mean cloud star formation efficiency of a complete sample with increasing cloud mass . the decrease occurs since it is much more likely for a large cloud to suffer a collision with a smaller one , because of the cloud mass spectrum . it is the smaller cloud which then determines the amount of resultant star formation . an observational lower limit to hii region luminosities , included in the analysis , implies a constant minimum mass of the smaller cloud , which drives the mass dependence of @xmath90 . stochastic models predict a similar , but less steep , decline due to the shape of the complete region in the @xmath90 versus @xmath76 parameter space . for stochastic models , at high cloud masses , if @xmath194 is independent of @xmath76 , the efficiencies tend towards a constant value . improved data samples should allow for discrimination between the models . the theory for collision induced star formation requires modification where there is a tight spatial correlation of star formation with spiral structure . spiral density waves decrease the local value of @xmath6 , and often @xmath16 in the arm region and @xmath214 in the inter - arm region ( e.g. kuno et al 1995 ) . two scenarios are possible . in the first , the rate limiting step is the formation of bound clouds , occurring exclusively in the arms , after which the clouds form stars at a fast rate ( e.g. via gravitational collapse of magnetically supercritical clouds or through rapid collisions ) and all the bound gas is involved in star formation . equation ( [ wyse ] ) is then a better description of the galactic sfr . in the second scenario , bound cloud formation is fast and is not the rate limiting step . a reservoir of bound gas clouds exists in the galaxy , including in the inter - arm regions . spiral arms now act to concentrate the spatial distribution of gas clouds , and the collision rate is enhanced in the arms . the individual collision rate for a particular cloud is still described by equation ( [ taucoll ] ) , but the overall sfr is modulated by the length of time the gas clouds spend in the arm region ( related to @xmath215 and the width and pitch angle of the arm ) and the degree of spatial concentration of clouds in the arm ( related to the strength of the spiral density wave ) . in m51 , which has strong , well - defined spiral arms , there is evidence that the cloud collision times in the arm are short compared to the arm crossing time ( kuno et al 1995 ) , thus favoring the first scenario . however , the similarity of the global star forming properties of galaxies with and without large scale density waves argues against bound cloud formation being controlled exclusively by spiral arms . deciding which scenario is the correct description , or if both processes operate at some level , requires study of the arm to inter - arm gas distributions and cloud collision timescales in larger samples of galaxies , which exhibit a tight correlation of star formation with spiral arms . obviously , neither scenario can explain the star formation observed in the galaxies without strong spiral structure . the @xmath88 and @xmath126 data of the circumnuclear disks of starbursts make up most of the dynamic range for k98 s schmidt law relationships ( figures [ fig : ken1 ] and [ fig : ken2 ] ) . however , much less is known about the details of star formation occurring in these regions than in the outer regions of galactic disks . downes & solomon ( 1998 , ds98 ) present observations and models of 10 circumnuclear disks . their key findings , relevant to the theory of collision induced star formation are the following : most of the circumnuclear disks are in the flat rotation curve regime ; @xmath17 ; the disks are thin and modeled without need to invoke large - scale non - axisymmetric features , such as bars or spiral arms ; much of the star formation is associated with very large bound gas clouds with maximum masses @xmath216 and sizes ( @xmath217 pc ) consistent with their confinement by tidal shear forces . thus our principal assumptions are met . the gas disks are predominantly molecular , including the inter - cloud medium , with the gas mass approximately @xmath218 of the dynamical mass . note , larger gas masses ( by a factor of @xmath219 ) are derived if the standard co / h@xmath23 conversion factor for virialized gmcs in normal galaxies is used ( as in the analysis of k98 ) . in fact , much of the co luminosity of the circumnuclear disks comes from the non - virialized molecular inter - cloud medium . since @xmath17 and not @xmath220 , the gas is probably not collapsing on the free - fall timescale ( @xmath221 yrs ) . this is consistent with downes & solomon s estimate that about half the gas is converted into stars over 10 orbital periods ( @xmath222 yrs ) . this efficiency per orbital period is similar to that of normal galactic disks ( k98 ) , motivating a unified theory to describe both regimes . the collision time for a typical cloud mass , which we take to be of order @xmath223 , is a few million years . the one inconsistency between our theory and the results of ds98 , is their estimate that only @xmath224% of the gas mass is in bound clouds . they base this estimate on observations of hcn which requires densities @xmath225 for excitation . however , the fact that @xmath138 implies that any overdense region on the critical scales will become gravitationally bound . a large fraction of the gas may be at densities less than that required for strong hcn emission , and yet still in bound clouds . circumnuclear starburst disks and the star forming regions of normal galactic disks may represent the extremes of a continuous family of states . our theory of collision induced star formation can be applied in both situations . as one moves inwards from the outer disk , the gas surface densities and the molecular fractions of the bound clouds increase , but self - regulation of star formation maintains @xmath226 . the typical cloud mass , @xmath76 also appears to increase . if this is set by the magnetic critical mass , @xmath227 , of the inter - cloud medium then , from equation ( [ m_b ] ) , @xmath228 must increase as @xmath10 decreases . the typical cloud mass may also be affected by changes in the collision time . this is much shorter ( @xmath229 yrs ) in circumnuclear disks than in the typical star - forming locations of normal disks ( @xmath230 yrs ) , particularly in comparison to the stellar evolutionary and cloud destruction timescales . this may increase @xmath76 in circumnuclear disks to be several times @xmath227 . we have invoked star formation triggered by cloud - cloud collisions to explain global star formation rates in disk galaxies and circumnuclear starbursts . previous theories based on the growth rate of gravitational perturbations ignore the dynamically important presence of magnetic fields . theories based on triggering by spiral density waves fail to explain star formation in systems without such waves . furthermore , observations suggest gas and stellar disk instabilities are decoupled . star formation resulting from cloud collisions has been proposed in the past ( e.g. scoville et al 1986 ) , but rejected because of supposedly long collision timescales . however , gammie et al ( 1991 ) show the collision rate of self - gravitating particles in a differentially rotating disk is much larger than that of particles in a box . collision rates are enhanced because particles collide at the shear velocity of encounters with initial impact parameters of order two tidal radii ( typically a few hundred parsecs for gmcs ) . gravitational focusing further increases the cross - section . also , the small scale height of gmcs implies essentially two dimensional interactions in the plane of the disk , increasing the collision rate relative to that for three dimensions . we calculate collision timescales short enough to allow a viable theory of collision induced star formation to be considered . in summary , in this model , self - gravitating gas disks fragment into bound gas clouds . this process is driven either by gravitational , thermal or parker instabilities , or the influence of stellar spiral density waves on the gas . these bound clouds , either atomic or molecular , are relatively long - lived , being supported by static and turbulent magnetic pressure . the latter may be produced by dynamically - regulated low mass star formation ( mckee 1999 ) . we hypothesize a fraction of cloud collisions lead to compression of localized regions of the clouds . these regions , if magnetically supercritical , collapse rapidly to form stars , including high mass ob stars . the bulk of galactic disk stars are thought to form via this `` burst''-mode ( lada et al 1993 ) . thus , the rate limiting step for star formation is not the formation of bound clouds , but the compression of these , or parts of these , in cloud - cloud collisions . therefore at any particular time , most of the bound gas is not actively undergoing star formation . specifically , we have considered an idealized , single mass population of gravitationally bound gas clouds , orbiting in an axisymmetric , thin disk . using the result of gammie et al ( 1991 ) for the cloud velocity dispersion , we predict radial gas distributions , dependent on the toomre @xmath6 stability parameter ( equation [ radialgas ] ) . applying our principal hypothesis , that cloud collisions trigger the majority of disk star formation , using the collision cross - section results of gammie et al ( 1991 ) and with the assumption star formation self - regulates ( @xmath17 ) , we predict enhanced cloud collision rates and a sfr law of the form @xmath231 ( equation [ sfr3 ] ) . for flat rotation curves ( @xmath73 ) , this result is in agreement with the disk averaged data of k98 ( figure [ fig : ken2 ] ) . although uncertain , our estimates of the total sfr in the milky way and for typical starburst systems are consistent with observations . we predict a b - band tully - fisher relation of the form @xmath232 , in agreement with observations ( burstein et al 1995 ; strauss & willick 1995 ) . this theory is to be further scrutinized to discriminate between it and other star formation mechanisms . to this end we have proposed several tests . we predict statistically enhanced sfrs in regions of large negative circular velocity gradients , where the shear rate is increased , and regions of increased cloud velocity dispersion . similarly , decrements are predicted in regions of large positive circular velocity gradients , which reduce the amount of shear . future observations ( e.g. martin & kennicutt 2000 ) of sfr , gas and circular velocity profiles of large samples of disk galaxies should allow for statistically significant tests of our proposed sfr law , and in particular the dependence on the circular velocity gradients and cloud velocity dispersion . however , these tests will be complicated by possible variations in the likelihood of collision induced star formation with collision velocity . the results of numerical simulations may be necessary to account for this effect . we also predict star formation efficiency , @xmath90 , linearly averaged , decreases with increasing cloud mass as @xmath233 . figure [ fig : eff ] shows model predictions for @xmath90 , logarithmically averaged over its distribution , and comparison to observations . larger and deeper surveys of hii regions and gmcs , including their atomic components , are required to improve the significance of this test . undoubtedly our model is an extremely simplified description of the actual star formation process . we have presented an idealized theory in which all star formation is triggered by cloud collisions , however other processes , such as spontaneous star formation , self - triggering and triggering by density waves undoubtedly operate at some level . for the results of the collision induced theory to be valid , we require that the majority of ( high mass ) star formation is initially triggered by this process . the basic theory needs modification where there is a tight correlation of star formation with large scale density waves , allowing for the duration clouds spend in the density wave , and the degree of spatial concentration . the theory can be improved by numerical calculation of collision rates in a many body system , rather than relying on simple two body interaction rates . numerical simulation of cloud collisions ( e.g. klein & woods 1998 ) may provide insight into the details of how a magnetically supercritical region can be produced from the collision of two magnetically subcritical clouds . the parameter space for the outcome of collisions with different initial conditions is also being probed by simulation ( klein , private communication ) . these simulations will constrain the probability , @xmath92 , for star formation to result from typical cloud - cloud collisions . this theory can be applied to analytic models ( e.g. shore & ferrini 1995 ; silk 2000 ) and simulations ( e.g. curir & mazzei 1998 ; weil , eke & efstathiou 1998 ) of disk galaxy formation and evolution , for comparison to cosmological sfr data . we thank chris mckee for many hours of stimulating discussion and much input . we also thank leo blitz , andrew cumming , alex filippenko , rob kennicutt , richard klein , chris matzner , antonio parravano , evan scannapieco , joe silk and an anonymous referee for helpful comments . bertoldi , f. , & mckee , c.f . 1992 , , 395 , 140 + binney , j. , gerhard , o.e . , stark , a.a . , bally , j. , & uchida , k.i . 1991 , , 252 , 210 + blitz , l. 1990 , _ the evolution of the interstellar medium _ , ed . blitz l. , asp press : san francisco , 273 + blitz , l. , & williams , j.p . 1999 , proc . : _ the physics of star formation and early stellar evolution ( crete ii ) _ , astro - ph/9903382 + block , d.l . , bertin , g. , stockton , a. et al 1994 , , 288 , 365 + burkert , a. & lin , d.n.c . 1999 , submitted to + clemens , d.p . 1985 , , 295 , 422 + curir , a. , & mazzei , p. 1998 , new astr . , 4 , 1 + das , m. , & jog , c.j . 1996 , , 462 , 309 + downes , d. , wilson t.l . , bieging , j. & wink , j. 1980 , , 40 , 379 + downes , d. , & solomon , p.m. 1998 , , 507 , 615 + elmegreen , b.g . 1985 , in _ protostars & planets ii _ , eds , black , d. , & matthews , m. , university of arizona press , tucson , 33 + elmegreen , b.g . 1989 , , 338 , 178 + elmegreen , b.g . 1991 , , 378 , 139 + elmegreen , b.g . 1993 , , 411 , 170 + elmegreen , b.g . 1994 , , 425 , l73 + elmegreen , b.g . & clemens , c. 1985 , , 294 , 523 + elmegreen , b.g . , & lada , c.j . 1977 , , 214 , 725 + elmegreen , d.m . , & elmegreen , b.g . 1986 , , 311 , 554 + field , g.b . , & saslaw , w.c . 1965 , , 142 , 568 + franco , j. , shore , s.n . , & tenorio - tagle , g. , , 436 , 795 + gammie , c.f . 1996 , , 462 , 725 + gammie , c.f . , ostriker , j.p . , & jog , c.j . 1991 , , 378 , 565 + greaves , j.s . , & white , g.j . 1991 , , 248 , l27 + grosbol , p.j . , & patsis , p.a . 1998 , , 336 , 840 + gsten , r. , & mezger , p.g . 1982 , vistas astr . , 26 , 159 + hasegawa , t. , sato , f. , whiteoak , j.b . , & miyawaki , r. 1994 , , 429 , l77 + heiles , c. , goodman , a.a . , mckee , c.f . , & zweibel , e.g. 1993 in _ protostars and planets iii _ , eds , levy , e.h . , & lunine , j.i . , university of arizona press , tucson , 279 + ho , l.c . , filippenko , a.v . , & sargent , w.l.w . 1997 , , 487 , 591 + ikuta , c. & sofue , y. 1997 , , 49 , 323 + jog , c.j . & solomon , p.m. 1984 , , 276 , 114 + jog , c.j . 1996 , , 278 , 209 + kenney , j. 1997 , iau symp . , 184 + kennicutt , r.c . 1989 , , 344 , 685 + kennicutt , r.c . 1998 , , 498 , 541 + klein , r.i . , & woods , d.t . 1998 , , 497 , 777 + knapp , g.r . , stark , a.a . , & wilson , r.w . 1985 , , 90 , 254 + kuno , n. , nakai , n. , handa , t. , & sofue , y. 1995 , , 47 , 745 + kuno , n. , tosaki , t. , nakai , n. , & nishiyama , k. 1997 , , 49 , 275 + kwan , j. , & valdes , f. 1987 , , 315 , 92 + lada , e.a . , strom , k.m . & myers , p.c . in _ protostars and planets iii _ , eds , levy , e.h . , & lunine , j.i . , university of arizona press , tucson , 245 + larson , r.b . 1981 , , 194 , 809 + larson , r.b . 1988 , in _ galactic and extragalactic star formation _ , pudritz & m. fich , dordrecht : kluwer , 435 + larson , r.b . 1992 , in _ star formation in stellar systems _ , ed . g. tenorio - tagle , m. prieto , & f. sanchez , cambridge : cambridge univ . press , 125 + lattanzio , j.c . , monaghan , j.j . , pongracic , h. , & schwarz , m.p . 1985 , , 215 , 125 + liszt , h.s . , & burton , w.b . 1996 , in _ unsolved problems of the milky way _ , eds . blitz , l. , & teuben , p. , kluwer : dordrecht + martin , c.l . , & kennicutt , r.c . 2000 , in preparation + mccall , m.l . , & schmidt , f.h . 1986 , , 311 , 548 + mckee , c.f . 1999 , conf . proc . : _ the physics of star formation and early stellar evolution ( crete ii ) _ , astro - ph/9901370 + mckee . c.f . , & williams , j.p . 1997 , , 476 , 144 + mestel , l. 1985 , in _ protostars & planets ii _ , eds , black , d. , & matthews , m. , university of arizona press , tucson , 320 + moriarty - schieven , g.h . , andersson , b .- g . , & wannier , p.g . 1997 , , 475 , 642 + myers , p.c . , & goodman , a.a . 1991 , , 373 , 509 + oort , j.h . 1954 , , 12 , 177 + pandey , a.k . , paliwal , d.c . , & mahra , h.s . 1990 , , 362 , 165 + quirk , w.j . 1972 , , 176 , l9 + sakamoto , k. 1996 , , 471 , 173 + sanders , d.b . , clemens , d.p . , scoville , n.z . , & solomon , p.m. 1986 , , 60 , 1 + schmidt , m. 1959 , , 129 , 243 + scoville , n.z . , sanders , d.b . , & clemens , d.p . 1986 , , 310 , l77 + scoville , n.z . , yun , m.s . , clemens , d.p . , sanders , d.b . , & waller , w.h . 1987 , , 63 , 821 + seiden , p.e . , & schulman , l.s . 1990 , adv . , 39 , 1 + shore , s.n . , & ferrini , f. 1995 , fund . phys . , 16 , 1 + silk , j. 1997 , , 481 , 703 + silk , j. 2000 , , submitted + solomon , p.m. , rivolo , a.r . , barrett , j. , yahil , a. 1987 , , 319 , 730 + spitzer , l. 1978 , _ physical processes in the interstellar medium _ , wiley : new york + stark , a.a , & brand , j. 1989 , , 339 , 763 + taniguchi , y. , & ohyama , y. 1998 , , 509 , l89 + thornley , m.d . , & mundy , l.g . 1997a , , 484 , 202 + thornley , m.d . , & mundy , l.g . 1997b , , 490 , 682 + toomre , a. 1964 , , 139 , 1217 + truelove , j.k . , klein , r.i . , mckee , c.f . , holliman , j.h . , howell , l.h . , & greenough , j.a . 1997 , , 489 , l179 + ungerechts , h. , & thaddeus , p. 1987 , , 63 , 645 + wada , k. , & norman , c. 1999 , , 516 , l13 + wang , b. , & silk , j. 1994 , , 427 , 759 + weil , m.l . , eke , v.r . , & efstathiou , g. 1998 , , 300 , 773 + williams , j.p . , & maddalena , r.j . 1996 , , 464 , 247 + williams , j.p . , & mckee , c.f . 1997 , , 476 , 166 + wiseman , j. , & ho , p.t.p . 1994 , in _ asp conf . 65 , clouds , cores and low mass stars _ , ed . clemens & r. barvainis , san francisco : asp , 396 + wiseman , j. , & ho , p.t.p . 1996 , , 382 , 139 + womack , m. , ziurys , l.m . , & sage , l.j . 1993 , , 406 , l29 + wyse , r.f.g . 1986 , , 311 , l41 + wyse , r.f.g . , & silk , j. 1989 , , 339 , 700 + cccc @xmath56 & obs./magnetic critical mass & @xmath234 & @xmath235 + @xmath10 & observation & @xmath236 & @xmath237 + @xmath9 & observation & @xmath238 & @xmath239 + @xmath63 & @xmath240 & @xmath241 & @xmath241 + @xmath62 & observation & @xmath242 & @xmath243 + @xmath244 & alfven velocity & @xmath245 & uncertain + @xmath246 & @xmath247 & @xmath248 & @xmath249 + @xmath250 & @xmath251 & @xmath252 & @xmath253 + cccc orbital period , @xmath254 & @xmath255 & @xmath256 & @xmath257 + free - fall , @xmath46 & @xmath258 & @xmath259 & @xmath260 + atomic to molecular , @xmath20 & @xmath261 & @xmath262 & @xmath263 + ambipolar diffusion , @xmath45 & @xmath264 & @xmath265 & @xmath266 + collision , @xmath89 & @xmath267 & @xmath268 & @xmath269 + destruction , @xmath270 & williams & mckee ( 1997 ) & @xmath271 & uncertain + lifetime , @xmath272 & @xmath273 & @xmath274 & @xmath275 + alfven crossing , @xmath276 & @xmath277 & @xmath278 & uncertain + impact time , @xmath279 & @xmath280 & @xmath281 & @xmath282 +

andreev bound states ( abss ) are well - defined many - body quantum states that emerge from the hybridization of individual quantum dot ( qd ) states with a superconductor and exhibit very rich and fundamental phenomena . we demonstrate several new electron transport phenomena mediated by abss that form on three - terminal carbon nanotube ( cnt ) qds , with one superconducting ( s ) contact in the center and two adjacent normal metal ( n ) contacts . three - terminal spectroscopy allows us to identify the coupling to the n contacts as the origin of the andreev resonance ( ar ) linewidths and to determine the critical coupling strengths to s , for which a ground state ( or quantum phase ) transition in such s - qd systems can occur . in addition , we ascribe replicas of the lowest - energy abs resonance to transitions between the abs and odd - parity excited qd states , a process we call excited state abs resonances . in the conductance between the two n contacts we find a characteristic pattern of positive and negative differential subgap conductance , which we explain by considering two nonlocal processes , the creation of cooper pairs in s by electrons from both n terminals , and a novel transport mechanism called resonant abs tunneling , possible only in multi - terminal qd devices . in the latter process , electrons are transferred via the abs without effectively creating cooper pairs in s. the three - terminal geometry also allows spectroscopy experiments with different boundary conditions , for example by leaving s floating . surprisingly , we find that , depending on the boundary conditions and the device parameters , the experiments either show single - particle coulomb blockade resonances , abs characteristics , or both in the same measurements , seemingly contradicting the notion of abss replacing the single particle states as eigenstates of the qd . we qualitatively explain these results as originating from the finite time scale required for the coherent oscillations between the superposition states after a single electron tunneling event . these experiments demonstrate that three - terminal experiments on a single complex quantum object can also be useful to investigate charge dynamics otherwise not accessible due to the very high frequencies .

1612.01201

nanoscale electronic devices in contact with superconducting contacts ( s ) exhibit a large variety of fundamental physical phenomena and play , for example , a central role in schemes for quantum computation.@xcite if s is in contact with the many quantum channels of a normal metal ( n ) , phase - coherent andreev reflections lead to electron pairing and an induced superconducting gap in n. this proximity effect has recently been demonstrated also for one - dimensional semiconducting nanowires,@xcite where strong spin - orbit interactions can give rise to majorana bound states.@xcite if s is strongly coupled to a single channel quantum dot ( qd ) , new subgap eigenstates form , which are known as andreev bound states ( abs).@xcite abss carry the supercurrent in josephson junctions,@xcite and thus constitute a model system to investigate the superconducting proximity effect in qds.@xcite abss might also be exploited as andreev quantum bits @xcite and have recently attracted considerable attention in both , theoretical @xcite and experimental work.@xcite previous experiments on abss were focused on qds with two contacts,@xcite and only few were possible in multi - terminal devices.@xcite three - terminal abs devices allow for new transport mechanisms , so that nonlocal processes like cooper pair splitting@xcite compete with local mechanisms like andreev tunneling.@xcite such mechanisms are expected to result in new effects like the triplet blockade @xcite , characteristic patterns in the electrical conductance,@xcite and other nonlocal effects.@xcite in addition , even the most basic abs characteristics are expected to be determined by the barrier strengths , such as the abs resonance broadening@xcite and whether a quantum phase transition in the s - qd many - body ground state occurs as a function of the gate and bias voltages . however , the coupling strengths can only be accessed unambiguously in three terminal geometries.@xcite here we report experiments on three - terminal qd devices fabricated on carbon nanotubes ( cnts ) with one central superconducting contact inducing abss on the qd . a scanning electron microscopy ( sem ) image is shown in fig . we investigate three devices with different coupling strengths @xmath0 to s to investigate several of these open questions and demonstrate a series of new effects and measurement configurations . we use pb as superconductor , which results in a large energy gap @xmath1 and thus in a high relative spectroscopic resolution,@xcite crucial for our experiments . the paper is structured as follows : in sec . [ sec : abs_basics ] we summarize our intuitive picture of abss and abs - mediated electron transport in two - terminal devices . [ sec : fab ] describes the sample fabrication and measurement setup . then we characterize each of the investigated devices by transport spectroscopy in sec . [ sec : abs - devicecharacteriation ] , from which we extract the tunnel coupling strengths to the individual contacts and demonstrate that @xmath0 determines the system ground state and abs dispersion , while the coupling strengths to the normal metal contacts , @xmath2determine the spectroscopic width of the andreev resonances ( ars).@xcite one device shows replicas of the lowest - energy abs resonance at higher energies , which we ascribe to transitions to excited odd - parity qd states in sec . [ sec : abs - addbias - excitedar ] . in sec . [ sec : abs - serialtransport ] , we analyze the transport between the two normal metal contacts mediated by abss , in which competing local and nonlocal transport mechanisms give rise to a characteristic pattern of positive and negative differential conductance . these findings are well captured by a simple rate equation model , which allows us to identify cooper pair splitting@xcite and a new three - terminal subgap process we call _ resonant abs tunneling_. in sec . [ sec : abs - floatings - proximity ] we show that depending on the boundary conditions imposed in the experiments , the measured conductance either exhibits abs or coulomb blockade characteristics , or both at the same time , which we tentatively attribute to finite frequency coherent oscillations between the single particle basis states when an abs is excited by single electron tunneling . for illustration and to establish some terms used below , we consider a qd with a single , spin - degenerate level strongly tunnel coupled to a superconductor and weakly to a normal contact . the former couples even charge states of the isolated qd by exchanging cooper pairs to form new eigenstates of the s - qd system . in the limit of a large superconducting energy gap , @xmath3 ( superconducting atomic limit ) , the emerging abs can be written as a superposition of the even ( empty and doubly occupied ) charge states of the isolated qd , i.e. , as a singlet eigenstate @xmath4 , with @xmath5 and @xmath6 the gate - dependent bogoliubov- de gennes ( bdg ) amplitudes and @xmath7 the corresponding eigenenergy . the orthogonal eigenstate at higher energies reads @xmath8 . in contrast , the odd charge states remain unperturbed doublets @xmath9 , with @xmath10 and energies @xmath11.@xcite the low - energy excitation spectrum of the s - qd system is depicted schematically in figs . [ fig2](a ) and ( b ) , for the case of the doublet being the ground state ( gs ) , and the abs the excited state ( es ) . due to the different parity of the doublet and the abs , one can directly probe the excitation spectrum by single electron spectroscopy using a weakly coupled normal metal tunnel probe ( n ) in an n - qd - s geometry.@xcite a current through such a device is only possible if the bias matches the andreev addition energy @xmath12 , i.e. if the electrochemical potential @xmath13 of n exceeds ( lies below ) the excitation energy @xmath14 ( @xmath15 ) with respect to the electrochemical potential in s , @xmath16 . as illustrated in figs . [ fig2](a ) and ( b ) , charge is transported by an electron tunneling from n into the @xmath17 part of @xmath18 , which excites ( e ) the s - qd system from the gs @xmath9 to the es @xmath18 . in a two - terminal device , the system can relax ( r ) back to the gs only by absorbing a second electron from n at a negative energy @xmath15 tunneling into the @xmath19 part of @xmath18 , and the transfer of a cooper pair to s. the probability of this process cycle scales with @xmath20.@xcite for a reversed bias , a cooper pair is removed from s. since in this process two electrons are transferred from n to s we call the resulting subgap conductance features _ andreev resonances _ ( ars ) , which can be seen as a generalization of andreev reflection found in non - interacting systems . excites the s - qd system ( e ) and a subsequent electron at @xmath15 allows the system to relax again ( r ) , thereby adding a cooper pair to s. ( b ) energy diagram of the excitation and relaxation processes . ( c)-(e ) abs resonance energy @xmath21 for @xmath22 as a function of the level position @xmath23 for @xmath24 , @xmath25 and @xmath26 , respectively . ] the ars show a characteristic gate and bias dependence in the shape of `` loops '' that directly reflect the competition between superconducting pairing and the coulomb repulsion on the qd.@xcite while deep in the even qd charge states the gs of the system is the singlet @xmath18 ( @xmath27 ) , in the odd charge state the charging energy @xmath28 favors the doublet @xmath9 as gs ( @xmath29 ) , in competition with the pairing induced by @xmath30 . as illustrated for the limit @xmath22 in figs . [ fig2](c)-(e ) , one finds for small enough @xmath0 [ fig . [ fig2](c ) ] a gate - tunable transition from @xmath18 to @xmath9 at gate voltages where the ar loops intersect ( @xmath31 ) . in contrast , for large @xmath0 the ar loops do not intersect and the gs is always a singlet , as shown in fig . [ fig2](e ) . for @xmath32 the critical parameter is @xmath33 with a gs transition occurring for @xmath34,@xcite see fig . [ fig2](d ) . the limit for a phase transition should be reduced for finite @xmath1 due to a competition between the superconducting pairing and the formation of a kondo singlet with the quasi - particles at @xmath1.@xcite for @xmath35 theoretical results suggest the following relation for a gs transition to occur : @xmath36 where the kondo temperature @xmath37 characterizes the coupling to s at the electron - hole symmetry point . the number @xmath38 varies considerably in the literature , between @xmath39 and @xmath40 in ref . , @xmath41 in ref . and @xmath42 in ref . ( values corrected for different definitions of @xmath43 ) . in the normal state we expect that the kondo temperature , @xmath43 , is obtained by replacing @xmath0 by the total coupling @xmath44 , where we already included the coupling to a third normal metal terminal . though not experimentally demonstrated , yet , the spectroscopic width of an ar line is expected to be determined by the finite coupling of all normal terminals to the qd , @xmath45.@xcite we have fabricated cnt three - terminal devices with a lead ( pb ) based central @xmath46 nm wide pd / pb / in ( @xmath47 nm ) s contact,@xcite two pd n contacts and two sidegates ( sgs ) using close to residue free electron beam lithography.@xcite the typical critical perpendicular field for these s contacts is @xmath46mt and we apply @xmath48mt for our normal state experiments . the highly p - doped si / sio@xmath49 substrate serves as a backgate ( bg ) . an sem image of a typical device is shown in fig . [ fig1 ] , which also illustrates our measurement setup : a dc voltage , @xmath50 , with a superimposed ac modulation of @xmath51 is applied to s , while measuring the variations @xmath52 in the currents @xmath53 through the normal terminals n1 and n2 using standard lock - in techniques , which results in the differential conductances @xmath54 . in addition , we can set the potentials on n1 and n2 by the voltages @xmath55 and @xmath56 , respectively ( not shown ) , or leave them floating individually by the switches sw@xmath57 outside the cryostat . we define the voltage differences @xmath58 , for @xmath59 and @xmath60 . similarly , we can also apply the bias to n1 while measuring the current variations in s and n2 , or leaving s floating . all measurements were performed in a dilution refrigerator at a base temperature of @xmath61mk . we discuss three cnt devices a , b and c , each exhibiting subgap ars . in device a , the pd wetting layer was @xmath62 nm thick , while in the devices b and c it was @xmath63 nm . the room temperature resistances between an n terminal and s were @xmath64 ( sample a ) and @xmath65 ( samples b and c ) . in this section we investigate abs mediated transport between n and s terminals in the three devices and take full advantage of the three - terminal geometry . primarily , we focus on extracting the superconducting energy gap @xmath1 , the charging energy @xmath28 and the individual tunnel coupling strengths of the three contacts to the qd and related them to the observed ar resonance patterns and the ar broadening . the relevant data for device a are shown in fig . [ fig : abs - a - characterization ] , device b in fig . [ fig : abs - b - characterization ] and device c in fig . [ fig : abs - c - characterization ] and will be discussed in the following subsections . in all figures we plot the differential conductances @xmath66 ( left , a and c ) and @xmath67 ( right , b and d ) as a function of a gate voltage and the bias applied to s , @xmath50 , while keeping @xmath68 and @xmath69 . the top panels ( a and b ) show the data for s in the normal state obtained by applying a magnetic field of @xmath70 t perpendicular to the substrate , and the bottom panels ( c and d ) the measurements with s in the superconducting state at @xmath71 . the relevant extracted and derived device parameters are summarized in table [ tab : abs - devices ] . in app . [ app : supplabs ] we demonstrate in detail that the transport between all three terminals of device a is governed by a single qd . this is also evident in figs . [ fig : abs - a - characterization](a - b ) , which shows the same coulomb blockade ( cb ) diamond pattern ( dashed lines ) in the conductance @xmath66 and @xmath67 as a function of the bias and the voltage @xmath72 applied to both sgs , with an even - odd shell - filling sequence in both simultaneously recorded conductance maps . in the odd diamonds we find horizontal conductance ridges ( k , white arrows ) due to zeeman - split kondo resonances.@xcite for increasing magnetic fields ( not shown ) , we observe a zeeman - splitting of these ridges with a g - factor of @xmath73 . using additional measurement configurations,@xcite e.g. with the bias applied to an n terminal , we find that the kondo resonance solely originates from the s contact ( not shown ) we first analyze the normal state data in figs . [ fig : abs - a - characterization](a ) and ( b ) . from the cb resonances we deduce a leverarm @xmath74 of the combined sidegates , a charging energy of @xmath75mev , and from excited state and inelastic cotunneling lines the lowest two orbital energies @xmath76mev and @xmath77mev . the three - terminal geometry allows one to determine the coupling strengths @xmath78 of the individual contacts to the qd by fitting a breit - wigner lineshape@xcite to different two - terminal conductances.@xcite this procedure is explained in more detail in app . [ app : supplabs ] . from these fits , we obtain @xmath79 , @xmath80 , @xmath81 and @xmath82 and thus @xmath83ev for resonance 1 in fig . [ fig : abs - a - characterization](a ) . for resonance 2 we obtain @xmath84 , @xmath85 , @xmath81 and @xmath86 and thus @xmath87ev , and for resonance 3 @xmath88 , @xmath89 , @xmath90 and @xmath91 , so that @xmath92ev . we note that generally @xmath93 . as a control and to establish another method used below , we compare the kondo temperature estimated from the kondo resonance width and find a very good agreement for all three resonances with @xmath94 at the electron - hole symmetry point in the middle of the odd diamonds.@xcite for example , for resonance 2 we obtain @xmath95 ( @xmath96mk ) . and @xmath67 of device a as a function of the bias applied to s , @xmath50 , and the voltage @xmath97 applied to both sidegates , for @xmath98 and the external magnetic field @xmath70 t . ( a - b ) or @xmath71 ( c - d ) . the white arrows in ( a - b ) point out zeeman - split kondo resonances ( k ) and the numbers indicate the zero - bias cb resonances used for fitting . yellow and orange arrows highlight excited state andreev resonances . ] if s is superconducting , as shown in figs . [ fig : abs - a - characterization](c - d ) , @xmath66 and @xmath67 both exhibit a gap @xmath99mev , with pronounced subgap resonances for @xmath100 . also here we find the same conductance features ( though of differing amplitudes ) in @xmath66 and @xmath67 . we discuss the higher energy replicas of the low energy ar pointed out by yellow and orange arrows in figs . [ fig : abs - a - characterization](c - d ) in sec . [ sec : abs - addbias - excitedar ] . the subgap features are fully consistent with ars : for even spin states ( no kondo ridge in the normal state ) we observe large loops , while for odd occupation the loops are smaller and intersect inside the normal state cb diamonds . this dispersion is characteristic for ars with a relatively small coupling to s. for example , comparing the extracted system parameters for resonance 2 results in @xmath101 , i.e. the large-@xmath1 limit , in which @xmath102 suggests a gate tunable gs transition , consistent with the loop structure observed in the experiments . we note that also the strong - coupling expression predicts a gs transition , @xmath103 for all @xmath38 in the literature . similar results for the other resonances are summarized in tab . [ tab : abs - devices ] and are also consistent with the observed loop structures . the full width at half the maximum ( fwhm ) of the ars in this device is @xmath104 , much smaller than the total coupling strength @xmath105mev , but consistent with the coupling @xmath106 to the two normal leads . this demonstrates that @xmath107 fully accounts for the abs excitation life time . and @xmath67 of device b as a function of @xmath50 and @xmath108 at @xmath109 , for @xmath70 t ( a - b ) and @xmath71 ( c - d ) . the white arrows in ( a ) label kondo ( k ) ridges , and dashed lines qualitatively sketch the cb diamonds , where small ( large ) diamonds correspond to odd ( even ) occupation . gs denotes the qd groundstate in ( c ) , and the white arrow points to a second andreev resonance visible at a higher bias @xmath110 . ] in device b a single qd is formed mainly between s and n1 , while the cnt segment between s and n2 shows an essentially gate independent conductance in an open regime and acts essentially as a contact to the qd . this is shown in more detail in app . [ app : supplabs ] . in fig . [ fig : abs - b - characterization ] we therefore plot @xmath66 and @xmath67 as a function of @xmath50 and @xmath108 to keep the other side unperturbed . in the normal state of s , @xmath66 in fig . [ fig : abs - b - characterization](a ) exhibits a pattern typical for a qd strongly coupled to the leads , with an even - odd filling that can still be deduced from cotunneling lines . the cb diamonds are outlined by dashed lines . in contrast , @xmath67 plotted in fig . [ fig : abs - b - characterization](b ) shows only a small conductance modulation , as expected for highly transparent barriers . in @xmath66 we find kondo ridges in the odd charge states ( k , arrows ) , which are strongly broadened so that the zeeman splitting can not be resolved . the horizontal lines at @xmath111mv in the middle of the conductance map ( yellow arrows ) do not agree with the expected zeeman splitting of a ( spin ) kondo resonance and @xmath112 and are probably due to inelastic cotunneling , or due to the valley kondo effect.@xcite from the charge stability diagram in fig . [ fig : abs - b - characterization](a ) we estimate a leverarm of @xmath113 , a charging energy of @xmath114mev , and a level spacing of @xmath115mev , consistent with a qd smaller than in device a and located between s and n1 . due to the pronounced kondo resonances and the strong coupling to the leads , it is difficult to extract the tunnel coupling constants directly . we therefore turn around the procedure established for device a and first estimate the coupling strength @xmath116 from the half - width at half maximum ( hwhm ) of the kondo ridges k1 and k2.@xcite we obtain @xmath117mev ( @xmath118k ) for k1 , and estimate @xmath119mev from @xmath94.@xcite similarly , for k2 we find @xmath120 mev ( @xmath121k ) and @xmath122mev . these values are consistent with the cb resonance width at finite bias . @xmath66 and @xmath67 for s in the superconducting state are plotted in figure [ fig : abs - b - characterization](c - d ) , where we find pronounced ars in @xmath66 for bias voltages @xmath50 below the superconducting transport gap , @xmath123mev . a weak modulation at the same voltages can also be found in @xmath67 . the ar loop in the left - most odd state ( k1 in the normal state ) is considerably smaller than the loops in device a , and the right - most ar loop ( k2 in the normal state ) does not exhibit a gs transition at all . the ar between k1 and k2 ( arrow ) exhibits two resonances related to qd excited states and consistent with normal state cotunneling lines , as discussed in more detail in sec . [ sec : abs - addbias - excitedar ] . ' '' '' device & @xmath1 ( mev ) & @xmath28 ( mev ) & @xmath124 ( @xmath125ev ) & @xmath30 ( @xmath125ev ) & @xmath126 & @xmath33 & @xmath127 & gs transition + ' '' '' a & 0.95 & 2.5 & 54/47/33 & 150/173/118 & 6.3/5.5/8.1 & 0.06/0.07/0.05 & @xmath128 & y / y / y + ' '' '' b & 0.65 & 3.0 & 100/200 & 700/1500 & 0.93/0.43 & 0.23/0.50 & 0.29/1.05 & y / n + ' '' '' c & 0.42 & 2.4 & @xmath129 & @xmath130 & 0.60/0.54 & 0.29/0.33 & 0.57/0.69 & crit . / n + we now use the ar linewidth to estimate the remaining tunneling parameters . equating the ar linewidths with @xmath107 as established for device a , we find for resonance k1 @xmath131 and @xmath132mev , with a corresponding kondo temperature for the s contact of @xmath133ev . with @xmath134 , this resonance is neither in the large-@xmath1 , nor in the strong - coupling limit . however , both expressions still correctly predict a gs transition : @xmath135 in the large-@xmath136 limit and @xmath137 for all mentioned @xmath38 in the strong coupling limit . similarly , we find for resonance k2 @xmath138ev and @xmath139mev , with @xmath140ev . here , @xmath141 , on a scale on which the use of expression [ kondo_temp ] is still acceptable . this yields @xmath142 for all mentioned @xmath38 , consistent with a suppression of the gs transition . we note that with @xmath143 the large-@xmath136 limit would predict touching ar loops , on the boundary of the phase transition , in contradiction to the experiment . qualitatively , the larger coupling of the k2 state results in the suppression of the gs transition , whereas a transition is expected for k1 . and @xmath67 of device c as a function of @xmath50 and @xmath108 at @xmath109 , for @xmath70 t ( a - b ) and @xmath71 ( c - d ) . the black arrows in ( a ) label kondo ( k ) ridges , and dashed lines sketch the cb diamonds . @xmath144 highlight the two superconducting transport gaps . ] in contrast to the devices a and b , device c shows clear characteristics of a double qd with strong inter - dot coupling ( see app . [ app : supplabs ] for details ) . by only varying sidegate 1 and keeping @xmath145 , we mostly tune qd1 between n1 and s , while the conductance through qd2 between n2 and s is varied only around the maximum of a qd resonance ( app . [ app : supplabs ] ) . for this reason qd2 can essentially be seen as a contact to qd1 . figures [ fig : abs - c - characterization](a - b ) show @xmath66 and @xmath67 as a function of @xmath50 and @xmath108 for s in the normal state , with strong cotunneling lines and kondo resonances in the odd charge states ( k1 and k2 ) . from the charge stability diagram interpolated in fig . [ fig : abs - c - characterization](a ) we estimate a leverarm of sg1 to qd1 of @xmath146 , a charging energy of @xmath147mev and a level spacing of @xmath148mev for qd1 . similar as for device b , we estimate @xmath116 from the hwhm of the kondo resonance and obtain @xmath149mev ( @xmath150k ) and @xmath119mev for the resonance k1 . for resonance k2 we find @xmath151mev ( @xmath152k ) and @xmath153mev . for s in the superconducting state , @xmath66 and @xmath67 plotted in figs . [ fig : abs - c - characterization](c - d ) both show pronounced ars . in contrast to the previous devices , we find _ two _ horizontal conductance maxima , one at @xmath154mev most prominent in @xmath66 , and @xmath155mev more clearly visible in @xmath67 , which we tentatively interpret as two superconducting transport gaps in the individual cnt arms . surprisingly , the abs loops connect smoothly , but the right - most loops are bounded by @xmath156 , while the left - most clearly reaches up to @xmath157 . currently we do not have a good explanation for these findings , but speculate that the qd wave function might be coupled inhomogeneously to s , e.g. at two separate places with two distinct effective coupling constants . for the present investigation the relevant findings are that the abs loops exactly touch for the k1 state and that for k2 the loops are slightly separated in energy , i.e. that the k1 parameters correspond to the critical values of the abs phase transition and that the gs transition is suppressed for k2 . the latter observations compare well to the extracted tunnel couplings : for device c , we find an ar line width of @xmath158mev for both resonances and estimate @xmath159mev at the position of the kondo resonance k1 and @xmath160mev for k2 . using @xmath156 as the relevant energy gap , we find @xmath161 for k1 and @xmath162 for k2 , both in a range for which it is acceptable to use eqn . ( [ kondo_temp ] ) . the latter results in @xmath163 for k1 , consistent with a ` touching ' of the ar loops if @xmath41 in eq . ( [ kondo_temp ] ) , and @xmath164 for k2 , suggesting a close but avoided ar loop crossing and no gs transition . we note that the ratios used for the large-@xmath1 limit , @xmath165 for k1 and @xmath166 for k2 , would wrongly predict a gs transition for both charge states . the results of this section are collected in table [ tab : abs - devices ] and compared to the ar loop structures found in the experiments . our experiments clearly suggest a gs transition limit of @xmath41 when using eq . ( [ kondo_temp ] ) . in addition , we find that the ar broadening is determined by the coupling to both normal contacts . in sec . [ sec : abs - devicecharacteriation ] we found multiple replicas at higher energies for all ars in device a and one replica in the central ar of device b. the spacings of @xmath167 and @xmath77mev between these replicas are very similar to the spacings of the lowest excited states in the normal state measurements of this device . we find no such replicas for device c , for which the orbital energies are larger than the superconducting gap , @xmath168 . we note that the replicas are not consistent with inelastic andreev tunneling,@xcite since they show a clear gate dependence following the lowest ar , we find no negative differential conductance between the resonances , and they are broadened with increasing temperature . in the following , we exploit the third qd terminal to show that the replicas are caused by abs mediated transport via excited states of the doublet @xmath9 in the odd qd state , a process we call excited state ars . and @xmath67 of device a at @xmath169 as a function of @xmath170 and a combined sidegate voltage @xmath97 at @xmath171 . white arrows denote transport transport resonances linear in the gate voltage between the n contacts , which partially overlap with excited state ars ( orange arrows ) . ] to experimentally identify the origin these replicas , we perform conductance measurements in a configuration different to the one in sec . [ sec : abs - devicecharacteriation ] . here we keep the dc parts @xmath172 and apply the dc bias @xmath170 to n2 , while modulating @xmath173 by the ac voltage @xmath174 ( see also fig . [ fig1 ] for orientation ) . we again define the conductances @xmath175 from the current modulations in the normal terminals 1 and 2 . figure [ fig : abs - a - exp - offsetbias ] shows @xmath66 and @xmath67 as a function of @xmath97 and @xmath176 . intuitively , one might expect that @xmath66 is independent of @xmath170 , which is the case for local processes when neglecting gating effects by @xmath170 . surprisingly , we find both , abs mediated ( ` curved lines ' ) and standard qd resonances ( ` straight lines ' ) in these data sets . in the picture of @xmath3 this is not trivial and we will discuss this finding in more detail in sec . [ sec : abs - floatings - proximity ] . here , we exploit that the width of both types of transport resonances between n1 and n2 is determined by @xmath177 only , and not by @xmath178 as in the normal state ( see sec . [ sec : abs - devicecharacteriation ] ) . this allows us to resolve also excited state resonances between the two n contacts . in fig . [ fig : abs - a - exp - offsetbias ] we find both , straight lines ( labeled by white arrows ) and ar with a curved dispersion ( orange arrows ) , best visible in @xmath67 . the straight excited state resonances have a similar spacing as the corresponding excited state ars and sometimes even originate from the same position in these plots . excited state ars can be understood on the same footing as the gs ars in the @xmath22-limit ( sec . [ sec : abs_basics ] ) . in addition to the one - electron state @xmath9 we also consider the excited state @xmath179 , separated in energy by @xmath180 . as depicted in fig . [ fig : abs - excitedar](a ) for @xmath18 being the gs , the system can now also be excited to @xmath179 ( blue arrows ) in addition to @xmath9 ( red arrows ) . the corresponding transition energies @xmath181 ( blue lines ) and @xmath21 ( red lines ) are plotted in fig . [ fig : abs - excitedar](c ) . @xmath181 replicates the shape of @xmath21 at higher energies , as observed in the experiments . we call the resonances @xmath181 _ excited state andreev resonances _ , because they correspond to a first order transition between an abs and an excited single - particle state . this situation of @xmath18 being the gs results in the excited state ar highlighted by yellow arrows in the experiments of fig . [ fig : abs - a - characterization](c - d ) . also consistent with the experiment is that the expected @xmath181 line ends at the intersection with @xmath21 in fig . [ fig : abs - excitedar](b ) , best visible in the central resonance of fig . [ fig : abs - b - characterization](c ) . this can be understood qualitatively by introducing a ( fast ) energy relaxation from @xmath179 to @xmath9 , which we call `` thermal relaxation '' ( tr ) to distinguish it from the parity changing relaxation by single electron tunneling . if @xmath18 is the gs this results only in a second relaxation channel from @xmath179 to the gs , as illustrated in fig . [ fig : abs - excitedar](a ) . however , if @xmath9 is the gs and @xmath18 lies energetically midway between @xmath182 and @xmath183 or higher , as shown in fig . [ fig : abs - excitedar](b ) , @xmath182 can be excited by single electrons at @xmath181 , but the system relaxes fast to @xmath9 , from where the energy @xmath184 is required to reach @xmath18 to close the charge transport cycle . therefore we expect for these gate voltages only one resonance at @xmath185 , but with an increased amplitude , because the additional transport channel via @xmath179 is allowed for @xmath186 . and @xmath179 with a level spacing of @xmath180 and the transition energies @xmath21 and @xmath181 . tr abbreviates thermal relaxation . ( a ) shows the processes for @xmath18 being the gs , ( b ) the ones for @xmath18 set between @xmath9 and @xmath179 . ( c ) @xmath21 and @xmath181 as a function of @xmath187 for @xmath188mev , @xmath189mev and @xmath190mev . @xmath191 denotes the different charge states . ] in three - terminal qd devices , one can expect novel subgap transport mechanisms due to `` non - local '' transport processes that require two n and one s terminal , and their interplay with local effects involving only two terminals.@xcite in particular , transport between two n contacts can be mediated by abss.@xcite to perform this type of experiments , we again use a different measurement configuration than in the previous sections , with the ac bias @xmath174 and the dc bias @xmath55 applied to n1 , while measuring the current variation @xmath192 the other n terminal n2 . this can be done while also recording the current variation @xmath193 in s at a fixed potential @xmath173 , or while leaving s floating ( see sec . [ sec : abs - floatings - proximity ] ) . we again define the differential conductances @xmath194 . and @xmath67 as a function of the applied bias voltage @xmath195 and the gate voltage @xmath108 , for ( a - b ) device b and @xmath196 , ( c - d ) device c and @xmath197mv . arrows and dashed rectangles mark regions of pronounced conductance sign changes . ] in fig . [ fig : abs - serial - sgnchange](a - b ) we plot @xmath198 and @xmath67 of device b as a function of @xmath55 and the gate voltage @xmath108 , with @xmath196 . in addition to the typical ar pattern in @xmath198 already discussed for fig . [ fig : abs - b - characterization ] , we find in @xmath67 pronounced resonances at the same gate and bias voltages . this illustrates again that n2 is a third terminal to the qd with abss . in contrast to the ars in @xmath199 , the ar amplitudes in @xmath67 show pronounced sign changes , from positive to negative differential conductance with similar magnitudes @xmath200 at positions symmetric around the electron - hole ( e - h ) symmetry points . the sign is inverted at the e - h symmetry point , at the singlet - doublet gs transitions , and for a reversed bias . this is best visible in the regions marked by rectangles in fig . [ fig : abs - serial - sgnchange](b ) . the central gs transition deviates from this description , most likely due to the overlapping second ar ( see sec . [ sec : abs - devicecharacteriation ] ) , but also exhibits systematic sign changes . for the devices a and c we find similar characteristics as for device b. as an example , we plot @xmath198 and @xmath67 of device c in fig . [ fig : abs - serial - sgnchange](c - d ) , where we added a dc bias @xmath197mv to s. this bias results in an offset of the features in @xmath198 and @xmath67 , where we again find the characteristic sign changes in @xmath67 , labeled by white arrows in fig . [ fig : abs - serial - sgnchange](d ) . from these and a series of other measurements with different @xmath50 we conclude that the sign changes are robust against @xmath50 and that all measurements can be understood considering potentials relative to the one on s. to describe the transport through an s - qd system , we introduce a rate equation model similar to ref . , using the @xmath22 limit to describe the abss and the doublet states , as shown in fig . [ fig : abs - rateeqschem](a ) . the excitation and relaxation rates depend on the transport processes , illustrated in fig . [ fig : abs - rateeqschem](b ) . in contrast to the two - terminal case , we now consider a coupling to a third normal metal terminal n2 with a strength @xmath201 , in addition to the terminal s ( @xmath30 ) and n1 ( @xmath202 ) . for @xmath203 we need to consider only transport mediated by the abs . the bias @xmath195 is applied to n1 as in the experiment , while s and n2 are kept at @xmath204 . the excitation and relaxation rates , @xmath205 and @xmath206 , denote the total rates by which the system is changed from the gs to the excited state ( es ) and back , and will be determined below . the steady state occupation probabilities @xmath207 and @xmath208 of the gs and es are then found by setting @xmath209 which yields @xmath210 and @xmath211 . the rates @xmath205 and @xmath206 contain all excitation and relaxation processes , which can be viewed in the simplified picture of sequential individual tunneling events , as illustrated in fig . [ fig : abs - rateeqschem](b ) . in the corresponding rates we use the superscript @xmath212 to denote the contacts n1 ( n2 ) and @xmath213 to state whether an electron is added ( removed ) from the s - qd system . the excitation and relaxation rates can then be calculated using fermi s golden rule.@xcite for example , if @xmath9 is the gs and @xmath18 the es , see fig . [ fig : abs - rateeqschem](c - d ) , the rate @xmath214 for the excitation by an electron tunneling in from n1 is @xmath215{+1e}\ket{-}:\ ; t_{\rm e}^{1+}=\gamma_{\rm 1 } \underbrace{\vert\braket{-|d_{\bar{\sigma}}^{\dagger}|\sigma}\vert^{2}}_{v^{2 } } f_{1}(\zeta),\ ] ] where @xmath216 is the fermi distribution in contact n1 ( n2 ) , @xmath217 ( @xmath218 ) the creation ( annihilation ) operator for an electron on the qd with spin @xmath183 ( opposite spin @xmath219 ) , and @xmath6 and @xmath5 are the bdg amplitudes of the abs . similarly , we find for the relaxation rate caused by electrons leaving to n2 @xmath220{-1e}\ket{\sigma}:\ ; t_{\rm r}^{2-}=\gamma_{\rm 2 } \underbrace{\vert\braket{\sigma|d_{\bar{\sigma}}|-}\vert^{2}}_{v^{2}}(1-f_{2}(\zeta)).\ ] ] ( @xmath206 ) . ( b ) schematic of the three - terminal n - qd - s system , with a faint superimposed s contact , and a negative bias voltage @xmath195 applied to n1 . transport is only considered through the ar @xmath221 , with color - coded rates @xmath222 . @xmath223 , @xmath224 and @xmath225 denote the technical current direction in the respective contacts ( arrows ) . ( c - d ) intuitive diagrams of the possible excitation ( c , red ) and relaxation ( d , blue ) processes and non - vanishing transition matrix elements , if @xmath9 is the gs . we consider only first order transitions ( + 1e/-1e ) with the energy conditions @xmath226 . ] we note that this latter process is possible only in a three - terminal device . all such excitation and relaxation processes are depicted schematically in fig . [ fig : abs - rateeqschem](c - d ) , which allows to intuitively deduce all tunnel rates . the total excitation and relaxation rates then read @xmath227 for the singlet @xmath18 being the gs , all rates can be obtained from eq . ( [ eq : abs - rates ] ) by replacing @xmath5 by @xmath6 , because the initial and final states in the transition matrix elements are interchanged . the total currents into n1 and n2 are then given by @xmath228 \\ & i_2= \frac{e}{h}\left [ ( t_{\rm e}^{2 + } - t_{\rm e}^{2- } ) p_{\rm gs } + ( t_{\rm r}^{2 + } - t_{\rm r}^{2- } ) p_{\rm es}\right ] , \end{split } \label{eq : abs - modelcurrents}\ ] ] and @xmath229 . the differential conductance measured at contact @xmath230 can then be obtained as @xmath231 for the experiments described in this section . ) and ( [ eq : abs - rates ] ) in the @xmath232 limit , for @xmath189mev , @xmath188mev and @xmath233mev . ( a ) @xmath66 , ( b ) @xmath198 and ( c ) @xmath67 as a function of @xmath195 applied to n1 and the qd orbital energy @xmath187 , for @xmath234 , as in the experiment . to simulate a broadening of the ar , a small finite temperature of @xmath235k was assumed . ( d ) schematic of the `` nonlocal '' transport processes for @xmath236 . the red arrow depicts `` resonant abs tunneling '' , i.e. a direct transfer of one electron from n1 to n2 via the abs , whereas the blue arrows show an inverse cps process , i.e. the `` nonlocal '' creation of a cooper pair in s. ] the resulting three conductances are plotted in fig . [ fig : abs - rateeqmodel ] as a function of @xmath55 and the normalized gate - tunable qd orbital energy @xmath187 . while @xmath66 plotted in fig . [ fig : abs - rateeqmodel](a ) was not measured in the experiment , @xmath199 in fig . [ fig : abs - rateeqmodel](b ) exhibits the familiar ars with solely positive conductance values . more importantly , fig . [ fig : abs - rateeqmodel](c ) shows @xmath67 , which compared with the experiment , for example in fig . [ fig : abs - serial - sgnchange](b ) . we find that the model at least qualitatively reproduces the experimental findings , with sign changes at the e - h symmetry point ( @xmath237 ) and for a reversed bias , and clearly highlighting the gs transition . similarly , for @xmath18 being the gs for all gate voltages ( large @xmath30 , not shown ) , we find sign changes only at the e - h symmetry point , consistent with the experiments shown in fig . [ fig : abs - serial - sgnchange](b ) and ( d ) . in addition , the sign changes in the model are independent of the applied voltages @xmath50 or @xmath170 , as in the experiment . due to the fermi functions , the only non - vanishing rates at low temperatures are @xmath214 , @xmath238 , @xmath239 and @xmath240 for a forward bias @xmath241 , as depicted in fig.[fig : abs - rateeqschem ] . for example , the local andreev processes between n1 and s scale with @xmath242 . more interestingly , the sign changes in @xmath67 are caused by the competition between processes comprising the two relaxation processes @xmath239 and @xmath240 , which remove or add an electron from the system at different energies , respectively . in this case the current in n2 is proportional to the difference between the bdg amplitudes , @xmath243 , with a sign change at the gs transition ( transition from @xmath244 to @xmath245 ) , or for a reversed bias @xmath195 . the conductance @xmath67 thus reflects directly the gradual charging of the abs with decreasing gate voltage , which evolves from an excess charge of @xmath246 in the @xmath247 state to an average charge of @xmath248 in the @xmath249 state . in a more coherent picture the process combining an electron tunneling into the system from both , n1 and n2 , ( @xmath214 combined with @xmath240 ) corresponds to the ` nonlocal ' creation of a cooper pair in s , often called ( inverse ) cooper pair splitting ( cps ) . similarly , the coherent tunneling into and out of the system at the same energy ( @xmath214 combined with @xmath239 ) corresponds to the direct transfer of one electron from n1 to n2 , mediated by the abs . this new process we call _ ` resonant abs tunneling'_. these ` nonlocal ' transport processes involve all three terminals and are illustrated in fig . [ fig : abs - rateeqmodel](d ) . for these processes we expect the rates @xmath250 for inverse cps and @xmath251 for resonant abs tunneling . this interpretation is also supported by recent more fundamental calculations,@xcite all in good agreement with our experiments and the simple model . we point out that to account for the similar magnitudes of the positive and negative differential conductance in @xmath67 , it is necessary that cps and resonant abs tunneling are of similar strength . in this section we present and compare measurements with different boundary conditions imposed to the qd terminals . for example , figs [ fig : abs - serial - floatings](a - c ) show the differential conductance @xmath252 between the two normal metal contacts ( see also in the previous section ) with the superconductor s left floating . the data for the devices a , c and b , are plotted as a function of the bias and gate voltage in the same gate voltage intervals as discussed in the previous sections . the data sets are ordered according to increasing @xmath30 , as found in sec . [ sec : abs - devicecharacteriation ] . such experiments are equivalent to ref . , where abss in a proximitized s - qd inas nanowire were investigated in the context of majorana bound states . for device a , with a weak coupling to s and a small @xmath33 ratio , we find cb diamonds very similar to the normal state . specifically , there are no ars , nor a suppression of @xmath252 related to a superconducting gap . ars were , however , clearly observed when the superconductor was grounded ( see fig . [ fig : abs - a - characterization ] ) , and both types of resonances were found in the experiments of fig . [ fig : abs - a - exp - offsetbias ] . in contrast , devices c and b , both with a larger coupling to s , show clear signatures of ars and the energy gap . however , in these experiments no sign changes in @xmath252 occur and we observe only positive differential conductance features , in stark contrast to s being on a fixed potential ( fig . [ fig : abs - serial - sgnchange ] ) . for device b , this experiment results in essentially the same ar gate and bias dispersion as when measured between an n and the s contact ( fig . [ fig : abs - b - characterization ] ) , whereas for device c we find additional curved resonances at higher energies ( cf . [ fig : abs - c - characterization ] ) . we note that the ars are centered around zero bias , which suggests a negligible change in the electrical potential of s. the experiments on devices b and c are straight forward to understand qualitatively based on our previous picture of abss : we assumed that the cnt segment between n1 and n2 forms a qd in the normal state of s. in the superconducting state , the eigenstates of the segment develop into abss that determine all subgap transport . if s is at a fixed potential , it can take up ( or emit ) cooper pairs , which allows cooper pair splitting as a process and , in unison with resonant abs tunneling , results in the pattern of positive and negative differential conductance . if s is floating , the experiments show that @xmath253 , since the ars are centered around zero bias . since cps and the local andreev processes both add charges to s , which would change the electrical ( though not the chemical ) potential of s , these processes exactly compensate each other , to satisfy the boundary condition . this leaves only resonant abs tunneling as a subgap transport process between n1 and n2 , which does not change the charge on s and results in a positive @xmath252 , mapping out the abs excitation energy . additional features and different amplitudes compared to the traditional ars are to be expected , since here only electrons at one energy are involved . the interpretation of the experiments on device a are more problematic . we find in the conductance between n1 and n2 the standard qd resonances when s is floating , while if s is at a fixed potential modulated by the ac voltage , abs signatures occur simultaneously with the standard qd features , see fig . [ fig : abs - a - exp - offsetbias ] in sec . [ sec : abs - addbias - excitedar ] . in the two - terminal measurements between n and s , we only find abs related signals . we note that both , the coulomb diamonds and the ars , are not offset in bias , suggesting that the cnt segment is at zero potential . these experiments seem quite contradictory , since one would expect that the eigenstates of the s - qd system do not depend on the boundary conditions of the measurement . as discussed for samples b and c , the different processes are sensitive to different boundary conditions , but simultaneous observation of standard coulomb blockade and abs features can not be resolved with such arguments alone . this suggests that our picture of abs mediated transport is not complete and only holds for a strong enough coupling to s. in this picture we assume that the qd develops new eigenstates due to the coupling to the unperturbed superconductor . one might expect that a continuous lowering of the coupling strengths results in a continuous transition from the abs to the bare qd states . our results , however , seem to suggest that in some parameter regime both sets of states are available for different transport processes . we speculate that for low enough @xmath0 the tunneling of an electron on and off the qd from n terminals is faster than the exchange of cooper pairs with s , which is necessary for the hybridization of the qd with s. right after the first tunneling event the system is in the state @xmath17 , which is a coherent superposition of the system eigenstates @xmath18 and @xmath254 , i.e. , @xmath255 ( @xmath256 ) . therefore we expect coherent oscillations between @xmath19 and @xmath17 in an abs excited by single electron tunneling events , with a characteristic frequency essentially determined by @xmath0 . this oscillation is expected to result in a suppression of the waiting time distribution for short times,@xcite and might be responsible for our findings here : in a three - terminal device an excitation can not only be absorbed by forming a cooper pair , as in a two - terminal device , but also by tunneling out into the third terminal ( resonant abs tunneling ) . if this happens faster than half a coherent oscillation , the tunneling electron stays in the bare qd state during the time it resides on the cnt segment . assuming that an oscillation by half a period is required to transfer a cooper pair to s , we find the relation @xmath257 as a necessary condition to observe the bare qd resonances , in good agreement with device a , but not for the other two devices with considerably larger @xmath0 and a faster oscillation rate relative to @xmath258 . our results demonstrate that a three - terminal qd can give access to the fast electron dynamics , in our example of the abs coherent oscillations with frequencies @xmath259 in the range between @xmath260ghz ( device a ) to @xmath261ghz ( device b ) , by the implicit comparison of the tunnel rates to the third terminal . in summary , we investigate electronic transport mediated by andreev bound states in a three - terminal s - qd device and identify the coupling to the normal metal leads as the main source for the spectroscopic broadening of the andreev resonances , and establish how the coupling to the superconductor determines the ground state of the system . in addition , we present ` excited state andreev resonances ' at higher energies , with transitions between abss and odd parity excited qd states . we also report pronounced sign changes in the abs - mediated transport between the two normal metal contacts , which we explain in an intuitive rate equation model as resulting from the competition between the nonlocal creation of a cooper pair in s and the new process of ` resonant abs tunneling ' , i.e. , the subgap transport of single electrons through the s - qd system , only allowed in multi - terminal devices . surprisingly , we find that depending on the imposed boundary conditions in the experiments it is possible to observe either abss or coulomb blockade resonances , or both in the same experiment , which we tentatively attribute to the competition between coherent oscillations in the abs ( exchange of cooper pairs with s ) and the relaxation of the system by single electron tunneling into the second normal terminal . we believe that experiments on complex quantum systems using multiple terminals provide a novel and clear experimental probe for many old and new phenomena , here for example the onset of the superconducting proximity effect on a qd and the formation of many - body quantum states . such experiments also give direct access to the strength of the coupling between s and the qd , and a way to probe time dependent phenomena like the finite - frequency coherent oscillation between superposition states , without resorting to high - frequency and time domain experiments . in particular , we envisage experiments in similar structures with gate - tunable tunnel barriers@xcite to investigate in more depths the presented physical mechanisms . we thank a. levy yeyati , j. schindele and p. makk for fruitful discussions . this work was financially supported by the swiss national science foundation ( snf ) , the swiss nanoscience institute ( sni ) , the swiss nccr qsit , the erc project quest and the eu fp7 project se@xmath262nd . in this appendix we analyze additional data of devices a , b and c. in particular , we discuss the electronic configuration of each device , using conductance maps as a function of both sidegate voltages . figures [ fig : abs - a - oneqdobject](a ) and ( b ) show the measured differential conductance @xmath66 and @xmath67 of device a as a function of the sidegate voltages @xmath108 and @xmath263 , for the ac bias applied to s , but with @xmath264 , and in the normal state of the device at @xmath265 t . these data are consistent with a single qd@xcite , for example because we observe the same resonance lines in both , n1 and n2 , and when we apply the ac bias to s. in addition , we find one dominant slope of the cb resonances that react similarly to both sidegates . there are only vague hints at a more complex confinement potential , possibly due to disorder or potential fluctuations on the substrate , e.g. a slight conductance modulation with some features stronger ( weaker ) in one ( the other ) arm , and slight changes in the slopes or the spacing between neighboring resonances . the electronic configuration we qualitatively deduce from these findings is sketched in the inset of fig . [ fig : abs - a - oneqdobject](a ) . in particular , in the region studied in the main text , marked by an orange line in fig . [ fig : abs - a - oneqdobject](b ) , the assumption of a single qd is justified . in fig . [ fig : abs - a - oneqdobject](c ) , we plot the differential conductance in the normal state of s along the yellow dashed line of fig . [ fig : abs - a - oneqdobject](b ) , as a function of the @xmath50 and one sidegate voltage only . here , we also clearly observe the same diamond structure in both arms of the device , further supporting this claim . and @xmath67 of device a as a function of the sidegate voltages @xmath108 and @xmath263 , for @xmath266 , at an external magnetic field of @xmath265 t applied out - of - plane . the orange [ dashed yellow ] line in ( b ) indicate the studied gate voltages in the main text [ in fig . the inset in ( a ) sketches the assumed electronic qd configuration of the device . ( c ) @xmath66 and @xmath67 as function of @xmath50 and @xmath263 only , at @xmath265 t and @xmath267 . ] ( a ) and @xmath67 ( b ) of resonance 1 ( main text ) as a function of the combined sidegate voltage @xmath97 , at @xmath266 and @xmath70 t . red ( blue dashed ) lines represent the best fits obtained for a breit - wigner@xcite lineshape ( thermally broadened cb resonance@xcite ) . ] next , we briefly demonstrate the fitting procedure to determine the individual @xmath78 of the contacts . in fig . [ fig : abs - a - fits ] , the simultaneously measured zero - bias conductance @xmath268 and @xmath269 of resonance 1 ( main text ) is plotted as a function of the gate voltage @xmath97 in the normal state of the device . fits with a breit - wigner line - shape for a three - terminal device in the lifetime - broadened limit ( @xmath270 ) and in the single - level transport regime ( @xmath271 ) with the gate - tunable position @xmath272 of the resonance , @xcite @xmath273 agree very well with the data , and yield @xmath274 , @xmath275 and @xmath276 . from these equations , all parameters are determined for @xmath277 . this assumption can be directly justified from a similar analysis of measurements with the bias applied to n1 , while measuring @xmath278 and @xmath279 ( not shown ) . figure [ fig : abs - b - gategatemap](a ) and ( b ) show @xmath66 and @xmath67 of device b as a function of the sidegate voltages @xmath108 and @xmath263 , for the ac bias applied to s , but with @xmath264 , and in the normal state of the device at @xmath70 t . in the conductance map of @xmath66 , we observe a single dominant slope which is tuned mostly with @xmath108 , suggesting a qd located between s and n1 . in contrast , @xmath67 plotted in fig . [ fig : abs - b - gategatemap](b ) shows a very high conductance with a weak and slow amplitude variation as function of @xmath263 , characteristic for a more open cnt regime with highly transmissive contacts . in particular , @xmath67 never approaches zero . imprints of the resonances from @xmath66 can be observed in @xmath67 , which we ascribe to resistive and capacitive cross - talk.@xcite hence , we assume an electronic configuration of the cnt device as schemtically depicted in the inset of fig . [ fig : abs - b - gategatemap](b ) , with a larger qd on the left side of the device mostly tunable by @xmath108 , and an ` open ' cnt lead to the right . and @xmath67 of device b as a function of the sidegate voltages @xmath108 and @xmath263 , for @xmath266 , at @xmath70 t . the orange line in ( a ) indicates the studied gate voltage region in the main text . inset in ( b ) : assumed electronic configuration of the cnt device . ] and @xmath67 of device c as a function of the sidegate voltages @xmath108 and @xmath263 , for @xmath266 , at @xmath70 t ( a - b ) and at @xmath71 ( c - d ) . white dashed arrows in ( a , b ) denote different slopes consistent with resonances of qd1 or qd2 . the white rectangle marks the region studied in ( c , d ) . in ( c ) , the orange line corresponds to the gate voltage region studied in the main text , and dashed lines denote the charge states of qd2 ( even / odd ) . ] in fig . [ fig : abs - c - gategatemap ] , @xmath66 and @xmath67 of device c are plotted as a function of the sidegate voltages @xmath108 and @xmath263 , for the ac bias applied to s , but with @xmath264 , for s in the normal state at @xmath70 t ( a - b ) and in the superconducting state at @xmath71 ( c - d ) . in contrast to device a and b , a charge stability diagram with anti - crossings characteristic for a double quantum dot with strong inter - dot coupling and hybridization is observed in the normal state of s. in fig . [ fig : abs - c - gategatemap](a ) and ( b ) , the dominant conductance lines with different slopes in @xmath66 and @xmath67 are consistent with resonances of qd1 or qd2 . an apparent smearing of the resonances in every other charge state , both for qd1 and qd2 , is due to pronounced kondo ridges . in particular , the broad resonances in fig . [ fig : abs - c - gategatemap](a ) which we ascribe to qd1 indicate an odd charge state with a kondo ridge . this is more obvious when one closely inspects the region indicated by a dashed rectangle in fig . [ fig : abs - c - gategatemap](a ) in the superconducting state of s , which is plotted in fig . [ fig : abs - c - gategatemap](c ) . here , one observes a pair of andreev resonance lines in the even charge state of qd2 ( yellow arrow ) , corresponding to the two singlet - doublet gs transitions observed in the odd charge states of qd1 . these gs transitions vanish close to the boundary and in the odd charge state of qd2 , which can for example be seen by following the andreev resonances marked with a yellow arrow . in the main text , we focus on andreev states in qd1 for a fixed even charge state of qd2 , but close to its charge degeneracy point . the corresponding gate voltage region is indicated by an orange line in fig . [ fig : abs - c - gategatemap](c ) .

we propose a general framework for parameter - free identification of a class of dynamical systems . here , the propagator is approximated in terms of an arbitrary function of the state , in contrast to a polynomial or galerkin expansion used in traditional approaches . the proposed formulation relies on variational data assimilation using measurement data combined with assumptions on the smoothness of the propagator . this approach is illustrated using a generalized dynamic model describing oscillatory transients from an unstable fixed point to a stable limit cycle and arising in nonlinear stability analysis as an example . this 3-state model comprises an evolution equation for the dominant oscillation and an algebraic manifold for the low- and high - frequency components in an autonomous descriptor system . the proposed optimal model identification technique employs mode amplitudes of the transient vortex shedding in a cylinder wake flow as example measurements . the reconstruction obtained with our technique features distinct and systematic improvements over the well - known mean - field ( landau ) model of the hopf bifurcation . the computational aspect of the identification method is thoroughly validated showing that good reconstructions can also be obtained in the absence of of accurate initial approximations . * keywords : * hydrodynamic instabilities , reduced - order models , mean - field models , variational data assimilation , adjoint - based optimization , * ams subject classifications : * 93a30 , 65k10 , 76d25

1209.4416

in this study we consider the problem of parameter - free identification of a class of dynamical systems . the approach we propose is derived from a general method for the reconstruction of the constitutive relations in systems described by partial differential equations ( pdes ) which was initially introduced in @xcite and further developed in @xcite , see also @xcite the idea is that , given an autonomous evolution equation @xmath0 for some quantity @xmath1 ( defined in a finite or infinite dimension with @xmath2 denoting time ) , one seeks to optimally reconstruct the flux function @xmath3 , so that the system outputs best match , in a suitably defined sense , with the measurements available . as shown schematically in figure [ fig : f ] , the originality of this approach is that the function @xmath3 is reconstructed directly as a _ continuous _ object , rather than employing a truncated polynomial or other galerkin expansion . in @xcite we reviewed the mathematical foundations and some computational aspects of this approach applied to the reconstruction of state - dependent transport coefficients in a class of systems described by pdes . in the present investigation we adapt this method to the problem of identification of the propagator function @xmath4 , @xmath5^t$ ] , in a finite - dimensional dynamical system @xmath6 with the state vector @xmath7^t \in \rr^n$ ] . to fix attention , instead of working with an abstract formulation , we will focus on a specific problem concerning a three - dimensional ( @xmath8 ) model characterized by oscillatory dynamics with a stable limit cycle . in many real - life physical problems with infinite - dimensional state spaces and evolution described by pdes such systems arise as reduced - order models providing low - dimensional approximate description in the neighborhood of fixed points and limit cycles . in some cases propagator @xmath9 may be derived from the description of the full plant based on first principles . in general , however , the reduced - order model is inferred from experimental or numerical data . such reduced models are of paramount importance as test - beds for the development of our understanding of system dynamics . they are also useful as low - cost surrogates guiding optimization and real - time control design for expensive full - scale models . in all cases , typical model identification is generally performed in three steps : ( 1 ) selection of the state space @xmath10 which is large enough to capture the behaviour of interest and at the same time sufficiently small to allow one to exploit the analytical / numerical advantages of the surrogate plant ; ( 2 ) structure identification of the propagator @xmath11 , e.g. , determination of the polynomial degree of its components @xmath12 , @xmath13 ; and ( 3 ) parameter identification , e.g. , inference of the polynomial coefficients from time - resolved trajectories @xmath14 . these different steps are illustrated schematically in figure [ fig : f]a . the challenge of this approach is to find the right balance between the robustness of the model identification , requiring only a few tunable parameters , and a good accuracy for which a larger number of parameters is typically needed . identification problems can also be solved using variational techniques and this is the approach we will pursue in the present study , cf . figure [ fig : f]b . similar methods have been developed for a broad range of problems in both the finite and infinite - dimensional setting , including flow control in fluid mechanics @xcite , data assimilation in meteorology @xcite and geophysics @xcite to mention just a few application areas . the related problem of state estimation is usually solved using various filtering approaches such as the kalman filter @xcite . we introduce now our model . the oscillatory fluctuation is parameterized by @xmath15 where @xmath16 is the imaginary unit , @xmath17 is the amplitude of the fluctuation ( `` @xmath18 '' means `` equal to by definition '' ) and @xmath19 the corresponding phase , while the base - flow deformation is characterized by a single parameter @xmath20 . following the mean - field theory @xcite , we make the following assumptions about the structure of the system : 1 . in the plane @xmath21 the system exhibits an unstable fixed point at the origin and an attracting limit cycle , 2 . the state variable @xmath20 is `` slaved '' to @xmath22 and @xmath23 , i.e. , @xmath24 , and 3 . the dynamics is phase - invariant , i.e. , the descriptor system depends only on @xmath25 . [ ass1 ] as regards the time @xmath26 , we will assume that @xmath27 $ ] for some @xmath28 . as a general form of a dynamical system consistent with assumption [ ass1 ] we will consider [ eq : descriptorsystemb ] @xmath29 or , equivalently [ eq : descriptorsystema ] @xmath30 where @xmath31 and @xmath32 . subsequently , we will also use the notation @xmath33^t$ ] and @xmath34 $ ] , where @xmath35 } r(t)$ ] . we note that the equations governing @xmath22 and @xmath23 , or equivalently @xmath25 and @xmath36 , do not depend on @xmath20 . equations and describe the dependency of a slow variable on the fluctuation amplitude , and the usefulness of these algebraic equations will become clear in the context of the mean - field models discussed in section [ toc : meanfieldsystem ] . in and the functions @xmath37 , @xmath38 , are assumed sufficiently regular to make the systems well posed ( the question of the regularity of functions @xmath39 , @xmath38 , will play an important role in our approach and will be addressed in detail further below ) . without loss of generality , we will also assume that @xmath40 . system , or , is complemented with the initial condition , respectively , @xmath41^t$ ] and @xmath42 , @xmath43 . dynamical systems of the type or arise commonly as a result of various rigorous and empirical model - reduction strategies in diverse application areas such as fluid mechanics @xcite , thermodynamics @xcite and phase transitions @xcite . the main contribution of this work is development of a computational technique allowing one to reconstruct functions @xmath44 , @xmath45 and @xmath46 in in a very general form based on some measurements . the key idea is to formulate a least - squares minimization problem in which one of the functions @xmath39 , @xmath38 , is the control variable . then , a variational gradient - based approach can be employed to find the optimal solution in a suitable function space . in the present investigation , we will consider system as a reduced - order model of hydrodynamic instabilities in open shear flows obtained using a suitable galerkin projection to fix attention without losing generality . although this problem is very well - studied , our method provides a systematic refinement of the state - of - the - art mean - field model , which is the second main contribution of this study . details of this problem are introduced in the next section . in the following section we develop our model identification approach , whereas in section [ toc : results ] we present a number of computational results concerning identification of the model for the system considered in section [ toc : modelidentification ] . then , in section [ toc : discussion ] , we analyze the computational performance of our method and discuss the improvements it offers over the predictions of some standard models applied to the problem in question . summary and conclusions are deferred to section [ toc : conclusions ] , whereas in appendices [ toc : pod ] , [ toc : regularity ] and [ toc : perturbation ] we collect some technical results . in this section we define a model identification problem associated with the transient two - dimensional ( 2d ) cylinder wake which is a well - studied hydrodynamic instability @xcite . this flow is a representative example of phenomena characterized by the hopf bifurcation with an unstable fixed point and a stable limit cycle . additional examples in this category include the rossiter modes of the flow over a cavity @xcite and other shear flows @xcite . first , in section [ toc : wake ] , we describe the initial - boundary value problem for the infinite - dimensional navier - stokes equation which is a system of coupled pdes representing the conservation of mass and momentum in the motion of viscous incompressible fluid . then , in section [ toc : galerkinexpansion ] , a low - dimensional galerkin expansion is recalled which reduces the kinematic description down to three modes whose amplitudes serve as the state variables @xmath22 , @xmath23 and @xmath20 . it will be demonstrated that the system governing these variables satisfies in fact assumptions [ ass1 ] and is in the form . special forms of this system arising as models in numerous applications are discussed in section [ toc : meanfieldsystem ] , whereas in section [ toc : measurements ] we describe the measurements used as the basis for the reconstructions . the 2d flow around a circular cylinder is described here in the cartesian coordinate system @xmath47 . the origin @xmath48 coincides with the center of the cylinder , the @xmath49-coordinate points in streamwise direction , while @xmath50 represents the transverse coordinate . the velocity field @xmath51 has components @xmath52 and @xmath53 aligned with the @xmath49-axis and @xmath50-axis , respectively , whereas @xmath54 represents the pressure field . the oncoming flow velocity is denoted by @xmath55 and the cylinder diameter by @xmath56 . the newtonian fluid is characterized by uniform density @xmath57 and kinematic viscosity @xmath58 . the flow properties depend on the reynolds number @xmath59 . here , we consider the case of @xmath60 which is far above the critical reynolds number of @xmath61 characterizing the onset of vortex shedding ( i.e. , the hopf bifurcation ) @xcite , and far below the transitional reynolds number of 187 which marks the onset of three - dimensional instabilities @xcite . in the following , all quantities are assumed to be non - dimensionalized with @xmath55 , @xmath56 and @xmath57 . the flow is considered in a rectangular domain surrounding the cylinder @xmath62 and its evolution is described by the incompressible navier - stokes equation [ eqn : incompressibleflow ] @xmath63 the boundary conditions for the velocity comprise the no - slip condition at the cylinder , the free - stream condition at the inlet boundary , the free - slip condition at the lateral boundaries and the no - stress condition at the downstream outflow boundary . the initial condition for the velocity at @xmath64 is the unstable fixed point @xmath65 of navier - stokes equation perturbed by the real part of the corresponding most unstable stability eigenmode @xmath66 @xmath67 the field @xmath66 is normalized to have unit @xmath68 norm . the initial - boundary - value problem is numerically integrated with a finite - element method based on an unstructured grid and details are provided in @xcite . figure [ fig : flows ] depicts three flow snapshots corresponding to the initial condition , an intermediate transient state and the flow approaching the limit cycle . [ fig : flows ] \(a ) + ( initial condition ) , ( b ) @xmath69 ( intermediate transient state ) , and ( c ) @xmath70 ( periodic solution at the limit cycle ) . the domain shown represents the computational domain @xmath71 which also coincides with the region where the galerkin expansion is defined . the flow patterns are visualized using streamlines ( which are defined as the level sets of the streamfunction @xmath72 related to the velocity components via @xmath73 and @xmath74 ) . the cylinder is indicated by the black circle . , title="fig:",width=264 ] + ( b ) + ( initial condition ) , ( b ) @xmath69 ( intermediate transient state ) , and ( c ) @xmath70 ( periodic solution at the limit cycle ) . the domain shown represents the computational domain @xmath71 which also coincides with the region where the galerkin expansion is defined . the flow patterns are visualized using streamlines ( which are defined as the level sets of the streamfunction @xmath72 related to the velocity components via @xmath73 and @xmath74 ) . the cylinder is indicated by the black circle . , title="fig:",width=264 ] + ( c ) + ( initial condition ) , ( b ) @xmath69 ( intermediate transient state ) , and ( c ) @xmath70 ( periodic solution at the limit cycle ) . the domain shown represents the computational domain @xmath71 which also coincides with the region where the galerkin expansion is defined . the flow patterns are visualized using streamlines ( which are defined as the level sets of the streamfunction @xmath72 related to the velocity components via @xmath73 and @xmath74 ) . the cylinder is indicated by the black circle . , title="fig:",width=264 ] below we review approximate modelling approaches typically employed to obtain low - dimensional descriptions of the dynamics described by , cf . @xcite . the quantities corresponding to the limit cycle will be denoted with the superscript @xmath75 and in the present problem in which the limit cycle is stable we will therefore have @xmath76 . the transient wake is usually characterized by the superposition of a time - varying symmetric base flow @xmath77 with the length of the recirculating region decreasing with time as the vortex shedding develops and an antisymmetric oscillating field @xmath78 representing the vortex shedding . the change of the base flow is accurately captured by a single `` shift mode '' @xmath79 @xcite , while the oscillation field can be approximated by the first two proper orthogonal decomposition ( pod ) modes @xmath80 , @xmath81 , computed at the limit cycle @xcite . hereafter , the subscript @xmath82 will denote various quantities related to the shift mode @xmath79 . some information about the construction of orthogonal bases using the pod approach is presented in appendix [ toc : pod ] . the resulting truncated galerkin expansion thus takes the form [ eqn : meanfieldexpansion ] @xmath83 one typically ignores deformations of the oscillatory modes during the transient . these deformations do not significantly alter the model predictions and do not have any effect on the proposed model identification approach . on the other hand , inclusion of this effect in our model would significantly complicate the study and is outside the scope of this investigation . identifying @xmath84 and assuming phase - invariant behavior , we note that the evolution of the galerkin expansion coefficients is governed by a system in the form , provided that @xmath85 , corresponding to an unstable fixed point at the origin , and @xmath86 and @xmath87 , corresponding to a locally attracting limit cycle at @xmath88 , cf . assumption [ ass1](a ) . moreover , from the assumptions made in section [ toc : introduction ] , it follows that @xmath89 . introducing a number of further simplifications one arrives at two well - known reduced models , the mean - field model and the landau model . since they provide a point of reference for our test problems , we describe them briefly below . the mean - field model of oscillatory flow instabilities @xcite explicates the mechanism of amplitude saturation in the mean - field deformation . this deformation is quantified by the amplitude @xmath90 of the shift mode , cf . , which is slaved to the fluctuation level @xmath25 . for the soft ( supercritical ) hopf bifurcation , the resulting evolution equations read [ eqn : meanfieldsystem ] @xmath91 \ > r{(t ) } , \\ \label{eqn : meanfieldsystem_2 } \dot \theta{(t ) } & = & \omega_1 + \gamma_{\delta } a_{\delta}({t } ) , \\ \label{eqn : meanfieldsystem_3 } a_{\delta}({t } ) & = & \alpha_{\delta } \ > r^2{(t)}.\end{aligned}\ ] ] the ordinary differential equations correspond to the navier - stokes equation linearized around the time - varying base flow with @xmath90 as the order parameter . the third algebraic equation represents the reynolds equation linking the mean flow to the reynolds stress generated by the fluctuations described by the first equation . evidently , equations are a particular case of with @xmath92 , @xmath93 and @xmath94 . the parameters @xmath95 , @xmath96 , @xmath97 , @xmath98 , @xmath99 of may be derived from the galerkin approximation described in section [ toc : galerkinexpansion ] , or identified from the solutions of the navier - stokes equation via suitable fitting . the periodic solution of is given by [ eqn : meanfieldsystemsolution ] @xmath100 at the limit cycle the mean flow is predicted to have a vanishing growth rate @xmath44 . this marginal stability property of mean flows has been conjectured by malkus @xcite and is corroborated by the global stability analysis of the navier - stokes equation @xcite . derivation details and a further discussion of the properties of the mean - field model can be found in @xcite . the landau equation for the supercritical hopf bifurcation is a corollary to the mean - field model and constitutes a prototype evolution equation for self - amplified , amplitude - limited oscillations . it is obtained by substituting in : [ eqn : landauequation ] @xmath101 here , we require @xmath102 to ensure a stable limit cycle with a positive angular velocity in the @xmath21 plane , cf . assumption [ ass1](a ) , whereas @xmath103 may vanish , be positive or negative . evidently , equations correspond to with @xmath104 and @xmath105 . therefore , the landau equation is also easily recognized as a particular case of and , while the mean - field equation is an example of and . the dependency of the growth rate and frequency on the shift mode amplitude @xmath90 in may in principle be recovered from by inverting to give @xmath106 and substituting @xmath107 into . this inversion assumes a monotonous dependence of @xmath90 on @xmath25 which is observed in actual wake data . for completeness , we note that the least - order galerkin model introduced in @xcite includes a fast - dynamics equation for @xmath90 . this dynamic equation is well represented by an inertial manifold @xmath108 of the form or which ignores short transients . the landau equation may also be derived from the first principles , e.g. , using a center - manifold reduction method , see , e.g. , @xcite . the parameters may be identified from solutions of the navier - stokes equation with @xmath95 and @xmath96 obtained as the real and imaginary part of the most unstable eigenvalue associated with the hopf bifurcation , whereas the nonlinearity parameters @xmath109 and @xmath103 can be inferred from the post - transient amplitude and frequency , @xmath110 and @xmath111 , respectively . in section [ toc : results ] we will demonstrate using our proposed approach how the structure of landau model could be modified to better reproduce the actual behavior . the state in our approximate model is characterized by three time - dependent mode amplitudes : @xmath22 , @xmath23 and @xmath20 and as `` measurements '' we will consider the functions @xmath112 , @xmath113 and @xmath114 which are obtained for @xmath115 $ ] by solving initial - boundary - value problem - followed by projection , in terms of the inner product defining the pod analysis ( cf . appendix [ toc : pod ] ) , of the resulting time - dependent velocity field @xmath116 on the modes @xmath117 , @xmath118 and @xmath79 . the same procedure applies when the velocity field comes from time- and space - resolved measurements . in either case , determination of the shift mode @xmath79 requires access to the unstable equilibrium @xmath65 , cf . , which typically needs to be obtained numerically . the mean - field model may also be constructed directly from experimental measurements . let @xmath119 be , for instance , a pressure or hot - wire signal with a dominant harmonic component corresponding to the vortex shedding . let @xmath120 be the short - time mean value , e.g. , a one - period average , and @xmath121 be the fluctuation . next , let @xmath22 and @xmath23 be the local cosine and sine component of the fluctuation . these variables may be obtained from a hilbert transform or , more robustly , from a morlet transform of the data . finally , we identify @xmath122 , where the tunable constant shall ensure the correct fixed - point behavior , i.e. , @xmath123 when @xmath124 . then , @xmath22 , @xmath23 and @xmath90 may be approximated by the mean - field model . in this case , it is not required that the fixed point be actually reached for the identification of the model . we only need a transient with a range of @xmath25 values over which the functions @xmath39 are identified . in the next section we introduce a computational approach allowing one to optimally identify the functions @xmath125 , @xmath126 and @xmath127 in a suitable class , so that the predictions of system best match in the least - squares sense the data @xmath128 , @xmath129 and @xmath130 . we note that , alternatively , a system of evolution equations for @xmath22 , @xmath23 and @xmath20 may be obtained by substituting ansatz into momentum equation , and some comparisons between such an approach and the results obtained using the model identification method developed here will be drawn in section [ toc : discussion_physical ] . the task of identifying the functions @xmath39 , @xmath38 , such that the output of system matches certain `` measurements '' is an example of an inverse problem @xcite . what makes this problem somewhat different from typical inverse problems is that the functions sought have the form of `` constitutive relations '' , in the sense that they depend on the _ state variables _ ( i.e. , the dependent variables in the problem ) , rather than the independent variables . more specifically , in the problem considered here @xmath39 , @xmath38 , depend on @xmath131 as opposed to @xmath26 . non - parametric formulations of such inverse problems have received some attention in the context of systems described by pdes @xcite , but we are not aware of similar approaches applied to the state - space description of dynamical systems . we will look for functions @xmath39 , @xmath38 , as elements of the sobolev space @xmath132 of continuous functions with square - integrable gradients on @xmath133 which is the `` identifiability '' region defined in section [ toc : introduction ] , i.e. , the interval spanned by the state variable @xmath134 during the system evolution , see also @xcite . some additional remarks concerning the regularity of functions @xmath44 and @xmath45 are presented in appendix [ toc : regularity ] . in the reconstruction problems considered in this study , the boundary behavior of the reconstructed functions will have to be restricted in different ways . this reflects the fact that system with the reconstructed functions @xmath39 , @xmath38 , should exhibit the behavior dictated by assumption [ ass1](a ) in specific regions of the phase space , namely , at the equilibrium @xmath124 and at the limit cycle @xmath135 . there is some flexibility as regards possible choices and the boundary conditions we adopt will reproduce the behavior described by the mean - field model introduced in section [ toc : meanfieldsystem ] . more specifically , * at the origin @xmath124 * * the jacobian of the right - hand side ( rhs ) in equation should be given by @xmath136 which is a priori unknown and is to be determined as a part of the solution of the reconstruction problem ; on the other hand , the jacobian of the rhs in equation should vanish ; this is achieved when the derivatives of both functions @xmath44 and @xmath45 are set to zero at the origin , i.e. , @xmath137 * at the limit cycle @xmath138 * * the rhs of should vanish resulting in the vanishing of @xmath44 , i.e. , @xmath139 * * in regard to equation , we will prescribe a given slope @xmath140 of the rhs , i.e. , @xmath141 we note that , while the boundary behavior described by at the equilibrium and the limit cycle is the same as in the mean - field model ( cf . section [ toc : meanfieldsystem ] ) , the behavior of the constitutive relations @xmath44 and @xmath45 for intermediate values of the state magnitude @xmath142 can be arbitrary and will be determined using our optimal reconstruction procedure . no restrictions are placed on the boundary behavior of function @xmath46 , cf . . a convenient way to solve inverse problems is to formulate them as suitable optimization problems @xcite . for each of the functions @xmath39 , @xmath38 , we thus define the corresponding cost functional @xmath143 as [ eq : j ] @xmath144 ^ 2 \ , dt , \label{eq : j1 } \\ \j_2(g_2 ) & : = \frac{1}{2 } \int_0^t \left [ e^{\imath \theta(t ) } - e^{\imath \tilde{\theta}(t ) } \right]^2 \ , dt , \label{eq : j2 } \\ \j_3(g_3 ) & : = \frac{1}{2 } \int_0^t \left [ g_3(r(t ) ) - \ta_{\delta}(t ) \right]^2 \ , dt , \label{eq : j3 } \end{aligned}\ ] ] where @xmath145 , @xmath146 and @xmath114 are the `` measurements '' obtained as described in section [ toc : measurements ] . the length @xmath147 of the assimilation window will be chosen sufficiently long to allow the transient trajectory to settle on the limit cycle . in and the functions @xmath134 and @xmath148 are related to @xmath44 and @xmath45 via system . the optimal reconstructions @xmath149 , @xmath150 , and @xmath151 are defined as solutions of the following problems [ eq : p ] @xmath152 below we describe in detail a gradient - based approach to solution of problem @xmath153 . problem @xmath154 has a similar structure and the solution method is essentially the same with some modifications described hereafter . in problems @xmath155 and @xmath156 it is assumed that the functions , respectively , @xmath45 and @xmath44 are fixed . problem @xmath157 has a different structure warranting a separate solution approach which will be described further below . the minimizer @xmath149 is characterized by the first - order optimality condition @xcite requiring the vanishing of the gteaux differential + @xmath158 $ ] , i.e. , @xmath159 where @xmath160 is an arbitrary perturbation direction . the ( local ) minimizer can be computed with the following iterative procedure @xmath161 where @xmath162 represents the initial guess , @xmath163 denotes the iteration count and @xmath164 is the gradient of cost functional @xmath165 . the length @xmath166 of the step is determined by solving the following line minimization problem @xmath167 which can be done efficiently using standard techniques such as brent s method @xcite . for the sake of clarity , formulation represents the steepest - descent method , however , in practice one typically uses more advanced minimization techniques , such as the conjugate gradient method , or one of the quasi - newton techniques @xcite . evidently , the key element of minimization algorithm is the computation of the cost functional gradient @xmath168 . it ought to be emphasized that , while the governing system is finite - dimensional , the gradient @xmath168 is a function of the state magnitude @xmath25 and as such represents a continuous ( infinite - dimensional ) sensitivity of cost functional @xmath169 to the perturbations @xmath170 . the fact that the control variable @xmath44 , and hence also the gradient @xmath171 , are functions of the _ state _ variable , rather than the independent variable ( figure [ fig : vars]a ) , will result in cost functional gradients with structure rather different than encountered in typical optimization problems for differential equations ( see @xcite for some related questions arising in pde optimization problems ) . in order to identify an expression for the gradient @xmath171 , we proceed by computing the gteaux differential of the cost functional @xmath172 r'(g_1;g'_1)\ , dt = \int_0^t \frac{r - \tr}{r } \bxi^t\bxi ' \ , dt = \big\langle \nabla\j_1(g_1 ) , g'_1 \big\rangle_{\x(\i ) } , \label{eq : dj}\ ] ] where we used the identity @xmath173 and the last equality is a consequence of the riesz representation theorem @xcite with @xmath174 denoting an inner product in the hilbert space @xmath175 ( to be specified later ) of functions defined on @xmath133 . the perturbation variable @xmath176 is a solution of the following perturbation problem ( see appendix [ toc : perturbation ] for a derivation ) [ eq : dxi ] @xmath177 \bxi'(t ) + \bi \ , \bxi(t)\ , g'_1 \nonumber \\ & = : \ba(\bxi(t ) ) \ , \bxi ' + \bi \ , \bxi(t)\ , g'_1 , \label{eq : dxi1 } \\ \bxi'(0 ) & = 0 , \label{eq : dxi2}\end{aligned}\ ] ] where @xmath178^t$ ] , @xmath81 . we note that gteaux differential is not yet in the form consistent with the riesz representation , since the perturbation @xmath160 does not appear in it as a factor , but is hidden on the rhs in perturbation equation . a standard technique to convert gteaux differential to the riesz form is based on the _ adjoint _ variable @xmath179\rightarrow \rr^2 $ ] . taking the inner product ( in @xmath180 ) of @xmath181 with equation , integrating over @xmath182 $ ] and then integrating by parts we obtain @xmath183 \bxi ' - \bxi\ , g'_1\big\}\ , dt \\ = & \int_0^t \ , \left ( \bxi'\right)^t \big\{-\dot{\bxi}^ * - \left [ g_1(r ) \bi + \bi\ , \bxi\ , ( \bnabla g_1(r))^t + g_2(r ) \bj + \bj\ , \bxi\ , ( \bnabla g_2(r))^t \right]^t \bxi^ * \big\ } \ , dt \\ & + \big [ ( \bxi^*)^t \bxi'\big]_{t=0}^{t = t } - \int_0^t \ , \left ( \bxi^*\right)^t \ , \bi \ , \bxi(t)\ , g'_1 \ , dt . \end{split } \label{eq : id1}\ ] ] defining the _ adjoint _ system as [ eq : axi ] @xmath184^t \bxi^*(t ) + \frac{r - \tr}{r } \bxi , \label{eq : axi_1 } \\ \bxi^*(t ) & = 0 , \label{eq : axi_2}\end{aligned}\ ] ] we reduce relation to @xmath185 we note that , although @xmath160 already appears as a factor in expression , this expression is still not in the riesz form , since the integration is with respect to the time @xmath186 , whereas in the inner product @xmath174 defining the riesz representer integration is with respect to the measure @xmath187 defined on the interval @xmath133 ( connection between the different integration variables is illustrated schematically in figure [ fig : vars]b ) . the two variables are related via the following transformation @xmath188 where we used the identities @xmath189 and @xmath190 , cf . . denoting the trajectory in the state space @xmath191 } \ \bxi(t ) \in \rr^2 \right\}$ ] , and combining with we obtain the expression @xmath192 which is already in the required riesz form . in the line integral over the contour @xmath193 and the definite integral over the interval @xmath133 are equal , because for dynamical system points on the contour @xmath193 with the magnitude @xmath194 are unique , so that the map @xmath195 is one - to - one ( in more general situations when this is not the case , or when the reconstructed function depends on more than one state variable , e.g. , both @xmath196 and @xmath197 here , the change of variables needed to obtain the riesz form will be more complicated and one has to employ more general techniques such as those developed in @xcite ) . while this is not the gradient we will use in actual computations , we will first obtain an expression for the @xmath198 gradient which in the next subsection will be used as the basis for the calculation of gradients defined in the sobolev space @xmath132 . thus , setting @xmath199 in , we obtain from @xmath200 expression is validated computationally in section [ toc : validation ] . as regards problem @xmath156 , the optimality condition takes the form , cf . and , @xmath201 where we used the identity @xmath202 . following the same steps as described above , we obtain an expression for the cost functional gradient in the form , however , the adjoint system satisfied by @xmath203 has now a different source term on the rhs [ eq : axi2 ] @xmath184^t \bxi^*(t ) + \frac{\sin(\theta - \tilde{\theta})}{r^2 } \bj\ , \bxi(t ) , \label{eq : axi2_1 } \\ \bxi^*(t ) & = 0 . \label{eq : axi2_2}\end{aligned}\ ] ] in regard to problem @xmath204 , using change of variables , we can rewrite as @xmath205 ^ 2 \ , dr . \label{eq : j3b}\ ] ] thus , in optimization problem we look for a function @xmath206 which is as close as possible ( in a weighted @xmath198 topology ) to a given function @xmath207 , this problem , in fact , does not have a solution because of the density of the function space @xmath132 in @xmath208 , cf . however , it is possible ( and satisfactory from the application point of view ) to `` solve '' problem approximately by finding a @xmath209 such that @xmath210 is sufficiently small . such an approach is described in section [ toc : sobolev ] . in this section we describe how sobolev gradients @xmath211 , @xmath81 , used in minimization algorithm for problems @xmath155 and @xmath156 can be obtained from . in addition to enforcing smoothness of the reconstructed functions , this formulation allows us to impose the desired behavior at the endpoints of the interval @xmath133 , cf . - , via suitable boundary conditions . we begin by defining the @xmath212 inner product on @xmath133 as @xmath213 where @xmath214 is a parameter with the meaning of a `` length scale '' . it is well known @xcite that extraction of cost functional gradients in the space @xmath212 with the inner product defined as in can be regarded as low - pass filtering the @xmath198 gradients with the cut - off wavenumber given by @xmath215 . as regards the behavior of the gradients @xmath216 at the endpoints of the interval @xmath133 , we can require the vanishing of either the gradient itself or its derivative @xmath217 , and the boundary conditions we prescribe correspond to relations introduced as a part of the formulation of optimization problems , cf . assumption [ ass1](a ) . as regards the boundary data at @xmath135 ( i.e. , at the limit cycle ) , in problem @xmath156 we use @xmath218 which ensures that the property @xmath219 of the initial guess @xmath220 remains unchanged during iterations . identifying expression with inner product , cf . , integrating by parts and using the boundary conditions mentioned above we obtain the following elliptic boundary - value problem on @xmath133 defining the sobolev gradient @xmath216 [ eq : gradjh1 ] @xmath221 where the expression for @xmath222 is given in . as concerns problem @xmath204 , we propose to reconstruct @xmath209 directly ( i.e. , without iterations ) by solving the following problem [ eq : gradj3 ] @xmath223 which , except for the boundary condition at @xmath138 , has an identical structure as . the superscript in @xmath224 indicates dependence of the solution on the parameter @xmath225 . we note that as @xmath226 the left - hand side ( lhs ) in approaches the identity transformation which means that @xmath227 , so that also @xmath228 , as @xmath226 . since solutions of system are not defined for @xmath229 , we will obtain our approximate reconstruction as @xmath230 for some small value of @xmath225 . results concerning model identification for the system described in section [ toc : modelidentification ] are presented in section [ toc : results ] , whereas in sections [ toc : validation ] and [ toc : discussion_computational ] we analyze certain computational aspects of the method . in this section we present results concerning the solution of model identification problems @xmath155 , @xmath156 and @xmath204 , cf . , for the system introduced in section [ toc : modelidentification ] . motivated by practical considerations , we make the following in the solution of problems @xmath155 and @xmath156 we set in system [ eq : g1g2 ] @xmath231 [ ass2 ] which means that in the reconstruction of @xmath45 we use the best available estimate of @xmath44 obtained from the solution of problem @xmath155 . the choice of @xmath45 has no effect on the solution of problem @xmath155 and hence without loss of generality we can adopt . as the gradient descent algorithm in problems @xmath155 and @xmath156 we use the polak - ribiere version of the nonlinear conjugate gradient method @xcite in which the `` momentum '' term is reset to zero every 20 iterations and iterations are declared converged when @xmath232 , @xmath81 . in the system described in section [ toc : modelidentification ] the limit cycle is characterized by @xmath233 , whereas the length of the time window is chosen as @xmath234 which is long enough to allow the transient to settle on the limit cycle ( see figure [ fig : ra1a3]a below ) . we have successfully solved problems @xmath155 , @xmath156 and @xmath204 using different combinations of numerical parameters , and the parameters used to obtain the results presented in this section are summarized below . systems and were solved using matlab subroutine ode45 with an adaptive time - stepping . unless stated otherwise , the integrals defined on the interval @xmath182 $ ] , cf . , were discretized using @xmath235 equispaced points . the interval @xmath133 was discretized using @xmath236 equispaced points , and boundary - value problems and were approximated using the second - order finite differences . the length - scale parameter appearing in and was @xmath237 in problems @xmath155 and @xmath156 , and @xmath238 in problem @xmath204 . the initial condition @xmath239 for system and the initial guesses @xmath162 and @xmath220 for reconstruction algorithm must be chosen so that the magnitudes @xmath240 , @xmath115 $ ] , span the entire interval @xmath133 , as otherwise the sensitivities ( gradients ) can not be properly defined for _ all _ values of @xmath25 ( we refer the reader to @xcite for a discussion how this limitation can be overcome in some cases ) . since our goal is now to assess possible improvements to mean - field model , we will use it with the coefficients determined as discussed in section [ toc : meanfieldsystem ] as the initial guess for the reconstructions , so that @xmath241 respectively , for problems @xmath155 and @xmath156 . it is clear that initial guesses satisfy properties with @xmath242 . as the initial condition @xmath239 for we used a small perturbation around the fixed point at the origin . our main results are presented in figure [ fig : g123 ] where we show the optimal reconstructions @xmath243 , @xmath38 , and compare them against the left - hand sides of equations , all shown as functions of the state magnitude @xmath25 . for completeness , the lhs of equations are shown as functions of time @xmath27 $ ] in figure [ fig : g12_vs_t ] ( in the case of @xmath149 , cf . figures [ fig : g123]a and [ fig : g12_vs_t ] , the lhs of equation is additionally divided into @xmath25 ) . in figures [ fig : g123]a , b we also indicate the rhs of mean - field model which were used as the initial guesses for the reconstructions . we see in figures [ fig : g123 ] that , as expected , the reconstructed constitutive relations @xmath243 , @xmath38 , smoothly approximate the left - hand sides of the corresponding equations evaluated using the measurements . considered as functions of @xmath25 , these left - hand sides are multi - valued which is a consequence of the fact that , due to the oscillations of the measurement data at the limit cycle ( see figure [ fig : ra1a3]a below ) , the map @xmath244 is not one - to - one . in figures [ fig : g123]a , b we observe systematic deviations of the optimal reconstructions @xmath149 and @xmath150 from the corresponding functions in mean - field model . in addition , based on the reconstructions @xmath149 and @xmath150 we can obtain estimates of two important quantities , namely , the growth rate of the instability at the origin given by @xmath245_{r=0 } = \hat{g}_1(0 ) = 0.1576 $ ] , and the oscillation frequency at the limit cycle given by @xmath246 . these numbers should be compared with , respectively , 0.151 and 1.036 obtained as discussed in section [ toc : meanfieldsystem ] and used in mean - field model . finally , in figure [ fig : ra1a3 ] we compare the outputs from system obtained using the mean - field model and the reconstructions @xmath243 , @xmath38 , against the corresponding measured quantities ( as regards the time - history of the state variables , we do not show @xmath247 , as it has qualitatively very similar behavior to @xmath248 already shown in figure [ fig : ra1a3]b ) . in figures [ fig : ra1a3]a , b ( see , in particular , the insets ) we note that the evolution of @xmath134 and @xmath248 obtained using the optimal reconstructions @xmath149 and @xmath150 is much closer to the measured quantities than the evolutions computed using mean - field model . we remark , however , that the measurements @xmath249 shown in figure [ fig : ra1a3]a reveal some high - frequency oscillations which are not captured by the trajectory @xmath134 obtained using the optimal reconstruction @xmath149 . these oscillations reflect a phase dependence in the behavior of the solutions of the original navier - stokes equation , an effect which is by construction excluded from ansatz , cf . assumption [ ass1](a ) . as a consequence , @xmath149 can convey phase - averaged information only . we will return to this problem again in section [ toc : discussion_physical ] . as regards the results shown in figure [ fig : ra1a3]c , we see that , while the optimal reconstruction @xmath250 is quite smooth ( figure [ fig : g123]c ) , the quantity @xmath251 exhibits oscillations absent in the original measurement data @xmath252 . this effect as well is a consequence of the lack of phase - dependence in ansatz and . in this section we analyze a number of computational and physical modelling aspects of the proposed approach which can be important in applications . we begin by examining the accuracy of the cost functional gradients in section [ toc : validation ] , followed by a study of the robustness of iterations in section [ toc : discussion_computational ] and conclude with some insights about physical modelling in section [ toc : discussion_physical ] . a key element of optimization algorithm are the cost functional gradients and a standard approach to their validation consists in computing the directional gteaux differential @xmath253 , @xmath81 , for some arbitrary perturbations @xmath254 in two different ways , namely , using a finite - difference approximation ( with step size @xmath255 ) and using the inner product of the adjoint - based gradient with the perturbation @xmath254 , namely riesz representation , and then examining the ratio of the two quantities , i.e. , @xmath256 } { \int_{\i } \bnabla^{l_2 } \j_i ( r ) \ , g'_i(r ) \ , dr } , \quad i=1,2 , \label{eq : kappa}\ ] ] for a range of values of @xmath255 . if the gradient @xmath257 is computed correctly , then for intermediate values of @xmath255 , @xmath258 will be close to the unity . remarkably , this behavior can be observed in figures [ fig : kappa]a , b corresponding to problems @xmath155 and @xmath156 over a range of @xmath255 spanning about 8 orders of magnitude . the quantity shown in figures [ fig : kappa]a , b is latexmath:[$\log| \kappa_i(\epsilon ) - 1 digits to which the two ways to evaluate @xmath253 in agree . furthermore , we also observe that refining the resolution @xmath260 of the time interval @xmath182 $ ] yields values of @xmath258 closer to the unity . the reason is that in the `` optimize - then - discretize '' paradigm adopted here such refinement of the discretization leads to a better approximation of the continuous gradient . as can be expected , the quantities @xmath258 deviate from the unity for very small values of @xmath255 , which is due to the subtractive cancellation ( round - off ) errors in finite - differencing , and also for large values of @xmath255 , which is due to the truncation errors , both of which are well - known effects . in this section we focus on the effect that the choice of initial guess @xmath261 , @xmath81 , has on the reconstructed functions @xmath149 and @xmath150 . we note that , given the nonlinearity of system , optimization problems @xmath155 and @xmath156 may be nonconvex and optimality conditions and characterize minimizers which are only _ local_. thus , different initial guesses may in principle give rise to different reconstructions and convergence to a global minimum can not be a priori assured . we investigate this issue in figures [ fig : gig]a and [ fig : gig]b where we show the reconstructions obtained , respectively , in problems @xmath155 and @xmath156 using different initial guesses generally much worse than mean - field model used in section [ toc : results ] . as regards problem @xmath155 , we note in figure [ fig : gig]a that accurate reconstructions are obtained using even relatively poor initial guesses @xmath162 . on the other hand , in figure [ fig : gig]b we see that in problem @xmath156 the reconstruction fails for a less accurate initial guess @xmath220 . these two examples are representative of the behavior we generally observed in our calculations and we conclude that problem @xmath155 appears more robust with respect to the choice of the initial guess than problem @xmath156 . we also noted that in both problems convergence tends to be more sensitive to the values assumed by the initial guesses @xmath162 and @xmath220 at @xmath124 and @xmath135 than to their behavior for intermediate values of @xmath25 . results from figure [ fig : gig ] are corroborated by the corresponding histories of the cost functionals in figures [ fig : jig]a and [ fig : jig]b . we note that in the case of the poorest initial guess in problem @xmath156 , cost functional @xmath262 reveals hardly any decrease with the iterations at all . while in the cases of successful reconstructions the cost functionals @xmath263 and @xmath262 drop over several orders of magnitude , they never attain values lower than @xmath264 . this is a consequence of the phase - dependent behavior of the measurements which can not be resolved using ansatz in the form , cf . assumption [ ass1](a ) , see also the inset in figure [ fig : ra1a3]a . inverse problems of the type considered here often tend to be ill - posed , in the sense that small perturbations to the data , for example due to noise , may result in significant changes in the computed solution . suitable regularization , for instance , using tikhonov s technique @xcite , may be required to stabilize the solution procedure in such situations . to focus attention in the present study , we concentrated on the structure of the gradients and did not investigate the effect of noise on the reconstructions , hence such regularization was not necessary . we refer the reader to @xcite for a thorough analysis of regularization applied to a related reconstruction problem . in this section we propose some physical interpretation of the numerical reconstruction results from section [ toc : results ] . we note in figures [ fig : g123 ] and [ fig : ra1a3 ] that the identified phase - invariant oscillation model is in fact in remarkably good agreement with the data obtained from the solution of the navier - stokes equation . a small difference between the model and the data is visible as wiggles due to the second harmonic present in the measurements which violates the phase - invariance assumed in our model ansatz , cf . assumption [ ass1](a ) . this difference can be easily removed by a simple pre - processing of the measurement data . referring to appendix [ toc : pod ] , we note that the pod eigenvalue @xmath265 , representing the variance of @xmath22 , is larger than the eigenvalue @xmath266 which represents the variance of @xmath23 . the following rescaling transformation [ eqn : a1a2rescaling ] @xmath267 ensures equipartition of energy in the new variables @xmath268 and @xmath269 while conserving the total energy in both modes . this rescaling effectively removes the second harmonics from @xmath270 , @xmath81 , which could be used as new inputs for the reconstruction . however , we refrained from applying in the computations reported in section [ toc : results ] in order to show the power of the proposed identification method to deal with data which can not be perfectly matched by the model . another observation concerning figure [ fig : g123 ] is the significant deviation of reconstructed functions @xmath149 and @xmath150 from the parabolic mean - field relations . evidently , higher - order corrections , such as @xmath271 , @xmath272 , etc . , are required for a better agreement between the identified propagators @xmath273 , @xmath81 , and the expansions used in the mean - field model . the information which higher - order terms ought to be included in the model as opposed to an a priori fixed polynomial expansion used typically in model identification is therefore the unique advantage of the proposed identification strategy . we note that odd powers of @xmath25 can be excluded by phase - invariance considerations . in addition , with our reconstruction method we were able to determine more accurate values of the instability growth rate at the origin and the oscillation frequency at the limit cycle than used in mean - field model . we stress that in fact such seemingly insignificant modifications of the structure of the reduced - order model may already affect its utility for various control applications . these results also shed light on the validity of mean - field model . initially , the mean - field theory @xcite was derived to be valid near the onset of the oscillation only . we probed the applicability of this model by applying it at a reynolds number @xmath274 which is more than twice the critical value of @xmath61 . hence , the deviation of @xmath149 and @xmath150 from does not invalidate the mean - field theory . one reason for this deviation is the change of the structure of the vortex street during the transient . the optimal oscillatory modes deform from the stability eigenmodes into the pod modes while the fluctuation center moves upstream and the frequency and wavenumber increase @xcite . similarly , the mean - field correction ( the shift mode , cf . ) changes during the transient @xcite which has a noticeable effect on the mean - field model @xcite . finally , the results concerning the identified descriptor system are also relevant to the empirical 9-dimensional galerkin model accounting for the base - flow variation and for the first four harmonics @xcite . the initial exponential growth of the first harmonic is limited by the base - flow variation ( which reduces the production of fluctuation energy ) and , to a lesser extent , by the energy transfer from the first into higher harmonics . the energy transfer may be accounted for by an energy - dependent eddy viscosity in the mean - field system . under certain assumptions ( see , e.g.@xcite ) , a generalized landau equation @xmath275 can be derived , where @xmath95 denotes the growth rate near the fixed point @xmath124 and @xmath109 , @xmath103 characterize the damping from the @xmath276-th and from higher harmonics , respectively . by carefully comparing the 9-dimensional galerkin model with the identified phase - invariant system it may be therefore possible to determine the values of @xmath109 and @xmath103 , or even to correct the powers of the new terms in . a complete derivation and an in - depth discussion of this problem is outside the scope of the present study . we have proposed and validated a novel method for model identification which is an adaptation of an approach already used in the context of systems described by pdes @xcite . as indicated in figure [ fig : f ] , we depart from the traditional approach of ( 1 ) characterizing the propagator of the dynamical system in a parameter space and then ( 2 ) performing a parameter identification . thus , arbitrary polynomial expansions of the propagator may be performed a posteriori ( following the solution of the optimal reconstruction problem ) at a practically vanishing cost . in addition , the performance of parametric models may easily be assessed and , if necessary , improved by introducing additional terms motivated by the form of the reconstructed constitutive relation . the method is applied to a three - dimensional descriptor system with three a priori undetermined relations describing the fluctuation growth , frequency and mean - field correction as functions of the fluctuation energy . as a benchmark problem we chose the onset of laminar von krmn vortex shedding behind a circular cylinder . results of a direct numerical simulation are transcribed into the mode amplitudes of a minimal 3-state galerkin model @xcite which are then captured with remarkable accuracy by our identified descriptor system . the form of the reconstructed system is marked by a noticeable departure from the mean - field models and may therefore guide the refinement of the latter by inclusion of higher - order terms . we emphasize that the usefulness of reduced - order models for flow control applications may in fact depend on such differences . as regards future research directions , while the present results offer a proof of the concept for the proposed approach based on a rather well - understood example , the key question is extension of this method to the identification of models with more complicated structure featuring , for example , multiple time scales , state space of a higher dimension , coexistence of several oscillation frequencies , non - trivial phase dependence and higher - dimensional inertial manifolds . as regards the first issue , one can consider a modification of our model problem with assumption [ ass1](b ) revised to allow @xmath277 to be a `` fast '' variable . while in such setting our computational approach would formally remain unchanged ( except that problem @xmath204 would be replaced with a problem similar to @xmath155 or @xmath156 ) , it is interesting how it would actually perform in practice . dealing with some of the other aspects will require formulation of the reconstruction problems in terms of propagator functions depending on more than just one state variable ( @xmath25 in the examples considered in the present study ) . this will , in turn , lead to a number of interesting questions at the level of numerical analysis and scientific computing related to the evaluation of the cost functional gradients . an emerging application which involves some of the aforementioned extensions is related to the question of optimal parametrization of subgrid turbulence representations which is an important open problem in theoretical fluid mechanics @xcite . in the context of galerkin reduced - order models , it may take the form of an additional dissipative term with the magnitude proportional to an `` eddy viscosity '' @xmath278 @xmath279 where @xmath280 represents the stabilizing viscous term of the navier - stokes equation in the galerkin system . in general , this term can be proven to be energy dissipative for all orthonormal systems of modes and a large class of boundary conditions . for some analytical modes , e.g. , the stokes modes , it can be shown that matrix @xmath281 is diagonal and negative - definite . there is abundant evidence @xcite that nonlinear closure strategies perform better as regards stabilization of system . assuming @xmath282 gives rise to an identification problem analogous to @xmath155 and @xmath156 , and one can use the algorithm described in section [ toc : computation ] to determine optimal closure strategies leading to the best possible reconstruction of the available data . preliminary identification results already obtained based on a reduced - order model with dimension @xmath283 applied to a complex mixing - layer flow are quite encouraging and reveal some nontrivial physical insights . they will be reported in the near future upon completion of the study . we also remark that a surprisingly large set of modelling and control problems can be cast in a similar form of function identification of a descriptor system @xmath284 for reasons of simplicity , let us assume that function @xmath11 is known and that function @xmath285 needs to be determined . if @xmath286 characterizes the slow modes , then @xmath287 represents the inertial manifold to be identified from a given system trajectory @xcite . if @xmath286 represents high - frequency components , such as the parameters of a subgrid turbulence representation described above , then their functional dependence on the state variable @xmath288 may also be inferred with our approach . on the other hand , if @xmath286 denotes the actuation amplitudes , as in numerous wake flow stabilization studies @xcite , then @xmath289 represents a full - state feedback control law . in principle , this control law may as well be identified from desired trajectories @xmath14 . the authors acknowledge the funding and excellent working conditions of the chair of excellence closed - loop control of turbulent shear flows using reduced - order models ( tucorom ) of the french agence nationale de la recherche ( anr ) and hosted by institute pprime . the first author is , in particular , grateful for the hospitality of this chair of excellence at institute pprime where most of this work was carried out . we also thank the ambrosys ltd . society for complex systems management and the bernd noack cybernetics foundation for additional support . we appreciate valuable stimulating discussions with markus abel , robert niven , michael schlegel and gilead tadmor as well as the local tucorom team : jean - paul bonnet , laurent cordier , thomas duriez , peter jordan , vladimir parezanovic and andreas spohn . in this appendix we describe the proper orthogonal decomposition ( pod ) employed in section [ toc : galerkinexpansion ] to construct a low dimensional model from the simulation data . it is closely related to other techniques of data analysis known as the principal component analysis or , in the discrete setting , the singular value decomposition . the starting point are snapshots of the velocity field @xmath290 , @xmath291 , @xmath292 , cf . . these snapshots are sampled at times @xmath293 uniformly spaced over one period of oscillation and form a statistically representative ensemble for the considered first and second moments . the goal is to construct a least order galerkin expansion @xmath294 with base mode @xmath295 and @xmath296 space - dependent expansion modes ( basis functions ) @xmath297 with the corresponding mode amplitudes @xmath298 which will result in a minimum - norm average residual @xmath299 of the snapshot ensemble ( see , e.g. , @xcite ) . the base mode is necessary so that galerkin expansion satisfies inhomogeneous boundary conditions for arbitrary mode amplitudes . for instance , expansion captures the prescribed oncoming flow velocity regardless of the values of @xmath300 . the galerkin expansion and its residual are embedded in the hilbert space @xmath301 of square - integrable vector fields . the inner product of two elements @xmath302 is defined by @xmath303 where @xmath304 denotes the standard euclidean inner product and @xmath305 an infinitesimal volume element of the domain @xmath71 . the associated norm thus is @xmath306 we search for empirical modes @xmath80 , @xmath307 , where @xmath308 , which minimize the average residual of the galerkin expansion of the snapshots @xmath309 with optimal mode amplitudes @xmath310 in the sense of the @xmath198 norm , i.e. , @xmath311 this problem is solved by the snapshot pod @xcite and the base mode is the mean of the snapshots @xmath312 the pod modes arise from the correlation matrix @xmath313 of the snapshot fluctuations @xmath314 we note that @xmath315 is a positive semi - definite grammian matrix . hence , the eigenvalue problem @xmath316 yields an orthonormal set of real eigenvectors @xmath317^t $ ] , @xmath318 , with non - negative eigenvalues which can be ordered as @xmath319 the last equality arises from the fact that @xmath320 vectors span a subspace of maximum dimension @xmath321 . hence , @xmath320 snapshots define only @xmath321 pod modes and the corresponding amplitudes . these are given by @xmath322 the pod modes form an orthonormal basis in @xmath68 , @xmath323 , @xmath324 , while their amplitudes have vanishing means @xmath325 , @xmath326 , and diagonal second moments @xmath327 , @xmath324 . the average is to be understood in terms of the snapshot ensemble ( see , e.g. , ) . the eigenvalue @xmath328 can be interpreted as twice the fluctuation energy contained the @xmath329-th mode . by the one - dimensional embedding result @xmath330 @xcite , we note that the reconstructed function will be hlder - continuous with @xmath331 . thus , it will not meet the assumptions of the picard - lindelf theorem @xcite , and in principle will ensure existence only , without uniqueness , of solutions of equation . in order to ensure the lipschitz - continuity ( @xmath332 ) of @xmath44 , which would also guarantee uniqueness of solutions of , we would need to reconstruct @xmath44 as an element of sobolev space @xmath333 , because then @xmath334 . while there are no fundamental difficulties here ( we would need to replace inner product with the corresponding definition in @xmath333 ) , we will refrain from this in the actual computations in section [ toc : results ] in order to keep the approach as simple as possible . nevertheless , in all problems we treated with the proposed approach the reconstructed functions possessed the lipschitz regularity which was verified a posteriori by performing suitable grid - refinement studies . in this appendix we present a derivation of perturbation equation . obtaining this equation is made somewhat more involved by the fact that the perturbation variable @xmath44 is itself a function of the state magnitude , i.e. , @xmath335 . we assume here that only @xmath44 is perturbed while @xmath45 remains fixed , with the opposite case leading to essentially the same calculations . by substituting , respectively , @xmath336 and @xmath337 into equation , we obtain @xmath338 and @xmath339 , where @xmath340 and @xmath341 are the corresponding solutions and @xmath342 , @xmath343 . taking the difference of these two equations and defining @xmath344 . we obtain @xmath345 where we also set @xmath346 in the first term on the rhs . as regards the terms denoted @xmath347 in , they are transformed as follows using the fundamental theorem of calculus for line integrals and the change of variables @xmath348 for @xmath349 $ ] @xmath350 where we also denoted @xmath351 . the integrand expression on the rhs in is then expanded in the taylor series around @xmath352 @xmath353 where the component notation was used for clarity with @xmath354 , @xmath355 and @xmath356 denoting the components of vectors @xmath357 , @xmath358 and @xmath176 . plugging expansion into expression we obtain @xmath359 \ , \bxi ' + \o\left ( |\bxi'|^2 \right ) . \end{aligned } \label{eq : a2}\ ] ] noting that term @xmath360 in transforms in an analogous way to @xmath347 in , using these results in , assuming smallness of @xmath176 and dropping terms of order quadratic and higher we finally arrive at perturbation equation . j. watson . on the non - linear mechanics of wave disturbances in stable and unstable parallel flows . the development of a solution for plane poiseuille flow and for plane couette flow . , 9:371389 , 1960 . m. morzyski , w. stankiewicz , b. r. noack , f. thiele , and g. tadmor . generalized mean - field model for flow control using continuous mode interpolation . in _ 3rd aiaa flow control conference _ , san francisco , ca , usa , 5 - 8 june 2006 , 2006 . invited aiaa - paper 2006 - 3488 . j. gerhard , m. pastoor , r. king , b. r. noack , a. dillmann , m. morzyski , and g. tadmor . model - based control of vortex shedding using low - dimensional galerkin models . in _ 33rd aiaa fluids conference and exhibit _ , orlando , florida , usa , june 2326 , 2003 , 2003 . paper 2003 - 4262 .

we investigate the migration of low - mass planets ( and ) in accretion discs threaded with a magnetic field using 2d mhd code in polar coordinates . we observed that , in the case of a strong azimuthal magnetic field where the plasma parameter is , density waves at the magnetic resonances exert a positive torque on the planet and may slow down or reverse its migration . however , when the magnetic field is weaker ( i.e. , the plasma parameter is relatively large ) , then non - axisymmetric density waves excited by the planet lead to growth of the radial component of the field and , subsequently , to development of the magneto - rotational instability , such that the disc becomes turbulent . migration in a turbulent disc is stochastic , and the migration direction may change as such . to understand migration in a turbulent disc , both the interaction between a planet and individual turbulent cells , as well as the interaction between a planet and ordered density waves , have been investigated . = 1 accretion , accretion discs magnetic fields mhd waves planets and satellites : dynamical evolution and stability planet - disc interactions

1508.03241

the migration of protoplanets embedded in accretion discs has been extensively studied in recent years ( see , e.g. , review by @xcite ) . a low - mass protoplanet excites density waves in the disc at the lindblad resonances and migrates toward the star if no other torques are exerted ( e.g. , @xcite ) . however , a corotation torque originating near the planet can slow or even reverse the migration ( e.g. , @xcite ) . another possible cause of slowed or reversed migration is the excitation of magnetic resonances in the disc when a relatively strong ( @xmath4 , where @xmath5 and @xmath6 are the pressure and magnetic field in the disc , respectively ] ) azimuthal magnetic field is present @xcite . these magnetic resonances can halt or reverse the migration of the planet if the magnetic field is sufficiently strong and it has a steep gradient toward the star the action of this mechanism has been shown in numerical simulations by @xcite . on the other hand , accretion discs threaded with a weak magnetic field are prone to the magneto - rotational instability ( mri ; @xcite ) . numerical simulations of planet migration in a turbulent disc have shown that the semimajor axis of the planet changes stochastically due to the planet s interaction with the turbulent cells in the disc . the migration rate can increase or decrease , and the direction of the migration can reverse @xcite . an initial goal of this paper was to more deeply investigate the effect of magnetic resonances on the migration of a low - mass embedded planet following the methods used by @xcite , albeit that we explored a larger parameter space with regards to the mass of the planet and the density distribution in the disc , as well as the magnetic field strength and its distribution in the disc . while exploring this expanded parameter space , we noticed the development of mri - driven turbulence in the disc in many cases . in particular , when the magnetic field is weak , non - axisymmetric density waves excited by the planet lead to perturbations in the initially - azimuthal magnetic field and , subsequently , to growth of the radial component of the magnetic field near the planet . the azimuthal field grows with time due to the stretching of the radial field lines by the differential rotation of the keplerian disc . as a result , an mri - type instability develops and propagates to larger distances from the planet , and the disc becomes turbulent . the planet interacts with the turbulent cells , and the torque associated with this interaction is much larger than the torque seen prior to the development of turbulence ( i.e. , the torque due to magnetic resonances ) . therefore , we also studied the parameter space describing the transition from a laminar to a turbulent disc , as well as the migration of a planet in a turbulent disc in detail . we first describe how the migration rate and direction of a low - mass planet are affected by the planet mass , as well as the distribution of surface density and magnetic field in the laminar disc . we also show how these parameters alter the time at which the onset of turbulence occurs in magnetized discs . finally , to better understand the interaction between the planet and turbulent cells in the disc , we study the torque on the planet by ( 1 ) an ordered , low - amplitude density wave generated at the inner boundary and ( 2 ) a high - amplitude wave that is excited in a turbulent mhd disc . the plan for this paper is as follows . in [ sec : theoretical background ] we overview the theory of the different sources of torque on the planet . in [ sec : model ] we describe our physical model and numerical setup . in [ sec : parameter space ] we describe our main parameters . we describe our test simulations in the hydrodynamical disc in [ sec : migration hydro ] and simulations in a laminar magnetic disc in [ sec : magres migration ] . migration in turbulent discs is shown in [ sec : migration turbulent ] , and migration under the influence of density waves is shown in [ sec : migration in waves ] . we conclude in [ sec : conclusions ] . in this section , we discuss the causes of inward and outward migration in hydrodynamic and mhd discs . in the absence of a magnetic field in the disc , the migration of the planet is determined by the lindblad and corotation torques @xcite . the lindblad torque is generated when the planet excites @xmath7-armed waves in the disc with `` orbital '' frequencies @xmath8 . the dispersion relation for these waves in a low - temperature , low - mass disc is @xmath9 where @xmath10 is the epicyclic frequency in a keplerian disc . substituting @xmath11 , the locations of the lindblad resonances are @xmath12 for each value of @xmath7 ( except @xmath13 ) , there is one resonance located closer to ( farther from ) the star than the planet called the inner ( outer ) lindblad resonance . the inner lindblad resonances exert a positive torque on the planet and `` push '' the planet outward , while the outer lindblad resonances have the opposite effect . the total lindblad torque ( also known as the `` differential lindblad torque '' ) determines the migration direction in the absence of other torques . the differential lindblad torque is negative , resulting in overall inward migration of the planet if there are no other sources of torque . however , there is typically a corotation torque exerted on the planet as well . the physics of the corotation torque and its effect on planet migration has been studied by number of authors ( e.g. , @xcite ) . in an isothermal disc , both the magnitude and sign of the corotation torque depend on the gradient of the surface density at the corotation radius . in strongly magnetized laminar discs , magnetic waves can be excited , and torques associated with the magnetic waves can affect a planet s migration . ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite investigated the propagation of waves in a magnetized disc in which the magnetic field is purely azimuthal and found that there are two singular radii at which the frequency perturbation in a frame rotating with the fluid matches that of a slow mhd wave propagating along the field lines ; these radii define the locations of the magnetic resonances . the inner and outer magnetic resonances are denoted similarly to lindblad resonances , with the inner magnetic resonances @xmath14 , and the outer magnetic resonances @xmath15 . @xcite derived the following dispersion relation : @xmath16 here , @xmath17 is the alfvn speed , given by @xcite as @xmath18 , where @xmath19 is the vertically - integrated square of the magnetic field . the sound speed @xmath20 , where @xmath21 . @xcite determines the strength of the magnetic field via @xmath22 , which is evaluated at the location of the planet .. ] the locations of the resonances when the disc is keplerian ( @xmath23 ) and thin ( @xmath24 ) are given by @xmath25 where the thickness of the disc @xmath26 and the plasma parameter @xmath27 are evaluated at @xmath28 . as the field becomes weaker ( i.e. , @xmath29 ) , the magnetic resonances converge toward the corotation radius . we suggest that @xmath30 and find the position of the inner and outer magnetic resonances to be @xmath31 @xcite showed that the waves associated with the magnetic resonances can exert a positive torque on the planet that is larger in magnitude than the differential lindblad torque if the magnetic field increases toward the star . @xcite performed simulations of a planet s migration in a magnetized disc and found that the magnetic resonances can slow or stop the migration of a planet . in particular , they investigated the migration of a @xmath0 planet embedded in a disc that is threaded with an azimuthal magnetic field with @xmath32 with @xmath33 initially . they took an initial value of @xmath34 at the planet s location and found that , depending on the steepness of the magnetic field distribution ( defined by @xmath35 ) , the planet s migration slowed or reversed in direction . @xcite studied the effects of an mhd corotation torque on a low - mass planet in a 2d laminar disc with a weak azimuthal field threading the disc . the field was not strong enough to generate an appreciable torque from magnetic resonances , and it was not strong enough to dominate the hydrodynamic corotation torques , but a `` torque excess '' attributed to the presence of the magnetic field was found . the mri arises under conditions in which a weak magnetic field threads the disc and the radial component of the field can be stretched and enhanced by the differential rotation in the disc . such a seed field can either be an axial field , perpendicular to the disc @xcite , or an azimuthal field @xcite . below , we briefly summarize the conditions for the onset of the mri instability in these two cases . _ axial seed magnetic field . _ @xcite considered the case in which an axial magnetic field , @xmath36 , threads a keplerian disc that rotates with an angular speed @xmath37 . for axisymmetric perturbations of the disc , for which @xmath38 $ ] and @xmath39 $ ] , and for perturbations proportional to @xmath40 , one finds the dispersion relation @xmath41^{1/2},\ ] ] where @xmath42 is the alfvn velocity , and @xmath43^{1/2}\ ] ] is the radial epicyclic frequency of the disc . in order for the perturbation to fit within the vertical extent of the disc , one needs @xmath44 , where @xmath45 is the half - thickness of the disc and @xmath46 is the isothermal sound speed in the disc . for most conditions , the disc is thin , with @xmath47 or @xmath48 . evidently , instability can occur if @xmath49 , which happens if @xmath50 . for a keplerian disc , this corresponds to @xmath51 . therefore , the above - mentioned condition that @xmath52 implies that instability occurs only for @xmath53 . _ toroidal seed magnetic field . _ @xcite studied the linear mhd stability / instability of a thin keplerian disc with a toroidal magnetic field @xmath54 . this case is more complicated than the case of a vertical field @xcite . the complication in the toroidal field case results from the presence of both the mri instability and the buoyancy instability of the toroidal field @xcite . the buoyancy instability is triggered by an azimuthally - dependent , radially - localized vertical displacement in the plasma and the toroidal field @xcite . these authors find that the mri instability in a thin disc with an embedded toroidal magnetic field shows up in localized perturbations ( i.e. , whose wavelengths are small compared with @xmath55 ) , @xmath56 , under the conditions @xmath57 and @xmath58 , where @xmath59 . this is the same as the condition for the mri instability in a disc threaded by a vertical field with @xmath60 and @xmath61 . therefore , perturbations of the azimuthal field may also lead to mri turbulence . work done by @xcite and @xcite showed the existence of additional mhd corotation torques in mri - turbulent discs using 3d mhd simulations . furthermore , @xcite investigated planet migration in an mri - turbulent disc and observed stochastic migration . however , the origin of the torque on the planet in a turbulent disc is not well studied . we utilize the mhd equations to numerically evaluate the perturbative effect of the planet on the disc : 1 . continuity equation ( conservation of mass ) @xmath62 where @xmath63 is the surface density ( with @xmath64 the volume density ) , and @xmath65 and @xmath66 are the radial and azimuthal velocities , respectively . 2 . radial equation of motion ( conservation of momentum ) @xmath67 \\ \nonumber & + & \frac{1}{r}\frac{\partial}{\partial\varphi}\left(\sigma v_{r}v_{\varphi}-\frac{\psi_{r\varphi}}{4\pi}\right ) \\ & = & \frac{\pi}{r } + \frac{\psi_{rr } + \psi_{\varphi\varphi}}{8\pi r } - \sigma\frac{gm_{\star}}{r^{2 } } + \sigma w_{r } , \label{eqn : radial_motion } \end{aligned}\ ] ] where @xmath68 is the surface pressure ( with @xmath5 the volume pressure ) , @xmath69 is the radial force exerted on the disc by the planet ( per unit area of the disc ) , and @xmath70 , @xmath71 , and @xmath72 are magnetic surface variables that are discussed in more detail in [ subsec : mag_variables ] . azimuthal equation of motion ( conservation of angular momentum ) @xmath73 \\ \nonumber & + & \frac{1}{r}\frac{\partial}{\partial\varphi } \left(\sigma v_{\varphi}^{2 } + \pi + \frac{\psi_{rr}+\psi_{\varphi\varphi}}{8\pi } - \frac{\psi_{\varphi\varphi}}{4\pi}\right ) \\ & = & \sigma w_{\varphi } , \label{eqn : az_motion } \end{aligned}\ ] ] where @xmath74 is the azimuthal force exerted on the disc by the planet ( per unit area of the disc ) . radial induction equation @xmath75 where @xmath76 and @xmath77 are magnetic surface variables , also discussed in [ subsec : mag_variables ] . azimuthal induction equation @xmath78 6 . entropy balance equation @xmath79 where @xmath80 is a function analogous to entropy , and we use @xmath81 so that our disc is isothermal . we chose an isothermal disc to ease comparisons of our results with the results of other authors ( e.g. , by @xcite ) . moreover , viscosity terms were added to the equations of motion following the @xmath82 prescription of @xcite ( see details in @xcite ) . we use a very small viscosity ( @xmath83 ) , analogous to @xcite , to isolate the effects of slow magnetosonic waves in the disc by smoothing out the effects of fast magnetosonic waves . the `` volume '' values for the radial and azimuthal magnetic fields are given by @xmath84 and @xmath85 . their vertically - integrated counterparts may be defined similarly to @xmath86 and @xmath87 as @xmath88 respectively . there are also terms in the mhd equations involving magnetic flux or energy that involve products of these variables : @xmath89 , @xmath90 , and @xmath91 . we define the following vertically - integrated quantities for these products , @xmath92 respectively . as shown in eqns . ( [ eqn : radial_motion ] ) - ( [ eqn : az_induction ] ) , the induction equations use @xmath76 and @xmath77 , while the equations of motion use @xmath70 , @xmath71 , and @xmath72 . as such , we need a way to relate @xmath93 and @xmath94 . we can do this by introducing a `` magnetic '' thickness of the disc , @xmath95 . using the definitions of @xmath93 and @xmath94 , with @xmath95 , we find that @xmath96 we suggest that the value of @xmath97 is the same in all three relations . by relating @xmath93 and @xmath94 in this way , we can define a `` surface magnetic field '' , @xmath98 , such that the mhd equations are parameterized with respect to a single magnetic field variable , @xmath99 then , the magnetic terms in equations ( [ eqn : radial_motion ] ) - ( [ eqn : az_induction ] ) take their usual form . the radial equation of motion becomes @xmath100 \nonumber \\ & + & \frac{1}{r}\frac{\partial}{\partial\varphi } \left(\sigma v_{r}v_{\varphi}-\frac{\mathfrak{b}_{r}\mathfrak{b}_{\varphi}}{4\pi}\right ) \nonumber \\ & = & \frac{\sigma v_{\varphi}^{2}}{r } + \frac{\pi}{r } + \frac{\mathfrak{b}_{r}^{2 } + \mathfrak{b}_{\varphi}^{2}}{8\pi r } \nonumber \\ & - & \sigma\frac{gm_{\star}}{r^{2 } } + \sigma w_{r } , \label{eqn : radial - eqn - motion - final}\end{aligned}\ ] ] and the azimuthal equation of motion becomes @xmath101 \nonumber \\ & + & \frac{1}{r}\frac{\partial}{\partial\varphi } \left(\sigma v_{\varphi}^{2 } + \pi + \frac{\mathfrak{b}_{r}^{2 } + \mathfrak{b}_{\varphi}^{2}}{8\pi } - \frac{\mathfrak{b}_{\varphi}^{2}}{4\pi}\right ) \nonumber \\ & = & \sigma w_{\varphi}. \label{eqn : az - eqn - motion - final}\end{aligned}\ ] ] the radial and azimuthal induction equations become , respectively , @xmath102 and @xmath103 we calculate the equations of motion in the stellar reference frame , which is not inertial because the star also revolves about the center of mass of the system . so , an inertial force term is added to the equation of motion for both the disc and the planet . assuming that the inertial acceleration is only due to the gravitational attraction between the star and the planet ( but not the disc ) , the inertial force per unit mass ( i.e. , acceleration ) is @xmath104 to describe the gravitational influence of the planet on the disc , we use a gravitational potential similar to that used by @xcite , @xmath105 where @xmath106 is the gravitational smoothing length , and @xmath26 is the scale height of the disc . the total force per unit mass is @xmath107 the components of this force in polar coordinates , @xmath108 and @xmath109 , are used in eqs . ( [ eqn : radial - eqn - motion - final ] ) and ( [ eqn : az - eqn - motion - final ] ) . the force exerted _ on _ the planet by a particular fluid element with mass @xmath110 , is the acceleration given in eqn . ( [ eqn : tot - force - per - unit - mass ] ) , with opposite sign , multiplied by the mass of the fluid element , @xmath111 we then calculate the total force exerted on the planet by the disc by integrating over the disc within the computational domain , @xmath112 which we use to find the position ( @xmath113 ) and velocity ( @xmath114 ) of the planet at each time step via the planet s equation of motion : @xmath115 the gravitational torque in the @xmath116 direction on the planet is the sum over the torques from each fluid element : @xmath117_{z}.\ ] ] we also calculate the planet s orbital energy and angular momentum per unit mass ( e.g. , @xcite ) via @xmath118 respectively . we use these relationships to calculate the semimajor axis and eccentricity of the planet s orbit at each time step , @xmath119 respectively . the initial conditions for the surface density , surface pressure and surface magnetic field " are defined to be power laws @xmath120 where @xmath121 is a characteristic radius in the disc ; in most of our simulations @xmath122 is the initial location of the planet . additionally , @xmath123 , @xmath124 , and @xmath35 are the power laws in the density , pressure , and magnetic field distributions , respectively . we suggest that the initial pressure distribution is similar to the initial density distribution ( i.e. , @xmath125 ) . the initial sound speed in the disc at @xmath121 , defined as a fraction of the keplerian speed at this radius , @xmath126 , is @xmath127 where @xmath128 . the initial surface pressure at @xmath121 is @xmath129 . the initial plasma parameter at @xmath121 is then defined as @xmath130 note again that this plasma parameter , @xmath131 , is twice as large as the plasma parameter defined in @xcite . we determine the surface magnetic field distribution via @xmath132 we determine the initial equilibrium in the disc from the force balance in the radial direction : @xmath133 it is satisfied if the initial distribution of the azimuthal velocity has the form @xmath134 . \label{eqn : init - az - vel - dist}\ ] ] for all of the simulations presented here , we took @xmath135 . the thickness of the disc at @xmath121 is @xmath136 . the smoothing radius of the gravitational potential is @xmath137 . an initial plasma parameter of @xmath138 is taken for analysis of migration in laminar mhd discs , while we increase @xmath131 up to @xmath139 to study migration in a turbulent disc . we consider the migration of a @xmath0 planet in most of the simulations , as well as a @xmath1 planet in a few simulations . l l l + @xmath140 & + @xmath141 & + @xmath142 & + @xmath143 & + @xmath144 & + @xmath145 & + @xmath146 & + + + + & @xmath147 au & @xmath148 au + @xmath141 & @xmath149 km s@xmath150 & @xmath151 km s@xmath150 + @xmath142 & @xmath152 days & @xmath153 days + @xmath143 & @xmath154 & @xmath154 + @xmath144 & @xmath155 g @xmath156 & @xmath157 g @xmath156 + @xmath145 & @xmath158 kg & @xmath159 kg + @xmath146 & @xmath160 @xmath161 s@xmath162 & @xmath163 @xmath161 s@xmath162 + + + + & @xmath147 au & @xmath148 au + @xmath141 & @xmath149 km s@xmath150 & @xmath151 km s@xmath150 + @xmath142 & @xmath152 days & @xmath153 days + @xmath143 & @xmath164 & @xmath164 + @xmath144 & @xmath165 g @xmath156 & @xmath166 g @xmath156 + @xmath145 & @xmath167 kg & @xmath168 g + @xmath146 & @xmath160 @xmath161 s@xmath162 & @xmath163 @xmath161 s@xmath162 + our simulations are performed using dimensionless units @xmath169 where @xmath170 are the reference units . we first choose some reference distance , @xmath140 . the results of our simulations are applicable to multiple regions in a protoplanetary disc , because @xmath140 can be chosen to correspond to different regions of the disc . in table [ table : units ] , we show examples of reference values for scale distances of @xmath147 au and @xmath148 au . the thickness of the disc is @xmath171 . we take the mass of the star , @xmath172 , and determine the reference velocity , which is the keplerian velocity at @xmath140 , @xmath173 . the reference time is @xmath174 , and we use the reference orbital period @xmath175 as the unit of time in our plots . we next define the reference mass of the disc @xmath176 : @xmath177 . we then introduce the reference surface density , @xmath144 , such that @xmath178 . when the disc is homogeneous ( i.e. , @xmath33 ) , @xmath143 is the mass of the disc inside @xmath179 . the reference surface pressure is @xmath180 . the reference surface magnetic field is derived from the condition @xmath181 . hence , @xmath182 . taking into account our definitions of the surface magnetic field ( see eqns . [ eqn : phi - def ] and [ eqn : psi - def ] ) , we obtain the reference volume magnetic field : @xmath183 . the reference torque per unit mass is defined as @xmath184 . these reference units are used to convert equations from dimensionless units . we show examples in table [ table : units]b . in our model , we use a small value of the dimensionless density @xmath185 ; as a result , our dimensional characteristic disc mass is @xmath186 . the other characteristic values are closer to realistic values and are much smaller than the reference values shown in table [ table : units]b . we show in table [ table : units]c the typical dimensional values corresponding to our simulations . one can see that , in the case of @xmath187 au , the surface density and other parameters are close to those expected in real protoplanetary discs . the values shown for @xmath188au are too large , but we keep these model parameters for all distances in order to compare our results with those in @xcite and others . from this point forward , we will use only dimensionless units and remove tildes from variables . we use a polar grid that is uniform in the @xmath189 direction . in the @xmath55 direction , however , the grid is non - uniform ; the size of the grid cells increases with radius such that the sides are approximately square - shaped throughout the disc . @xmath190 is defined to be the number of radial grid cells , and @xmath191 is the number of azimuthal grid cells . our grid consists of @xmath192 and @xmath193 cells . our inner boundary is at @xmath194 , and our outer boundary is at @xmath195 . we apply a wave damping procedure near both boundaries similar to that used in 3.1.3 of @xcite to avoid high - amplitude wave reflections that can overwhelm the migration signal . we perform damping after every time step for @xmath196 , and @xmath197 , where @xmath198 and @xmath199 . we calculate the velocities ( @xmath65 , @xmath66 ) , as well as @xmath86 , @xmath87 , and @xmath200 , using a method similar to @xcite , @xmath201 & r < r_{\rm damp , in } \\ \mathcal{j}_{\rm out } + ( \mathcal{j } - \mathcal{j}_{\rm out } ) \exp\left [ -\left(\frac{r - r_{\rm damp , out}}{\delta_{\rm out}}\right)^{2}\right ] & r > r_{\rm damp , out } \\ \mathcal{j } & { \rm otherwise } , \\ \end{cases}\ ] ] where @xmath202 , @xmath203 , and @xmath204 . furthermore , @xmath205 and @xmath206 are the values of @xmath207 at the inner and outer disc boundaries , respectively . we calculated a number of models at different initial parameters , which are described below ( see also table [ table : simulation_names ] ) : * simulation type : * the simulations are performed in a hydrodynamic disc or an mhd disc , or to study the interaction between the planet and waves in the disc . * planet mass : * we explored two different planet masses @xmath0 and @xmath1 . * surface density exponent , @xmath123 : * this defines the initial surface density distribution in the disc , according to @xmath208 . * surface magnetic field exponent , @xmath35 : * this defines the initial surface magnetic field distribution in the disc , according to @xmath209 . * matter - to - magnetic pressure ratio , @xmath131 : * this defines the value of @xmath131 at the initial location of the planet . * initial planet location , @xmath210 : * this defines the initial orbital radius of the planet . the last column of table [ table : simulation_names ] shows the names of the simulations presented in subsequent sections . these names are referenced in the figure captions and related discussion . l l l l l l l & @xmath211 & & & & & + type & ( @xmath212 ) & @xmath123 & @xmath35 & @xmath213 & @xmath214 & name + hydro & @xmath215 & @xmath216 & @xmath217 & @xmath217 & @xmath218 & h5n-1r1 + hydro & @xmath215 & @xmath219 & @xmath217 & @xmath217 & @xmath218 & h5n-0.5r1 + hydro & @xmath215 & @xmath220 & @xmath217 & @xmath217 & @xmath218 & h5n0r1 + hydro & @xmath215 & @xmath218 & @xmath217 & @xmath217 & @xmath218 & h5n1r1 + hydro & @xmath221 & @xmath216 & @xmath217 & @xmath217 & @xmath218 & h20n-1r1 + hydro & @xmath221 & @xmath219 & @xmath217 & @xmath217 & @xmath218 & h20n-0.5r1 + hydro & @xmath221 & @xmath220 & @xmath217 & @xmath217 & @xmath218 & h20n0r1 + mhd & @xmath215 & @xmath220 & @xmath220 & @xmath218 & @xmath218 & m5n0k0@xmath31r1 + mhd & @xmath215 & @xmath220 & @xmath220 & @xmath222 & @xmath218 & m5n0k0@xmath32r1 + mhd & @xmath215 & @xmath220 & @xmath218 & @xmath222 & @xmath218 & m5n0k1@xmath32r1 + mhd & @xmath215 & @xmath220 & @xmath222 & @xmath222 & @xmath218 & m5n0k2@xmath32r1 + mhd & @xmath215 & @xmath220 & @xmath220 & @xmath223 & @xmath218 & m5n0k0@xmath310r1 + mhd & @xmath215 & @xmath220 & @xmath220 & @xmath224 & @xmath218 & m5n0k0@xmath3100r1 + mhd & @xmath215 & @xmath218 & @xmath220 & @xmath222 & @xmath218 & m5n1k0@xmath32r1 + mhd & @xmath215 & @xmath218 & @xmath218 & @xmath218 & @xmath218 & m5n1k1@xmath31r1 + mhd & @xmath215 & @xmath218 & @xmath218 & @xmath222 & @xmath218 & m5n1k1@xmath32r1 + mhd & @xmath215 & @xmath218 & @xmath218 & @xmath223 & @xmath218 & m5n1k1@xmath310r1 + mhd & @xmath215 & @xmath218 & @xmath222 & @xmath222 & @xmath218 & m5n1k2@xmath32r1 + mhd & @xmath221 & @xmath220 & @xmath220 & @xmath222 & @xmath218 & m20n0k0@xmath32r1 + mhd & @xmath221 & @xmath220 & @xmath218 & @xmath222 & @xmath218 & m20n0k1@xmath32r1 + mhd & @xmath221 & @xmath220 & @xmath222 & @xmath222 & @xmath218 & m20n0k2@xmath32r1 + waves & @xmath215 & @xmath220 & @xmath217 & @xmath217 & @xmath222 & w5n0k0r2 + waves & @xmath215 & @xmath220 & @xmath220 & @xmath224 & @xmath222 & w5n0k0@xmath3100r2 + orbits of the planet . the color background shows the surface density distribution . the @xmath225 lindblad resonances are indicated by the solid black circles.,width=264 ] as a first step , we investigated the migration of a planet in a hydrodynamic disc with an initially homogeneous density distribution ( @xmath226 ) ; these simulations are used as a base for studying migration in magnetic discs . [ figure : hydro - sigma ] shows the surface density distribution in the disc after @xmath223 orbits of the planet . the planet excites two density waves at the inner and outer lindblad resonances ( which are shown as black circles in the figure ) . as a next step , we considered different initial density distributions in the disc , @xmath227 including those where the density is homogeneous , @xmath226 , increases towards the star , @xmath228 , or decreases toward the star , @xmath229 . [ figure : hydro-5me-20me ] ( left panel ) shows the variation of the planet s semimajor axis over time for these cases . we observed inward migration when @xmath226 and more rapid inward migration when @xmath228 . when @xmath230 , the migration almost stalls ( where only slow outward migration has been observed ) , and we observed more rapid outward migration when @xmath231 . the migration rate and its direction are determined by the cumulative value of the lindblad and corotation torques , as described in sec . [ section : hydro_torque ] . when the density increases towards the star ( @xmath228 ) or the density is constant in the disc ( @xmath226 ) , the lindblad torque is larger than the corotation torque , and the planet migrates towards the star . however , when the density _ decreases _ towards the star , @xmath231 , the corotation torque is larger than the lindblad torque , and the planet migrates outwards . in the case of a more shallow positive density distribution , @xmath230 , the lindblad and corotation torques are almost equal , and the cumulative torque is small . the migration due to both the lindblad and corotation torques has been studied in semi - analytical models by @xcite . in two dimensions , they obtained the torque on the planet from the disc : @xmath232 where @xmath82 defines the slope of the density distribution ( @xmath233 ) , and @xmath234 is the isothermal sound speed in the disc at the orbital radius of the planet . this implies that the total torque on the planet in a two - dimensional disc is zero when @xmath235 , suggesting that 1 . the torque on the planet is positive , and the planet migrates outward , when @xmath236 ; 2 . the torque on the planet is zero , and the planet s migration halts , when @xmath235 ; and 3 . the torque on the planet is negative , and the planet migrates inward , when @xmath237 . the value of @xmath230 obtained from our simulations is very close to the value @xmath238 corresponding to zero torque derived by @xcite for two dimensions . we also performed simulations of the migration of a @xmath1 planet . [ figure : hydro-5me-20me ] ( right panel ) shows that the planet migrates inward when @xmath226 ; almost no migration is observed when @xmath230 ( as with the @xmath0 planet ) ; and the migration direction is outward when @xmath231 . these results are in accord with that of a @xmath0 planet , while the migration rate is faster in the @xmath1 case . this is expected according to eqn . ( [ eqn : gamma - total ] ) , which states that the total lindblad torque increases with the mass of the planet . our simulations of hydrodynamic discs are in accord with theoretical studies ( e.g. , @xcite ) and simulations performed by others ( see review by @xcite ) . _ right bottom panel : _ the one - dimensional surface density distribution , taken along the horizontal dashed line in the left two panels . the vertical dashed line shows the location of the planet [ figure : sigma - bf ] we performed a series of simulations to investigate the influence of an ordered azimuthal magnetic field in the disc on a planet s migration . we started by investigating migration in discs with a homogeneous surface density distribution ( i.e. , @xmath226 ) , with initial and boundary conditions very similar to those used by @xcite . in particular , we considered an isothermal disc ( i.e. , @xmath239 ) and set the strength of the magnetic field near the planet such that @xmath240 . our goal was to see whether our code can reproduce the results presented by @xcite . first , we investigated a homogeneous magnetic field ( i.e. , @xmath241 ) . [ figure : sigma - bf ] shows an example simulation after 10 orbital periods of the planet . we observed that the planet excites ring - like waves at which the azimuthal magnetic field is stronger and the surface density is lower than in other nearby regions in the disc . these locations correspond to the magnetic resonances that are excited by slow magnetosonic waves propagating along the field lines . [ figure : sigma - bf ] shows two rings of lower density ( left panel ) , corresponding to two rings of enhanced azimuthal field ( middle panel ) . the position of the magnetic resonances is similar to that found in the simulations by @xcite , and they correspond to the theoretical resonance locations predicted by @xcite ( see the dashed - line circles in fig . [ figure : sigma - bf ] ) . the right panel shows the linear distribution of @xmath85 and @xmath86 in the radial direction . according to the theory presented by @xcite , these waves exert a torque on the planet that can reverse the planetary migration direction if the magnetic field distribution is steep enough . we varied the steepness of the magnetic field , @xmath35 ( see eq . [ eqn : init - power - laws ] ) , taking @xmath242 and @xmath222 , and we obtained different migration rates . [ figure : sax - time - ndiff ] ( left panel ) shows that , for a constant magnetic field distribution ( @xmath241 ) , the planet migrates inwards but more slowly than in the purely hydrodynamic case . when @xmath243 , the planet slowly migrates outward , while the planet migrates outward more rapidly when @xmath244 . these simulations confirm the result presented by @xcite : in discs with a relatively strong azimuthal magnetic field , the magnetic resonances can slow or reverse a planet s migration . next , we investigated the action of the magnetic resonances at different surface density distributions . first , we took a density distribution such that the density in the disc increases towards the star , @xmath228 , and repeated the above simulations at @xmath245 . [ figure : sax - time - ndiff ] ( middle panel ) shows that the planet migrates inward in all three cases and , therefore , that the torque from the magnetic resonances is small compared with the differential lindblad torque . the migration rate in the case of @xmath243 almost exactly coincides with the purely hydrodynamic case . for a very steep magnetic field distribution , @xmath244 , the accretion rate is only slightly slower than for @xmath243 . overall , we conclude that , in the case of this density distribution ( @xmath228 ) , the positive torque associated with the magnetic resonances is not strong enough to overcome the negative differential lindblad torque . in another example , we considered a disc whose density decreases towards the star , @xmath246 . this situation is possible , for example , at the inner edge of the disc where the expanding magnetosphere or erosion of the disc may push the inner disc away from the star ( e.g. , @xcite ) . when @xmath230 , the migration rate in the hydrodynamic case is very low ( see fig . [ figure : hydro-5me-20me ] , top panel ) because the negative differential lindblad torque is approximately compensated by the positive corotation torque . for this surface density distribution ( @xmath230 ) , the positive torque associated with the magnetic resonances leads to outward migration of the planet , even for a flat magnetic field distribution , @xmath241 ( see fig . [ figure : sax - time - ndiff ] , right panel ) . for a steeper magnetic field distribution , @xmath243 , the planet migrates outward even more rapidly . it appears , then , that magnetic torques may play a significant role at the disc - cavity boundaries , as well as other regions where the surface density in the disc is flat or decreases towards the star . we conclude that the rate and direction of migration are determined by the steepnesses of both the magnetic field distribution and the surface density distribution . if the surface density _ increases _ radially towards the star ( e.g. , @xmath228 ) , then the magnetic resonances do not exert enough torque to drive outward migration for any value of @xmath35 . by contrast , when the surface density in the disc _ decreases _ toward the star ( e.g. , @xmath230 ) , the torque from the magnetic resonances is large enough to drive outward migration for different values of @xmath35 . when the surface density is constant ( i.e. , @xmath226 ) , the magnetic resonances drive outward migration when @xmath247 , and inward migration when @xmath241 . the above analysis shows that the effect of an azimuthal magnetic field on a planet s migration strongly depends on both the surface density and magnetic field distributions in the disc . it is also interesting to investigate whether the magnetic resonances also depend on the mass of the planet . to investigate this issue , we simulated the migration of a @xmath1 planet in a magnetized disc with a flat surface density distribution ( @xmath226 ) and different steepnesses in the distribution of the magnetic field ( @xmath248 ) . [ sax - n0k_diff - bet2 - 20me ] ( left panel ) shows that , for a constant magnetic field in the disc , @xmath241 , the migration rate is slower than in the hydrodynamic case . for a steeper magnetic field distribution , @xmath243 , the inward migration is even slower , and the direction of the migration reverses at an even steeper field distribution , @xmath244 . these results are in general agreement with those obtained for a smaller - mass planet : the magnetic resonances are strong enough to reverse the migration . note that , at @xmath244 , the outward migration of the more massive planet is much faster than the migration of the lower - mass planet . however , longer simulation runs have shown that , for a @xmath1 planet , the laminar stage of the disc does not last long , @xmath249 planetary orbits . at later times , the disc becomes turbulent . [ sax - n0k_diff - bet2 - 20me ] ( right panel ) shows that , after a relatively brief stage of slow migration in the laminar disc , the migration becomes stochastic in the turbulent disc . we also see such a transition to stochastic migration for a @xmath250 planet , though at much later times ( @xmath251 when @xmath226 , @xmath241 , and @xmath240 ) . in both cases , the disc becomes turbulent , and the semimajor axis varies in time stochastically due to the planet s interaction with turbulent cells in the disc . we found that the observed turbulence is an interesting phenomenon and investigated the migration of planets in turbulent discs . earlier , such migration was studied by @xcite in 3d simulations ; they observed that the migration becomes stochastic and the migration rate may be strongly modified or reversed due to interaction with turbulent cells in the disc . in this paper , we performed 2d simulations in polar coordinates , and we see similar stochastic migration of the planet due to interaction with turbulent cells in the disc . while 2d simulations of the mri are somewhat restricted , they allow us an opportunity to investigate the details of the interaction between the planet and inhomogeneities in the disc . below , we investigate migration in turbulent discs . in this section , we investigate the transition from the laminar to the turbulent disc , the formation of turbulent discs , and the migration of a low - mass @xmath252 planet in discs with different strengths of the magnetic field . according to the theory of the mri ( see sec . [ sec : condition for mri ] ) , instability is expected in magnetized discs with @xmath253 ; that is , in discs in which the alfvn velocity @xmath254 is smaller than the sound speed @xmath46 ( e.g. , @xcite ) . below , we show the results of simulations at different initial values of @xmath131 ( see sec . [ sec : migration turb beta ] ) and the details of migration in the turbulent disc ( see sec . [ sec : migration in turb disc ] ) . we studied the migration of the planet in discs with different initial plasma parameters , @xmath255 . [ figure : sax - n0k_diff - bet2 - 5me ] ( left panel ) shows that the disc is initially laminar , and the planet migrates smoothly . this period of smooth migration is longest when @xmath256 and @xmath222 , but it is shorter for @xmath257 and even shorter at @xmath139 . this is understandable , because the magnetic field is more easily tangled by the disc matter when the matter strongly dominates , such as when @xmath257 and @xmath224 . the right panel of the same figure shows the semimajor axis of the planet at later times . one can see that the disc is still laminar when @xmath256 . in all other cases , the disc becomes turbulent . note that the migration often changes direction from inward to outward and vice versa ; this reflects the stochastic nature of the migration process in a turbulent disc , as discussed below in sec . [ sec : migration in turb disc ] . in this section , we consider the development of the mri , as well as the migration of a planet in a turbulent disc , in greater detail . as a base , we consider a model where the initial value of plasma parameter is large , @xmath139 ( model m5n0k0@xmath3100 ) , so that the mri instability starts easily . we observed that the origin of the radial component of the field , which is required for the instability , is in the fact that the planet excites non - axisymmetric waves in the disc . these waves lead to non - axisymmetric motion of the gas in the disc and , subsequently , to the formation of a radial component of the magnetic field , which is further stretched by the differential rotation in the disc . [ figure : mri - startup-3 ] shows how parts of the initially azimuthal magnetic field near the planet ( left panel ) acquire a radial component and are subsequently stretched by the differential rotation in the disc , eventually forming a loop ( middle panel ) . later , this process occurs at larger distances from the planet , many more field lines are stretched in the radial direction ( such that different inhomogeneities and loops form ) , and the disc becomes globally inhomogeneous and turbulent . [ figure : mri - analysis - n0k0-bet100-r10 ] ( left panel ) shows that the disc consists of azimuthally - stretched turbulent cells . the middle panel of the same figure shows that the total magnetic field , @xmath258 , becomes strongly inhomogeneous , with some regions having the original polarity and others reversing polarity . the distribution of the plasma parameter , @xmath131 , shows that the simulation region splits into regions that are either magnetically- or matter - dominated ( right panel ) . therefore , in the mri regime , the disc becomes strongly inhomogeneous both in the density and in the magnetic field . we investigated the density distribution , torque , and semimajor axis evolution of the planet in this case in greater detail . [ figure : mri - bet100-torques - sax-10f ] ( top panels ) shows the surface density distribution in the disc at @xmath259 and @xmath260 . at @xmath261 , the disc is still laminar and two lindblad density waves are clearly seen . however , after @xmath262 , non - axisymmetric motions start to `` tangle '' the field lines of the weak magnetic field , and an mri - type turbulence gradually develops . mri - type turbulence is observed during the time interval @xmath263 . the middle column ( @xmath264 ) shows small - scale turbulent cells that persist for many orbits . however , later ( at @xmath265 ) , the turbulent cells become larger in size because `` islands '' of stronger magnetic field also become larger in size , and often one or two main density waves form in the inner parts of the disc . the beginning of this process is seen in the right column of fig . [ figure : mri - bet100-torques - sax-10f ] . we suggest that the finite life of the mri turbulence and formation of these larger - scale waves may be connected with the 2d nature of our mri turbulence . however , the low - amplitude mri turbulence proceeds over long periods of time , which is sufficient to study the migration of a planet in the turbulent disc . formation of large - scale waves is possible in realistic discs ; we use these waves to study the interaction of a planet with waves in the disc in sec . [ sec : migration in waves ] . the second row in fig . [ figure : mri - bet100-torques - sax-10f ] shows the 1d density distribution along the line connecting the planet and the center of the star at the same times as the top row . when the disc is laminar , we see only a small bump in density associated with matter accumulation near the planet . however , as the disc becomes more and more turbulent , larger and larger variations in the surface density of the disc are observed . the 3rd row in fig . [ figure : mri - bet100-torques - sax-10f ] shows the torques acting on the planet at the same times as the top two rows . when the disc is laminar ( left column ) , the planet migrates due to the excitation of density waves at the lindblad resonances : the inner torque is positive and smaller than the negative outer torque , so that the total torque is negative ( the blue line in fig . [ figure : mri - bet100-torques - sax-10f ] ) . in this case , the planet slowly migrates inward ( see the slow variation of the semimajor axis up to time @xmath266 in the bottom row ) . however , when the disc is turbulent ( @xmath267 ) , we observe that both the inner and outer torques can be either positive or negative ; they both vary rapidly and the resulting torque also changes sign . the _ magnitudes _ of these turbulent torques are much larger than those seen in the laminar case ( note the torque magnitudes shown on the @xmath268-axes in the third row of fig . [ figure : mri - bet100-torques - sax-10f ] ) . the torques become stochastic and correspond to the interaction of the planet with individual turbulent cells . the variation of the sign of the net torque shows also that the total averaged torque may be either negative or positive ( i.e. , that the direction of migration may change ) . in the shown example , the average total torque is negative and the planet migrates inward overall . note that , in other cases , the direction of the migration may be either inward or outward ( see , e.g. , figs . [ sax - n0k_diff - bet2 - 20me ] and [ figure : sax - n0k_diff - bet2 - 5me ] ) . the interaction between a planet and turbulent cells in a disc is a complex process . the planet interacts with a set of turbulent cells gravitationally , but the lindblad density waves are not homogeneous . the closest cell may contribute to the torque more strongly than more remote cells , and the torque becomes more stochastic . additionally , the planet passes through individual cells , and each cell may exert a corotation torque on the planet . it is difficult to track the interaction of a planet with individual turbulent cells . that is why we developed conditions in which a planet interacts with ordered inhomogeneities ( in the form of waves in the disc ) . we consider two types of waves : ( 1 ) low - amplitude ordered waves generated in a hydrodynamic disc by a force at the inner boundary ( sec . [ subsec : low - amp waves ] ) , and ( 2 ) high - amplitude waves that form at later times in simulations using an mhd disc ( sec . [ subsec : high - amp waves ] . ) to better understand the interaction between a planet and individual turbulent cells , we created a model in which an ordered density wave is generated at the inner disc boundary by a periodic force that decreases with the radius as @xmath269 . this force generates density waves with a small amplitude , for which the density contrast between the wave and the disc is small , about @xmath270 . the density wave then propagates through the simulation region . we placed a planet at an initial radius of @xmath271 , away from the action of this force at the boundary . we observed that the planet moves faster than the wave and it interacts differently with different parts of the wave . [ figure : wave - weak - rp2 ] ( top panels ) shows slices of the surface density at @xmath272 . we show the moments when the planet is located at the inner or outer edge of a wave ( i.e. , where the surface density either decreases or increases toward the star , respectively ) . the corotation torque is larger than the differential lindblad torque when the density slope in the disc corresponds to that in equation ( [ eqn : gamma - total ] ) ; that is , the slope of the density distribution is either positive or only slightly negative toward the star ( see sec . [ sec : migration hydro ] ) . this is expected when the planet is at the inner edge of a wave . the middle row of fig . [ figure : wave - weak - rp2 ] shows the torques acting on the planet ; the times from the top panels marked with vertical dashed lines . the total torque has maxima at @xmath273 and @xmath274 . the top panels show that , at these moments of time , the planet is located at the inner edge of the wave and , therefore , that the positive torque acting on the planet is the corotation torque ( which appears due to the positive density gradient at the inner edge of the wave ) . the bottom panel of fig . [ figure : wave - weak - rp2 ] shows that the planet migrates inward overall , because the differential lindblad torque dominates overall . however , at @xmath273 and @xmath275 , the planet migrates outward due to the temporarily dominant corotation torque . at @xmath276 and @xmath277 , the planet is located at the outer edge of the density wave , where the density increases towards the star . at these moments of time , the total torque is negative and the planet migrates inward . we suggest that , at the outer edge of a wave , a planet excites density waves at the lindblad resonances , and the differential lindblad torque drives overall inward migration . however , when the planet is at the inner edge of the wave , the positive corotation torque is large , and the planet migrates outward . in this subsection , we analyze the passage of a planet through waves with much higher amplitudes . these high - amplitude waves often form at later times in simulations of mri - turbulent discs , where the non - axisymmetry of the gravitational potential leads to the formation of the inner density waves . we chose the model where the initial plasma parameter @xmath278 and a planet is placed at @xmath271 . at later moments in time , after the wave forms , the planet migrates to a radius of @xmath279 . [ waves - rp2-bet100 - 5me ] ( top panels ) shows the density distribution in the wave and the position of the planet for several representative moments in time , where the planet is located at the inner edge of a wave ( left and right panels , @xmath280 ) , the planet is in the middle of a wave ( @xmath281 ) , and the planet is in the low - density region ( @xmath282 ) . the density contrast between the wave and the rest of the disc is @xmath283 , which is about @xmath223 times higher than the low - amplitude waves discussed in the previous subsection . the middle and bottom rows of fig . [ waves - rp2-bet100 - 5me ] show the torques acting on the planet and the variation of the planet s semimajor axis in time . the dashed vertical lines show the four moments in time corresponding to the top four panels . one can see that , at @xmath284 , the planet is located at the inner edge of the wave , where the density decreases towards the star , and the positive corotation torque is expected to be larger than the differential lindblad torque . indeed , the middle row shows that the total torque is positive at this moment of time , and the planet migrates outward . at the second moment in time , @xmath281 , the planet is located in the middle of the density wave . the middle row shows that , at this moment , a strongly negative torque dominates , and the planet migrates inward . this large negative torque is primarily comprised of the differential lindblad torque . this torque is proportional to the surface density in the disc , and therefore it is large when the planet is located inside the high - density wave . the positive corotation torque is relatively small . at @xmath282 , the planet is located away from the wave in the low - density part of the disc , and both torques are small . at the last considered moment ( @xmath285 ) , the planet is again at the inner edge of the wave , where the corotation torque is positive and larger than the differential lindblad torque , and the planet migrates outward . in this example , the density wave is an analog of a large turbulent cell , where the total torque is either positive or negative and acts onto the planet depending on the position of the planet relative to the wave . we expect that the interaction with smaller - sized turbulent cells is similar to the observed interaction with mhd waves , but that the duration of the interaction is shorter . as a result of such interactions , a torque is exerted on the planet during short intervals of time , and the planet s semimajor axis varies stochastically under the action of these torques . based on our observations with both the low- and high - amplitude waves we can schematically describe the interaction between a planet and a turbulent cell ( see fig . [ fig : planet - cell ] ) . when the planet is at the outer edge of the cell ( the side that is farther from the star ) , the density is increasing toward the planet ( @xmath286 ) , and so we expect the torque to be negative and the planet to migrate inward . conversely , when the planet is on the inner edge of the cell ( the side that is closer to the star ) , the density decreases toward the star ( @xmath287 ) , and so we expect the torque to be positive and the planet to migrate outward . this inward - outward migration does not `` trap '' the planet , because the cell is moving away from the star , so this is a transient process . we investigated the migration of a low - mass planet ( @xmath288 ) in magnetized discs using a godunov - type hlld mhd code in polar coordinates @xcite . the initial surface magnetic field is azimuthal , with different radial distributions @xmath289 , and different strengths determined by the initial plasma parameter at the planet s location , @xmath131 , which varied between @xmath256 and @xmath224 . we also varied the initial radial surface density distribution in the disc , @xmath227 . our main conclusions are as follows : 1 . in strongly - magnetized discs ( @xmath290 ) , and where the density distribution in the disc is flat ( @xmath226 ) , the planet s migration is strongly influenced by magnetic resonances , which are excited near the planet and exert a positive torque on the planet . the migration slows down when the magnetic field distribution is flat ( @xmath241 ) ; it slows down more strongly when the magnetic field increases towards the star ( @xmath243 ) ; and the migration reverses when the field steeply increases towards the star ( @xmath244 ) . these results are in accord with theoretical predictions by @xcite and simulations by @xcite . 2 . compared with @xcite , we investigated the effect of magnetic resonances on the migration of a planet in discs with different density distributions . we observed that the steepness of the density distribution strongly influences the rate and direction of the migration . when the density increases towards the star ( @xmath228 ) the planet migrates inward at any steepness of the magnetic field ( @xmath248 ) and the effect of the magnetic resonances is negligibly small . this is because , at larger density steepness , the negative differential lindblad torque is much larger than the positive corotation or magnetic torques . in the opposite situation , when the density in the disc decreases toward the star ( @xmath230 ) , and the positive corotation torque almost balances the negative lindblad torque , the role of the positive magnetic torque becomes very significant . the planet migrates outward due to magnetic resonances at all values of the magnetic field steepness ( @xmath245 ) . experiments with a larger mass planet , @xmath1 , show that the action of the magnetic resonances is similar to that in the case of a lower - mass planet : the positive torque increases with the steepness of the field , as predicted by the theory . we also observed that , at @xmath244 , the more massive planet migrates outward more rapidly than the lower - mass planet . this is in accord with eqn . ( 65 ) of @xcite , in which the migration time scale is shown to be @xmath291 . we investigated weakly - magnetized discs with initial plasma parameter @xmath292 . we observed that non - axisymmetric motions in the disc lead to the formation of a radial component of the magnetic field , its stretching by the differential rotation in the disc , and subsequent mri- driven turbulence in the disc . interaction between the planet and turbulent cells leads to stochastic migration , similar to that observed in 3d discs by , e.g. , @xcite . we investigated the transition from the laminar to turbulent disc that often starts near vicinity of the planet , where non - axisymmetric density waves are excited by the planet . the turbulence starts more rapidly when the planet is more massive . the torques acting on the planet are larger in turbulent discs than in the laminar discs , leading to more rapid migration . however , the direction of the migration is also stochastic , and outward migration is frequently observed . 6 . to understand how planets interact with individual turbulent cells , we investigated the propagation of a planet through density waves in the disc . we observed that a planet experiences a strong negative torque when it is located inside the wave or at the outer edge of the wave ( where the density increases towards the star ) . however , when the planet is located at the inner edge of the wave ( where the density decreases towards the star ) , then it experiences a positive corotation torque , and it migrates away from the star . we conclude that the stochastic motion of the planet is connected with the alternating action of these positive and negative torques . we gratefully acknowledge support from the nasa research opportunities in space and earth sciences ( roses ) origins of solar systems grant nnx12ai85 g . we also acknowledge support from grants fap-14.b37.21.0915 , ss-1434.2012.2 , and rfbr 12 - 01 - 00606-a . avk was supported by the russian academic excellence project 5top100 " . resources supporting this work were provided by the nasa high - end computing ( hec ) program through the nasa advanced supercomputing ( nas ) division at the nasa ames research center and the nasa center for computational sciences ( nccs ) at goddard space flight center . armitage p.j . & rice w.k.m . 2006 in a decade of extrasolar planets around normal stars : proceedings of the space telescope science institute symposium held in baltimore , maryland , may 2 - 5 , 2005 , ed . m. livio , k. sahu , & j. valenti ( baltimore : stsci )

the effect of stress - triaxiality on growth of a void in a three dimensional single - crystal face - centered - cubic ( fcc ) lattice has been studied . molecular dynamics ( md ) simulations using an embedded - atom ( eam ) potential for copper have been performed at room temperature and using strain controlling with high strain rates ranging from/sec to/sec . strain - rates of these magnitudes can be studied experimentally , _ e.g. _ using shock waves induced by laser ablation . void growth has been simulated in three different conditions , namely uniaxial , biaxial , and triaxial expansion . the response of the system in the three cases have been compared in terms of the void growth rate , the detailed void shape evolution , and the stress - strain behavior including the development of plastic strain . also macroscopic observables as plastic work and porosity have been computed from the atomistic level . the stress thresholds for void growth are found to be comparable with spall strength values determined by dynamic fracture experiments . the conventional macroscopic assumption that the mean plastic strain results from the growth of the void is validated . the evolution of the system in the uniaxial case is found to exhibit four different regimes : elastic expansion ; plastic yielding , when the mean stress is nearly constant , but the stress - triaxiality increases rapidly together with exponential growth of the void ; saturation of the stress - triaxiality ; and finally the failure .

cond-mat0310541

ductile fracture of metals commonly occurs through the nucleation , growth and coalescence of microscopic voids . @xcite much can be learned about the ductile fracture through the study of these voids . a particularly interesting case is the dynamic fracture of ductile metals , @xcite in which the strain rates are so high that processes such as diffusion operating on relatively long time scales may be neglected , while inertial effects become relatively important . void growth is driven by the need to relax tensile stress that builds up in the system , and to minimize the associated elastic energy . the material around a void deforms plastically in order to accommodate the void growth . naturally , the plastic deformation results from a local shear stress , which may arise from the applied stress , but it also may arise from the stress field of the void even if the applied stress is hydrostatic . so the expectation is that the evolution of the plastic zone , and hence the growth of the void , is influenced by the degree of stress triaxiality ; _ i.e. _ , the ratio of the mean ( hydrostatic ) stress to the shear stress . it is this relationship that we study here by varying the triaxiality of the loading . in particular , we conduct simulations in which one , two or three directions of the system are expanded , producing a state of uniaxial , biaxial or triaxial strain , respectively . variation in the triaxiality of the strain causes variation in the triaxiality of the stress state , where it should be noted that uniaxial ( biaxial ) strain does not imply pure uniaxial ( biaxial ) stress . besides dynamic crack propagation experiments , dynamic fracture can be measured for instance in shock physics or spallation experiments , to which the simulations performed here are compared . various techniques are employed to generate the shock waves : hopkinson bar , gas gun , high explosives , and laser ablation . with hopkinson bar the strain - rates @xmath2 usually are of the order @xmath3/sec , in gas gun of @xmath4/sec , with high - explosives even higher strain - rates can be produced , and with lasers strain - rates exceeding @xmath0/sec are attained . in a gas gun for instance the fracture results from essentially one - dimensional shock loading . two compressive shock waves are generated by the impact of a flier on a metal target , propagate away from each other , reflect from opposite free surfaces becoming tensile release waves and finally come into coincidence again . if the combined tensile stress exceeds the rupture strength of the material , the metal fails , after some incubation time , producing a fracture surface . in strong shocks , a scab of material may spall from the back side of the target and fly off . spallation experiments @xcite for single and polycrystal copper report spall strength values of @xmath5 gpa at strain - rates @xmath6/sec , and scaling between the spall strength and strain - rate , @xmath7 . in this study of dynamic fracture in ductile metals at high strain - rates ( @xmath0 @xmath1/sec ) we have concentrated on void growth starting from a single crystal copper lattice containing an infinitely weakly bound inclusion or a pre - existing nanoscale void . the lattice is initially free of other defects . we have focused on the effect of stress - triaxiality on void growth . in some fracture experiments , for example in necking and cup - cone fracture , @xcite the uniaxial strain produces a stress state that transitions rapidly to triaxial state due to the plastic flow during the course of loading . it is during the triaxial phase that void growth and failure take place . because of the connection with shock experiments , the stress triaxiality study done here is carried out using strain control , and it is the strain that is varied from uniaxial to biaxial to triaxial . much of the damage modeling of metals has been carried out at mean - field or continuum level based on constitutive theories . the continuum models concentrate especially on two areas : macroscopic crack growth phenomenon @xcite and studies of porosity , _ i.e. _ behavior of an array of voids , at sub - grain level during loading . @xcite in the latter area for example the locus of yield surfaces in stress space has been studied , which is related to the question of the effect of stress - triaxiality studied here . of the void growth studies especially the gurson model @xcite is commonly used to model cavitation ( the development of porosity ) at the sub - grid level in what is termed as damage modeling . these continuum calculations often assume that the matrix material , where the voids are embedded , is elastically rigid and plastically incompressible , and the dilation of the void - matrix aggregate is completely due to the void growth . of particular interest , and relevance in terms of this study , is a single crystal plasticity study of void growth . @xcite the calculations are typically done by determining approximate solutions for integrals of incremental equations of virtual work using the finite element method . continuum modeling has been used to study some of the phenomena addressed in this paper such as the effect of triaxiality on void growth and void shape changes . @xcite the validity of the approximate solutions of incrementals limits the strain - rates to be rather low compared to the strain - rates used in this study . in order to characterize the void growth not only with macroscopic quantities and at the continuum level , but to investigate what happens at the atomistic level , we have employed molecular dynamics ( md ) simulations . md simulations enable us to see what the effects are on the void surface at the single atom level , when it grows while the total system yields . this article presents work , which is an extension to the work done earlier by some of this article s authors of void growth in a single crystal copper with hydrostatic loading , void nucleation and growth in single and polycrystalline copper . @xcite molecular dynamics simulations of void growth in single crystal copper have also been conducted by other groups for slab geometries of interest in the semiconductor chip metalization problem . @xcite the effectively two - dimensional , thin film systems are in contrast to three dimensional bulk systems studied here . in some cases md simulations of plasticity and crack propagation have been as large as billion atoms . @xcite this article is organized as follows . it starts in section [ method ] with an overview of the md method used and the simulations which have been carried out in this study . exploration of the results of the simulations starts in section [ stressstrain ] by the study of the mean or hydrostatic stress versus strain as well as the deviatoric part of the stress tensor , von mises stress , which is used to measure the shear stress , and stress - triaxiality versus strain for all the simulated strain - rates and modes of expansion . section [ plasticstrain ] concentrates on the macroscopic plastic quantities , such as mean and equivalent plastic strain , plastic work and its relation to the temperature . the evolution of the void in terms of its growth and shape changes is studied in section [ voidevol ] . section [ summary ] summarizes the results and compares different measured quantities with each other concentrating on one of the simulations , uniaxial strain with strain - rate @xmath8/sec . the article is concluded with discussions of the results and suggestions for future studies in section [ discussion ] . in this atomistic - level study of void growth , the simulations have been done using empirical embedded - atom ( eam ) potentials in classical molecular dynamics @xcite following the scheme developed earlier . @xcite the copper eam potential we have used is due to oh and johnson . @xcite the system , in which the simulations are done , is a three - dimensional single - crystal face - centered - cubic ( fcc ) lattice in a cubic box with @xmath9 faces . periodic boundary conditions are used in all the three directions so that there are no free boundaries in the system apart from the void . equivalently the system can be imagined to consist of an infinite periodic array of voids . note that periodic boundary conditions have also been used in continuum models of void growth , but in the continuum modeling of void growth in isotropic materials the calculations are done in a reduced cell , which exhibits one quarter of the box in two dimensions , and one eighth in three dimensions , and the behavior of other areas are derived from the symmetries . we use the full cubic box because the cubic symmetry present in the continuum is broken in md at finite temperature , and processes such as dislocation nucleation at the void surface would be over - constrained in a reduced simulation box . in the simulations , the system is brought to thermal equilibrium at room temperature , @xmath10 k , with a commonly used thermostat @xcite and at ambient pressure , @xmath11 mpa , keeping the volume constant . after that a spherical void is cut in the middle of the system , later the thermostat is turned off , and the dilational strain is applied uniformly with a constant strain - rate @xmath2 . the removal of the atoms in the spherical region may be considered to simulate the instantaneous separation of the matrix material from an infinitely weakly bound inclusion . the uniform expansion in these strain - controlled simulations is applied through rescaling the coordinates as in the parrinello - rahman method . @xcite technically the three cartesian coordinates of the atoms are rescaled to the unit - box , each coordinate @xmath12 . when calculating the forces and velocities , as well as updating the new positions of the atoms , the unit - box is multiplied by a diagonal scaling matrix @xmath13 , where @xmath14 s are the simulation box s side lengths , to compute the true positions of the atoms , @xmath15 this scaling matrix @xmath16 is updated each time - step , when the load is applied , by multiplying the initial matrix @xmath17 with the sum of the unit matrix and the strain matrix @xmath18 , @xmath19 for our purposes the strain - rate matrix @xmath20 is always diagonal , since neither rotation nor simple shear type strains are studied . in the triaxial case all the terms in the diagonal are equal ; in the uniaxial there is a single non - zero term ; and in the biaxial case two of the three diagonal terms differ from zero and are equal . prior to expansion the system is cubic , its scaling matrix @xmath17 is diagonal , and all the terms are equal and correspond to the equilibrium size at ambient pressure . hence the scaling matrix @xmath16 remains diagonal throughout the simulation , and the strain in each case is in a @xmath21 direction . in fracture and plasticity simulations the first quantity to consider is the stress - strain behavior . with the strain as an input parameter , here we have to measure the stress . in this study of the stress - triaxiality we are interested in both mean and shear stresses . therefore the whole stress tensor @xmath22 is needed . the stress tensor ( the negative of the pressure ) can be calculated atomistically on each time - step using the virial formula : @xcite @xmath23 the first term in the stress tensor is the kinetic contribution of atoms denoted with @xmath24 and having masses @xmath25 and momenta @xmath26 . the second term , a microscopic _ virial _ potential stress , consists of sums of interatomic forces @xmath27 of atom pairs @xmath28 with corresponding distances @xmath29 . it should be noted , that here and in the rest of the article @xmath24 and @xmath30 denote the atoms , and @xmath31 and @xmath32 the cartesian coordinates . note that the thermal stress is included , although in practice in these simulations it contributes less than 1 gpa , less than 10% of the yield stress value , and never dominates the changes in stress . typically in the simulations carried out here , the cube consists of 60 fcc unit cells in each direction , giving 864 000 atoms . the equilibrium side - length of such a copper system is @xmath33 nm at room temperature and ambient pressure . the radius of the spherical void cut from the system , unless otherwise noted , is 0.1 of the side - length of the box ; thus @xmath34 nm . after the void is cut , there are 860 396 atoms in the system . the relatively inexpensive potential used enables us to do extensive simulations in time . a single time - step takes typically about 40 sec of cpu - time in a system with 864 000 atoms in a linux workstation with intel xeon 1700mhz processor . the longest calculation required 835 050 time - steps corresponding to 5.6 nanoseconds . the time - step was @xmath35 femtoseconds . as mentioned earlier , in order to study the effects of the stress - triaxiality and different modes of expansion on the void - growth , we have applied three different types of expansion , namely uniaxial , biaxial , and triaxial . the strain - rates used for each of the three modes of expansion are @xmath36/sec , @xmath37/sec , @xmath38/sec , @xmath8/sec , and @xmath0/sec . for the lowest strain - rate , the md code was parallelized in order to take advantage of massively parallel computers . the parallelization was done using a spatial domain - decomposition , and was shown to scale nearly linearly up to 128 processors . the parallel code was used in the case with 835 050 time - steps mentioned above , for example . for comparison in the elastic regime , we have also performed simulations without a void in all three modes of expansion . these simulations have been used to determine the bulk , elastic stress - strain response of the eam copper and hence the elastic constants . without a void , the system is not so strain - rate and system size dependent , at least up to the point of failure , so the so - called `` no void '' simulations have been performed with a smaller system size , 45 fcc - cells in each direction ( 364 500 atoms ) and at the single strain - rate @xmath39/sec . a uniaxial study of the 60@xmath40 system size , but with a smaller initial void radius of @xmath41 nm , was carried out with the strain - rate @xmath42/sec in order to study the void - size dependence . in this case the system with the void contains 863 543 atoms . it should be mentioned , too , that all of the intermediate strain - rate simulations ( @xmath43/sec and @xmath38/sec ) expansion were not started from equilibrium conditions at @xmath44 mpa , but from systems expanded previously at the strain - rate @xmath39/sec . these simulations have been restarted well before yielding , when the system s behavior is rate independent , and relaxed for 2000 time - steps , or 13.4 picoseconds , without expansion before continuing the expansion at the intermediate strain rates . the energy is conserved during the relaxation in md simulations . these restarts have been accomplished at strain values @xmath45 % , @xmath46 % , and @xmath47 % in uniaxial , biaxial , and triaxial cases , respectively . let us begin to explore the results of the md simulations by looking at the stress - strain curves . figure [ fig1 ] shows these curves for each of the modes of expansion at all the strain - rates computed . the data from `` no void '' cases are also plotted . the mean or hydrostatic stress , @xmath48 is plotted to indicate the principal impetus for void growth . note that the strain is the _ engineering strain _ , defined as the expanded system size divided by the original system size minus one . in the uniaxial and biaxial cases the strains are the principal strain values @xmath49 in the direction of the strain , such as @xmath50 , @xmath51 in the uniaxial case , and @xmath52 , @xmath53 in the biaxial case . hence the strain is the value of a non - zero diagonal term of @xmath54 in eq . ( [ eq_update ] ) . the mean strains @xmath55 are 1/3 and 2/3 of the plotted uniaxial and biaxial strains , respectively . thus the total volumetric strain - rates are not the same in the different expansion modes . in the triaxial cases the plotted and the mean strain values are the same . the shape of the stress - strain curves do not differ much depending on the modes of expansion in these cases , at least when plotted versus mean strain . independent of the strain - rate , and whether with or without a void , the stress - strain curves lie essentially on top of each other during elastic expansion , _ i.e. _ , the initial smooth behavior when the system is still recoverable and has not deformed plastically . the stress - strain curve starts to deviate from the trend of the elastic behavior at a specific , `` critical '' point which we call here a _ yield point_. in other quantities we measure a change in behavior happens at a specific point , too , and as we shall see later the critical or yield points mostly coincide with each other , _ i.e. _ , their strain values are approximately the same independent of from which quantity we derived it . the same point is also the one when the void starts to grow , which is the primary mechanism for plasticity in this study . here we define numerically the yield point of the cases with void by comparing their stress - strain curves with the reference `` no void '' curve , which behaves elastically beyond the yield points of the other cases [ cf . [ fig1](a ) inset ] . ultimately the no void case does fail by homogeneous nucleation of voids , and this is the reason for the drop in the mean stress . there is a small offset between cases with a void compared to the case without a void due to the elastic relaxation of the void . the value of the stress at the yield point in the cases with void is lower with lower strain - rate , and thus the strain to yield is also lower . in each of the modes of expansion , the stress at the yield point for the strain - rate @xmath56/sec is close to the value to which the higher strain - rates converge . of course , at much lower strain rates the physics changes , new mechanisms become active , so this value need not hold for arbitrarily low strain rates . however , it is noteworthy that the stress at the yield point is not scaling with strain , contrary to the experimental finding for the spall strength explained in section [ intro ] . overshooting , the phenomenon that the maximum stress is much higher than the stress at the yield point , is evident here for the higher strain rates . the scaling of the spall strength versus the strain - rate @xcite with an exponent 0.2 is reproduced here when one compares the maximum stress values instead of the stresses at the yield point for strain - rates @xmath57/sec , @xmath37/sec , and @xmath1/sec , since then the exponent is @xmath58 , lowest for the uniaxial case and highest for the triaxial case . on the other hand the stress value at the yield point , which is at the same time the maximum stress , when @xmath56/sec is very close to the value of 6 - 8 gpa the spall strength scaling predicted from the lower strain - rates mentioned in section [ intro ] . it should be noted also , that since we are limited to finite , fairly small , system sizes , at late stages of the stress - strain curves , at the failure , the data is not realistic anymore . the reason is that at the plastic part of the stress - strain behavior when the void grows , it also emits dislocations , and in a finite system with periodic boundaries , when the dislocations have traveled long enough , they propagate through the boundaries and reenter from the other side . in the picture where we have a periodic array of voids in an infinite system this means that the voids are so close to each other that they start to interact . in reality voids are never arranged in a perfect cubic lattice structure and in symmetric positions with respect to each other , and thus the interactions of the voids in the simulations with their periodic images are just an unphysical finite size effect . in the shear stress or more precisely in the deviatoric part of the stress tensor @xmath59 plotted in fig . [ fig2 ] , a much bigger difference is seen between the modes of expansion than in the mean stress . for the deviatoric part of the stress @xmath59 we use _ von mises stress _ : @xmath60^{1/2 } , \label{eq_misesstress}\ ] ] where @xmath61 is the second invariant of the stress deviator @xmath62 . @xcite thus von mises stress reads : @xmath63^{1/2 } \\ & = & \left [ \frac{1}{2 } \sum_{\alpha > \beta } ( \sigma_{\alpha \alpha } - \sigma_{\beta \beta})^2 + 3 \sum_{\alpha > \beta } \sigma_{\alpha \beta}^2 \right]^{1/2}. \end{array } \label{eq_misesstress2}\ ] ] while the mean stress at the yield point gets a value of about @xmath64 gpa when loaded with strain - rate @xmath65/sec in each of the three modes of expansion , von mises stress has a value of @xmath66 gpa and 0.7 gpa in the uniaxial and biaxial cases , respectively . in the triaxial case it should be zero by symmetry , and the difference from zero , representing symmetry - breaking effects , is small . thus the loading differences between the modes of expansion are quantified in von mises stress . after the onset of plasticity or the void growth , von mises stress gets a value of about @xmath67 gpa , 0.2 gpa , and 0.1 gpa , in the uniaxial , biaxial , and triaxial cases , respectively , independent of the strain - rate , but with significant fluctuations in this regime . in this period also dislocations move under the action of the shear stress until the stress has dropped to the point that it is no longer sufficient to move a dislocation through the forest of dislocations . the final value of the shear stress corresponds to the flow stress , and they are close to the tensile strength values of copper , 200 - 400 mpa , quoted in the literature . @xcite although von mises stress captures differences between the loading modes quite well , an even better quantity to study is the ratio between hydrostatic and shear stresses , the _ stress - triaxiality _ @xmath68 which has been plotted in fig . [ fig3 ] . in the uniaxial case the stress - triaxiality starts from the value @xmath69 and slowly decreases linearly to a value @xmath70 until the onset of rapid growth at the yield point . after the rapid increase the stress - triaxiality saturates at @xmath71 . the stress - triaxiality in the biaxial case starts with a much larger value than in the uniaxial case . it begins at @xmath72 and increases linearly to a value @xmath73 at the yield point where it grows rapidly to a value of @xmath74 . we have noted the correspondence of the final von mises stress and the flow stress above . similarly the spall strength provides an experimental measure of the mean stress that can be supported in void growth . thus the stress - triaxiality values can be compared with the ratio of the spall and the tensile strength of copper . previously quoted literature values for them are 6 - 8 gpa and 200 - 400 mpa , respectively , giving for their ratio values between 15 and 40 , and thus comparable with the stress - triaxiality values here . the comparison is not fully rigorous , but it provides an indication of how reasonable the final stress triaxiality values are in terms of experiment . since the stress - triaxiality is the mean stress divided by von mises stress , which is equal to zero in the triaxial case until the yield point and very small even after that , the stress - triaxiality is diverging and therefore not plotted here in that case . the stress - triaxiality values in the uniaxial and biaxial cases at the elastic part of the simulation is compared here also with the values one get from the elasticity theory : @xcite @xmath75 @xmath76 in the uniaxial case and @xmath77 in the biaxial case . the literature values for the elastic constants of copper are @xmath78 168 gpa and @xmath79 121 gpa . @xcite thus @xmath80 and @xmath81 in the uniaxial and biaxial cases , respectively , which compare quite well with the simulations presented here . when the elastic constants are derived from the stress - strain curves as @xmath82 , they are close to the actual experimental values : @xmath83 162 gpa , @xmath84 121 gpa and @xmath83 168 gpa , @xmath84 124 gpa in the uniaxial and biaxial cases , respectively . [ cols="<,^,^,^,^,^,^,^",options="header " , ] the critical mean stress , von mises stress , and the stress - triaxiality values where their behaviors start to deviate compared to the elastic ones , or what we call yield points , are summarized in table [ tab1 ] for the strain - rates @xmath85/sec , @xmath86/sec , @xmath8 , and @xmath0/sec of the principal strains . in the case of the highest strain - rate @xmath87/sec , the shapes of the stress - strain curves are so much rounded due to over - shooting , that there is no clear point , where the stress - strain curve deviates from the elastic behavior , and thus our definition of the yield point is no longer suitable . in comparing the mean strain values @xmath55 ( as in the table ) at the onset of plasticity for a particular strain rate , one finds that the uniaxial expansion always starts to yield at the least strain , and the hydrostatic expansion , at the greatest strain . there are two effects that contribute to the increase in the plastic threshold as the triaxiality increases . first , the shear component of the applied stress contributes to the resolved shear stress and lowers the threshold for heterogeneous nucleation of dislocations at the void surface . @xcite and second , the volumetric strain - rate is lowest in the uniaxial case and the highest in the hydrostatic case . strain - rate hardening then leads to an increase in the stress value at the onset of plasticity as the triaxiality increases . the difference between the critical strain values when defined as when a behavior deviates from the elastic behavior is nearly negligible and thus independent of whether one uses the criterion from the hydrostatic stress , von mises stress or stress - triaxiality curves . the differences reflect mainly the difficulties in defining the point what we call the yield point . however , we will see later that if the mean stress and von mises stress start to deviate from the elastic behavior with the same ratio as they have during elastic expansion , the stress - triaxiality may deviate a bit later than the other quantities . we shall see later , too , that the onset of plasticity defined from these quantities is very close to where the void starts to grow . after sufficient expansion , the system yields and the mean stress is observed to drop with respect to the elastic response . then as the simulation box continues to expand , the stress remains roughly constant until the precipitous drop at final failure . in the region of increasing strain but roughly constant mean stress , most of the strain is in the form of plastic strain , a macroscopic measure of the plastic , permanent and irrecoverable deformations in the system . in this section we study the macroscopic quantities of plasticity such as mean and equivalent plastic strains as well as the plastic work , and in the next section in more detail what are the actual plastic deformations visible in the void . the concomitant dislocations related to void s shape and volume changes are studied elsewhere . @xcite in deriving the plastic strain here it is assumed that the tetragonal symmetry is approximately preserved and thus the off - diagonal terms of the stress tensor are negligible . following the literature we separate the strain increment @xmath88 into elastic and plastic parts . @xcite thus by definition the plastic strain increment becomes @xmath89 where @xmath90 is the total increment of the strain . below we use @xmath2 instead of @xmath91 since @xmath92 can be divided from both sides of eq . ( [ eq_plastic ] ) . the total strain increment is an input parameter in these strain - controlled simulations . it is given by the strain - rate matrix @xmath20 . the compliance @xmath93 relating the elastic strain increment to the stress increment is derived from the stress - strain curves in the elastic region by : @xmath94 the stress matrix @xmath22 is calculated each time - step using eq . ( [ eq_stresstensor ] ) . the elastic compliance tensor @xmath95 is retrieved from the elastic part of the stress - strain curve of the cases without the void as follows . due to the nonlinearity of a stress - strain curve we have not only retrieved the slope of it , which would give @xmath96 , where @xmath97 is the bulk modulus , but fitted a fourth order polynomial to the strain versus stress curve , whose derivative gives us @xmath98 . this is done separately for the uniaxial , biaxial , and triaxial no - void cases , and the respective curves are used for the cases with the void . it should be mentioned , too , that in the derivation of the bulk modulus the mean total logarithmic strain is used instead of the engineering principal strain used in the plots of this article . similarly the term @xmath99 is derived using a fourth order polynomial in the mean strain versus von mises stress curve giving @xmath100 . note , that when @xmath101 is derived from the plot using mean strain there are prefactors 1/3 and 2/3 for @xmath102 in the uniaxial and biaxial cases , respectively . using formulas which relate @xmath103 and @xmath104 to @xmath105 and @xmath106 in cubic crystals , @xcite and the correspondence between elastic constants and moduli we get for @xmath103 and @xmath104 @xmath107 using the @xmath108 as @xmath109 and @xmath110 as @xmath111 due to the symmetry and neglecting off - diagonal terms , which are small compared to the diagonal ones , we get all the necessary terms for @xmath112 , and thus @xmath113 from eq . ( [ eq_eepsilon ] ) . note , that since the @xmath114 is used as a parameter instead of @xmath22 , the von mises stress must be mapped with the mean stress when finding the corresponding @xmath101 . this was done again by fitting the von mises vs. mean stress curves with fourth order polynomials . subtracting the elastic strain from the total strain as in eq . ( [ eq_plastic ] ) we get the mean plastic strain increment @xmath115 which time - integral is plotted in fig . [ fig4 ] for all the loading modes and strain - rates . in these plots one sees that after the yield point the mean plastic strain first increases roughly exponentially , although the region is too small to be definitive , and thenceforth roughly linearly . note that the mean plastic strain is not the equivalent plastic strain commonly used in plasticity , which will be defined below , but a measure of the porosity . this will be studied in the next section , where the mean plastic strain will be compared with the growth of the volume of the void . we turn now to the quantification of the dislocation flow , conventionally computed at the continuum level as the second invariant of the deviatoric plastic strain , the _ equivalent plastic strain_. typically in the case of tetragonal total strain , the equivalent plastic strain rate would be calculated as : @xmath116 the equivalent plastic strain is calculated in turn as @xmath117 in practice this formula for the equivalent plastic strain is problematic in md for several reasons . first , the time and length scales in md are much shorter than those assumed in continuum formulations of plasticity . the time scale is a problem because dislocation flow becomes partially reversible at short enough time scales . thermal fluctuations cause reversible oscillations of dislocations and fluctuations in the local elastic strain . to the contrary , the integrand in eq . ( [ eq_equivalent_t ] ) is positive definite , as appropriate for plastic deformation that is cumulative even when reversed . in practice , the application of eq.([eq_equivalent_t ] ) in md gives a result dominated by the fluctuations for small time increments ; in fact , in our attempt to apply the formula to the md deformation every 10 time steps , the contribution of the fluctuations was 22 times as large as the applied total mean strain ( these values are obtained from the biaxial case with strain - rate @xmath43/sec ) . the formula must be modified to be insensitive to thermal fluctuations . second , the formula for the equivalent plastic strain assumes isotropic plasticity in the following sense . in isotropic plasticity , the plastic flow is driven by the shear stress quantified by the von mises stress . the equivalent plastic strain is conjugate to the von mises stress , and therefore takes on a particular signficance in the theory . implicit is the assumption , for example , that slip systems that experience the same shear stress will exhibit the same plastic strain . this assumption is violated in md for two reasons . once again , the thermal fluctuations may cause the initiation of flow on one glide system before that on a symmetrically related system . this effect is observed in our md simulations . typically the symmetry is restored after a brief period , but because the plastic strain is cumulative , the symmetry breaking fluctuation is never eliminated from the plastic strain ( [ eq_equivalent_t ] ) . second , a more mundane reason the assumption of isotropy fails is that the single crystal systems are anisotropic , both because of the specific glide planes involved and because of the elastic constants , especially in copper . it may be possible to rectify these problems while retaining the basic formulation of the equivalent plastic strain , for example through a suitable multi - resolution calculation of the integral ( [ eq_equivalent_t ] ) . we have made several attempts at a new formulation , but we were not able to develop a satisfactory algorithm , providing a meaningful measure of the plastic strain on md time scales based on the equivalent plastic strain integral ( [ eq_equivalent_t ] ) . we found that we could eliminate the anomalies due to fluctuations or the anisotropy , but not both simultaneously in a robust manner . we have therefore turned to a different quantification of the plastic strain . certainly , the full deviatoric plastic strain tensor is a measure of the plastic flow , conjugate to the deviator stress . its rate of increase is given by the traceless part of eq . ( [ eq_plastic ] ) . typically , the rate would be integrated in a cumulative fashion , but we will not do so . the nature of our simulations is such that at the continuum level plastic flow is only expected in one direction , so any sign reversal may be attributed to fluctuations . we then calculate @xmath118 where the plastic strain rate is given by eq . ( [ eq_plastic ] ) . we emphasize again that no absolute value is taken , so fluctuations cancel . then in order to have a scalar quantification of the plastic strain , we compute the @xmath119 invariant , normalized as the equivalent plastic strain would be : @xmath120 we must stress that this quantity is not equal to the equivalent plastic strain commonly used in plasticity , except in the extraordinary case of monotonic isotropic plasticity . it is not conjugate to the von mises stress , for example , in our md simulations . nevertheless , it is a useful qualitative measure of the degree of plasticity , and it allows us to compare the plastic response as the system is loaded in different ways and we call it here _ equivalent plastic strain _ for simplicity . the evolution of the equivalent plastic strain during uniaxial and biaxial expansion is plotted in fig . [ fig5 ] . in the triaxial case it is essentially zero , as expected by symmetry , and therefore it has not been plotted . in practice , the stress during triaxial expansion has only a negligibly small fluctuating shear component , so the calculated elastic shear strains are very small , too . since neither the box strain nor the elastic strain has an appreciable shear component , the equivalent plastic strain is found to be zero . now , once the tensors for both the stress and the plastic strain are derived , ( actually only the diagonal terms of the plastic strain are needed ) , the plastic work can be calculated , @xmath121 ( see fig . [ fig6 ] ) . it should be compared with the temperature from the same simulations , fig . note , that in these simulations , when the dilational strain is applied , the thermostat is turned off and thus the temperature is allowed to change . first the system cools in the elastic regime due to adiabatic cooling on expansion , but when plastic deformations begin , work is done in the system resulting in heating . we find that the increase in plastic work does not match exactly with the temperature . in principle we expect several effects to contribute to this difference : the surface energy of the void , the defect formation energies for dislocations and point defects , further adiabatic cooling , and any error in calculating the elastic energy or strain from the stress . using the best data available to bound the contributions from surface , defect and adiabatic cooling energies , we find that there remains an energy deficit that we attribute to an error in the calculated elastic energy . the error comes from the use of the average stress despite stress inhomogeneity in the system due to the void : in the plastic work the product of plastic strain and stress is calculated with averaged quantities , while in the temperature the product is calculated at level of each atom and averaged afterward . we now consider the volume and shape evolution of the void . during the md simulations undergoing expansion , the surface of the void is determined by finding individual atoms that belong to the surface . this is done by creating a fine two - dimensional mesh , in which each mesh point corresponds to spherical angular coordinates @xmath122 . an atom is found to represent the surface at each point of the mesh , with some atoms representing multiple mesh points . in particular , taking the origin to be the center of the void , within the solid angle associated with each mesh point , the atom that is closest to the origin is defined to be the surface atom at that mesh point . there are , however , some uncertainties related to this method . if the mesh is too dense with its size diverging it can capture almost all the atoms in the system . on the other hand if it is too sparse , it may neglect some surface atoms , especially when void is anisotropic , non - spherical , and has some sharp edges in it . therefore we introduced a width to each of the atoms by drawing a circle around it that implies a width @xmath123 to the angles , so that one atom can occupy several mesh points in a fine mesh . we have typically 75 - 100 points for each angular coordinate , giving a total of 5625 - 10000 mesh points . in the surface of the void there are typically few thousand atoms . besides introducing the width to the atoms , we also select atoms based on their radial distance : if an atom has much larger radial distance @xmath124 compared to its neighbors it is neglected in order not to capture atoms that do not belong to the surface . once the surface atoms are identified , the surface is tessellated using a generalization of the delaunay triangulation method . @xcite the delaunay triangulation is an optimal triangulation of a collection of points in our case atoms on the plane . it is optimal roughly in the sense that the aspect ratio of the triangles is as near to unity as possible ; more precisely , the delaunay theorem guarantees that there is a unique ( up to degeneracy ) triangulation such that if any triangle in the triangulation is circumscribed by a circle , none of the other points will be in the interior of the circle . the delaunay theorem , as formulated , does not apply to points on a curved surface . in fact , there appears to be a topological obstruction to the existence of a unique , optimal triangulation on a closed surface when the euler character is non - zero . nevertheless , it is possible to extend the delaunay triangulation algorithm to achieve a locally optimal triangulation almost everywhere . the approach we have taken is to project the points patchwise onto flat surfaces . in particular , stereographic projections are used to project the upper and lower hemispheres separately onto planes . cylindrical coordinates are used to project the equatorial region to a cylinder . the delaunay algorithm is used to triangulate each of these projected regions . the patches overlap at latitudes of @xmath125 where the projections are not too distorted . the three patches are sewn together using a simple advancing front triangulation at the boundaries . using this triangulation , the volume of the void can be calculated precisely by summing up the volumes of the tetrahedra with one apex at the center of the void and the opposing face on the void surface . as we shall see below , the void shape evolves to be far from spherical . therefore the approximation of the void surface by triangles captures the shape better than just assuming it to be spherical and using only the solid angles and radial distances of a sphere , when calculating the volume of the void . an advantage of this method is that if some atom , which should be taken into account , is missing from the surface of the void , its position is filled with the triangles created by its neighboring atoms , and thus the `` hole '' is well approximated by its neighbors . in fig . [ fig8 ] the porosity or the ratio between the volume of the void and the total volume of the system , @xmath126 is plotted for a fraction of the simulations of the strain - rates @xmath36/sec , @xmath37/sec , @xmath38/sec , and @xmath8/sec . it should be noted that in order to get information about the positions of the atoms for the strain values of the interest ( _ i.e. _ , close to and after the onset of plasticity ) the calculations were restarted from already expanded system . after the restart the expansion was applied with the same strain - rate as earlier but now to the already expanded system , thus the strain - rate was increased by a few percent compared to the original [ since @xmath127 in eq . ( [ eq_update ] ) was the restart value ] . therefore these simulations are not precisely comparable with the ones plotted in figs . [ fig1]-[fig7 ] , where the continuum quantities are shown . in these plots , as for the mean plastic strain , in most cases first the void grows exponentially and then ( if the calculation has been carried out that long ) it changes to a linear growth , see especially fig . [ fig8](b ) and the strain - rates @xmath37/sec and @xmath128/sec there . the cross - over to the linear growth happens at the same point as the rate of decrease of the mean stress slows . although it is beyond the scope of this article to go into the analysis , the reduction in the growth rate coincides with the point at which the dislocation density along the shortest distance between the void and its periodic image ( at the apices of the faceted void , cf.section [ voidshape ] ) reaches saturation , and the nature of the dislocation activity changes dramatically . this can be interpreted as a finite size effect as the void approaches the boundary of the simulation box or as a start of the coalescence process of the void with its own periodic image . the void - void interactions and the coalescence process of two voids in a less restrictive geometry will be presented elsewhere . @xcite the shapes of the porosity curves as a whole can be compared with the mean plastic strain plotted in fig . although the volume of the void is not calculated throughout the whole simulation , one sees easily , that there is correspondence between mean plastic strain and the volume of the void . there is of course an offset at the elastic part of the simulations , since mean plastic strain equals zero then , but the initial volume of the void is finite . the correspondence will be revisited and studied more carefully in section [ summary ] . however , it can be concluded already here that the macroscopic quantity mean plastic strain captures the microscopic behavior of the void growth very well . effects such as the excess volume associated with defects are negligible . this also means that the matrix material is plastically incompressible , the dilation comes from the void growth , and thus it is consistent with the gurson type of continuum models . @xcite let us now look at the shape evolution of the void in more detail . in fig . [ fig9 ] snapshots of the void are shown from uniaxial expansion at the strain - rate @xmath43/sec . there are several interesting aspects in the snapshots . in the first two snapshots at strains @xmath129 and @xmath130 when the system still behaves elastically , the void is expanded in the @xmath131-direction , which is the direction of the strain . however , after that the void makes a rapid shape change and becomes more extended in the transverse @xmath132 and @xmath133-directions , _ i.e. _ , the strain - free directions , than @xmath131-direction . this prolate - to - oblate transition may be counterintuitive , but the behavior has been observed previously in continuum calculations,@xcite and it has been related to the appearance of shallow dimples in the fractography studies of ductile fracture surfaces in low triaxiality conditions . see also studies of non - spherical voids . @xcite for example , budiansky _ et al . _ , ref . , investigated void shape change in a non - linear viscous plastic flow model . they explained the oblate growth of voids under uniaxial loading as due to a non - linear amplification of the shear stress on the surface of the void , with the maximal void growth rate at the locations of maximal von mises stress : the equator . their analysis does not apply directly to our simulations since they neglect elasticity , and the non - linear viscous solid model they have used is not expected to be a precise description of the plastic flow early in our md simulation when the prolate - to - oblate transition takes place . furthermore , it is not clear from our simulations what value should be assigned to the strain - rate exponent that controls the non - linearity in the model of budiansky _ et al_. , although a large value is reasonable . despite these differences , the same localization of plastic flow to the equator of the void and the transition to an oblate shape does appear in both our simulations and the viscous solid model of budiansky _ et al_. following some additional expansion beyond the transition , the void begins to become faceted , as visible in the last three snapshots . it should be mentioned that the anisotropy visible in this uniaxial case is less pronounced in the biaxial case . the cases with the hydrostatic loading are the most isotropic and the octahedral shape , somewhat visible in fig . [ fig9](f ) , becomes more pronounced . @xcite the octahedral shape has been seen in spallation experiments in the fcc metal aluminum,@xcite and also in experiments on the equilibrium void shape of another material , silicon , too , where it has been used to calculate anisotropic surface energies through an inverse wulff construction . @xcite in void growth associated with dynamic fracture in copper , several effects contribute to the faceting : the low surface energy and high ad - atom energy of the \{111 } surfaces common to fcc metals , the high anisotropy of the copper elastic constants ( a=3.21 ) , and the \{111 } dislocation glide systems . these effects are analyzed in detail elsewhere . @xcite in order to characterize the shape change of the void not only qualitatively and visually , but also quantitatively , multipole moments of the void shape are calculated . to the best of our knowledge , this is the first time that multipole moments have been used to characterize surface shape . they are a powerful way to quantify the evolution of the complex surface containing thousands of atoms , and they are suitable for use in continuum models and experimental void characterization as well . using spherical harmonics , @xmath134 expressed as polynomials of cartesian coordinates , in contrast to more commonly used trigonometric forms , @xcite we are able to define different multipole moments of the void based on its surface atoms : @xmath135 where the mean square radius @xmath136 . this is in contrast to the volume integral more commonly used when calculating multipole moments . the axial index of the moment ranges @xmath137 , and for each @xmath138 except @xmath139 , @xmath140 has both real and imaginary parts . here we concentrate on @xmath141 , and 4 . only the positive moments of @xmath138 are calculated , since the negative ones are related by @xmath142 in all 24 different terms are calculated . the polynomial forms used here of the moments are listed in the appendix . the moments @xmath140 are not rotationally invariant , but depend on the way the coordinates @xmath131 , @xmath132 , and @xmath133 are chosen . the set of @xmath143 moments at fixed @xmath14 form an irreducible representation of the so(3 ) rotation group , and are mixed by rotations according to the usual transformation rules . they may be combined into a single rotationally invariant combination for each @xmath14 according to @xmath144^{1/2 } , \label{eq_ql}\ ] ] see _ e.g. _ ref . . their use drastically reduces the amount of data to be shown . only the positive @xmath138 s are needed for @xmath145 due to the square of the norm of @xmath140 and eq . ( [ eq_rel ] ) . technically the calculation of the multipole moments has been done using the information about the shape of the void obtained from the surface triangulation procedure explained earlier . as in the calculation of the void volume , some refinements have been introduced to reduce the uncertainty in the values of the moments that arises from single atoms near the threshold for inclusion as surface atoms . these borderline atoms can appear intermittently on the void surface during the growth , and the tessellation is used to minimize their effect on the moments . in calculating the volume of the void the triangulation gave one face of the tetrahedra that acted as small volume elements for the total volume . here the triangulation is used to weight the atoms by the amount of solid angle associated with each surface atom . in particular , each triangle in the tessalation contributes one third of its projected solid angle to each of the three atoms that make up its vertices , where the solid angle of a triangle is computed using the formula @xmath146 , where @xmath147 and @xmath148 for @xmath149 ( mod 3 ) and where @xmath150 is the unit vector in the direction of the @xmath151 vertex of the triangle . @xcite the weight of each atom is the sum of these solid angle contributions . this reduces the sensitivity of moments to the atomic discretization of the surface , since evanescent atoms that occasionally appear and disappear from the fluctuating surface only make a small , local change to the value of @xmath152 . it may be of interest to note that in the course of the development of these surface multipole moments , several different variations on the definition of the moments were tried . the definition presented here ( [ eq_qlm ] ) produced substantially less noise ( up to a factor of 5 less noise ) than the other definitions we tried , even though they all showed the same trends in the shape evolution . using these weights for the atoms and after normalizing each atom s @xmath153 coordinate by its distance @xmath154 from the center of the void all the terms in eqs . ( [ eq_ql1])([eq_ql4 ] ) , and eq . ( [ eq_ql ] ) , are calculated . the center of the void is defined to be the point where the three components of @xmath155 , as given by eq . ( [ eq_ql1 ] ) , are zero . due to space limitations , only a fraction of the multipole moment data is shown here . in fig . [ fig10](a ) the quadrupole moments @xmath156 for all the positive @xmath138 of the void are shown in the uniaxial case for the strain - rate @xmath43/sec . this is the same simulation as the snapshots in fig . indeed , the quadrupole moments are able to represent numerically the shape changes one sees in the snapshots . in figs . [ fig9](a ) and ( b ) at strains @xmath129 and @xmath130 the void is elongated to the direction of the load , which is visible as @xmath157 between strains @xmath158 [ fig . [ fig9](c ) ] and @xmath159 [ fig . [ fig9](d ) ] the void is extended more transverse to the direction of the strain , thus @xmath160 , and later it starts to becomes more of octahedral shape and the absolute value of @xmath161 saturates . figures [ fig10](b)(d ) show the rotational invariant multipole moments @xmath145 , eq . ( [ eq_ql ] ) , in cases with uniaxial , biaxial , and triaxial loading , respectively . each of the cases has strain - rate @xmath43/sec . in the plots it is clear that the behavior that the quadrupole moment has first a non - zero value and then makes a rapid dip but returns back to a non - zero value due to the transverse elongation is the strongest in the uniaxial case . on the other hand the octahedral shape measured by the @xmath162 is more pronounced in the biaxial and triaxial cases than in the uniaxial case as explained qualitatively earlier . hence we find that the multipole moments introduce a good method to measure the shape changes of the void . the non - zero values for @xmath163 , as well as @xmath161 in other cases than uniaxial , indicate that the void is not ( cubically ) symmetric in these simulations . it should be mentioned that these first four moments are enough to characterize the shapes of the void and the higher moments contain little relevant information . this was checked by creating a three dimensional surface based on the moment values and drawing it in the same figure with the actual positions of the surface atoms using a standard commercial visualization program . the surfaces overlapped very well . based on the data shown earlier in this article for the shape and volume changes of the void as well as the stress - strain behavior and the stress - triaxiality , it is evident that the uniaxial loading raises many interesting aspects to be studied in more detail . therefore we now concentrate on the uniaxial case when summarizing how the void evolves and how the evolution is related to the stress - triaxiality as the system is expanded . by plotting most of the measured values together in one figure for the uniaxial simulation at the strain - rate @xmath43/sec , it is possible to compare the evolution sequence and investigate causality , see fig . [ fig11](a ) . for clarity , we have chosen not to plot many quantities in the figure , _ e.g. _ plastic strain , plastic work , and temperature . however , their concomitant behaviors are included in the verbal explanation below and shown in previous figures . the data shown in this figure are from the restarted simulation ( see the explanation near fig . [ fig8 ] ) , as are the data in figs . [ fig8]-[fig10 ] . the mean stress and stress - triaxiality data are from that simulation , too , and thus are different from the data shown in figs . [ fig1](a ) and [ fig3](a ) . in any case , the overall behavior stays the same as well as the other details as the system size , _ etc . _ the evolution of the void and the system s stress - strain behavior can be divided in three or even four different regions . the first region is when the system expands elastically . the mean and von mises stresses increase smoothly , nearly linearly , and the stress - triaxiality stays nearly constant . through the elastic region the void volume fraction remains nearly constant , too . it is not exactly constant , since due to the free surface of the void , the elastic expansion is a bit greater at the surface of the void compared to the total system . trivially the mean and equivalent plastic strains as well as the plastic work are equal to zero in the elastic region , and temperature decreases in the system . the quadrupole moment has a non - zero value because of the elongation in the direction of the strain . the second region begins at what we call the yield point , _ i.e. _ , the onset of rapid growth of the void facilitated by plastic deformation . heterogeneous nucleation of dislocations at the void surface is the primary mechanism for plasticity in the simulation , and thus it is at this point that the measured quantities start to deviate from their elastic behavior . the mean stress begins to plateau here , but fluctuating somewhat . the early departure from elastic behavior prior to the plateau is much less pronounced . the change in the void shape begins just at the point the mean stress deviates from elastic behavior : @xmath164 goes rapidly from the positive value acquired during elastic expansion to a negative value , _ i.e. _ , from a prolate shape ( elongated in the direction of the strain ) to an oblate shape ( expanded in the transverse directions ) . @xmath161 , on the other hand , drops from a positive value , almost reaching zero at the prolate - to - oblate transition point ( @xmath165% ) and rising even larger value after that [ in fact , the oblate shape is somewhat more pronounced than the earlier prolate shape , seen as a larger absolute value of @xmath164 in fig . [ fig10](a ) ] . when @xmath161 starts to drop @xmath162 starts to rise . then after the prolate - to - oblate transition point , @xmath162 begins to saturate . at a strain of @xmath166 5.55% , @xmath161 is 1.5 times as large as its value at the end of the elastic phase ( @xmath166 5.25% ) . mean plastic strain , equivalent plastic strain , plastic work , and temperature increase together with porosity . their increase starts immediately at the yield point , _ i.e. _ , when the mean stress first deviates from the elastic behavior . a bit later than the plastic strain , equivalent plastic strain , plastic work , and temperature , the stress triaxiality increases simultaneously with the first substantial drop in the von mises stress . the fact that the increase in stress - triaxiality follows later is dependent on how the ratio between mean stress and von mises stress develops , as discussed earlier . in fig . [ fig3](a ) and in the data reported in table [ tab1 ] the stress - triaxiality started to increase simultaneously with the mean and von mises stresses deviating from the elastic behavior . the increase of stress - triaxiality is caused by von mises stress plummeting in contrast to nearly constant mean stress . the drop in von mises stress follows from the flow of dislocations nucleated at the void and from the relaxation of the shear stress of the system due to the flow . the third region is when the void fraction , mean plastic strain , equivalent plastic strain , plastic work , and temperature switch from rapid increase to a linear growth or even saturate . subsequently the increase of the stress - triaxiality slows down and plateaus . the value at the plateau is related in continuum models to the ratio of the mean stress threshold for void growth to the flow stress . at the plateau von mises stress saturates at a small value , close to the tensile strength , and the shape of the void starts to become more of octahedral shape although having a non - zero quadrupole moment , too . hence at the second and third regions the mean stress is nearly constant , but von mises stress and the stress - triaxiality changes . a conclusion might be that once the threshold for void growth is reached , the population of dislocations rises sufficiently to relax the shear stress quite effectively and it drops to a low value ( the flow stress ) ; the mean stress , on the other hand , plateaus since it is relaxed by void growth and requires that the stress at the void surface be sufficiently high to continue to nucleate dislocations . the fourth region is the failure , when the system breaks , and it is not studied here . in order to see if the rapid changes studied above are due to the smallness of the size of the void we have done one additional simulation with a system in which the initial void radius is half that in the other simulations ; otherwise , the system size is the same , see fig . [ fig11](b ) . for the small void simulation , uniaxial strain at a strain - rate @xmath43/sec has been used . here the difference is that the quadrupole moment suffers stronger fluctuation due to the smallness of the void , where each of the surface atom contribute more to its value and therefore is even more sensitive to the selection criterion of surface atoms , and also the shape changes are harder to determine based on the @xmath164 behavior . the other main difference is that the growth of the void is linear all the time . also the changes in stress - triaxiality in fact advances the porosity when saturating , and the mean stress does not fluctuate but drops more rapidly ( this can be compared with the case without the void , where the mean stress drops abruptly ) . we have also compared the mean plastic strain and the void volume fraction calculations . in continuum solid mechanics , it is assumed that solid materials are plastically incompressible . any local dilation , as indicated by a change in the mean strain , is attributed either to elastic dilation or to a change in the porosity of the material , where the porosity is equated to the void volume fraction . the porosity @xmath167 and the mean plastic strain @xmath168 are then related according to the equation @xcite @xmath169 where the dots denote time derivatives . integration with respect to time , porosity from @xmath170 to @xmath167 , and mean plastic strain from zero to @xmath168 , gives @xmath171 where @xmath170 is the initial porosity . it is interesting to check whether this relationship holds for the md simulation , where other effects such as excess volume associated with dislocation cores , vacancies or other defects could require corrections . in comparing the porosity inferred from the mean plastic strain and that calculated directly as the void fraction , the agreement is very good . the trends are in excellent agreement , but there is a small discrepancy between the curves , so that the porosity from the mean plastic strain is overestimated . we believe that the discrepancy arises because of the void surface . in calculating the void fraction , we have defined the void surface to pass through the center of the surface atoms . however , the properties of the surface atoms are distinct from the bulk atoms . therefore , there is some ambiguity in where the surface should be placed , and a small uniform shift @xmath172 of the surface radially into the bulk is enough to account for the discrepancy . in fig . [ fig12 ] we have plotted the comparison of the porosity from the mean plastic strain , eq . ( [ eq_porosity ] ) , and from the void fraction , eq . ( [ eq_void_frac ] ) , using a constant radius increase @xmath173 , where @xmath174 is the lattice constant for the void volume calculation . the correction for the void size , @xmath175 , is a fit parameter and it varies for different strain - rates and slightly for different loading modes , but is always positive and of the order of the lattice constant , @xmath174 . it should be noted , too , that by taylor expanding eq . ( [ eq_porosity ] ) and discarding the higher order terms , it becomes @xmath176 indicating a linear correspondence between @xmath167 and @xmath168 as long as the void fraction is small . in this article void growth in copper has been studied in a high range of strain - rates at the atomistic level . the model has been designed to simulate the growth of a void nucleating from a very weakly bound inclusion during strain - controlled dynamic fracture . in order to see the effect of various modes of expansion and the related stress - triaxiality , three different modes have been applied , namely uniaxial , biaxial , and triaxial . the molecular dynamics method developed here has been shown to be efficient enough to explore the different loading conditions and strain - rates varying over four orders of magnitude . a uniform expansive loading of a system with periodic boundary conditions has been implemented using a well defined scaling matrix method . for the longest calculations , the md method was parallelized successfully . the macroscopic observables mean stress , von mises stress , stress - triaxiality , mean plastic strain , equivalent plastic strain , plastic work , and temperature have been calculated and compared with the microscopic quantities measured at the atomistic - level , such as the volume of the void and its shape change . a method to describe the shape changes in the void is introduced and employed , namely calculations of the multipole moments of the void based on spherical harmonics in polynomial , not trigonometric , form . when calculating the volume of the void with an unknown shape or defining solid surface for the multipole moment calculation a useful method , namely optimal triangular tessellation , has been introduced . this method has been extended from the usual planar case to non - planar objects such as the surface of the void . when the different measured quantities are compared with each other during an md simulation in uniaxial expansion , it is found that at early stages of plasticity von mises stress , and thus also stress - triaxiality , plays a more significant role to the void growth and its shape change than expected . on the other hand , most of the macroscopic plastic quantities as mean and equivalent plastic strain as well as plastic work and temperature , seem to be more dependent on the simultaneous saturation of the mean stress . these calculations show a counter - intuitive behavior , observed previously in continuum void growth modeling , @xcite that a prolate - to - oblate transition occurs . when the system starts to yield , the expansion of the void switches from its original elastic extension in the direction of the load to transverse plastic expansion . the yield stress values for the lowest strain - rates @xmath0/sec are in reasonable agreement with the experimentally measured spall strength . @xcite the fact that mean plastic strain can be mapped to the growth of the void is consistent with continuum models . @xcite this study leaves many open questions . for instance related to the void growth are the dislocations , which occur when the system yields . since the fcc crystal studied here is perfect apart from the void , the dislocations form from void s surface . they are also responsible for its growth by carrying material away . thus the characterization of plasticity surrounding a growing void at the level of dislocations should be investigated , too , especially with respect to the stress - triaxiality . these investigations are underway . @xcite their results are beyond the scope of this article , other than to mention that the identification of the yield point in this article does indeed correspond to the point of initial nucleation of dislocations . another topic that is beyond the scope of this article and needs further investigation , but is closely related to the studies here , is the quantitative connection between the shear stress , and thus the mode of the loading , and the onset of the void growth and the resulting change in the stress state . other areas where this study can be extended are different materials including different lattice structures such as body - centered cubic ( bcc ) lattices ; in the uniaxial case other orientations of the lattice as @xmath177 and @xmath178 ; continuously changing stress - triaxiality in order to create the full yield surface to the stress space similarly as in gurson type of continuum studies ; @xcite to include grain boundaries , defects , pre - existing dislocations , several voids , etc . this work was performed under the auspices of the u.s . of energy by the university of california , lawrence livermore national laboratory , under contract number w-7405-eng-48 . the authors would like to thank dr . richard becker bringing to our attention refs . and that find the prolate - oblate transition in continuum modeling . the 24 different polynomial terms of the multipole moments used in this study are listed below . @xcite here the conventional notation is used , so that the principal axis of the coordinates is the @xmath133-direction . when these formulas are used in interpreting the void shape evolution , the principal axis is the direction with uniaxial loading , _ i.e. _ , the @xmath131-coordinate in the article . similarly the load free directions @xmath132 and @xmath133 correspond to @xmath131 and @xmath132 below .

we describe methods for proving upper and lower bounds on infinite - time averages in deterministic dynamical systems and on stationary expectations in stochastic systems . the dynamics and the quantities to be bounded are assumed to be polynomial functions of the state variables . the methods are computer - assisted , using sum - of - squares polynomials to formulate sufficient conditions that can be checked by semidefinite programming . in the deterministic case , we seek tight bounds that apply to particular local attractors . an obstacle to proving such bounds is that they do not hold globally ; they are generally violated by trajectories starting outside the local basin of attraction . we describe two closely related ways past this obstacle : one that requires knowing a subset of the basin of attraction , and another that considers the zero - noise limit of the corresponding stochastic system . the bounding methods are illustrated using the van der pol oscillator . we bound deterministic averages on the attracting limit cycle above and below to within 1% , which requires a lower bound that does not hold for the unstable fixed point at the origin . we obtain similarly tight upper and lower bounds on stochastic expectations for a range of noise amplitudes . limitations of our methods for certain types of deterministic systems are discussed , along with prospects for improvement .

1512.05599

in the study of dynamical systems with complicated and possibly chaotic dynamics , average quantities are often of more interest than any particular solution trajectory . this is partly because of the difficulty of computing a trajectory precisely and partly because average quantities are more important in many applications . for instance , one might seek time - averaged drag forces in a model of an oil pipeline or ensemble - averaged temperatures in a stochastic climate model . one way to estimate averages quantitatively is to integrate the dynamical system numerically and average over the resulting particular solution . such direct numerical simulations are often straightforward , but the accuracy of the result is not guaranteed unless errors are rigorously controlled . moreover , even a perfectly accurate solution does not give information about the different trajectories that result from different initial conditions . finally , the computational cost of computing well - converged averages is often prohibitive , especially in systems that are high - dimensional or stochastic . a complementary approach , which we pursue here , is to prove bounds on average quantities directly from the governing equations . an advantage over numerical integration is that bounds can be proven without knowing any solution trajectories . furthermore , they can be proven for all possible trajectories at once , or for all trajectories within a given region of state space . on the other hand , it is generally difficult to prove bounds that are tight enough to give good estimates of the average quantities being bounded . the aim of this work is to develop methods for proving bounds that are tight , meaning that the upper and lower bounds are equal or nearly so . we assume that the quantity of interest can be described by a function @xmath0 , where @xmath1 is the state vector of a dynamical system , and we seek to bound averages of @xmath2 . in deterministic systems , we consider averages over infinite time , @xmath3 \,dt.\ ] ] if the above limit does not exist it can be replaced by limit superior or inferior for the upper and lower bound problems , respectively . the value of @xmath4 depends in general on which trajectory @xmath5 is being averaged over . in stochastic systems with a stationary probability distribution @xmath6 , we consider stationary ensemble averages , @xmath7 one obstacle to proving tight bounds is , for reasons described shortly , the need to determine whether certain complicated expressions are sign - definite . as proposed in @xcite , this can be done systematically with computer assistance for finite - dimensional systems with polynomial dynamics . the main idea , described further in [ ss : sosboundsreview ] , is to construct a polynomial whose non - negativity implies the desired bound . by the methods of sum - of - squares ( sos ) programming @xcite , a sufficient condition for this non - negativity can then be posed as a semidefinite program ( sdp ) . in deterministic systems there is a second obstacle to proving tight bounds . it is generally easiest to construct bounds that hold for all possible initial conditions . sometimes this is desired , but other times one is interested in a particular local attractor , and bounds holding globally are not generally tight for averages over a local attractor . in this sense , the local bound is spoiled by any other invariant structure in the state space , such as another attractor or an unstable fixed point . we pursue two ways of obtaining tight bounds specific to a local attractor . the first is to enforce conditions that imply the bound only on an absorbing set around the attractor , thereby omitting other invariant structures . the second is to add noise to the system and prove bounds for ensemble averages in the vanishing noise limit . if the system is stochastically stable , then under certain conditions this limit will agree with the corresponding deterministic time average @xcite . note that these ideas are applicable irrespectively of any special structure in the system ( such as hamiltonian ) , and can in principle be applied to systems of arbitrarily high but finite dimension . throughout this work , we illustrate the methods described using the van der pol oscillator @xcite , which can be written as @xmath8 where a dot denotes @xmath9 . the parameter @xmath10 sets the strength of the nonlinear damping . there is a limit cycle that attracts all trajectories except the unstable fixed point at the origin ( figure [ f : vdpphaseportrait ] ) , and the global invariant set is composed of the limit cycle and the fixed point . the system is a standard example of a nonlinear oscillator and has been studied extensively , including with stochastic forcing @xcite . here we find nearly tight bounds on averages of @xmath11 both with and without noise . the rest of this work is organized as follows . section [ ss : sosboundsreview ] reviews the framework of @xcite for bounding deterministic averages and uses it to find upper bounds on @xmath12 in the van der pol system . section [ ss : localbounds ] extends the framework to give attractor - specific bounds , which we use to find lower bounds on @xmath12 over the van der pol limit cycle . these lower bounds are larger than zero and thus do not apply to the unstable fixed point . the bounding methods for stochastic dynamical systems are described in [ ss : generalnoiseformulation ] , and they are specialized to the case of small and vanishing noise in [ ss : smallnoise][ss : vanishingnoise ] . the methods of these sections give bounds on @xmath13 in the van der pol example for a range of noise amplitudes . section [ s : furthercomments ] discusses the limitations of our methods and gives ideas for improvement , and [ s : conclusion ] offers concluding remarks . consider an autonomous dynamical system @xmath14 in which all trajectories @xmath5 remain bounded as @xmath15 . we assume nothing special about the structure of @xmath16 except that the system is bounded , although eventually we restrict attention to polynomial @xmath16 . we wish to prove upper and lower bounds on @xmath4 , the average of a function @xmath0 over a trajectory @xmath5 . in applications , @xmath2 can be chosen as a quantity of interest for the system . unless the system has a single attractor that is globally asymptotically stable , different trajectories can give different values of @xmath4 . in this section we seek global bounds on @xmath4 , meaning they hold for every possible value of @xmath4 . suppose we wish to prove a constant lower bound @xmath17 on all possible @xmath4 , @xmath18 central to our method will be a suitably chosen differentiable _ storage function _ @xmath19 . no matter what @xmath20 is chosen , @xmath21 along any bounded trajectory because @xmath22-v\left[{{\mathbf{x}}}(0)\right]\right ) = 0.\ ] ] the desired bound is thus equivalent to the inequality@xmath23 for any differentiable @xmath19 . the above time average can not be evaluated without knowing trajectories @xmath5 , but a sufficient condition for the inequality to hold is for it to hold pointwise for all @xmath24 . thus , we will have proven that @xmath25 if we can find any differentiable @xmath19 such that @xmath26 an upper bound @xmath27 can be proven in a similar way by reversing the inequality sign in the pointwise sufficient condition . the following proposition summarizes both conditions . [ t : proposition1 ] let @xmath28 with @xmath29 be a dynamical system whose trajectories are bounded forward in time , let @xmath0 be a scalar function and let @xmath30 be its time average defined as in . if there exist differentiable functions @xmath31 , @xmath32 , and constants @xmath33 , @xmath17 such that @xmath34 then along any trajectory @xmath5 , @xmath35 [ r : remark1 ] in the statement of proposition [ t : proposition1 ] we have assumed that the time average @xmath30 exists . should the limit in not converge , the proposition holds if we take the limit superior when computing the upper bound , and the limit inferior when computing the lower bound . that is , for any trajectory @xmath5 , @xmath36\,dt \leq \limsup_{t\to\infty } \int_{0}^{t } { \varphi}[{\mathbf{x}}(t)]\,dt \leq u.\ ] ] see @xcite for a discussion of when infinite - time averages do or do not converge . there are two difficulties in applying the above proposition to yield good bounds @xmath33 and @xmath17 . the first is choosing the storage functions @xmath37 and @xmath38 . the second is checking whether @xmath39 and @xmath40 for candidate storage functions and bound values . these difficulties can be prohibitive in general , but the task is greatly simplified when @xmath16 and @xmath2 are polynomials of the state variables @xmath41 . in the rest of this work we assume that @xmath16 and @xmath2 are polynomials . if the chosen @xmath37 and @xmath38 are also polynomials , then so are @xmath42 and @xmath43 . checking the sufficient conditions and thus amounts to verifying the non - negativity of polynomial expressions . while this is an np - hard problem , the computational complexity can be significantly reduced by replacing the conditions @xmath44 and @xmath45 with the stronger conditions that @xmath43 and @xmath46 belong to the set @xmath47 of polynomials that are sums - of - squares ( sos ) . a polynomial @xmath48 is said to be sos if it can be expressed as the sum of squares of some other polynomials that is , if there exist polynomials @xmath49 such that @xmath50 to prove the bounds @xmath33 and @xmath17 it suffices to find polynomials @xmath37 and @xmath38 such that @xmath51 and @xmath52 . the best bounds that can be proven in this framework are @xmath53 for the storage functions @xmath37 and @xmath38 , we must specify polynomial ansatze with undetermined coefficients . the decision variables in the upper bound optimization , for instance , are @xmath33 and the coefficients of @xmath37 . [ [ computational - methods . ] ] computational methods . + + + + + + + + + + + + + + + + + + + + + + optimization problems with sos constraints such as can be solved numerically using the methods of sos programming . the main idea behind sos programming is that every polynomial can be represented as a symmetric matrix ( after defining a suitable polynomial basis set ) , and this matrix can be positive semidefinite if and only if the polynomial admits a sos decomposition @xcite . constraints involving sos polynomials , including additional equality and inequality constraints , can be posed as equality and inequality conditions on symmetric matrices . an optimization problem with constraints of this type is known as a semidefinite program ( sdp ) . a number of efficient computer solvers are available for sdps ( e.g. @xcite ) , and the software packages yalmip @xcite or sostools @xcite can assist in formulating sos constraints as sdps . more details on convex optimization and the link between sos polynomials and sdps can be found in @xcite , while examples using sos programming to study dynamical systems can be found in @xcite . for our numerical implementation , we used the sos module of yalmip @xcite to transform sos optimization problems into sdps . to solve the sdps , we used the multiple - precision solver sdpa - gmp @xcite for the following reasons . first , the sdps we solved were badly conditioned even for modest polynomial degrees , and none of the standard double - precision solvers we tested converged reliably . this issue could be resolved by carefully rescaling the system , but there is no systematic rule to determine a suitable rescaling . second , working in multiple - precision offers a simple check on results : increase the precision , and confirm that the bounds @xmath33 or @xmath17 change very little . we have done this for all bounds reported here . appendix [ s : rigorousbounds ] describes additional checks on numerical results , including how sdp computations could be part of fully rigorous computer - assisted proofs . [ [ ss : exampledeterministicvdp_ub ] ] example : deterministic upper bounds for the van der pol limit cycle . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + to illustrate the application of proposition [ t : proposition1 ] , we have computed upper bounds @xmath54 in the van der pol system for various @xmath55 and four different polynomial degrees of @xmath37 . figure [ f : ubvdpsdpagmp ] shows the resulting @xmath33 , along with estimates of @xmath4 obtained by integrating over the limit cycle using the fourth - order runge - kutta method with fixed time steps . as expected , increasing the degree of the storage function @xmath37 gives a better ( smaller ) upper bound @xmath33 . for a given degree , the bounds become less tight as @xmath56 is raised because the shape of the limit cycle is more complicated ( cf . figure [ f : vdplimitcycles ] ) . the trivial lower bound @xmath57 is the best _ global _ lower bound possible for the van der pol system . it can not be improved because it is saturated by the trajectory staying at the unstable fixed point @xmath58 . however , if one is interested only in the trajectories that tend to the stable limit cycle , then the lower bound of zero is not tight . raising the lower bound requires a result that does not apply globally , but instead applies only to a subset of all possible trajectories . for the van der pol oscillator computed with the upper bound problem of for different degrees of @xmath37 . the time average @xmath30 obtained by numerical integration ( n.i . ) of the system is also shown.,scaledwidth=60.0% ] often , one is interested in the average of a function @xmath0 over only a subset of all possible trajectories , such as those tending to a particular attractor . if a bound on @xmath4 over such trajectories is to be tight , it should not apply to trajectories that tend to a different local attractor , nor should it apply to trajectories on unstable invariant structures that are not part of the attractor of interest ( such as unstable fixed points , unstable limit cycles , or basin boundaries ) . however , global bounds of the kind derived using proposition [ t : proposition1 ] are unlikely to be tight for a particular attractor since they must obey @xmath59 where @xmath5 is any trajectory of the dynamical system . suppose we wish to bound possible values of @xmath4 , not for all trajectories but only for trajectories that eventually enter and remain inside a given absorbing domain @xmath60 . to compute @xmath4 over any such trajectory , it suffices to begin averaging after the trajectory has permanently entered @xmath60 , so dynamics outside of @xmath60 can be ignored . this suggests the following modification of proposition [ t : proposition1 ] , where the inequality conditions are imposed only on @xmath60 and , consequently , the resulting bounds are proven only for trajectories that permanently enter @xmath60 . the comments made in remark [ r : remark1 ] still apply . [ t : proposition2 ] let @xmath28 be a dynamical system with @xmath29 , let @xmath0 be a scalar function , let @xmath30 be its time average defined as in , and let @xmath60 be an absorbing domain . if there exist differentiable functions @xmath31 , @xmath32 , and constants @xmath33 , @xmath17 such that @xmath61 then along any trajectory @xmath5 that permanently enters @xmath60 , @xmath35 applying proposition [ t : proposition2 ] to a particular system requires specifying an appropriate absorbing domain . for instance , if one is interested in a local attractor @xmath62 with basin @xmath63 , it suffices to choose any @xmath60 such that @xmath64 . bounds proven on @xmath60 then apply to all trajectories that approach the attractor or are part of the attractor , but they need not apply to trajectories outside @xmath63 . finding a mathematical description of @xmath60 for a given attractor is difficult in general , but this too can be done using sos programming @xcite . an example in which it is _ not _ difficult to choose @xmath60 is when the trajectory to be excluded from the bound is a repelling fixed point . in this case it suffices to choose @xmath65 for any small enough set @xmath66 around that point . if the absorbing domain can be specified as a semi - algebraic set that is , defined by a set of polynomial inequalities and equalities the conditions of proposition [ t : proposition2 ] can be checked by sos programming using the _ generalized @xmath67-procedure _ ( * ? ? ? * lemma 2.1 ) . for instance , suppose @xmath68 for some polynomial @xmath69 ; proving a lower bound for trajectories entering @xmath60 calls for sos conditions that imply @xmath70 when @xmath71 . for this to be true , it suffices that there exist @xmath72 such that @xmath73 . strengthening these non - negativity constraints to sos constraints and making similar arguments for the upper bound gives the following two sos programs : @xmath74 where polynomial ansatze are specified for @xmath37 , @xmath38 , and @xmath75 , and their free coefficients are the decision variables . using similar ideas , the @xmath67-procedure can be generalized to semi - algebraic @xmath60 defined by multiple polynomial inequalities and equality constraints @xcite . [ [ example - deterministic - lower - bounds - for - the - van - der - pol - limit - cycle . ] ] example : deterministic lower bounds for the van der pol limit cycle . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let us revisit the lower bound on @xmath12 for the van der pol oscillator . suppose we want a bound that does not apply to the unstable fixed point at the origin but applies to the other trajectories , all of which approach the limit cycle . if such a lower bound is perfectly tight , it will equal the value of @xmath4 on the limit cycle . to apply proposition [ t : proposition2 ] , we must specify an absorbing domain @xmath76 that contains the entire limit cycle but omits the origin . as described in appendix [ s : vdpabsorbingdomainproof ] , the set @xmath77 . figure [ f : lbvdp ] shows the lower bounds obtained by solving the lower bound program of using the absorbing domains @xmath78 and @xmath79 . at a given polynomial degree , using the smaller absorbing domain @xmath79 gives a better bound . results would likely be improved further by using @xmath60 that closely approximate the attractor . this suggests a two - step procedure : using sos techniques to find the smallest possible @xmath60 around the attractor of interest @xcite , and then solving the sos bounding programs . consider a stochastic dynamical system @xmath80 in which the random trajectories @xmath81 are bounded almost surely as @xmath82 . the standard vector wiener process @xmath83 is pre - multiplied by the matrix @xmath84 that describes the relative effect of each noise component @xmath85 on each state variable @xmath86 , and the overall noise strength is scaled by @xmath87 . we assume for simplicity that the noise is additive , meaning that @xmath88 does not vary with @xmath24 , but the present analysis extends with only minor modifications to multiplicative noise in either the it or stratonovich interpretation . we also assume that the system has reached statistical equilibrium , in which case the probability density of its trajectories , @xmath89 , decays at infinity and satisfies the stationary fokker planck equation @xmath90 suppose we wish to prove a constant lower bound @xmath91 , where @xmath92 is the stationary expectation of @xmath0 defined as in . we assume that such a stationary average exists , which is true for all @xmath0 that do nt grow too fast as @xmath93 ( precise statements can be found in @xcite ) . the subscript on @xmath92 indicates its dependence on the noise strength @xmath94 . since @xmath95 , this is equivalent to proving @xmath96 as in the deterministic case , our method of proof relies on a suitably chosen differentiable storage function @xmath19 . for any @xmath20 that does not grow too quickly as @xmath97 , the stationary expectation @xmath98 is zero because @xmath99d{{\mathbf{x}}}\\ & = \int_{{\mathbb{r}}^n}v\,{\nabla}\cdot ( { \varepsilon}{\mathbf{d}}{\nabla}\rho - { { \mathbf{f}}}\rho)d{{\mathbf{x}}}\\ & = 0 . \end{aligned}\ ] ] the third line above follows from the stationary fokker - planck equation . the second line follows from integration by parts , assuming that the boundary terms vanish that is , assuming @xmath100 where @xmath101 is the outwards unit normal to the sphere @xmath102 , and @xmath103 is the surface element . the above condition holds in many cases , including any case where @xmath20 is polynomial and @xmath104 decays exponentially at infinity . when this condition holds , then so does the equality , hence the inequality is equivalent to @xmath105 the above expectation can not be evaluated without knowing @xmath104 , but it is sufficient for the inequality to hold pointwise for all @xmath24 . the lower bound @xmath91 is therefore proven if we can find any differentiable @xmath19 satisfying the boundary integral condition and the pointwise inequality @xmath106 the same argument with a reversed inequality sign gives a sufficient condition for an upper bound @xmath33 on @xmath92 . we summarize these results in the following proposition , which is the stochastic analog of proposition [ t : proposition1 ] . [ t : theorem3 ] let @xmath107 with @xmath108 , @xmath109 , @xmath110 be a stochastic dynamical system in a statistically stationary state with probability distribution @xmath6 . if there exist differentiable functions @xmath37 , @xmath38 and constants @xmath33 , @xmath17 such that [ e : stochasticinequalities ] @xmath111 where @xmath112 , and if @xmath37 , @xmath38 grow slowly enough at infinity to each satisfy , then @xmath113 the inequality conditions and were derived by a different method in @xcite for the case where @xmath114 is the identity matrix ( hence @xmath115 ) . the conditions differ from their deterministic counterparts and , respectively , only in the addition of the diffusive terms @xmath116 and @xmath117 . similar results for non - negative @xmath2 were also stated in @xcite . like the deterministic bounds of [ ss : sosboundsreview][ss : localbounds ] , the stochastic bounds and storage functions in proposition [ t : theorem3 ] can be found numerically for a given @xmath94 if @xmath16 and @xmath2 are polynomials . letting @xmath37 and @xmath38 be polynomials also , we replace and with stronger sos constraints to obtain the sos programs @xmath118\in \sigma , \end{gathered } & & \begin{gathered } \max_{v_l } \quad l \\ \text{s.t . } \ ; { \varepsilon}{\nabla}\cdot({\mathbf{d } } { \nabla}v_l ) + { { \mathbf{f}}}\cdot{\nabla}v_l + { \varphi}- l\in \sigma . \end{gathered}\end{aligned}\ ] ] since proposition [ t : theorem3 ] relies on the boundary integral condition , the @xmath37 and @xmath38 constructed by the above sos programs must satisfy this condition in order for @xmath33 and @xmath17 to be proven bounds . one way to guarantee this is to prove that @xmath104 decays exponentially as @xmath97 . [ [ ss : examplenoisyvdpnolog ] ] example : stochastic bounds for the van der pol oscillator . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + to illustrate the application of proposition [ t : theorem3 ] , we have bounded the stationary expectation @xmath119 for the stochastic van der pol oscillator @xmath120 with @xmath121 . for the @xmath88 matrix chosen above , the wiener processes @xmath122 acts on the @xmath123 component alone . this corresponds to a stochastic physical force , as seen by rewriting as a second - order equation for the position @xmath124 , @xmath125 we have computed bounds using the sos programs for noise amplitudes @xmath126 and three different polynomial degrees of @xmath37 , @xmath38 . figure [ f : noisyvdp ] shows the optimal bounds , along with several values of @xmath92 computed by numerically solving the stationary fokker - planck equation . our solution of the fokker - planck equation employed finite differences with operator splitting as in @xcite , and the system was evolved to steady state by implicit euler time stepping . the limiting expectation @xmath127 equals the deterministic average @xmath30 on the limit cycle , reflecting the fact that the van der pol system is stochastically stable . the upper bounds @xmath33 in figure [ f : ub_noisy_vdp ] are nearly tight at all noise strengths for @xmath37 of degree 10 and higher . they become less tight as @xmath94 increases , but they still correctly capture the increase in @xmath128 that occurs when stochastic forcing `` smears out '' the van der pol limit cycle . the lower bounds @xmath17 in figure [ f : lb_noisy_vdp ] are good for strong noise but not for weak noise . in fact , @xmath129 as @xmath130 for any fixed polynomial degree of @xmath38 . to see that this is inevitable , observe that the diffusive term @xmath117 in vanishes as @xmath130 if @xmath131 is bounded . as @xmath130 , the stochastic sos programs reduce to the deterministic sos programs , so the stochastic bounds can be no tighter than the global deterministic bounds . in the van der pol example , upper bounds are not affected by this phenomenon since the deterministic global upper bound on @xmath30 of [ ss : exampledeterministicvdp_ub ] is sharp and @xmath132 as @xmath130 . however , a tight lower bound on @xmath92 at small @xmath94 is impeded by the unstable deterministic trajectory at the origin , which saturates the deterministic global lower bound @xmath133 . in the deterministic case , a tight local lower bound was achieved in [ ss : localbounds ] by ignoring part of phase space , but we can not do so in the stochastic case since @xmath134 everywhere . instead , the diffusive term @xmath117 in the sos constraint of must remain @xmath135 near the origin as @xmath130 . clearly , this can not be accomplished with polynomial @xmath38 . the next section illustrates how a non - polynomial @xmath38 can be used instead to improve the lower bounds on @xmath92 at small @xmath94 . an unstable trajectory of the deterministic system @xmath28 can interfere not only with tight bounds on @xmath4 for a local attractor but also with tight bounds on @xmath92 in the corresponding stochastic system , at least when the noise is weak . this was illustrated in the previous section , where lower bounds on @xmath128 in the van der pol system were not tight : they approached zero as @xmath130 with polynomial @xmath38 of fixed degree . in this section we derive lower bounds on @xmath128 that remain tight when @xmath94 is small . we do this using a non - polynomial @xmath38 that depends explicitly on @xmath94 . the method described below applies to stochastic upper or lower bounds , provided that the deterministic trajectory interfering with the bounds is a _ repelling fixed point_. for concreteness , suppose we are trying to prove a tight lower bound on an expectation @xmath92 that is strictly positive , as in the van der pol example . suppose also that the global lower bound on @xmath4 in the corresponding deterministic system , @xmath28 , is zero and is saturated by the repelling fixed point @xmath136 . for the lower bound on @xmath92 to remain larger than the deterministic lower bound as @xmath137 , the stochastic bounding inequality must not reduce to its deterministic counterpart . this requires that the diffusive term @xmath117 remains commensurate with the other terms in , at least near the fixed point at the origin . the term @xmath117 can remain @xmath135 near the origin as @xmath130 if @xmath38 develops a _ boundary layer _ there . purely polynomial @xmath38 can have no such boundary layer , so we add a non - polynomial term and let @xmath138 + p({{\mathbf{x}}}).\ ] ] here , @xmath48 is a polynomial , @xmath139 is a tunable constant , and @xmath140 is a positive definite quadratic form , @xmath141 for a suitable symmetric , positive definite matrix @xmath142 ( denoted @xmath143 ) to be determined . when @xmath94 is small , the logarithmic term dominates @xmath38 near @xmath136 , but @xmath48 remains significant away from the origin . despite not being polynomial , the ansatz is suitable to derive a sos optimization problem for the bound because the relevant derivatives of @xmath38 are rational functions , @xmath144 substituting these expressions into the bounding inequality gives a rational inequality with the positive denominator @xmath145 . multiplying this rational inequality by the denominator we obtain the equivalent polynomial inequality @xmath146 where @xmath147 consequently , a lower bound @xmath17 on @xmath92 can be calculated for a fixed , small @xmath94 by solving the optimization problem @xmath148 the above sos program can produce lower bounds larger than zero for very weak noise strengths , meaning that these bounds are not spoiled by the fixed point at the origin . in fact , while the sos bounding program for polynomial @xmath38 reduces to its deterministic counterpart as @xmath130 , the above program does not . instead , the polynomial @xmath149 reduces to @xmath150 , retaining the term @xmath151 that does not appear in the deterministic program . this term is due to the boundary layer built into @xmath38 around the origin . [ [ choosing - the - quadratic - form - zeta . ] ] choosing the quadratic form @xmath152 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + if the coefficients of the positive quadratic form @xmath152 ( equivalently , the entries of @xmath142 ) are tuned at the same time as the other coefficients , then the sos constraint @xmath153 is quadratic in the decision variables , as opposed to linear . this produces a non - convex optimization problem for @xmath17 that is harder than a standard sdp and requires a bilinear solver . here we keep the problem convex by fixing a non - optimal @xmath152 in advance and tuning only the other coefficients in @xmath38 . even a non - optimal @xmath152 provides a good lower bound at small @xmath94 provided that , for reasons explained in [ ss : vanishingnoise ] , we have not only @xmath154 but also @xmath155 near the unstable fixed point . quadratic @xmath152 satisfying these two conditions can always be found if the point is a repeller but not if it is a saddle , which is why we have restricted ourselves to repellers . further discussion of how to choose or optimize @xmath152 is given in [ ss : vanishingnoise ] and appendix [ s : constructzeta ] . [ [ ss : examplenoisyvdpwithlog ] ] example : stochastic van der pol oscillator with weak noise . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we have used the methods of this section to derive lower bounds on @xmath128 for the stochastic van der pol oscillator . in the logarithmic ansatz for @xmath38 , we used the positive quadratic form @xmath156 that satisfies @xmath157 near the origin . figure [ f : noisyvdp_log ] shows the resulting lower bounds computed using the sos program for various degrees of the polynomial @xmath158 . when the noise amplitude is moderate or small , these bounds dramatically improve on the bounds of figure [ f : lb_noisy_vdp ] produced using purely polynomial @xmath38 . when @xmath159 , @xmath17 is indistinguishable from the deterministic _ local _ lower bounds found in [ ss : localbounds ] . in this section we propose a second method for bounding deterministic averages on a local attractor . while in [ ss : localbounds ] we did this by segmenting the phase space , here we do it by adding noise . many dynamical systems are stochastically stable , in the sense that the vanishing noise limit of a stationary expectation , @xmath160 , is equal to the deterministic average @xmath4 on a particular attractor @xcite . this is true in the van der pol example , where the vanishing noise limit of @xmath128 is equal to @xmath12 on the limit cycle . such correspondence can be exploited to bound @xmath4 on a local attractor : by proving bounds on @xmath92 that hold as @xmath130 , we obtain bounds that hold also for @xmath4 . in essence , the vanishing noise limit preserves the attractor while destroying unstable invariant structures . this idea was proposed in @xcite , though here we extend it by treating the @xmath130 limit rigorously , rather than numerically . we continue the analysis of [ ss : smallnoise ] , still assuming that @xmath136 is a repelling fixed point , @xmath161 , and @xmath162 on the attractor of interest . we suppose also that @xmath163 , meaning that the zero - noise limit of the expectation converges to the deterministic time average . rigorous statements about when this property holds can be found in @xcite and references therein . under these assumptions , we can obtain a lower bound @xmath164 that is tight for the local attractor by deriving a lower bound @xmath165 . to this end , we recall that although the polynomial inequality @xmath166 considered in [ ss : smallnoise ] suffices to prove the lower bound , we are really interested in proving the integral inequality . it is proven in appendix [ s : analyticepsproof ] , under mild assumptions on the stationary distribution @xmath104 , that holds in the limit @xmath130 if there exists any @xmath167 such that @xmath168 . dividing this relation by @xmath152 ( which is constructed to be positive definite ) and rearranging gives the sufficient condition @xmath169 \geq 0.\ ] ] since @xmath167 decreases @xmath17 by an arbitrarily small amount , we can let the bound be implied by a strict inequality and set @xmath170 . the lower bound we seek , @xmath171 , is thus returned by the sos optimization problem @xmath172 \begin{gathered } \text{s.t . } \quad \alpha\ , { { \mathbf{f}}}\cdot{\nabla}\zeta + \zeta \left ( { { \mathbf{f}}}\cdot{\nabla}p + { \varphi}- l \right ) \in \sigma,\\ { \mathbf{z}}\succ { \mathbf{0}}. \end{gathered } \end{gathered}\ ] ] like the finite - noise program , the above program can be made convex by fixing in advance the positive definite quadratic form @xmath152 . the chosen @xmath152 must satisfy @xmath155 near the origin . that is , it must satisfy @xmath173 where @xmath174 denotes the linearized dynamics near the origin . this condition is needed because the sos constraint becomes @xmath175 near the origin . the only way the above condition can be satisfied for @xmath17 larger than zero is if its first term is positive . appendix [ s : constructzeta ] gives a way of constructing an admissible @xmath152 when the unstable fixed point is repelling . when the unstable fixed point is a saddle , it is not generally possible to satisfy for positive definite @xmath152 . [ [ example - deterministic - bounds - for - the - van - der - pol - limit - cyclevanishing - noise - formulation . ] ] example : deterministic bounds for the van der pol limit cycle vanishing noise formulation . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + to demonstrate the methods of this section , we have computed lower bounds on the vanishing noise limit of @xmath128 for the van der pol system using the sos program . the results serve also as deterministic lower bounds on @xmath4 for all trajectories approaching the limit cycle ( strictly speaking , this conclusion requires proving that @xmath176 , which is outside our present scope ) . bounds of the latter type appear also in [ ss : localbounds ] , computed differently using the sos formulation . c | c c & @xmath177 & @xmath178 + + @xmath179 & @xmath180 & @xmath181 + @xmath182 & @xmath183 & @xmath184 + @xmath185 & @xmath183 & @xmath186 + [ table : table1 ] [ f : lb_zeronoise_vdp ] figure [ f : zeronoisevdp ] shows the bounds obtained by fixing the degree of the polynomial @xmath158 to 12 for the three different @xmath152 defined in table [ table : table1 ] . the condition is satisfied with @xmath187 by @xmath182 and @xmath185 when @xmath188 , and by @xmath179 when @xmath189 . the failure of the condition is why the bound using @xmath179 is poor for @xmath190 , while the bounds using @xmath182 or @xmath185 are poor near @xmath191 . taking the best of the three lower bounds at each @xmath56 gives a lower bound on @xmath12 within 1% of the value found by numerical integration over the entire range @xmath192 . despite our success in computing bounds for the van der pol oscillator , the study of more complicated dynamical systems presents several obstacles that require further investigations . we discuss some of these below , and put forward some preliminary ideas for future improvements . the techniques we have developed in this paper can in principle be applied to systems of arbitrarily high but finite dimension . however , computing tight bounds typically requires the use of polynomial storage functions of large degree . since the size of the sdp required to determine the existence of a sos decomposition for a polynomial of degree @xmath193 in @xmath194 variables grows as @xmath195 ( * * theorem 3.3 ) , relatively large sdps must be solved even for a low - dimensional system such as the van der pol oscillator . moreover , the sdps seem to be consistently ill - conditioned , and more so for larger systems . our present methods are therefore practical only for systems of moderate dimension . one way to improve the numerical conditioning of the sdps is to try to rescale the system . a preliminary investigation showed that rescaling the van der pol system such that the limit cycle is contained in the box @xmath196 ^ 2 $ ] dramatically improves the numerical conditioning , and bounds very similar to those reported above can be obtained without the need for multiple precision solvers . researchers studying other aspects of dynamics using sdps have had similar success rescaling the relevant dynamics to lie in @xmath196^n$ ] . however , the appropriate rescaling is not generally clear _ a priori_. there are several possible ways to reduce the computational cost of constructing bounds . one option is to exploit special structures in the sdps such as symmetries or sparsity ( e.g. @xcite ) . it might also be profitable to exploit special structure of the underlying dynamics , such as conserved quantities . finally , it is sometimes possible to work with a lower - dimensional truncation of a high - dimensional system and bound the errors introduced by the truncation ( e.g. @xcite ) . the methods developed here apply only to finite - dimensional systems , and more work is needed to extend them to partial differential equations ( pdes ) . one idea is to project the pde variables onto a finite galerkin basis . bounds can then be constructed using the finite - dimensional system , provided that the influence of the unprojected component can be controlled . this approach has been used successfully for nonlinear stability analysis of a fluid dynamical pde @xcite . a second idea is to bound integrals of the pde variables directly using the dissipation inequalities proposed in @xcite . in essence , these inequalities are integral constraints that replace the polynomial constraints of proposition [ t : proposition1 ] when a finite - dimensional system is replaced by a pde . despite recent advances in the numerical implementation of integral inequalities @xcite , however , this technique has so far been applied only to simple examples . if deterministic bounds are spoiled by a saddle point @xmath197 that is not embedded in the attractor of interest , the @xmath67-procedure can be used to remove a set containing @xmath197 that is disjoint from the attractor . however , it can be difficult to establish that @xmath197 is indeed separate from the attractor . if @xmath197 _ is _ embedded in the attractor , alternative methods must be found . the vanishing - noise approach of [ ss : smallnoise][ss : vanishingnoise ] seems promising , but the present formulation works only for repelling points , not saddle points . this is because requires that the polynomial @xmath198 increases along all trajectories near the unstable fixed point , but @xmath199 is monotonic and can not increase along trajectories on the stable and unstable manifolds of @xmath197 simultaneously . similarly , the logarithmic ansatz is not suitable when stochastic bounds with weak ( but finite ) noise are spoiled by a saddle point . in fact , the term @xmath200 in each of the inequalities in has opposite signs along the stable and unstable manifolds of @xmath197 unless @xmath201 changes rapidly , and any negative contribution must be balanced by large laplacians . this requires polynomials of large degree , making the sos programs intractable in practice . one possible solution is to find an ansazt for @xmath20 that , similarly to for repelling fixed points , lets stochastic bounds stay tight as @xmath130 by developing appropriate boundary layers . the following proposition suggests a way to deduce the ideal scaling of @xmath20 in this limit . [ t : theoremsc ] consider the inequality @xmath202 that is a sufficient condition for the lower bound @xmath203 . if @xmath204 is continuous and the stationary distribution @xmath6 is piecewise continuous , then the above inequality can be satisfied for the perfect lower bound @xmath205 only if @xmath206 wherever @xmath207 . this statement is proven by adding @xmath208 , which holds because @xmath17 is a perfect bound , and expression to find @xmath209 . since @xmath204 is continuous and non - negative , @xmath209 is possible only if @xmath210 wherever @xmath207 , as claimed . an analogous statement holds for the perfect upper bound . proposition [ t : theoremsc ] suggests that as @xmath38 is improved and @xmath17 gets closer to its perfect value of @xmath92 , the inequality @xmath211 gets closer in some sense to being an equality . this motivates us to consider solutions @xmath19 to the equation @xmath212 where @xmath205 and @xmath104 obeys the decay condition . a solution exists only when @xmath213 , as seen by taking the expectation of the equality and using . asymptotic analysis of @xmath19 in the above equality as @xmath130 might suggests a good ansatz for @xmath38 in the corresponding inequality . proposition [ t : theoremsc ] also provides a good way of assessing _ a posteriori _ whether the stochastic bounds obtained with sos optimisation are nearly sharp . this is useful , for instance , when data from numerical simulations are not available for comparison . for example , if a lower bound is close to the exact value of @xmath92 , @xmath38 should be close to the solution of whenever @xmath214 . in addition , an argument similar to the proof of proposition [ t : theoremsc ] for deterministic bounds shows that if an upper or lower bound is perfect , then @xmath215 for any trajectory @xmath5 on the attractor . if a good lower bound is proven using @xmath38 , for instance , then @xmath216 should be very close to @xmath217 on the attractor . for illustration , we have solved the lower bound program for the van der pol oscillator with @xmath48 of degree 6 and degree 12 . the degree-6 lower bound of @xmath218 is rather poor , while the degree-12 bound of @xmath219 is nearly tight . figure [ f : correctbehaviour ] shows the variation of @xmath220 along the limit cycle in each case and compares it to the value of @xmath217 along the limit cycle . the @xmath220 and @xmath217 curves match very closely in the degree-12 case but not in the degree-6 case . the bounds we have computed for the van der pol system all rely on the numerical solution of sdps , so they are subject to roundoff error . it is unlikely that roundoff error has led to any invalid bounds in our calculations because we have solved the sdps using multiple - precision arithmetic and checked the results by increasing the precision . nonetheless , such computations can not be considered rigorous to the standard of computer - assisted proofs because of the presence of roundoff errors . two different methods have been proposed for obtaining rigorous results from numerical sdp solutions . the first method is to project an approximate numerical solution onto an exact solution in terms of rational numbers @xcite . the second method is to use perturbation analysis , made rigorous by interval arithmetic , to construct a small interval around the approximate numerical solution in which the exact solution is guaranteed to lie . this approach has been implemented in the software vsdp @xcite , but the relevant sdps would need to be constructed manually since there is no sum - of - squares parser available that incorporates interval arithmetic . in this work we have presented computer - assisted methods for deriving bounds on average quantities in both deterministic and stochastic dynamical systems using sum - of - squares programming . we have given particular attention to proving bounds that apply only to trajectories approaching a particular attractor . one method is to use the @xmath67-procedure to omit segments of phase space on which the bounds do not need to hold . another strategy is to remove unstable invariant structures by adding noise to the system . this idea , proposed previously in @xcite , has been extended here by analyzing the weak and vanishing noise cases when the unstable structures to be omitted are repelling fixed points . these methods give improved bounds for weak but finite noise , and also in the rigorous limit of vanishingly weak noise . our methods have worked well when applied to the van der pol oscillator . the best deterministic and stochastic bounds proven throughout are summarized in figure [ f : bestbounds_vdp ] . in the deterministic case , we obtained upper and lower bounds on the infinite - time average of @xmath221 over the limit cycle that are all within 1% of the `` true '' values found by numerical integration . in the stochastic case , bounding the stationary expectation of @xmath11 at a variety of noise strengths , we obtained upper bounds all within 1% and lower bounds all within 10% of the `` true '' values found by solving the fokker - planck equation numerically . moving to dynamical systems other than the van der pol oscillator , whether the methods we have described can yield similarly tight deterministic and stochastic bounds depends on the details of those systems . if the dimension of a system is small enough for sos optimization to be computationally feasible , we expect that tight global bounds on deterministic averages can be obtained using the methods of [ ss : sosboundsreview ] . in fact , since the first draft of this work , the bounding techniques we have presented for deterministic systems have been successfully applied to the design of control systems for fluid flows @xcite . with stochastic forcing that is not too weak , we expect that fairly tight bounds on stationary expectations also can be obtained using the methods of [ ss : generalnoiseformulation ] . as the noise strength decreases , the tendency of unstable invariant structures to spoil tight bounds can be combatted by the methods of [ ss : smallnoise ] if these structures are repelling fixed points . new methods will be needed to maintain tight bounds as noise becomes weak in systems with other unstable structures , including saddle points . lastly , the methods of [ ss : localbounds ] can be used to bound averages on trajectories within an absorbing domain and , in particular , on a single attractor . this much is true for any dynamical system , but whether these bounds can be tight depends on the nature of the attractor . for instance , bounds for a chaotic attractor must apply not just to generic chaotic trajectories but also to all trajectories embedded within the attractor , such as saddle points and unstable orbits . what can be deduced about chaotic attractors is of particular interest if the methods presented here are to be applied to complex systems of physical and engineering relevance . as explained in [ ss : computerassistedproof ] , techniques are available to control the roundoff errors in the solution of the sdps and compute rigorous bounds . however , these may be impractical for large problems . we are also unaware of any parsers for sos programs that incorporate rigorous computations . for these reasons , we illustrate a method to produce rigorous bounds that can be implemented with existing software . for definiteness , we consider the upper bound problem in . the solution of the sos program consists of a polynomial storage function @xmath37 , a bound @xmath33 , a vector of monomials @xmath222 and a positive semidefinite , symmetric matrix @xmath223 such that @xmath224 since @xmath223 is positive semidefinite , @xmath225 is a sos polynomial and so is the right - hand side . we refer the reader to @xcite for further details . in practice , the decomposition is only approximate , and the error @xmath226\ ] ] is non - zero . however , according to theorem 4 in @xcite , the polynomial @xmath227 is still certifiably non - negative if @xmath228 where @xmath229 is the coefficient of @xmath230 of largest magnitude , @xmath231 is the smallest eigenvalue of @xmath223 and @xmath232 denotes the dimension of @xmath223 . unfortunately , the optimal solution of a sos problem rarely satisfies this condition , since @xmath233 for the optimal @xmath223 @xcite . consequently , we suggest to compute a slightly suboptimal @xmath33 by manually decreasing an initial guess @xmath234 using a sequence of feasibility problems , checking after each step that holds using rigorous computations such as interval arithmetic @xcite . the steps are outlined in algorithm [ algorithm1 ] . 0.5ex @xmath235 initial guess for the upper bound _ loop : _ @xmath236 construct a polynomial with variable coefficients @xmath237 construct the expression @xmath238 @xmath239 round the coefficients of @xmath20 and the entries of @xmath223 to @xmath193 decimal places check for the truncated @xmath20 , @xmath223 using rigorous numerics decrease @xmath33 and * goto * _ loop _ increase @xmath33 and * goto * _ loop _ increase @xmath33 and * goto * _ loop _ figure [ f : ubvdprigorous ] compares some preliminary bounds obtained with algorithm [ algorithm1 ] for the van der pol oscillator to the optimal ones . the results are only slightly suboptimal ; moreover the feasibility problems were better conditioned than the full sos optimization and could be solved with standard double - precision solvers such as sedumi @xcite . we remark that given the preliminary stage of this investigation , we avoided the technical step of computing rigorous roundoff errors for the eigenvalues of @xmath223 . instead , we checked condition using arbitrary precision arithmetic , increasing the precision until @xmath231 changed by less than 1% . although our results are only `` certified '' but not fully rigorous , we expect that similar results would be obtained when @xmath231 is computed rigorously . moreover , we also stress that algorithm [ algorithm1 ] was not very robust and required careful tuning ( e.g. in the choice of decimal figures to retain ) . a thorough investigation and resolution of these numerical issues is left for future work . to show that @xmath240 is an absorbing domain for the van der pol oscillator for @xmath241 , let us reverse the direction of time in and consider the energy @xmath242 of the system @xmath243 , for which the origin is a stable fixed point . one has @xmath244 meaning that @xmath245 when @xmath246 . therefore , any contour of @xmath247 which is contained in the strip @xmath246 defines the boundary of a region of attraction of the origin for the time - reversed oscillator . one concludes that in the original system all orbits ( except of course the fixed point at the origin ) will leave the ball @xmath248 if @xmath249 , i.e. @xmath250 is an absorbing domain . to prove a lower bound on @xmath92 we need to prove inequality . the ansatz for @xmath20 allows us to rewrite this condition as @xmath251 ^ 2 } \right\rangle_{{\varepsilon } } \geq 0,\ ] ] where @xmath252 is as in . since @xmath152 is quadratic in @xmath24 , it can be verified that @xmath253 is the dominant term in @xmath252 for any fixed @xmath254 in the limit @xmath137 . moreover , @xmath253 is the dominant term as @xmath255 because it contains the monomials of highest order . the condition @xmath256 then implies that @xmath257 in the limit @xmath137 for all @xmath24 , with the possible exception of a ball @xmath258 of radius @xmath259 , @xmath260 , where the integrand in becomes singular . for definiteness , we will write @xmath261 for some constant @xmath262 . thus , we conclude that @xmath263 ^ 2}d{{\mathbf{x}}}\geq 0.\ ] ] to prove that holds , we will now show that @xmath264 ^ 2 } d{{\mathbf{x}}}= 0.\ ] ] upon substitution of , the integral becomes @xmath265 d{{\mathbf{x}}}.\end{gathered}\ ] ] let us proceed term by term and let us assume that @xmath104 is bounded in @xmath258 uniformly as @xmath137 . this is a reasonable assumption , since we are assuming that the point @xmath136 is a repeller and hence does not belong to the attractor of the deterministic system on which we require the bound . since @xmath158 and @xmath2 are continuous , @xmath266 d{{\mathbf{x}}}\bigg\vert \leq \\ \frac{4\pi}{3}\,c^3{\varepsilon}^{3\eta } \ , \max_{b_r}\big\lbrace \rho { \vert { \varepsilon}{\nabla}\cdot({\mathbf{d}}{\nabla}p ) + { { \mathbf{f}}}\cdot{\nabla}p + { \varphi}- l \vert } \big\rbrace \end{gathered}\ ] ] so @xmath267 d{{\mathbf{x}}}=0 . \ ] ] to study the other terms , we switch to polar coordinates , @xmath268 where @xmath269 $ ] , @xmath270 $ ] , @xmath271 $ ] and @xmath272 the quantity @xmath273 is a fixed real number . moreover , since @xmath152 is a homogeneous , positive definite quadratic form of @xmath24 and @xmath114 is positive semi - definite ( recall that @xmath274 ) we can write @xmath275 for some strictly positive function @xmath276 and non - negative function @xmath277 . let @xmath278 and @xmath279 then , we have @xmath280 & \leq { \varepsilon}\,{\vert \alpha { \nabla}\cdot({\mathbf{d}}{\nabla}\zeta ) \vert } \,\max_{b_r}(\rho ) \,2\,\pi^{n-1}\int_{r=0}^{r } \frac{r^{n-1}}{{\varepsilon}+r^2f^ * } dr . \notag\end{aligned}\ ] ] if @xmath281 , the last term can be integrated to give @xmath282 while when @xmath283 we can estimate @xmath284}\left(\frac{r^{n-1}}{{\varepsilon}+r^2f^*}\right ) \bigg\rbrace.\ ] ] it can be verified that the maximum of the last term is achieved at the endpoint @xmath285 . taking the limit @xmath137 shows that for all @xmath286 @xmath287 similarly , we can show @xmath288 \leq \begin{cases } \displaystyle { \varepsilon}\left [ \frac{{\vert \alpha \vert } \pi^{n-1 } g^ * } { 2f^{*2}}\,\max_{b_r}(\rho ) \,\log ( 1+c^4\,f^{*2}{\varepsilon}^{4\eta-2})\right ] , & n=2 \\[10pt ] \displaystyle { \varepsilon}^{1+\eta}\,\left [ 2c { \vert \alpha \vert}\pi^{n-1 } g^*\,\max_{b_r}(\rho ) \,\max_{r\in[0,r]}\left ( \frac{r^{n+1}}{{\varepsilon}^2 + r^4f^{*2}}\right)\right ] , & n\geq 3 \end{cases}\end{gathered}\ ] ] where , again , the last maximum is achieved at @xmath289 . we conclude that @xmath290 finally , since @xmath291 and @xmath292 is linear , the term @xmath293 is a polynomial of @xmath24 that only contains monomials of degree 2 and higher . consequently , we can write @xmath294 for some continuous functions @xmath295 whose usual @xmath296 norm @xmath297 is finite . each term in this series can be considered separately ; for each @xmath298 we have @xmath299 } \left ( \frac{r^{n+m}}{{\varepsilon}+ r^2f^ * } \right ) \bigg\rbrace , \notag\end{aligned}\ ] ] which tends to 0 as @xmath137 ( since @xmath286 , @xmath300 , and the last maximum is obtained at @xmath289 ) . we therefore conclude that @xmath301 combining , , and proves and , consequently , . note that the proof presented here is valid for systems of dimension @xmath302 or higher ; one - dimensional systems are not of interest as bounds can be computed analytically . if the unstable fixed point is a repeller and @xmath303 can be diagonalized , a quadratic form @xmath152 that satisfies can be constructed from the eigenvectors of @xmath303 . let @xmath304 denote the matrix of eigenvectors of @xmath303 and @xmath305 be the usual diagonal matrix of eigenvalues , such that @xmath306 note that @xmath305 is a positive definite matrix since the unstable point is repelling . letting @xmath307 , an appropriate choice of @xmath152 is @xmath308^t { \mathbf{u}}^{-1 } { { \mathbf{x}}}.\ ] ] in fact , we have @xmath309^t { \mathbf{u}}^{-1 } { { \mathbf{x}}}\\ & = 2{{\mathbf{x}}}^t [ { \mathbf{u}}{\mathbf{u}}^{-1}]^t\ , { \mathbf{j}}_0^t \ , [ { \mathbf{u}}^{-1}]^t { \mathbf{u}}^{-1 } { { \mathbf{x}}}\\ & = 2 { \mathbf{w}}^t [ { \mathbf{u}}^{-1 } { \mathbf{j}}_0 \,{\mathbf{u}}]^t { \mathbf{w } } \\ & = 2 { \mathbf{w}}^t \boldsymbol{\lambda}{\mathbf{w } } , \end{aligned}\ ] ] which is positive for @xmath310 because @xmath305 is positive definite . we conclude that holds for @xmath187 . note , however , that this is not the only possible choice of @xmath152 . the authors thank g. chini , c. r. doering , and a. wynn for their helpful comments and discussions . d.g . and s.c . are grateful for the hospitality of the institute for pure and applied mathematics , where this work was begun , and g.f . and d.g . are grateful for the hospitality of the geophysical fluid dynamics program at the woods hole oceanographic institution , where most of this work was carried out . during part or all of this work , g.f . was supported by imperial college london under the ic phd scholarship resfs g82059 , d.g . was supported by the us nsf mathematical physics award phy-1205219 , and d.h . and s.c . received funding from epsrc under the grant ep / j011126/1 and support in kind from airbus operation ltd . , eth zurich ( automatic control laboratory ) , university of michigan ( department of mathematics ) , and university of california , santa barbara ( department of mechanical engineering ) .

every singlet state of a quantum spin- system can be decomposed into a linear combination of valence bond basis states . the range of valence bonds within this linear combination as well as the correlations between them can reveal the nature of the singlet state , and are key ingredients in variational calculations . in this work , we study the bipartite valence bond distributions and their correlations within the ground state of the heisenberg antiferromagnet on bipartite lattices . in terms of field theory , this problem can be mapped to correlation functions near a boundary . in dimension , a non - linear model analysis reveals that at long distances the probability distribution of valence bond lengths decays as and that valence bonds are uncorrelated . by a bosonization analysis , we also obtain in despite the different mechanism . on the other hand , we find that correlations between valence bonds are important even at large distances in , in stark contrast to . the analytical results are confirmed by high - precision quantum monte carlo simulations in , and . we develop a single - projection loop variant of the valence bond projection algorithm , which is well designed to compute valence bond probabilities and for which we provide algorithmic details .

1311.3975

magnetic ordering is a primary concern of classical magnetism . long - range order can be disrupted by the presence of disorder or of frustration between the magnetic degrees of freedom . quantum fluctuations form also an important path to destroy long - range order , specifically antiferromagnetic ( af ) order . they are particularly strong for low values of the spin @xmath10 and in low dimensionality . the prototypical model to study quantum fluctuations is the @xmath11 af heisenberg model @xmath12 where @xmath13 denote nearest - neighbor ( nn ) sites on a hypercubic lattice , and @xmath14 the spin operator [ of amplitude @xmath15 on site @xmath16 . while long - range order in the ground state of eq . is prohibited in one dimension ( 1d ) @xcite , it can be shown rigorously @xcite that the system displays long - range order in @xmath17 for all @xmath18 . for @xmath11 , there is no exact proof but the numerical computations @xcite leave no doubt that this is also the case . in this paper , we deal only with this extreme quantum case @xmath11 and assume an even number @xmath19 of spins . the hamiltonian eq . exhibits su(2 ) symmetry , as it commutes with the total spin operator @xmath20 . the total spin quantum number @xmath21 is defined by the eigenvalue @xmath22 of the operator @xmath23 . for af interactions , the ground state for finite @xmath19 of the hamiltonian eq . on the hypercubic lattice can be proven @xcite to be a total singlet @xmath24 . natural objects to describe such a singlet state are _ valence bonds _ ( vbs ) , which are intrinsically non - magnetic and antisymmetric objects , with two spins @xmath0 coupled into a singlet . this notion was introduced in the early days of quantum mechanics @xcite and has regained interest thanks to the proposal that wave functions composed of vbs [ `` resonating valence bond ( rvb ) wave functions '' ] could describe a quantum spin - liquid phase without any magnetic order , for heisenberg models on frustrated lattices @xcite . the idea is that local ( typically nn ) superpositions of vbs can effectively accommodate the constraint imposed by frustration in the building blocks of such lattices and lower the energy of the full system . while the original suggestion @xcite of this scenario taking place on the triangular lattice is now abandoned ( since this lattice supports af long - range order @xcite ) , it stimulated many investigations which led to significant progresses in quantum magnetism . in fact , the existence of the rvb phase is established in several quantum dimer models @xcite , which are effective models of quantum antiferromagnets . furthermore , the usefulness of vb wave functions has been demonstrated for the understanding of the low - energy physics of heisenberg af models on other frustrated lattices , such as kagome @xcite or frustrated square lattices @xcite . these developments have confirmed the importance of the vb picture in understanding quantum antiferromagnets . much of the studies of quantum antiferromagnets based on the vb picture is aimed at quantum spin liquids without a long - range antiferromagnetic order . however , it should be noted that the set of vb states , once including longer - range vbs beyond nearest neighbors , is _ overcomplete _ in the total singlet sector @xcite . thus , any singlet state can be represented as a superposition of vb basis states . the finite - size ground state of the antiferromagnetic hamiltonian on the hypercubic lattice in @xmath7 dimensions is no exception , despite its long - range antiferromagnetic order . this somewhat counterintuitive fact is indeed consistent with the existence of symmetry - broken ground states in the thermodynamic limit , thanks to the asymptotic degeneracy in the thermodynamic limit @xmath25 of the `` anderson tower of states '' of nonzero total spin @xmath21 with the ground state @xcite . an antiferromagnet is called a bipartite antiferromagnet if the sites can be grouped into two sublattices , and the heisenberg exchange interactions exist only between two sites belonging to different sublattices . examples include the antiferromagnets on a hypercubic lattice with only nearest - neighbor interactions . it is then natural to consider a restricted set of vb basis states , bipartite vb basis states , in which only vbs connecting different sublattices are allowed @xcite . even the bipartite vb basis , if longer - range vbs are included , is overcomplete in the total singlet sector ( see _ e.g. _ ref . or ref . ) and thus can represent any singlet ground state of an @xmath11 antiferromagnet . in fact , in a pioneering work by liang , douot and anderson ( lda ) @xcite , the long - range - ordered af ground state of the heisenberg antiferromagnet on the square lattice was described in terms of bipartite vb basis states , including bipartite vbs of arbitrary length . in practice , this was done by defining a so - called amplitude product state , where the weight of every valence bond state is factorized into weights @xmath26 coming from contributing valence bonds of length @xmath27 . the original results of lda pointed to a minimal variational energy for the square lattice heisenberg model reached for a power law decay @xmath28 , with @xmath29 . the vb basis provides a new perspective for understanding quantum antiferromagnets , which is complementary to the more traditional @xmath30 basis . as an indication of its significance , a quantum monte carlo ( qmc ) algorithm based on the vb basis was proposed @xcite and proved to be useful . the more recent numerical results from the vb - based qmc simulation @xcite , obtained without assuming the power - law decay , imply @xmath31 rather than @xmath32 suggested by lda . this power - law decay and the value @xmath31 can be understood through a master equation @xcite or other mean - field @xcite approaches , which moreover predict @xmath33 to be the best variational ansatz for the ground state of eq . in dimension @xmath34 , within amplitude product states . these previous works were explicitly assuming a certain amplitude product variational ansatz for the antiferromagnetic ground state @xcite . however , one could instead also ask for the length distribution @xmath3 of valence bonds in the _ real _ ground state of heisenberg antiferromagnets . this question may at first glance seem ill defined since the expression of singlet states in terms of valence bonds is in general non - unique . fortunately , the occupation number of _ bipartite valence bonds _ is nevertheless still well defined @xcite . in this paper , we compute by analytical means the distribution of vb lengths @xmath3 for antiferromagnetic heisenberg models on hypercubic lattices in dimension @xmath34 . somewhat surprisingly , the question of vb distribution can be reduced to a problem of _ boundary field theory_. @xcite boundary field theory has been found useful , not only in boundary critical phenomena and quantum impurity problems , but also in recent topics such as entanglement entropy . @xcite the vb distribution may be added to the list of applications of boundary field theory . the key in the relation is the identity @xcite between the bipartite vb distribution and an overlap with the reference state , which is naturally related to boundary correlation functions . in principle , any field theory , which is an appropriate effective theory for the bulk , can be used to calculate the bipartite vb distribution with a certain boundary condition . the ground state and low - energy excitations of the heisenberg antiferromagnet on the hypercubic lattice in @xmath7 spatial dimensions are believed to be described by the o(3 ) non - linear @xmath2 model . in fact , in the ground state , the su(2 ) symmetry is spontaneously broken in the non - linear @xmath2 model in @xmath7 , as is the case in the heisenberg antiferromagnet on the hypercubic lattice . the effective field theory in the low - energy limit is then reduced to the field theory of two massless nambu - goldstone modes . we will apply such a field theory with boundary , in order to study the bipartite vb distribution in @xmath7 dimensions . on the other hand , in @xmath6 , the physics is quite different as there is no spontaneous breaking of the symmetry . the system belongs to the universality class of tomonaga - luttinger liquids , which can be described by a free - boson field theory in @xmath6 spatial dimension . again , together with an appropriate boundary condition , it is applied to the bipartite vb distribution in @xmath6 dimension . as a result , in @xmath35 dimensions , we find a power - law dependence @xmath36 , confirming previous results based on mean - field approximations @xcite as well as a previous numerical qmc estimate @xcite . the analytical predictions are compared in detail to the results of qmc simulations of the heisenberg model in dimension @xmath6 , 2 and 3 . while `` free - boson field theory '' is a common ingredient for the cases @xmath6 and @xmath7 , the correspondence between the field theory and the quantum antiferromagnet is quite different , reflecting the very different physics . in fact , the difference between @xmath6 and @xmath7 becomes apparent in the _ correlations _ between vbs , as follows . in the wave function studied by lda , it is assumed that the weight of each vb configuration can be factorized in products of weights carried by each vb individually : @xmath37 where @xmath38 denotes a vb configuration composed of different vbs @xmath39 formed between spins at sites @xmath16 and @xmath40 . the amplitude @xmath41 is often chosen as depending on the distance @xmath42 between @xmath16 and @xmath40 , _ i.e. _ , the `` length '' of the vb @xmath39 . the factorization of vb amplitudes means the absence of correlation among vbs . it is indeed a strong assumption in the ansatz lda wave function , and very few works @xcite to our best knowledge went beyond it until recently @xcite . in fact , it is very natural to ask whether this assumption is correct or can be justified for realistic ( short - range ) heisenberg models . the boundary field theory approach allows us to elucidate analytically the correlation among vbs . we find that , whereas there are nontrivial correlations between vb occupation numbers even at large distances for @xmath6 , no correlations are asymptotically present in @xmath1 . the analytical prediction is then carefully verified via a numerical investigation with qmc simulations . this difference between @xmath6 and @xmath1 is another manifestation of different physics , in terms of the vb picture . our results for @xmath1 also justify the use of factorized wave functions of the lda type . the plan of the paper is the following . in sec . [ sec : def ] , we review the notion of valence bond occupation and correlation in a singlet wave function , giving a pedagogical derivation of a simple formula to effectively count the `` average '' number of vbs shared by two sites . in sec . [ sec : qmc ] , we present a method to measure these quantities numerically using qmc algorithms in the vb basis , and discuss in particular a specific improved algorithm for doing so . we then consider the case of @xmath1 heisenberg antiferromagnets using a non - linear @xmath2 model description in sec . [ sec : sigma ] , computing the vb length distribution as well as correlations . we furthermore compare the analytical predictions for the heisenberg model on the square and simple cubic lattices with qmc results . in sec . [ sec : bos ] , we study the more complicated case ( from the analytical point of view ) of @xmath6 with bosonization , showing the presence of correlations between vbs . the numerical computations in @xmath6 are simpler and allow for an exhaustive comparison with bosonization results . we further discuss the implication of our results in sec . [ sec : discussion ] and give a conclusion . consider a bipartite lattice , such that interactions in the hamiltonian eq . occur only between sublattice @xmath43 and sublattice @xmath44 . every bipartite valence bond state @xmath45 on such a lattice may be written as @xmath46 with @xmath47 on sublattice @xmath43 and @xmath48 on sublattice @xmath44 . in this notation the pairs @xmath49 represent precisely the bipartite valence bonds @xmath50 that form the state @xmath45 . we may then define the occupation number @xmath51 of a valence bond state as @xmath52 note , that this definition is rather formal and not always very useful . using the nel state @xmath53 an interesting alternative expression can be given in the form of a scalar product @xcite : @xmath54 therein , the operator @xmath55 permutes the spins @xmath56 and @xmath57 within the nel state @xmath58 , as illustrated in fig . [ fig : occupation ] . notice that here we swap only a single spin ( @xmath59 ) , with @xmath60 and @xmath61 . obviously , a valence bond @xmath49 in @xmath45 contributes a factor of @xmath62 when overlapped with @xmath63 and @xmath64 for the overlap with @xmath65 . hence , a valence bond connecting @xmath66 and @xmath67 must yield a factor of @xmath68 in eq . , as required . on the other hand , if there are valence bonds @xmath69 that connect @xmath66 and @xmath70 ( see fig . [ fig : occupation ] ) , then the overlap with a triplet state vanishes @xmath71 , thus showing the equivalence of eq . with eq . . it is also straightforward to generalize eq . to arbitrary singlet states @xcite , which can always be expressed as ( albeit non - unique ) superpositions of bipartite vb states . writing a given singlet state @xmath72 as @xmath73 we can immediately generalize eq . to the `` average '' vb occupation number for @xmath72 as @xmath74 even though the expression eq . of the singlet state in terms of vb basis states is not unique , the vb occupation number eq . is unique and well defined . since every vb basis state has the same overlap with the nel state , that is , @xmath75 , eq . is thus an `` average '' defined with respect to the coefficients @xmath76 themselves as the weight , and not to their squares @xmath77 . if some of the coefficients @xmath76 can be negative , the average eq . may be ill defined . however , for the ground state of a bipartite quantum antiferromagnet , all the coefficients @xmath76 of the bipartite vb basis states are non - negative upon a proper choice of the overall phase factor , thanks to marshall s sign rule @xcite . thus the average occupation number is always well - defined for such a ground state and satisfies @xmath78 . we emphasize that this only applies to bipartite valence bonds @xcite . we define the single vb distribution function @xmath79 where two sites @xmath80 are separated by the vector @xmath81 . as we consider translationally invariant systems , @xmath82 should not depend on the choice of the site @xmath56 and is a function of the separation @xmath27 only . moreover , since the site @xmath56 must belong to one vb in any of the vb basis states , we find the sum rule @xmath83 this construction carries over immediately to the more general problem of simultaneous occupation numbers of multiple vbs . this can be done by choosing the set of sites @xmath66 and @xmath67 from each sublattice @xmath43 and @xmath44 , so that @xmath84 . then , the `` number of vbs connecting @xmath66 and @xmath67 '' can be obtained as @xmath85 we note that , in this quantity , one no longer sees which sites precisely in @xmath66 and @xmath67 are connected to each other . for example , with the choice of @xmath86 and @xmath87 , we measure the simultaneous occupation of two vbs connecting the two sets . this may be interpreted as a sum of the occupation numbers for the two possible vb configurations @xmath88 and @xmath89 : @xmath90 we assert that , however , the occupation numbers for the individual vb configurations @xmath88 or @xmath89 are not well defined , and only the combination eq . is uniquely defined . measuring valence bond occupation numbers or correlations is straightforward using eq . if one has direct access to the wave function studied , such as within exact diagonalization or density matrix renormalization group ( dmrg ) calculations ( see ref . for a more precise discussion and an example of such computations ) . since , however , we would like to study asymptotic , long - distance , properties of @xmath91 quantum antiferromagnets on large systems , these methods are not practical and we have to resort to stochastic qmc schemes . recent qmc state - of - the - art algorithms allow access to quantities in the valence bond language directly , by projecting out the ground state from some trial wave function . the trial wave function is usually chosen to be some special singlet state , such as the amplitude product state eq . , in order to achieve a significant reduction of computational cost @xcite . initially , two such schemes have been devised @xcite , which are often referred to as single projection and double projection , respectively . the choice between these two schemes depends on the observable we want to measure . observables that commute with the hamiltonian can be measured within the single projection scheme , where some power @xmath92 of the hamiltonian @xmath93 is applied to a singlet trial state @xmath94 , @xmath95 herein , @xmath96 is a reference state , which is in principle arbitrary . however , it turns out that the nel state @xmath97 is a particularly good choice , as it has the same overlap with all singlet states . furthermore , @xmath98 is chosen to be the energy of the maximally polarized state @xmath99 , such that the projection reveals the ground state . notice , that such a scheme requires the ground state @xmath100 to be an eigenstate of the observable @xmath101 . in cases where this does not apply we have to carry out the double projection scheme , where the ground state projection occurs on both sides , @xmath102 valence bond occupation , eq . , is a rather peculiar quantity , which measures properties of a given singlet state without being an observable . there are however different ways to generate related observables , which can be measured within double projection , _ e.g. _ , @xmath103 unfortunately , it turns out that simulations converge rather badly when carrying out a double projection scheme with these observables . this is due to the fact that the appropriate estimator fluctuates exponentially , like the one for fidelity @xcite and rnyi entropies @xcite and in contrast to the energy estimator and other quantum mechanical observables @xcite . on the other hand , one notices that eq . looks very similar to the single projection expression eq . , suggesting an implementation of this last scheme . indeed valence bond occupations ( through the related valence bond entanglement entropy @xcite ) have already been measured with the single projection method using a local update algorithm . however , whereas for the double projection method very efficient loop updates are known @xcite , this seems not to be the case for the single projection algorithm . we devote the next section to the description of such a method . in order to carry out a single projection scheme , we first rewrite expression eq . and decompose the projection step as a sum over all possible contributing bond operator strings @xcite @xmath104 every one of such operator strings is a product of bond singlet projectors , @xmath105 and @xmath106 is a multi - index that is summed over . notice , that the singlet projector @xmath107 on a given bond @xmath108 between sites @xmath16 and @xmath40 is related to the spin - spin interaction by @xmath109 with these notations and noting that the singlet trial state can be decomposed as a superposition of bipartite vb states @xmath110 , eq . reads @xmath111 where @xmath97 is used as the reference state and the sums over @xmath112 and @xmath16 can be sampled stochastically . the way this is done is illustrated in fig . [ fig : loopgraph ] , where an example bond operator string is explicitly written out . we use the following updates in order to sample the configuration space of different bond operator positions and valence bond states . , residing on @xmath113 sites ( drawn vertically ) . herein , the singlet projectors ( red shaded squares ) are intentionally represented by two valence bonds . notice that the nel state on the left ( represented by black and white circles ) does not close the red ( dashed ) loops , whereas the blue ( continuous ) loops are closed intrinsically or by the valence bond state ( yellow shaded rectangle ) on the right . it is very straightforward to see , that the above expression can be evaluated by sutherland s @xcite overlap rule @xmath114 , where @xmath115 denotes the number of _ closed _ loops and @xmath116 the _ total _ number of valence bonds . here we count @xmath117 closed loops and @xmath118 valence bonds . ] _ loop updates . _ the loop updates @xcite can be applied here with a slight modification , where loops passing through the nel state can not be flipped as in the double projection scheme @xcite . this can be understood in two different ways : first , in the original scheme loop flips correspond to a flip of underlying spin configurations . whereas every valence bond and every singlet projector contains two such configurations , the nel state is a single spin configuration . therefore , loops through the nel state exist only in one flavor and can not be flipped , in contrast to all other loops that exist in two flavors . another way of seeing this is by looking at the interpretation of sutherland s overlap rule @xcite described in fig . [ fig : loopgraph ] , which deals only with closed loops but not with open ones . the loop algorithm aims precisely to mimic the varying contributions from the number of loops , by rewriting the term @xmath119 as a sum over different configurations . allowing each closed loop to take two different flavors generates exactly the required sum , whose stochastic sampling is computationally cheaper than counting the number of closed loops in a given overlap graph . as a result , we keep the flavor of the red ( dashed ) loops in fig . [ fig : loopgraph ] constant , while we flip every blue ( continuous ) loop with a probability of @xmath120 between the two flavors . as in the original scheme @xcite , the different flavors represent a constraint for all other updates , which can be carried out in completely the same manner as in the original work . this includes _ operator updates _ , which are used to move the bond operators into different positions of the overlap graph , as well as _ state updates _ , in order to sample the different valence bond states @xmath121 . self - optimized amplitude product states @xcite turn out to be an advantageous choice for the single projection scheme as well , as can be seen in fig . [ fig : selftrial ] . we observe the same benefit for quantities such as eq . . the probabilities of both such updates remain the same as for the double projection scheme @xcite . sites versus the projection length @xmath92 in a single projection scheme . inset is a zoom on the low - energy part . for the self - optimized amplitude product state a significant convergence speed up is achieved ( error bars are smaller than the point sizes ) . ] _ measurements . _ as the sampling weight @xmath122 is simply given by @xmath123 where the remaining factors keep constant during the simulation , the estimator is given by @xmath124 notice that the evaluation of this last expression is very simple , since the denominator is a constant and cancels most of the terms in the numerator . at the end one only has to determine the projected valence bond state @xmath125 , which can directly be read out of the overlap graph ( fig . [ fig : loopgraph ] ) in the following way : two sites form a valence bond in @xmath125 if they are linked together by an open loop in fig . [ fig : loopgraph ] . graphically , one just needs to start from any initial site on the nel state side and follow the open loop until it reaches a second site on the nel state : the initial and second sites are then coupled in a valence bond in @xmath125 . the operator @xmath101 has then to be inserted between this resulting valence bond state and the nel state itself . in the case of valence bond occupations and valence bond correlations the operator @xmath101 is replaced by @xmath126 , which amounts to measure @xmath68 , if the projected valence bond state @xmath125 contains a valence bond @xmath49 , and @xmath127 if not . in practice , we measure a histogram of all valence bonds in @xmath125 , where we record the length @xmath128 of the valence bonds in case of valence bond occupations , or their lengths together with their separation distance in case of valence bond correlations . whereas in one dimension it is possible to collect all data for every single valence bond , in higher dimensions we will restrict ourselves to measuring only specific configurations , as the amount of data would otherwise not be manageable . let us consider the statistics of vb occupations eq . , in a field theory framework . we note that , when @xmath72 is a ground state , the average vb occupation number between sites @xmath56 and @xmath57 [ eq . ] for example is given as @xmath129 where @xmath130 is the initial state , which is arbitrary as long as the overlap @xmath131 does not vanish . this can be interpreted as a correlation function @xmath132 , in the presence of a boundary described by the `` boundary state '' @xmath133 @xcite . the boundary condition in the field theory , which exactly matches the nel state @xmath133 as the boundary state , is unknown . nevertheless , the large - distance asymptotic behavior of boundary correlation functions is governed by fixed points of boundary renormalization group flow . those boundary fixed points correspond to conformally invariant boundary conditions . thus , as long as we are concerned with the large - distance asymptotic behavior of the vb distribution , we can replace the boundary condition with the conformally invariant one corresponding to the infrared fixed point for the boundary rg flow starting from @xmath134 . in general , tracking the boundary rg flow is a difficult problem . however , in the circumstances we will consider , we can find a natural candidate for such a boundary condition . this approach is generally applicable to the evaluation of the amplitude ( inner product ) between a reference state and the groundstate of a quantum many - body system in any dimension , at least in principle . we note that this approach indeed has been employed @xcite in @xmath6 dimension , for problems closely related to ( but somewhat different from ) the present one . here we demonstrate that it is also useful in @xmath7 , and we will come back to @xmath6 in sec . [ sec : bos ] . the low - energy / large - distance effective theory of the heisenberg antiferromagnet on the hypercubic lattice in @xmath1 dimensions is the o(3 ) nonlinear @xmath2 model . it is defined by the lagrangian density @xmath135 ^ 2 , \label{nlsigma}\ ] ] where @xmath136 is a three - dimensional vector subject to the constraint @xmath137 the field @xmath136 describes the staggered component of the original spin operator : @xmath138 as we have discussed , the ground state of the heisenberg antiferromagnet in a @xmath7 hypercubic lattice generally displays a long - range antiferromagnetic order . this also implies the spontaneous breaking of the o(3 ) symmetry in the corresponding non - linear @xmath2 model , and the field @xmath136 is aligned to a particular direction . without losing generality , let us assume that @xmath136 is aligned in the @xmath139 direction . the small fluctuation around the `` vacuum '' can be described by the two components @xmath140 and @xmath141 , so that @xmath142 the lagrangian density eq . is now written as @xmath143 , \label{ngboson}\ ] ] where higher - order terms in @xmath140 and @xmath141 which come from the expansion of eq . are neglected . physically , eq . represents the theory of two nambu - goldstone modes which arise due to the spontaneous symmetry breaking . the ignored higher - order terms correspond to interaction between nambu - goldstone bosons . in the present case ( @xmath7 and @xmath144 ) , the interactions are irrelevant and the theory of free nambu - goldstone bosons is asymptotically exact in the large - distance limit , in the symmetry - broken phase . while , in general , the direction of ordering in the reference nel state can be arbitrary in eq . , in the present case we have to choose the reference nel state with the same ordering direction as in the ground state @xmath72 . otherwise , the overlap would vanish . thus , here we take the reference nel state with the ordering in the @xmath139 direction and use eq . . in terms of the free nambu - goldstone boson field theory eq . , eq . is proportional to @xmath145 ( a similar term with @xmath141 gives the same contribution ) @xcite . in the nel state , the spins are completely aligned antiferromagnetically along the @xmath139 axis . considering eq . , it is natural to expect that ( the infrared limit of ) the boundary condition corresponding to @xmath146 is the dirichlet boundary condition for @xmath140 and @xmath141 : @xmath147 imposing @xmath148 , which corresponds to the perfect nel order in the @xmath139 direction , at the boundary @xmath149 . thus the valence bond occupation number , in the large - distance limit , is proportional to the correlation function @xmath150 along the boundary with the dirichlet boundary condition . this correlation function appears to vanish , because @xmath140 is set to zero on the boundary . however , in adopting the boundary picture in the continuum limit , the operator @xmath140 does not have to be exactly at the boundary ; generally we expect that they would be rather located within a short distance @xmath151 , of the order of the ultraviolet cutoff , from the boundary . namely , we postulate that @xmath152 where @xmath153 is the expectation value in the presence of the boundary at @xmath154 , where the dirichlet boundary condition eq . is imposed . a similar trick , known as `` point - splitting technique '' is often used in applications of field theory . within the free nambu - goldstone boson theory eq . , correlation functions with the dirichlet boundary condition can be conveniently calculated by the method of mirror images . for example , . can be written as @xmath155 \left[m^x(\mathbf{r } , \epsilon ) - m^x(\mathbf{r } , - \epsilon)\right ] \rangle,\end{aligned}\ ] ] in terms of the correlation functions denoted by @xmath156 in the infinite plane . for a small @xmath151 , it can be expanded as @xmath157^{\frac{d-1}{2 } } } \big|_{\tau = \tau ' = 0 } \notag \\ & = { { \rm const.}}\frac{1}{|\mathbf{r}|^{d+1 } } . \label{vbdist - d - dim}\end{aligned}\ ] ] thus we have derived the power law length distribution of vbs , in agreement with several earlier works @xcite . the present approach clarifies universality and asymptotic exactness of this power law in quantum heisenberg antiferromagnets . moreover , it enables us to discuss correlations among vbs by analyzing the simultaneous occupation number of multiple vbs . as the simplest among such cases , let us discuss the simultaneous occupation number of two vbs as introduced in eq . , and whether there is any correlation between two vbs eq . can be written in terms of a four - point correlation function of @xmath140 and @xmath141 , through eq . . within the free nambu - goldsone boson theory eq . , thanks to wick s theorem , the four - point correlation function is given in terms of products of two - point correlation functions . we thus find @xmath158 where @xmath159 is the distance between sites @xmath160 and @xmath161 . namely , it is given simply by a sum of products of the average occupation numbers of the individual vbs . this implies that there is in fact _ no correlation _ in valence bond occupations between different pairs of sites , and occupation of each pair occurs _ independently_. this conclusion extends straightforwardly to correlations among occupations of an arbitrary number of bonds . it should be noted that the present field theory describes the asymptotic behavior only in the large - distance limit . the above result does not exclude correlations in a quantum antiferromagnet at short length scales . of valence bonds of length @xmath162 ( in all orientations ) on different clusters with @xmath163 sites . the dashed blue line is a guide to the eyes , showing the expected @xmath164 behavior from field theory . this scaling is observed as long as valence bonds are significantly shorter than the finite - size lattice sample . for very long valence bonds , the data exhibit a small upturn above this power law , as well as a slight anisotropy , due to the periodic boundary conditions . data have been produced for a projection length of @xmath165 for all clusters except the biggest one , for which a projection length of @xmath166 was chosen ( convergence was checked ) . error bars are smaller than the point sizes . inset : valence bond distribution @xmath167 in logarithmic polar coordinates with logarithmic color scheme for @xmath168 in the region @xmath169 ^ 2 $ ] . the data grid is illustrated by small black circles in the range @xmath170 ^ 2 $ ] , whereas the colored mesh represents interpolated data . outside this range the data grid is very dense and is therefore omitted . for short valence bonds the distribution appears to be rather isotropic , in agreement with the principal figure , whereas the boundary effects of the finite size sample can be observed for very long valence bonds . ] here we present the numerical results on the vb distribution for the @xmath11 heisenberg antiferromagnet on a @xmath17 square lattice . one first issue to check is whether the distribution of valence bonds with length @xmath128 is isotropic , as suggested by the field theory predictions . this can be carried out by simulating different two - dimensional clusters with periodic boundary conditions . here , we study clusters with @xmath171 , @xmath172 , @xmath173 and @xmath168 and provide an example of the distribution @xmath174 of valence bonds with length @xmath128 for the biggest cluster in the inset of fig . [ fig : prvbd2d ] . since we expect a power - law decay in @xmath162 , we choose logarithmic polar coordinates for the @xmath175-@xmath176 plane and a logarithmic color scheme for the probability @xmath174 . in the figure we see that the distribution seems indeed to be rather isotropic in the center , a behavior that is perturbed for large @xmath112 , due to the periodic boundary conditions . the isotropic behavior can also be observed in the main panel of fig . [ fig : prvbd2d ] , which shows the decay of the probability for measuring a valence bond with length @xmath112 . we observe that almost all points lie on the power law @xmath164 as predicted by the theory eq . . for very long valence bonds we can see the effect of the boundary as a deviation from the power - law , as well as through a very small anisotropy , characterized by a small spreading fan at large @xmath112 . the points on the envelope of this fan can be identified very accurately . whereas the upper envelope contains points of valence bonds with lengths @xmath177 and @xmath178 , the bottom envelope describes valence bonds of lengths @xmath179 . note , that there are no bipartite valence bonds of lengths @xmath180 . another important issue is that the raw data for sufficiently small @xmath112 in fig . [ fig : prvbd2d ] fall precisely on the same curve for all system sizes , without any extra renormalization . considering the original definition of the vb distribution @xmath181 , there is no obvious reason why @xmath181 is independent of the system size . this , however , can be understood rather naturally from the field theory description discussed in sec . [ sec : sigma ] . once the vb occupation number is mapped to a correlation function , the latter does not depend on the system size as long as the distance @xmath112 is much smaller than the system size @xmath182 . in the present problem , the correlation function is not of a standard type , but in the scaling limit it is nonetheless mapped to the boundary correlation function in the field theory . thus it should not depend on the system size , assuming that the point - splitting scale @xmath151 is insensitive to @xmath182 . the mapping to the field theory works only for a sufficiently large @xmath112 . nevertheless , the system - size independence of the vb distribution @xmath181 at the shortest distance @xmath183 can be understood even without the field theory mapping . that is , @xmath184 is related exactly to the ground state energy @xmath185 of the underlying hamiltonian eq . as @xmath186.\ ] ] since the ground state energy per site , @xmath187 , is an intensive quantity , so is @xmath184 . it is also natural to expect that @xmath181 for an arbitrary distance @xmath112 is also intensive , which is indeed seen in the numerical results . the sum rule eq . requires a constraint on the asymptotic power - law behavior @xmath188 at large distances , which follows from the field theory . let us assume that the power law is valid for @xmath189 , where @xmath190 is the ultraviolet `` cutoff '' scale ( usually of the order of the lattice constant ) above which the field theory applies . furthermore , here we assume that the proportionality constant @xmath191 is independent of the system size ( length ) @xmath182 . we first use the decomposition @xmath192 where both terms in the right - hand side are non - negative by definition . in a finite - size system , @xmath112 is actually cut off by @xmath182 . thus the second term , which is the large - distance part , is estimated as @xmath193 where @xmath194 . only the leading singular part in @xmath195 is retained in the integral in the last manipulation , which is valid only if @xmath196 . this shows that the total contribution from the large - distance part of @xmath197 is finite even in the thermodynamic limit @xmath195 and thus the sum rule eq . can be satisfied . if @xmath198 , on the other hand , the contribution from the large - distance part is divergent and consequently the sum rule will be violated . therefore , the sum rule eq . requires that the asymptotic large - distance power law should have either @xmath196 or the proportionality constant @xmath191 depending on the system size @xmath182 . our finding that @xmath199 and @xmath191 is system - size independent is consistent with this constraint . and @xmath67 in order to measure the correlation between two valence bonds ( shaded ellipses ) of lengths @xmath112 , drawing an angle of @xmath200 . notice that eq . yields one when two valence bonds connect @xmath66 and @xmath67 and vanishes otherwise . this is true for the illustrated vb configuration , as well as for the one that is sketched by the dashed lines . b ) the same configuration with two valence bonds drawing an angle of @xmath201 . ] second , we want to study vb - vb correlations in the ground state of the heisenberg antiferromagnet . in order to see whether correlations are present in two dimensions we choose the two cases that are depicted in fig . [ fig : dimers2d ] , since we intuitively expect correlations to be most pronounced when two valence bonds are very close to each other . we refer to these two cases as those where two valence bonds of equal length draw an angle of a ) @xmath200 and b ) @xmath201 . we shall here emphasize again that we can not directly measure the simultaneous occupation number of two vbs for specified pairs of the sites , but only the sum of occupation numbers corresponding to two possible vb patterns . in the present case , these patterns are drawn as shaded ellipses and dashed lines in fig . [ fig : dimers2d ] , respectively . in the case a ) @xmath200 , for example , we have two valence bonds of lengths @xmath202 and @xmath203 and two other valence bonds of lengths @xmath204 and @xmath205 . in the case b ) we encounter the same valence bond lengths , but at a different angle @xmath201 . let @xmath206 with @xmath207 represent the simultaneous vb occupation numbers in the two cases a ) and b ) , respectively . if there is no correlation between the two vbs , the normalized simultaneous vb occupation number on the pair should be given as @xmath208 where we used the isotropy of the single vb occupation number @xmath209 on a given pair of sites separated by @xmath210 . in fact , there are no other patterns that are expected to have exactly the same decomposition as in eq . . in order to verify eq . , we have to find the proper constant of proportionality . here it is important , that we used translational and rotational invariance in order to measure @xmath174 , @xmath211 and @xmath212 , accounting for a factor of @xmath19 in all cases . for this reason it is obvious that the appropriate constant of proportionality must be @xmath213 . therefore , we have to multiply the correlation data by the system size , when comparing different system sizes @xmath182 , since @xmath174 does not depend on @xmath182 . furthermore , patterns of @xmath211 appear only half as much as those for @xmath212 , due to rotational symmetry . between them , as illustrated in fig . [ fig : dimers2d ] . for comparison , we also show the sum of the products of corresponding single vb occupation numbers as in the right - hand side of eq . . data were obtained for clusters of @xmath163 sites with a projection length of @xmath165 ( convergence checked ) and error bars are smaller than the point sizes . the good agreement among the three different sets of data confirms the decomposition as in eq . , which indicates the asymptotic absence of the correlation between two vbs . this also implies the agreement with the field theory prediction @xmath214 , as shown . ] the properly normalized data are plotted in fig . [ fig : vbc2d ] , and show essentially the expected @xmath215 behavior . this power law is also obtained by the simple product of the pure valence bond occupations , which indicates the absence of correlations between vbs in two dimensions . interestingly , the very first data point for @xmath183 seems to be slightly above the predicted power law @xmath215 . in fact , this point is the only one where the correlation data are slightly above the sum of the products of the occupation numbers for the individual vbs . in the entire rest of the range , this is not the case . this implies the existence of a correlation among vbs at very short distances @xmath216 . such an effect seems to be in agreement with some recent results , which conclude that including correlations between valence bonds at very short distances can indeed improve the energy for variational calculations @xcite . in any case , this does not contradict the field theory analysis , which only applies to the asymptotic behavior at large distances . in fact , the present numerical analysis demonstrates that the field theory works very well above the rather short length scale @xmath217 . for larger @xmath112 comparable to the system size @xmath182 , we see the effects of the periodic boundary conditions , and the data do not follow the field theoretical prediction for the infinite system any more . we furthermore see a more pronounced splitting between the different data for @xmath212 , @xmath211 and the valence bond occupation product . however , these differences are very small ( of order @xmath218 ) compared to the correlations between short valence bonds . in order to check that the same field theory also applies to higher dimensions , we here present qmc results for the case @xmath219 . because simulations become more and more demanding as we increase the dimension , we restrict ourselves to the distribution of valence bonds @xmath181 , which is depicted in fig . [ fig : prvbd3d ] for samples of up to @xmath220 sites . furthermore , for the biggest cluster we could not achieve convergence of the data for @xmath221 ( fortunately , this is a region where boundary effects start to play a role and where deviations from field theory predictions are therefore expected ) . as expected , we observe the same behavior as in two dimensions , but with the exponent in the power law now being equal to @xmath222 . we furthermore observe that the distribution is still isotropic in three dimensions , as can be seen in the inset of fig . [ fig : prvbd3d ] , where a two - dimensional cut through the data is shown . of valence bonds of length @xmath223 ( in all orientations ) on different clusters with @xmath224 sites . the dashed blue line is a guide to the eyes , showing the expected @xmath225 behavior from field theory . notice that qualitatively the only difference from the case @xmath17 ( fig . [ fig : prvbd2d ] ) is the exponent in the observed power law , due to the higher dimension . data have been produced for a projection length of @xmath226 for the two smallest lattices and @xmath227 for @xmath228 ( convergence was checked ) and error bars are smaller than the point sizes . for @xmath229 we used a relatively small projection length of @xmath230 , resulting in non - converged data for @xmath221 ( highlighted in gray ) . inset : valence bond distribution @xmath231 in logarithmic polar coordinates with logarithmic color scheme for @xmath232 in the region @xmath169 ^ 2 $ ] within the y - z plane ( at @xmath233 ) . the data grid is illustrated by small black circles , whereas the colored mesh represents interpolated data . similarly as for @xmath17 , for short valence bonds the distribution appears to be rather isotropic , in agreement with the main panel , whereas the boundary effects of the finite - size sample can be observed for very long valence bonds . ] the physics of the @xmath11 heisenberg antiferromagnet in a @xmath6 chain is quite different from that of the same model in the @xmath7 hypercubic lattice . in @xmath6 , quantum fluctuations are so strong that the long - range antiferromagnetic order is absent even in the ground state . the ground state of the @xmath11 antiferromagnetic chain is critical : various correlation functions decay algebraically . the universal behaviors in the low - energy , long - distance regime are described @xcite as a tomonaga - luttinger liquid ( tll ) . this is a free - boson field theory defined by the lagrangian density @xmath235 the field @xmath236 obeys the equation of motion , which is nothing but the wave equation in one spatial dimension . as a consequence , @xmath236 can be decomposed as @xmath237 where @xmath238 are right - moving and left - moving components . we also introduce the dual field @xmath239 , which is defined by @xmath240 the spin operators are represented in terms of the boson field @xmath236 as @xmath241 where @xmath242 is a parameter of the theory called the compactification radius . for the isotropic heisenberg antiferromagnetic chain , @xmath243 despite the significant difference in physics between @xmath6 and @xmath7 , it is possible to construct a boundary field theory formulation of vb distributions in @xmath6 as well . first , let us identify the conformally invariant boundary condition , which would be the infrared fixed point for the given boundary state @xmath133 . in the reference nel state , the staggered component of @xmath30 is fixed to a constant . thus it is natural to expect that the dirichlet boundary condition @xmath244 is the appropriate boundary condition . the dirichlet boundary conditon is in fact conformally invariant . assuming that this is the case , the valence bond occupation number is just given by the correlation function @xmath245 along the boundary , where the dirichlet boundary condition @xmath246 is imposed . this problem is reduced to the correlation function of the vertex operators in the presence of the dirichlet boundary condition : @xmath247 . such boundary correlation functions have been studied in great detail in boundary conformal field theory . in the present case with the dirichlet boundary condition , the calculation can be quite easily done as follows . let us assume the system is defined on the half plane @xmath248 and the boundary is at @xmath249 , along the @xmath175 axis . because of eq . , imposing the dirichlet boundary condition @xmath246 at the boundary @xmath154 is equivalent to extending the chiral field @xmath250 to the other side of the boundary @xmath251 , where no field was originally defined , by the following relation : @xmath252 this means that the dual field at the boundary @xmath154 can be written entirely in terms of the chiral field @xmath250 . moreover , the correlation function of the chiral field can be evaluated in the infinite plane without boundary . combining this observation with eqs . and , the large distance asymptotic behavior of the valence bond occupation number is given as @xmath253 where @xmath254 now means the expectation value in the infinite plane without a boundary . the result can be readily obtained as @xmath255 where @xmath112 is the distance between the two sites @xmath40 and @xmath56 , and we have set @xmath256 for the isotropic heisenberg chain in the second line . despite the difference in the formulation , the final result for the single vb occupation number in @xmath34 dimensions can be summarized as @xmath257 , whether @xmath6 or @xmath7 . on the other hand , correlations among vbs reveal an interesting difference between @xmath6 and @xmath1 . although the effective field theory ( tll ) is also a free - boson field theory , ( the staggered part of ) the spin operator is expressed by a vertex operator ( exponential in the boson field ) , and not by the boson field itself . thus , while wick s theorem certainly applies to the tll , it does not mean the absence of correlations in multipoint correlation functions of _ spins _ , and consequently , among the vb occupations . in fact , the simultaneous occupation number of two vbs [ eq . ] can be evaluated exactly within the tll theory formulation eq . . following the same logic as in the single vb occupation number , we find @xmath258 where @xmath259 , @xmath260 , and @xmath261 again we have assumed @xmath256 for the su(2 ) symmetric heisenberg antiferromagnetic chain . when two bonds are separated by a large distance ( relative to the bond lengths ) , namely , in the limit @xmath262 we find @xmath263 this implies that @xmath264 namely , the correlation asymptotically vanishes , as the separation between two vbs is taken to infinity . this is consistent with our intuitive expectation . on the other hand , for general separation , our result exhibits a nontrivial correlation between vb occupation . a major feature of the correlation is that , because of the factor @xmath265 in the numerator , the simultaneous occupation amplitude vanishes as @xmath266 or @xmath267 . of valence bonds as a function of @xmath112 for one - dimensional lattices with @xmath182 sites as obtained from qmc calculations , versus field theory predictions . all chains were simulated for a projection length of @xmath268 and the convergence was checked ( error bars are smaller than the point sizes ) . note that we applied the conformal transformation @xmath269 , where @xmath175 is the measured valence bond length . ] let us now compare the theoretical predictions above with the numerical results obtained by qmc simulations . as we have seen earlier , qmc simulations are restricted to finite sizes , and the vb distribution functions inevitably deviate from the theoretical predictions which were made for an infinite system . fortunately , however , our theory for @xmath6 is based on the tll , which is a conformal field theory in @xmath270 dimensions . in this class of field theory , the finite - size effect is conveniently described by a conformal mapping @xcite . the rule of thumb is that , in order to obtain a correlation function in a finite system ( ring ) of length @xmath182 with the periodic boundary condition , a distance @xmath112 appearing in the corresponding correlation function in the infinite system should be replaced by the arc distance @xmath271 on the finite ring . in the one - dimensional case we also first check the length distribution @xmath272 of the valence bonds , as plotted in fig . [ fig : vbd ] . the simulations are carried out on the one - dimensional heisenberg chain with @xmath273 , @xmath274 and @xmath275 sites and periodic boundary conditions . we find a power - law decay @xmath276 , which agrees very well with the field theory prediction eq . with the conformal mapping . here again , all the data in fig . [ fig : vbd ] for different system sizes appear to collapse on a single curve , without any rescaling . that is , the normalization constant @xmath191 is independent of the system size . as we have discussed in sec . [ sec : qmc2d ] , this can be naturally understood from the mapping to the boundary correlation function and the intensiveness of the ground state energy per site . in fact , we can make the analysis in eq . more precise , by using the correlation function in a finite system of length @xmath182 obtained by the conformal mapping , as @xmath277 the leading contribution in the limit of @xmath278 , for @xmath279 , however remains the same as in eq . with @xmath6 . for @xmath280 , the large - distance contribution diverges , as in the case of @xmath281 for general dimension @xmath34 . our finding that @xmath191 is independent of the system size and that @xmath199 applies also to @xmath6 , again in consistency with the sum rule eq . . only at the shortest distance @xmath183 , there is a visible difference between the numerical result and the field theory prediction . as in @xmath7 , the disagreement between the data and the field theory prediction is not surprising , as the field theory describes only the long - distance asymptotic behaviors . the data imply that the ultraviolet cutoff scale @xmath190 , above which the field theory is valid , is small ( @xmath282 ) also in @xmath6 . ( separated by @xmath283 ) and @xmath67 ( separated by @xmath284 ) with the corresponding valence bonds of lengths @xmath285 and @xmath286 . b ) there is also another possible configuration of bipartite valence bonds with lengths @xmath287 and @xmath288 . ] we now study the correlations between vbs on the heisenberg chain , where we choose again the case of two equally long close valence bonds , as illustrated in fig . [ fig : dimers1d ] . again , there are two possible vb patterns and we choose @xmath289 as the varying parameter . the results are plotted in fig . [ fig : vbc1d ] . notice first of all , that eq . uses conformal coordinates , such that we expect a power - law decay as @xmath290 for sufficiently long valence bonds . this implies , however , a rather drastic decrease of probabilities , resulting in relatively rare events . this is indeed observed in the simulations , where probabilities of @xmath291 with pronounced error bars are encountered . nevertheless , we can confirm the predicted @xmath292 behavior and clearly exclude a possible @xmath293 power - law decay obtained from a simple product of valence bond occupations @xmath294 . this is a strong indication , that correlations are indeed present even at large distances , in one dimension . ) in the ground state of eq . on @xmath295 chains of different sizes @xmath182 . the measured correlations multiplied by system size ( data points ) clearly agree with the field theory predictions of a power - law decay @xmath296 ( red dashed curve ) . all data were obtained with projection lengths of @xmath268 ( convergence was checked ) and the error bars are smaller than the point sizes if omitted . notice , that we applied the conformal transformation @xmath269 , with @xmath175 being the measured distances . ] in order to further examine this behavior , we now study a case where three points of this four - point correlation function are fixed and one point is variable , as shown in fig . [ fig : vbcpumps ] . we plot the vb - vb correlations , as well as the product of the valence bond occupations , which both reproduce the theoretical predictions . in agreement with our naive intuition , both theories show the same behavior if we consider short valence bonds , separated by a rather large distance . this can be seen in the figure when @xmath297 or @xmath298 and means that no further correlations are present in such a case . however , the two curves differ considerably for @xmath299 , i.e. , when the two valence bonds start affecting each other . this confirms that in such a case correlation effects are clearly important in one dimension . sites for the two ( undistinguishable ) configurations @xmath300 and @xmath301 , with @xmath302 , @xmath303 , @xmath304 and @xmath305 varying . we show the product of valence bond occupations ( green squares ) , that fit to a theory without correlation as comparison . however , the measured correlations ( multiplied by the system sizes , shown as black circles ) rather correspond to a theory with correlations . notice , how the two theories agree asymptotically around @xmath306 and @xmath307 and disagree substantially at @xmath308 . the projection length is @xmath268 ( convergence was checked ) and error bars are smaller than point sizes . ] through a combination of analytical calculations and quantum monte carlo simulations , we have investigated the behavior of valence bond probability distributions and correlations in ground states of the antiferromagnetic heisenberg model on chain , square and cubic lattices . our analytic results provide a clear ground for the probability distribution @xmath3 for a valence bond to join two sites ( existing on different sublattices ) separated by @xmath81 to scale as @xmath309 where @xmath34 is the lattice dimension . this result , which was observed in previous numerical simulations and justified by a mean - field ansatz , is now understood within a more trustworthy analytical framework : bosonisation in 1d , and the non - linear @xmath2 model in @xmath91 . this will be useful for variational calculations , or for constructing guiding wave functions in monte carlo simulations , which aim at targeting an antiferromagnetic state on a bipartite lattice . moreover , our results provide a formal justification for the factorization ansatz eq . introduced by liang , douot and anderson , as long as a non - linear @xmath2 model description is valid . indeed , our results , corroborated by large - scale qmc calculations in the 2d case , clearly indicate the absence of correlations between valence bonds at long distances . for the quasi - long - range ordered ground state of the heisenberg chain , the situation is more subtle as correlations are present between valence bonds , as clearly observed in fig . [ fig : vbcpumps ] . this may be understood as a signature of the strong ( power - law - decaying ) dimer correlations which are known @xcite to be present in the heisenberg af chain . we would also like to point out that the existence of long - range antiferromagnetic order does not necessarily imply a distribution of valence bonds with a distribution law @xmath310 with no correlations . for instance , the wave functions composed of the equal - weight linear combinations of nearest - neighbor valence bonds on the simple cubic or diamond lattices have been shown to sustain af long - range order @xcite ( see also ref . ) . however , their description is most certainly beyond the non - linear @xmath2 model approach , which probably can not describe the dipolar dimer correlations observed in these wave functions @xcite . finally , we suggest several further investigations as possible extensions of our work . for one - dimensional systems , the bosonization analysis of sec . [ sec : bos ] carries over for the critical phase of the @xmath311 anisotropic spin chain . while the su(2 ) symmetry is lost in this more general case , analytical predictions for mixed expectation values of the type @xmath312 [ such as in eq . ] can be made and tested through , for instance , dmrg or qmc calculations . in general , the power - law exponent will depend on the anisotropy parameter through the compactification radius @xmath242 . another interesting problem would be to calculate analytically logarithmic corrections to the power - law decays of valence bond occupations and correlations in one dimension , which most certainly exist in the su(@xmath8 ) case . we note , however , that these are presumably small , hard to detect , effects as the qmc results are already very well described by the pure power - law decays . in dimensions larger than 1 , the non - linear @xmath2 model approach may also be used to describe the behavior of valence bond distributions and correlations at a quantum critical point between the antiferromagnet and a paramagnet , such as for a bilayer heisenberg model @xcite . this could be useful in explaining and improving variational approaches based on valence bonds that aim to describe this quantum phase transition ( such as , _ e.g. _ , ref . ) . more generally , the current formalism based on the overlap eq . could be used with any effective field theory which describes a particular quantum phase or a quantum critical point of quantum antiferromagnets on bipartite lattices . finally , it would be very interesting to see whether an approach similar to the one developed here could be applied to describe the `` spinon detection '' procedure recently advocated by tang and sandvik @xcite , which is also based on a valence bond description of a spin system with one or two unpaired spins ( _ i.e. _ , that do not belong to a valence bond ) . this is a more challenging case as one would first need to derive equations similar to eq . or for this situation . we thank matthieu mambrini for very insightful discussions and participation at an early stage of this project . this work was performed using hpc resources from genci - ccrt , genci - idris ( grants x2010050225 , x2011050225 , x2012050225 ) and calmip ( grants 2011-p0677 and 2012-p0677 ) and is supported by the french anr program anr-08-jcjc-0056 - 01 , the indo - french centre for the promotion of advanced research ( ifcpar / cefipra ) under project 4504 - 1 , and mext / jsps kakenhi grant nos . 20102008 and 25400392 . m. o. thanks laboratoire de physique thorique , irsamc , universit de toulouse and cnrs for hospitality during his visits , during which the present work was initiated and then completed . e. jordo neves and j. fernando peres , phys . a * 114 * , 331 ( 1986 ) ; i. affleck , t. kennedy , e.h . lieb and h. tasaki , phys . * 115 * , 477 ( 1988 ) ; f.j . dyson , e.h . lieb and b. simon , j. stat . * 18 * , 335 ( 1978 ) r.r.p . singh and d.a . huse , phys . * 68 * , 1766 ( 1992 ) ; b. bernu _ et al . _ , b * 50 * , 10048 ( 1994 ) ; l. capriotti , a.e . trumper , and s. sorella , phys . lett . * 82 * , 3899 ( 1999 ) ; s.r . white and a.l . chernyshev , phys . lett . * 99 * , 127004 ( 2007 ) f. alet _ et al . _ , lett . * 99 * , 117204 ( 2007 ) ; r.w . chhajlany , p. tomczak , and a. wjcik , phys . lett . * 99 * , 167204 ( 2007 ) ; a.b . kallin , i. gonzlez , m.b . hastings , and r.g . melko phys . * 103 * , 117203 ( 2009 ) ; h. tran and n.e . bonesteel , phys . b * 84 * , 144420 ( 2011 ) . following the logic here , exchanging @xmath313 and @xmath314 still gives the same result in the field theory . however , in the original formula eq . , the amplitude vanishes after the exchange of the spin operators @xmath313 and @xmath314 . this shows the presence of a subtlety in the mapping to the field theory . despite this , we believe that the asymptotic large - distance limit is correctly given by the present field - theory prescription , for the appropriate ( nonvanishing ) amplitude . this will be also supported by numerical calculations in sec . [ sec : qmc2d ] .

we have used a relatively long , contiguous vhf observation of a bright cosmic radio source ( cygnus a ) with the very large array ( vla ) through the nighttime , midlatitude ionosphere to demonstrate the phenomena observable with this instrument . in a companion paper , we showed that the vla can detect fluctuations in total electron content ( tec ) with amplitudes of tecu and can measure tec gradients with a precision of about tecu km . we detail two complementary techniques for producing spectral analysis of these tec gradient measurements . the first is able to track individual waves with wavelengths of about half the size of the array ( km ) or more . this technique was successful in detecting and characterizing many medium - scale traveling ionospheric disturbances ( mstids ) seen intermittently throughout the night and has been partially validated using concurrent gps measurements . smaller waves are also seen with this technique at nearly all times , many of which move in similar directions as the detected mstids . the second technique allows for the detection and statistical description of the properties of groups of waves moving in similar directions with wavelengths as small as 5 km . combining the results of both spectral techniques , we found a class of intermediate and small scale waves which are likely the quasi - periodic ( qp ) echoes that have been observed to occur within sporadic- ( ) layers . we find two distinct populations of these waves . the members of one population are coincident in time with mstids and are consistent with being generated within layers by the coupling instability . the other population seems more influenced by the neutral wind , similar to the predominant types of qp echoes found by the sporadic- experiments over kyushu ( seek * ? ? ? * ; * ? ? ? * ) . we have also found that the spectra of background ( i.e. , isotropic ) fluctuations can be interpreted as the sum of two turbulent components with maximum scales of about 300 km and 10 km .

1201.3874

ground - based remote sensing of the ionosphere is a rich field of study that has developed over several decades and now includes a host of different instruments including , but not limited to , gps , certo @xcite , and other radio beacon receivers , ionosondes , and radar / hf arrays . a powerful yet relatively underused resource for remote sensing are radio - frequency arrays , particularly those that operate in the vhf regime . designed as radio synthesis telescopes , they are chiefly used to observe cosmic sources . while such observations require detailed calibration schemes to remove the effects of the ionosphere , this calibration data is seldom used to actually study the ionosphere . these interferometers basically measure the time - averaged correlation of complex voltages from pairs of antennas which can be combined with fourier methods to make relatively high angular resolution images of cosmic sources ( for a detailed discussion of the methods involved , see * ? ? ? the correlated signals , or `` visibilities '' have an extra phase term added to them by the ionosphere proportional to the difference in the total electron content ( tec ) between the two antennas lines of sight . because of the relatively large collecting area of the individual elements ( usually dishes , dozens of meters across ) , and the brightness of many cosmic sources , these additional phase terms can typically be converted to differential tec ( @xmath7 ) measurements with a precision of @xmath8 tecu or better . in addition , the range of scales that can be probed with such interferometers ( dozens of meters to hundreds of kilometers ) and the virtual ubiquity of target sources available on the sky make them valuable assets for the exploration of ionospheric dynamics from very fine to medium size scales . consequently , radio arrays have been used , to a somewhat limited degree , for ionospheric studies . because of the relative stability of its electronics , quiet radio frequency interference ( rfi ) environment , and available vhf system ( bands at 74 and 330 mhz ) , the very large array ( vla ) has been used almost exclusively in these efforts . seminal experiments were performed by @xcite using 330 mhz vla observations of several sources to explore the environment of ionospheric waves above the vla . among other results , they discovered a new class of magnetic eastward directed waves that were later shown to actually be located within the plasmasphere @xcite . subsequent larger - scale investigations using data from the vla low - frequency sky survey ( vlss ; * ? ? ? * ) , a survey of the northern sky at 74 mhz , have shown that the median behavior of ionospheric fluctuations observed by the vla over several years is essentially consistent with turbulence @xcite . joint observations made with an all - sky optical camera and the vla at 74 and 330 mhz by @xcite demonstrated the vla s ability to aid in the exploration of the interaction between gravity waves generated lower in the atmosphere and relatively small - scale phenomena such as sporadic-@xmath4 . recently , from a campaign using the vla and the cosmic satellite , @xcite were able to detect a relatively rare instance of a southeastward - propagating traveling ionospheric disturbance ( tid ) . in a companion paper @xcite , we described in detail the methods for extracting ionspheric information from the calibration of vla vhf system data . we demonstrated that when observing a bright cosmic source , the vla vhf system was capable of achieving a typical @xmath7 precision of @xmath9 tecu . since arrays such as the vla are essentially only sensitive to the tec gradient , we also detailed techniques for measuring time series of the tec gradient to a precision of about @xmath1 tecu km@xmath2 . here , we seek to develop this effort further by presenting new spectral analysis techniques for these tec gradient measurements . we will demonstrate that these techniques are capable of detecting and characterizing several phenomena from medium ( @xmath10 km ) to small ( @xmath11 km ) scales while also providing a broader statistical description of the spectrum of tec fluctuations observable with the vla . the data used in this analysis came from a roughly 12-hour vla observation of one of the brightest cosmic radio sources , cygnus a ( or `` cyg a '' ; also known as 3c405 ) from a latitude and longitude of @xmath12 n and @xmath13 w ( vla program number ak570 ) . the observations were conducted during the night of 12 - 13 august , 2003 simultaneously at 74 and 327 mhz , both operating in dual polarization . the vla was in its `` a '' configuration spanning a circle with a diameter of about 40 km . the observations consisted of a 35-minute block of time , or `` scan , '' and a second 12-hour scan with a temporal sampling of 6.67 seconds . there was a moderate amount of geomagnetic activity ( @xmath14 index @xmath154 ) and the amount of solar activity was fairly average ( @xmath16 sfu ) . measurements of the tec gradient toward cyg a are described in detail in @xcite . in short , the ionosphere adds an extra phase term to the complex visibility measure for each pair of antennas which is proportional to the difference in the tec along the two lines of sight , or @xmath7 . thus , the interferometer is sensitive only to tec gradients and not absolute tec . however , the y - shape of the array ( northern , southeastern , and southwestern `` arms '' ) makes measuring and performing fourier inversions of the tec gradient difficult . this is in contrast to the standardized procedure of inverting visibilities to produce images of cosmic sources . this is because the visibilities are functions of the _ differential _ antenna positions and the tec gradient is a function of the _ actual _ antenna positions projected through the ionosphere . this is discussed in more detail by @xcite . because of the consequences of the y - shape of the vla , @xcite established two ad hoc approaches to measuring the tec gradient . the first method estimates the full two - dimensional gradient at each antenna using a second - order , two - dimensional polynomial ( taylor series ) on each time step . while insensitive to fine - scale structure significantly smaller than the array , @xcite showed that this method is effective in recovering the gradient associated with relatively large , long - period disturbances visible at dusk , dawn , and intermittently throughout the night . the second method numerically computes the projection of the tec gradient along each of the vla arms at each antenna and time step . while little directional information can be obtained with these measurements , @xcite showed that there are many significant fine - scale fluctuations , especially near midnight local time , that are detected with this method and missed / damped within the polynomial - based method . in the following section , we will detail the techniques we have developed to spectrally analyze the tec gradient time series produced by these two methods . to identify individual disturbances or sets of disturbances that passed over the array , we have developed the following fourier - based method which uses the polynomial - based tec gradient measurements . this method is partially based on that used by @xcite , but expands upon it to include some allowance for wavefront distortions and multiple wave fronts observed simultaneously . the basic concept is to use both the apparent motion of cyg a and the movements of the observed ionospheric disturbances to increase the spatial coverage of the vla . in other words , we will essentially convert temporal baselines into spatial ones , sampling the observed phenomena with a relatively long , 40 km - wide `` strip . '' since there are presumably several disturbances contributing to the tec gradient an each time step with a range of speeds , the temporal sampling can not be converted into the same set of spatial samples for all observed phenomena . instead , we must separate the time series for each antenna into different fourier modes and then analyze how the properties of each mode vary across the array to estimate the wavelength , speed , and direction of the contributing pattern(s ) . to achieve this , we first assumed that the tec fluctuations could be approximated as the sum of many individual oscillating modes , each having the following form @xmath17\ ] ] where @xmath18 and @xmath19 are the wavenumber and position vectors , respectively , @xmath20 is a complex amplitude which only varies perpendicular to @xmath18 ( i.e. , wavefront distortions ) , and the temporal frequency , @xmath21 , is assumed to be @xmath22 where @xmath23 is the wave speed . for a single fourier mode with temporal frequency , @xmath21 , the fourier transforms of the @xmath24 ( n - s ) and @xmath25 ( e - w ) partial derivatives are given by @xmath26 \mbox{e}^{- i \vec{k_{\nu } } \cdot \vec{r } } \\ f_{y}(\nu ) \simeq \left [ i k_{\nu , y } + \frac{\partial}{\partial y } a_{\nu}(r_{\perp } ) \right ] \mbox{e}^{- i \vec{k_{\nu } } \cdot \vec{r}}\end{aligned}\ ] ] from these equations , one can see that along a line parallel to @xmath18 , the @xmath24 and @xmath25 dependences of the phases of @xmath27 and @xmath28 are identical since both partial derivatives of @xmath29 are constant along this line . there may be other lines in the @xmath30 plane where this is true . for instance , in the absence of significant wavefront distortions , the @xmath24 and @xmath25 dependencies will be the same everywhere . however , in the general case , the line parallel to @xmath18 is the only one where these dependencies are required to be equal according to equations ( 2 ) and ( 3 ) . we have exploited this fact in our analysis to help determine the direction of @xmath18 for different fourier modes and consequently , the wavenumbers and velocities . to illustrate this better , we have displayed an example of a wave , including wavefront distortions , in fig . [ exwave ] . if one imagines this wave drifting over the vla at a constant speed , one can see that the time series measured at two separate points will have identical periods , but will be out of phase . for a constant speed , this phase difference will simply be proportional to the length of the separation between the points . however , if one examines the n - s and e - w partial derivatives of this wave along different position vectors , one can see that the wavefront distortions add additional variations in phase . we show this in the lower panels of fig . [ exwave ] where we have plotted the n - s and e - w partial derivatives , @xmath27 and @xmath28 , along three different vectors shown on the image of the wave with color - coded arrows . in all cases , @xmath27 and @xmath28 are out of phase with one another . this phase difference is constant for the vector that is parallel to the wavenumber vector but varies with distance along the other two vectors . therefore , when wavefront distortions are present , the time series phases will have the same spatial derivative for both @xmath27 and @xmath28 only when points along the direction of the wave are considered . while this derivation focuses on what one should expect for a single wave pattern , we must recognize that at a given temporal frequency , @xmath21 , there will likely be contributions from several waves . this can result from several factors including the finite size of the temporal window used for the fourier inversion to the existence of phenomena such as turbulent cascades which have distributions of spectral power that span a large range of frequencies . it this case , wave properties derived for a single mode using equations ( 2 ) and ( 3 ) will be weighted mean values with more weight given to larger amplitude fluctuations . in addition , even for a single dominant wave , the actual wavenumber and speed for a particular value of @xmath21 may change over the span of time used in the fourier inversion . therefore , one should take the measured values for these properties to be averages over either the temporal window used for the fourier transforms or the duration of the disturbance(s ) , whichever is shorter . thus , we may identify instances where single dominant waves are present by deriving wave properties using equations ( 2 ) and ( 3 ) for several temporal frequencies at each time step . this can be done using fourier transforms computed with a `` sliding '' window , one that is relatively wide but centered on each individual time step . the derived spectral power , direction , and speed can then be examined as functions of time and @xmath21 . any relatively large regions in the @xmath31,@xmath21 plane with relatively large power and uniform direction and speed are likely instances of dominant waves . the wave properties measured in all other locations in the @xmath31,@xmath21 plane are likely composite values from several waves and can be used to study wave properties on a statistical basis . using the polynomial - based tec gradient fits , we calculated @xmath27 and @xmath28 from equations ( 2 ) and ( 3 ) as functions of time by computing the discrete fourier transform ( dft ) of each polynomial coefficient within one - hour sliding windows for frequencies up to 12 hr@xmath2 ( or , periods @xmath32 minutes ) . we chose the one - hour width because it was the same used to de - trend the @xmath7 data ( see * ? ? ? because of the one - hour box width , we excluded the first 35 minute scan from this analysis . we also only computed @xmath27 and @xmath28 for times ranging from one half hour after the beginning of the second scan to one half hour before the end of the observing run so that a full hour of data could be used for each time step . the dft of each polynomial coefficient was then used to compute @xmath27 and @xmath28 at the locations of the antennas where the values of the polynomial fits are most reliable . upon inspection of this data , we found that the phases of both @xmath27 and @xmath28 at a single value of @xmath21 were typically well approximated with planes across the array . because of this , at each time step and value of @xmath21 , we fit planes to both these quantities . we then checked the combined ( in quadrature ) rms scatter about both of these fits against the rms difference between the phases of @xmath27 and @xmath28 to see if the @xmath24 and @xmath25 dependencies of these phases indeed differed across the array , indicating significant wavefront distortions . in the vast majority ( @xmath33% ) of cases , the rms difference between the phases of the two partial derivative dfts was larger than the combined rms scatter about the two planar fits ( a median of four times larger ) . in these cases , we assumed that the azimuth angle ( measured clockwise from north ) of the line where the two planes intersected was the azimuth angle of @xmath18 and that the derivative of the dft phase along this line gave @xmath34 . in the rare instance that there was no evidence of wavefront distortions , we computed a mean @xmath24 and @xmath25 slope for the dft phases from the two planar fits and computed the direction and magnitude of @xmath18 from these slopes . the upper panel of fig . [ waveprop ] displays the total power of the tec gradient in the direction of @xmath18 as a function of local time and temporal frequency . while seemingly random fluctuations are frequently apparent , there are several instances where one can see significant detections of waves above the background . most of these waves are seen between frequencies of 1 and 5 hr@xmath2 , or periods of 12 to 60 minutes . this is roughly consistent with what is typical for mstids . to isolate these detections , we computed a mask by measuring the median and median absolute deviation ( mad ) within elliptical `` annuli '' around each pixel that had dimensions of 1.5 hr in local time by 3 hr@xmath2 in @xmath21 . any pixel with a value more than three times the mad above the median for its annulus was considered a detection above the background . we used this mask to display both the azimuth angle for @xmath18 and the wave speed for significant detections in the middle and lower panels of fig . [ waveprop ] , respectively . the plots in fig . [ waveprop ] imply that the candidate mstids , i.e. fluctuations with periods @xmath3 minutes or greater , have estimated speeds between 100 and 200 m s@xmath2 , consistent with typical mstids detected at night during northern hemisphere summer @xcite . they also appear to move roughly westward / northwestward before midnight and toward the northeast after midnight . while @xcite showed that summer nighttime mstids in the northern hemisphere predominantly move toward the southwest , @xcite showed that near the west coast in california , they move almost due west and occasionally toward the northwest . thus , the waves observed before midnight are somewhat typical mstids . however , the northeastward directed waves seen after midnight are unusual and will be discussed in more detail below . while the results displayed in fig . [ waveprop ] are useful for identifying and examining instances of waves or groups of waves , the polynomial - based analysis can also be used to produce a more statistical description of the observed set of tec fluctuations . in fig . [ wavespec ] , we have plotted the mean power within bins of wavenumber , @xmath35 , for one - hour blocks of time . we note that these wavenumbers have not been corrected for any doppler shifts caused by the motion of the ionosphere pierce - points as the vla tracked the apparent motion of cyg a through the sky . however , the results shown in appendix a of @xcite demonstrated that the equivalent velocities at the estimated peak heights for the pierce - points are relatively small ( @xmath36 ) , except near the beginning and end of the observing run when cyg a was rising and setting . at these times , the main features detected in the data shown in fig . [ waveprop ] generally move roughly perpendicular to the pierce - point motions , implying that the doppler shifts were relatively small during these times as well . we have also plotted in fig . [ wavespec ] what we will refer to as `` noise - equivalent '' spectra . these spectra were computed by performing the polynomial fits to each of the four @xmath7 measurements ( two bands and two polarizations ) and computing the dft of the difference between those fits and the fits to the final @xmath7 . the wave analysis detailed above was then applied to these residual dfts to estimate the results one would achieve if analyzing fluctuations that were simply noise . the noise - equivalent spectra plotted in fig . [ wavespec ] are average spectra computed within each one - hour block and among the two bands and two polarizations . these spectra imply that this technique can indeed detect significant power of structures as small as about half the size of the array ( or , @xmath37 ) . the mstids visible in the upper panel of fig . [ waveprop ] are visible in most of the panels as significant bumps above the background at wavenumbers ranging from 0.0150.05 km@xmath2 , or wavelengths of about 130500 km , which is again roughly consistent with the known properties of mstids . few if any features are see at higher wavenumbers where the spectrum ranges from a simple power - law , reminiscent of turbulence , to relatively flat spectra , which may imply a population of intermediate - scale waves . to explore these data further , we have computed mean power spectra within bins of @xmath18 azimuth angle . we have displayed these as images in fig . [ wavespecpa ] where we have divided the data within the same wavenumber bin by the total power within that bin over all azimuth angles to enhance any detected features at higher wavenumbers . these images reveal that there are often intermediate - scale waves ( @xmath38 , or wavelengths @xmath39 km ) detected moving in some preferred direction . often , these waves appear to be moving in directions similar to detected mstids . this is especially apparent in the first three panels where one can see that even when the mstids have disappeared by a local time of @xmath40 , the roughly northward moving intermediate - scale waves persisted . the presence of strong mstids do not appear to be required for these intermediate - scale waves as they seem to exist during all time periods . however , the mstids broadly appear to have the effect of `` focussing '' the intermediate - scale waves into a particular direction . those smaller - scale waves which are directed parallel to coincident mstids may simply be small - scale structures within the mstids themselves . indeed , our spectral analysis method has indicated that significant wavefront distortions are common within the vla data . however , the smaller waves which do not travel in the same direction as mstids may be located within the @xmath4 region , most likely within sporadic-@xmath4 ( @xmath5 ) layers given the prevalence of this phenomenon during summer nighttime . in fact , @xcite showed that @xmath5 layers are likely the source of small - scale tec fluctuations often observed with the vla during summer nighttime . in addition , @xmath41-km - sized wave - like fluctuations have frequently been observed within @xmath5 layers as so - called quasi - periodic ( qp ) echoes ( e.g. , * ? ? ? * ; * ? ? ? if so , these waves may yield further insights into the coupling between the midlatitude nighttime @xmath4 and @xmath6 layers ( see * ? ? ? * ; * ? ? ? * ) as the mstids are likely located within the lower @xmath6 region ( see , e.g. , * ? ? ? this will be discussed further in 5 . to partially validate our technique for detecting and characterizing waves , we have analyzed all available gps data in the area for the time of our observations . from 1213 august , 2003 , there were six dual frequency gps receivers operating within the state of new mexico ( station codes azcn , nmsf , pie1 , sc01 , tcun , and wsmn ) with publicly available data in rinex format . in contrast , there are currently more than 30 such stations in operation which bodes well for future studies conducted with the new vhf system being developed for the expanded vla ( evla ) . data for each of the six stations operating in 2003 were obtained and processed with standard gps toolkit @xcite routines ( ` discfix ` and ` rescor ` ) to produce slant and vertical ( for a height of 400 km ) relative tec values ( i.e. , no bias correction was done ) for all available satellites . for each station / satellite pair , the slant tec data were de - trended according to @xcite where they de - trended similar data by subtracting from each time step the average of two values , each an interval @xmath42 before or after the time step . they noted that for a wave with a period @xmath43 , this technique effectively multiplies the wave amplitude by a factor of @xmath44 . they chose a value of @xmath45s based on physical arguments of the expected periods of mstids . we have instead chosen @xmath46s so that the de - trending technique was optimized for waves with periods similar to the mstids detected within our vla data . after the gps data were de - trended , they were corrected to vertical tec values using the corrections computed by ` rescor ` within gps toolkit . within two - hour blocks , starting at a local time ( relative to midnight , 13 august ) of @xmath47 , we then identified station / satellite pairs which had relatively contiguous data ( @xmath48% of the time interval covered ) from one half hour before and one half hour after the time block . we then computed the dft of the de - trended tec data within a one - hour sliding window to produce a spectrum at each time step within the two - hour block . these spectra are displayed in fig . [ exgps ] within each two - hour block for the 12 station / satellite pairs with ionospheric `` pierce - points '' closest to that associated with cyg a. nearly all of these spectra show some level of wave activity with periods of about 1260 minutes ( @xmath495 hr@xmath2 ) . those station / satellite pairs with pierce - points closest to that of cyg a in particular show very good agreement with the vla - derived spectrum displayed in the upper panel of fig . [ waveprop ] . this is especially true for the top three rows of fig . [ exgps ] for local times of @xmath47 to @xmath40 and @xmath50 to @xmath51 . unfortunately , the spacing among the gps stations was too large to be able to track individual waves . the closest separations among pierce - points is @xmath52 km when separations on the order of tens of kilometers are required to overcome the effects of wavefront distortions ( e.g. , * ? ? ? this makes it difficult to use the gps data to directly validate the directional information we have determined for the vla - detected waves displayed in the middle panel of fig . [ waveprop ] . however , the fact that the spectra shown in fig . [ waveprop ] and [ exgps ] observe many of the same waves is encouraging and at least partially validates our vla - based approach . while there were no ionosondes operating in the immediate vicinity of the vla , there were two relatively nearby at similar latitudes , one at dyess air force base ( afb ) in west texas ( @xmath53 n ; @xmath54 w ) , and one at point arguello in california ( @xmath55 n ; @xmath56 w ) . the contemporaneous data from these stations are useful for interpreting our vla observations . for instance , northward or northeastward - directed mstids , similar to those observed here at @xmath50@xmath51 and @xmath57@xmath58 , have been observed over japan with the superdarn hokkaido hf radar and an airglow imager by @xcite . like the northeastward - directed mstids described here , these were detected at midlatitudes during nighttime . @xcite noted that the roughly northward - directed mstids were observed with the superdarn array most frequently in may and august . they also noted that the northeastward - directed waves were coincident in time with a drop in the height of the f - region as indicated with nearby ionosondes data . to demonstrate that this may also be the case for our vla observations , we have plotted h@xmath59f for both the dyess afb and point arguello ionosondes as functions of vla local time in the upper panel of fig . [ ionos ] . the point arguello data show a substantial drop in h@xmath59f at around @xmath50 at the start of the first observed instance of northeastward directed waves . the dyess afb data shows a decline in h@xmath59f which begins around @xmath60 and reaches a minimum at about @xmath58 , corresponding to the middle of the second period of northeastward mstid activity . this indicates that it is plausible that we have observed a similar phenomenon with vla over new mexico as was found by @xcite over japan . the ionosondes data also serve to validate our claims that @xmath5 layers were present during our observations . in the lower panel of fig . [ ionos ] , we have plotted foes for both ionosondes stations . both stations observed e - region reflections with maximum frequencies between 2 and 6 mhz throughout the night , strongly indicating that @xmath5 layers were present . the most active time for the appearance of these layers seems to be between @xmath61 and @xmath58 vla local time , especially for the point arguello station . as evidenced by fig . [ wavespec ] and [ wavespecpa ] , relatively small - scale fluctuations are visible within the vla data throughout this time period . they are especially prominent from @xmath62 to @xmath50 where their presence has made the mean spectra shown in fig . [ wavespec ] virtually flat for wavenumbers @xmath63 km@xmath2 . while the fourier analysis detailed in 3 has yielded interesting results , it neglects the ability of the vla in its a configuration to detect fluctuations on scales as small as a few kilometers . to make full use of the data , we have developed a complementary spectral analysis which uses the projected tec gradients measured along each arm which are sensitive to fluctuations with sizes as small as the shortest antenna spacings within the vla . to analyze the projected gradients in a manner similar to the method used for the polynomial - based gradients , we first performed fast fourier transforms ( ffts ) of the times series for each antenna . this was done within a sliding window of approximately one hour ( 512 time steps , or 56.9 minutes ) centered on each time step . then , for a single arm , time step , and temporal frequency , we unwrapped the phases of the ffts along the arm and numerically computed the derivatives of these phases with respect to distance along the arm to obtain estimates of the ( mean ) projected wavenumber , @xmath64 , of each fourier mode . as noted in 3 , our fourier analysis of the polynomial fits revealed that significant wavefront distortions were common . because of this , we expect that these @xmath64 estimates are not reliable for tracking individual disturbances along any of the arms . however , since the effects of such distortions should average to zero over time , these estimates are sufficient to perform a statistical analysis of the overall population of waves seen by the vla from a few to hundreds of kilometers in wavelength . to begin such a statistical analysis , we binned the fft and @xmath64 data within one - minute intervals and computed the mean fft power within bins of @xmath64 for each arm . the resulting spectra are displayed in fig . [ armspec ] where positive projected wavenumbers correspond to directions outward from the center of the array , or `` up '' the arm and negative projected wavenumbers correspond to inward directions , or `` down '' the arm . since these are projected wavenumbers , instances where there are groups of waves moving in a particular direction will show up in these plots as over - densities in the positive ( negative ) direction for one arm , and in the negative ( positive ) direction for the other two . in cases where a group of waves moves nearly perpendicular to a particular arm , they will not show up at all in that arm s spectrum . a good example of this can be seen between roughly @xmath60 and @xmath51 local time ( highlighted with vertical white lines in the panels of fig . [ armspec ] ) . here , there is an obvious over - density on the positive side of the spectrum for the southwestern arm and mild over - densities on the negative sides of the spectra for the northern and southeastern arms . this implies that these fluctuations are moving almost directly toward the southwest . part of this group of waves can also be seen the panel of the upper right corner of fig . [ wavespecpa ] , but in fig . [ armspec ] , they can be seen to extend to wavenumbers of at least 0.7 km@xmath2 , or wavelengths at least as small as 9 km . to establish the minimum scales on which this analysis can be considered reliable , we have also computed noise - equivalent spectra using the full two - band , two - polarization data as we did with the polynomial - based fourier analysis in 3 . we have plotted in fig . [ armspec1hr ] mean spectra for each arm within one hour blocks of time along with the corresponding noise - equivalent spectra . in some instances , the spectra are always above the noise - equivalent spectra . however , in general , the spectra reach the noise level at projected wavelengths of about 5 km , or @xmath65 ( see the vertical dotted lines in fig . [ armspec1hr ] ) . this is in keeping with the mean antenna separation along the arms of about 2.5 km , implying an approximate nyquist sampling limit of 5 km . because of this , we consider these spectra most reliable for @xmath66 . we note that for most of these one - hour mean spectra , the relatively small - scale ( @xmath67 ) portions of the spectra that are above the noise level are roughly consistent with what is expected for @xmath5 layers . from the data presented by @xcite , a typical time series of @xmath5 activity would yield values for the spectral power ranging from about @xmath68 up to @xmath69 mtecu@xmath70 km@xmath71 hr@xmath70 . this is nearly exactly the range inhabited by the intermediate / small - scale regions of the one - hour mean spectra plotted in fig . [ armspec1hr ] . in addition , the contemporaneous ionosondes data presented in 3.3.2 and fig . [ ionos ] show that @xmath5 layers were likely present during the vla observations . for the remainder of the analysis , we found it useful to define three size classes for the detected waves : ( 1 ) medium scale , @xmath72 which includes the full range of mstid sizes , ( 2 ) intermediate scale , @xmath73 , which represents the remaining range of scales one can probe with our previous polynomial - based approach , and ( 3 ) small scale , @xmath74 representing the scales that can not be probed with the polynomial fits . we have re - displayed the arm - based spectra in fig . [ armspecmed ] with these three size classes marked . in this representation , we have divided the spectra by an estimate of the spectra of background fluctuations , i.e. , those that do not move in a preferred direction . we have estimated this background at each value of @xmath75 and time step using a combination of the spectra from all three arms . in this process , only one of the two directions , up or down the arm , was used for each arm . in each case , the direction with the lowest power was used as it was more likely to be dominated by isotropic fluctuations . the median power among all three arms ( within @xmath76 pixel in both time and @xmath75 ) was then taken to be the value for the background . for displaying purposes only , we also applied a three - pixel square median filter to the arm - based spectra and divided these by the estimated background spectra and have shown them in fig . [ armspecmed ] . in this representation , one can see that groups of waves at virtually all detectable scales are seen moving in a variety of preferred directions at different times . to exploit the arm - based spectra described above and displayed in fig . [ armspec][armspecmed ] to identify groups of waves moving in a particular direction , we have developed the following method . to begin , we computed a weighted mean value of @xmath64 for each size class , arm , and one - minute interval using the spectral power as the weight . if there is a significant population of waves within any of the three classes at a particular time step moving in a similar direction , these weighted mean values will follow a cosine as a function of the azimuth angle of the arms which will peak in the direction of the true wavenumber vector , @xmath18 . to detect such groups of waves , we have fit a simple cosine model to the weighted mean @xmath64 values at each time step , including the nearest two time steps so that the fit would be better constrained and so some level of temporal coherence would be imposed . we then compared the amplitude of the fitted cosine with the rms scatter about the fit among the nine values used ( i.e. , three arms and three time steps ) to assess the significance of the detection of a group of waves . we have plotted the azimuth angles for @xmath18 for both @xmath77 and @xmath78 detections for each of the three size classes in the panels of fig . [ armspecpa ] . the @xmath78 detections for the medium and intermediate scale waves have been re - plotted in the panels of the smaller size class(es ) for comparison . for reference , we have also shaded in grey time ranges where mstids where detected using the polynomial - based technique as described in 3 . in general , the @xmath18 directions for the medium - scale class agree with those seen for the detected mstids displayed in fig . [ waveprop ] , which serves to partially validate our arm - based method . most of the time , the azimuth angles for the intermediate and small classes roughly agree , indicating that they are likely part of the same distribution of waves . both often agree with the azimuth angles determined for the medium class , but are at times significantly different , most notably during two instances where mstids were detected . we have also determined approximate speeds for all of the detected groups of waves which we have plotted in fig . [ armspeckv ] with their values of @xmath79 taken from the above described cosine fits . the speeds were computed using the mean temporal frequency within bins of @xmath64 that were within @xmath80 km@xmath2 for the small class and @xmath81 km@xmath2 for the intermediate and medium classes of the cosine fit - determined value of @xmath79 from each arm . a spline fit to the data as a function of arm azimuth angle was then used to estimate the mean temporal frequency along the direction of @xmath18 where waves , rather than background fluctuations , are more likely to dominate . the speed was then computed using this frequency and the value of @xmath79 from the cosine fit . the speeds agree in general with what can be seen in fig . [ armspecpa ] , that the medium - scale class of waves are a somewhat distinct population in terms of direction and speed whereas the intermediate and small scale classes tend to have similar directions and speeds . to further explore the nature of the small - scale waves in particular , we have plotted the distribution of azimuth angles for the @xmath78 detections in fig . [ padist ] for the small - scale class only . from this it can be seen that the distribution has four distinct peaks at angles of roughly @xmath82 , @xmath83 , @xmath84 , and @xmath85 . for reference , we have also plotted the azimuth angles of the three vla arms ( both up and down the arms ) as vertical dashed lines . from this , one can see that the group of waves with a peak azimuth angle of @xmath84 were moving almost directly up the southwestern arm . this indicates that this group of waves is likely small in extent as well as in wavelength . this is because for groups of waves substantially smaller than the array , the arm - based method we described above will be biassed toward waves traveling along one of the three arms because the number of antennas that `` see '' these waves will be maximized in this case . in contrast , groups of waves which span all or nearly all of the array will essentially be seen by all the antennas for any direction and will experience no such bias . this appears to be the case for the group of waves moving roughly southeastward and for the waves with a peak azimuth angle of @xmath82 . the peak of the distribution for the southeastward - directed waves is somewhat skewed toward the azimuth of the southeastern arm . however , the bulk of these waves have azimuth angles between that of the southeastern ( up ) and northern ( down ) arms , indicating that any influence of the arm - based bias on these waves is minimal . the group of waves moving toward an angle of @xmath83 peak near but not at the azimuth angle of the southwestern arm , indicating that their measured distribution has been influenced by the arm - based bias , but to a lesser degree than those whose distribution peaks near @xmath84 . because of this bias , the paucity of waves directed either due west or east only indicates that there were few groups of small - wavelength waves moving in these directions which also spanned the array . however , the lack of small - scale waves seen directed northwest , due north , or due south ( i.e. , up the southeastern arm or up / down the northern arm ) indicates a genuine absence of such phenomena and demonstrates that the population of small - scale waves was far from isotropic . in terms of the implied orientation of the wavefronts , the peaks at @xmath83 and @xmath84 are essentially the same since they are roughly @xmath86 apart . they are both conspicuously close to the required orientation for the @xmath87 layer instability ( or , `` @xmath5li '' ) described in detail by @xcite . they demonstrated that an @xmath5 layer is unstable against perturbations with wavefronts aligned from northwest to southeast . in particular , for no meridional wind component , the optimum orientation is 35@xmath88 west of magnetic north . for the vla , where the magnetic declination is about 10@xmath88 , this corresponds to @xmath18 azimuth angles of either @xmath89 or @xmath90 . if there is a significant meridional wind component , the optimum angle can be as much as @xmath91 closer to due south , giving a range of optimum position angles of @xmath92 to @xmath89 or @xmath93 to @xmath90 . this is roughly consistent with the locations of the two largest peaks in fig . [ padist ] . however , the exact peak azimuth angles of these two distributions are not precisely constrained given the bias toward the direction of the vla arms described above . we can only say that they are generally moving either northeast or southwest . to illustrate the agreement with the predictions of the @xmath5li model , we have computed the azimuth angle ( at the vla ) dependence of the @xmath5li assuming a magnetic dip angle of @xmath94 according to @xcite , including a range of peak azimuth angles of @xmath91 . we have plotted a scaled version of the growth rate in fig . [ padist ] . from this curve , one can see that the bulk of the two groups of waves with azimuth angles near the optimum values for the @xmath5li are contained within the regions that are within @xmath95 of the maximum growth rate . if we consider only those waves located in time ranges where mstids were detected , the agreement with the predictions of the @xmath5li growth rate are even better ( see the red histograms in fig . [ padist ] ) . in fact , only a small fraction of such waves ( @xmath96% ) have azimuth angles where the @xmath5li growth rate is @xmath97 . in contrast , for all waves , about 16% lie outside this region . one can see from the red histograms in fig . [ padist ] that the azimuth angle distribution for the southwest - directed waves coincident with mstids is significantly skewed with a tail extending away from the azimuth angle of the southwestern vla arm . this indicates that the arm - based bias discussed above has a larger effect on these waves than the other southwest - directed waves which are not seen with mstids and have a more symmetric distribution . this further implies that these groups of waves and their northeast - directed counterparts are relatively small in extent as compared to the other detected groups of small - scale waves . we note that small - scale waves were detected during nearly all observed instances of mstids . this indicates that the preference for small - scale waves aligned northwest to southeast during mstid activity is not simply a product of the arm - based bias . in other words , there is no indication that there were periods of mstid activity where small - scale waves were not detected because they were not directed along a vla arm and their extent was too small to be detected with the arm - based method . this preference for northwest to southeast aligned , small - scale waves is again consistent with the notion that these waves are associated with the @xmath5li . this is because according to @xcite , the coupling between the perkins instability in the @xmath6 region and the @xmath5li is itself unstable . the growth rate for the coupled instability is maximized for instances where both the @xmath6 and @xmath5 layer perturbations are aligned from northwest to southeast . from fig . [ armspecpa ] , we can see that while the small - scale waves are so aligned , the mstids are not necessarily . however , we note that the range of allowed orientations for the perkins instability is much broader than that for the @xmath5li as it depends somewhat on the strength and orientation of the @xmath6 region neutral wind . we also note that models of this instability @xcite have shown that random perturbations within an @xmath5 layer are enough to form mstids in the lower / bottom part of the @xmath6 region through the @xmath4-@xmath6 coupling mechanism . the mstids can then , through this same coupling , help northwest to southeast aligned perturbations grow within the @xmath5 layer . @xcite confirmed that this process works most efficiently when conditions are right in the lower / bottom part of the @xmath6 region to form northwest to southeast aligned mstids . however , they also showed that while weaker , mstids of different orientations ( in their example , aligned north to south ) could be formed through this process . in this case , the @xmath5 layer waves that formed were still aligned northwest to southeast due to the more rigid directional constraints of the @xmath5li . therefore , the results show here are quite consistent with the predictions of @xmath4@xmath6 coupling in the nighttime , midlatitude ionosphere . the group of small - scale waves seen in fig . [ padist ] to have @xmath18 pointed to the southeast can not be explained with the @xmath5li . however , several authors have noted that qp echoes are often found with no preferred direction , or with directions inconsistent with the direction - dependent @xmath5li ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite found a tendency for the fronts of qp echoes to be nearly perpendicular to the direction of the neutral wind at the height of the @xmath5 layer(s ) . to explore this possibility , we have plotted the neutral wind velocity azimuth angle profile taken from publicly available gpi data - driven runs of the tiegcm code ( * ? ? ? * currently available at http://www.hao.ucar.edu/modeling/tgcm/ ) for the night of 13 august at @xmath98n and @xmath99w at one hour intervals in fig . [ windpa ] . in each panel , the @xmath18 azimuth angles of each @xmath100 detection of small - scale waves within the corresponding time bin is plotted as a vertical dotted line . to form and maintain an @xmath5 layer , a zonal wind shear is required with the westward moving wind at higher altitudes than the wind moving eastward . in the plots in fig . [ windpa ] , this occurs at altitudes where the neutral wind transitions from moving southeastward at lower altitudes to moving southwestward at higher altitudes . this usually occurs at heights between about 100 and 110 km which is typical for @xmath5 layers . the small - scale waves detected with @xmath18 toward the southeast are then consistent with having wavefronts nearly perpendicular to the neutral wind at the @xmath5 layer . we also note that the waves with @xmath18 pointing to the southwest which are not coincident with mstids may also be consistent with this same scenario , depending on the altitude at which the proposed @xmath5 layer has formed . @xcite concluded that these types of qp echoes may be formed either by interactions with gravity waves or via the kelvin - helmholtz instability . since these waves were detected during the period of lowest mstid activity ( see fig . [ armspecpa ] and [ padist ] ) , @xmath4@xmath6 coupling likely has little to do with the generation of these structures . gravity waves or kelvin - helmholtz instabilities are much more likely candidates for their generation mechanism . while our arm - based techniques have found evidence for groups of waves moving in preferred directions on all size - scales , these results do not exclude the existence of populations of structures moving in a wide enough range of directions as to appear isotropic within our analysis . to examine this possibility , we have looked to the background spectra constructed using the arm spectra described in 4.1 . these spectra should be reasonably free of the presence of directed waves and will provide insights into the quasi - isotropic set of background fluctuations . mean background spectra computed within one hour bins are plotted in fig . [ armspecbin ] . while some of these spectra may be reasonably approximated with single power - laws , many have the appearance of `` broken , '' or two - component power - laws . upon detailed inspection , we found that both the inner and outer most regions of the spectra were roughly @xmath101 , but with different slopes and offsets . this is intriguing since this is the dependence one expects for turbulent fluctuations . in particular , for a thin shell , @xcite predicted from @xcite that the spectrum of refractive index fluctuations gives a spectrum of wavefront phase variations @xmath102 . since we have examined the gradient of tec ( or , wavefront phase ) fluctuations , the fluctuations have been effectively multiplied by a factor of @xmath103 , and so the power spectrum is altered by a factor of @xmath104 , or @xmath101 . thus , the general shapes of the background spectra suggest they may be interpreted as the sums of two separate turbulent components . we have consequently fit a two - component turbulent model to each of the background spectra plotted in fig . [ armspecbin ] . to allow each spectrum to flatten above a maximum turbulent scale - length , @xmath105 , we assumed the following form for each component @xmath106^{- \frac{5}{6}}\ ] ] the resulting two - component fits are plotted with the spectra in fig . [ armspecbin ] ; the two components are plotted separately as well . in general , one can see that each spectrum is well represented by a component with a relatively large @xmath105 ( blue curves ) and one that flattens at much smaller scales ( higher wavenumbers ; green curves ) . to illustrate this quantitatively , we have plotted in fig . [ armspecfit ] the power at @xmath105 ( upper panel ) and @xmath105 ( lower panel ) for each of the two components as functions of local time . for both components , @xmath105 varies somewhat with time , but is typically @xmath107 km for the larger - scale turbulent component and roughly 10 km for the smaller - scale component . while the power at @xmath105 seems relatively stable with time for the larger component , the small - scale component shows a definite peak at a local time of about @xmath108 . this corresponds to the middle of the roughly three - hour lull in mstid activity that started around @xmath62 . since we also detected small - scale waves consistent with qp - echoes generated by gravity waves during this time , it is possible that this increase in power is the result of the same gravity waves influencing turbulent processes in the lower ionosphere / thermosphere where the role of ion - neutral coupling is relatively strong . our exploration of a long , vhf observation of cyg a with the vla has successfully demonstrated the power of this instrument to characterize a variety of transient ionospheric phenomena . for this observation , the typical @xmath109 uncertainty in the @xmath7 measurements was @xmath9 tecu , yielding more than an order of magnitude better sensitivity to tec fluctuations than can be achieved with gps - based relative tec measurements ( see , e.g. , * ? ? ? * ) . through our new spectral - based analysis , we have demonstrated the ability of the vla to detect and characterize individual instances of mstids as well as smaller - scale structures likely associated with @xmath5 layers . we note that since disturbances within the plasmasphere have been observed and in fact discovered with the vla , it stands to reason that some of the phenomena we have detected may be located within the plasmasphere as well . however , the phenomena discovered by @xcite have azimuth angles clustered around @xmath110 ( i.e. , near magnetic east for the vla ) , and few if any of the wave - like structures we have detected meet this criterion . therefore , the vast majority ( if not all ) of the phenomena we have observed were likely within the ionosphere . among these phenomena are mstids , small - scale wave - like phenomena consistent with @xmath5 layer disturbances , and turbulent fluctuations . both the mstids and the small - scale fluctuations were present intermittently throughout the night , and turbulent fluctuations were seen at all times , which is common within vhf observations with the vla @xcite . the mstids appear to change direction after local midnight from being directed generally westward to being directed toward the northeast ( see below ) . there is also a noticeable change in the turbulent activity near midnight , namely the spectral power of these fluctuations on small ( @xmath41 km ) scales peaks near midnight and gradually decreases toward dawn . before midnight local time , we observed instances of westward / northwestward moving mstids ( see fig . [ waveprop ] ) with an additional group moving westward observed briefly near @xmath111 ( see fig . [ armspecpa ] ) . while atypical for what is observed for most of north america ( see * ? ? ? * ) , the directions of these waves are similar to what is typically observed near the west coast in california during summer nighttime ( see * ? ? ? * ) , especially those directed closer to due west . small - scale waves were detected coincident with nearly all of these mstids moving toward either the northeast or southwest . given the difference in directions , it is unlikely that these are simply the signatures of wavefront distortions within the mstids themselves . as discussed above , the amplitudes and orientations of these small - scale waves are consistent with them being generated within @xmath5 layers via the @xmath5li . coupling between the @xmath5li and the perkins instability in the f region ( see * ? ? ? * ; * ? ? ? * ) may be what has influenced the mstids observed near dawn to move closer to northward as they were detected concurrently with northeastward - directed small - scale waves . in contrast , those waves seen near @xmath111 were moving closer to due west and were accompanied by small - scale waves moving toward the southwest . unusual , northeastward - directed mstids were observed near @xmath50 to @xmath60 and @xmath57 to @xmath58 . as discussed above , @xcite observed similar phenomena over japan at midlatitudes during the night when the height of the f region was seen to drop significantly . similarly , we have found evidence within data from relatively nearby ionosondes that suggests that the f region may have experienced a similar drop in height during the instances of northeastward - directed mstids observed with the vla . small - scale waves were also observed at the same times as these unusual mstids . they were also mostly directed toward the northeast , suggesting that they may represent small - scale structures within the mstids themselves . however , we note that ( 1 ) fig . [ armspeckv ] shows that they were moving significantly more slowly than the mstids , ( 2 ) the distribution of azimuth angles seen in fig . [ padist ] to peak near the azimuth angle of the vla s southwestern arm indicates a bias that only affects groups of waves that span an area smaller than the array , and ( 3 ) toward the end of the night , the small - scale waves change direction toward the southwest while the mstids remained northeastward directed . these factors seem to indicate that the small - scale waves are separate phenomena , similar to the other observed small - scale waves that are consistent with qp echoes generated within @xmath5 layers . it could be that coupling between these disturbances and the larger waves in the f region plays some role in the unusual direction of the observed mstids . in any case , a more thorough , statistical examination of these phenomena seems warranted . there are two distinct groups of small - scale waves which were detected with no coincident mstids . they were predominantly directed toward the southeast ( from @xmath62 to @xmath50 ) with a shorter period of activity of waves moving toward the southwest ( between @xmath112 and @xmath113 ) . the southeastward - directed waves in particular seem to be somewhat different from the other observed small - scale waves . given their orientations , they can not have been generated via the @xmath5li . they appear to be moving somewhat faster with typical speeds of about 150 m s@xmath2 while most of the other small - scale waves have speeds between 30 and 100 m s@xmath2 ( see fig . [ armspeckv ] ) . they also seem to be largely unaffected by the arm - based bias discussed in 4.3 , indicating that as a group , the likely span an area larger than that of the array . despite these differences , the southeastward - directed small - scale waves are generally consistent with qp echoes generated in @xmath5 layers , both in amplitude and wavelength . they are more specifically consistent with the qp echoes observed by the sporadic-@xmath4 experiments over kyushu ( seek * ? ? ? * ; * ? ? ? * ) , which were heavily influenced by the direction of the neutral wind at the height of the @xmath5 layer(s ) . the shift in direction at around @xmath61 from southwestward to southeastward may then be an indication of a change in @xmath5 layer height from above the wind sheer altitude to below it . this may imply the formation / dissipation of @xmath5 layers near this time . in addition , these waves occurred during the peak in small - scale turbulent activity as evidenced by the results shown in fig . [ armspecfit ] . this suggests that the same mechanism that generated these qp echoes also caused a significant turbulent disturbance in the lower ionosphere / thermosphere . since the rocky mountains lie largely to the north and northwest of the vla , this mechanism is likely gravity waves associated with wind flow over the rocky mountains . the authors would like to thank the referees for useful comments and suggestions . basic research in astronomy at the naval research laboratory is supported by 6.1 base funding . the vla was operated by the national radio astronomy observatory which is a facility of the national science foundation operated under cooperative agreement by associated universities , inc . 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 . coker , c. , s. e. thonnard , k. f. dymond , t. j. w. lazio , j. j. makela , and p. j. loughmiller ( 2009 ) , simultaneous radio interferometer and optical observations of ionospheric structure at the very large array _ rad . , 44 _ , rs0a11 cosgrove , r. b. , r. t. tsunoda , s. fukao , and m. yamamoto ( 2004 ) , coupling of the perkins instability and the sporadic _ e _ layer instability derived from physical arguments , _ j. of geophys . res . , 109 _ , a06301 dymond , k. f. , c. watts , c. coker , s. a. budzien , p. a. bernhardt , n. kassim , t. j. lazio , k. weiler , p. c. crane , p. s. ray , a. cohen , t. clarke , l. j. rickard , g. b. taylor , f. schinzel , y. pihlstrom , m. kuniyoshi , s. close , p. colestock , s. myers , and a. datta ( 2011 ) , a medium - scale traveling ionospheric disturbance observed from the ground and from space , _ rad . , 46 _ , rs5010 larsen , m. f. , d. l. hysell , q. h. zhou , s. m. smith , j. friedman , and r. l. bishop ( 2007 ) , imaging coherent scatter radar , incoherent scatter radar , and optical observations of quasiperiodic structures associated with sporadic @xmath4 layers , _ j. of geophys . , 112 _ , a06321 shiokawa , k. , y. otsuka , n. nishitani , t. ogawa , t. tsugawa , t. maruyama , s. e. smirnov , v. v. bychkov , and b. m. shevtsov ( 2008 ) , northeastward motion of nighttime medium - scale traveling ionospheric disturbances at middle latitudes observed by an airglow imager _ j. of geophys . res . , 113 _ , a12312 tolman , b. , r. b. harris , t. gaussiran , d. munton , j. little , r. mach , s. nelsen , and b. renfro ( 2004 ) , the gps toolkit : open source gps software , paper presented at 16th international technical meeting of the satellite division of the institute of navigation , long beach , calif . tsugawa , t. y. , y. otsuka , a. j. coster , and a. saito ( 2007 ) , medium - scale traveling ionospheric disturbances detected with dense and wide tec maps over north america , _ geophys . lett . , 34 _ , l22101 yokoyama , t. , d. l. hysell , y. otsuka , and m. yamamoto ( 2009 ) , three - dimensional simulation of the coupled perkins and @xmath5-layer instabilities in the nighttime midlatitude ionosphere _ j. of geophys . res . , 114 _ , a03308 yokoyama , t. and d. l. hysell ( 2010 ) , a new midlatitude ionosphere electrodynamics coupling model ( mieco ) : latitudinal dependence and propagation of medium - scale traveling ionospheric disturbances _ geophys . lett . , 37 _ , l08105 wang , w. , t. l. killeen , a. g. burns , and r. g. roble ( 1999 ) , a high - resolution , three - dimensional , time dependent , nested grid model of the coupled thermosphere - ionosphere , _ phys , 61 _ , 385397 , as a function of local time ( abscissa ) and temporal frequency ( ordinate ) of fourier modes within the @xmath7 data . middle : the azimuth angle of the wavenumber vector of each fourier mode with power significantly larger than background levels ( see 3.2 ) . lower : the velocity of each fourier mode , using the same masking as in the middle panel . in all panels , the white contours are of spectral power.,width=576 ] , for each of the vla arms . the noise - equivalent spectrum ( see 4.1 ) is also plotted in each panel as a grey curve . the vertical dotted grey lines denote values of @xmath115 which corresponds to the approximate nyquist sampling limit for the average antenna separation of 2.5 km along the vla arms ( in a configuration).,width=576 ] and @xmath78 detections ( see 4.2 ) for each of the three wavenumber classes . time ranges where mstids were detected from the data displayed in the upper panel of fig . [ waveprop ] are shaded in grey.,width=576 ] azimuth angles for @xmath100 detections of small - scale moving structures displayed in fig . [ armspecpa ] ( shaded grey histograms ) . the same distribution for those detections that were coincident with mstid detections ( see fig . [ armspecpa ] ) are displayed in red . the growth rate for the @xmath116 layer instability is plotted as a black curve ( see right ordinate for scale ) , allowing for a range in meridional wind components ( see 4.3 and * ? ? ? * ) .,width=576 ] azimuth angles for the @xmath100 detections of small - scale moving structures . the light - grey shaded regions indicate the typical range in altitudes for @xmath5 layers , 100110 km . each dark - gray horizontal line indicates the altitude where the wind direction changes from eastward at lower altitudes to westward at higher altitudes , i.e. , where the wind shear necessary for creating an @xmath5 layer exists.,width=576 ] , the power at the maximum wavelength ( upper ) and the maximum wavelength ( lower ) for each component as functions of local time [ see 4.4 and equation ( 4 ) ] . here , turbulent component 1 ( green curves in fig . [ armspecbin ] ) refers to the smaller - scale turbulent component and turbulent component 2 ( blue curves in fig . [ armspecbin ] ) refers to the larger - scale component.,width=576 ]

this paper illustrates how the tools of equilibrium statistical mechanics can help to explain a far - from - equilibrium problem : the jamming transition in frictionless granular materials . edwards ideas consist of proposing a statistical ensemble of volume and stress fluctuations through the thermodynamic notion of entropy , compactivity , , and angoricity , ( two temperature - like variables ) . we find that edwards thermodynamics is able to describe the jamming transition ( j - point ) . using the ensemble formalism we elucidate the following : _ ( i ) _ we test the combined volume - stress ensemble by comparing the statistical properties of jammed configurations obtained by dynamics with those averaged over the ensemble of minima in the potential energy landscape as a test of ergodicity . agreement between both methods supports the idea of `` thermalization '' at a given angoricity and compactivity . _ ( ii ) _ a microcanonical ensemble analysis supports the idea of maximum entropy principle for grains . _ ( iii ) _ the intensive variables describe the approach to jamming through a series of scaling relations as and . due to the force - volume coupling , the jamming transition can be probed thermodynamically by a `` jamming temperature '' comprised of contributions from and . _ ( iv ) _ the thermodynamic framework reveals the order of the jamming phase transition by showing the absence of critical fluctuations at jamming in observables like pressure and volume . _ ( v ) _ finally , we elaborate on a comparison with relevant studies showing a breakdown of equiprobability of microstates . the application of concepts from equilibrium statistical mechanics to out of equilibrium systems has a long history of describing diverse systems ranging from glasses to granular materials . for dissipative jammed systems particulate grains or droplets the key concept proposed by edwards is to replace the energy ensemble describing conservative systems by the volume ensemble . however , this approach alone is not able to describe the jamming point ( j - point ) for deformable particles like emulsions and droplets , whose geometric configurations are influenced by the applied external stress . therefore , the volume ensemble requires augmentation by the ensemble of stresses . just as volume fluctuations can be described by compactivity , the stress fluctuations give rise to an angoricity , another analogue of temperature in equilibrium systems . in the past 20 years since the publication of edwards work there has been many attempts to understand and test the foundations of the thermodynamics of powders and grains . three approaches are relevant to the present study : * * experimental studies of reversibility. * starting with the experiments of chicago which were reproduced by other groups , a well - defined experimental protocol has been introduced to achieve reversible states in granular matter . these experiments indicate that systematically shaken granular materials show reversible behavior and therefore are amenable to a statistical mechanics approach , despite the frictional and dissipative character of the material . these results are complemented by direct measurements of compactivity and effective temperatures in granular media . * * numerical test of ergodicity. * numerical simulations compare the ensemble average of observables with those obtained from direct dynamical measures in granular matter and glasses . these studies find general agreement between both measures and , together with the experimental studies of reversibility , suggest that ergodicity might work in granular media . * * numerical and experimental studies of equiprobability of jammed states. * exhaustive searches of all jammed states are conducted in small systems to test the equiprobability of jammed states , as a foundation of the microcanonical ensemble of grains . numerical simulations and experiments indicate that jammed states are not equiprobable . these results suggest that a hidden extra variable might be needed to describe jammed granular matter in contrast with the work in * 1 * and * 2*. the current situation can be summarized as following : when directly tested or exploited in practical applications , edwards ensemble seems to work well . these include studies where ensemble and dynamical measurements are directly compared , and recent applications of the formalism to predict random close packing of monodisperse spherical particles , polydisperse systems , and two and high dimensional systems . however , a direct count of microstates reveals problems at the foundation of the framework manifested in the breakdown of the flat average assumption in the microcanonical ensemble . in this paper we investigate the edwards ensemble of granular matter focusing on describing the jamming transition . a short version of this study has been recently published in . we employ a strategy that mixes the approaches * 2 * and * 3 * above . we first perform an exhaustive search of all jammed configurations in the potential energy landscape ( pel ) of small frictionless systems in the spirit of . we then use this information to perform a direct test of ergodicity in the spirit of . our results indicate : _ ( i ) _ the dynamical and ensemble measurements of presure , coordination number , volume , and distribution of forces agree well , supporting ergodicity . a microcanonical ensemble analysis supports also a maximum entropy principle for grains . _ ( ii ) _ intensive variables like angoricity , , and compactivity , , describe the approach to jamming through a series of scaling relations . due to the force - volume coupling , the jamming transition can be probed thermodynamically by a `` jamming temperature '' comprised of contributions from and . _ ( iii ) _ these intensive variables elucidate the thermodynamic order of the jamming phase transition by showing the absence of critical fluctuations above jamming in static observables like pressure and volume . that is , the jamming transition is not critical and there is no critical correlation length arising from a thermodynamic n - point correlation function . we discuss other possible correlation lengths . _ ( iv ) _ surprisingly , we reproduce the results of regarding the failure of equiprobability of microstates while obtaining the correct dynamics measurements as in . we then offer a possible solution to this conundrum to elucidate why the microstates seems to be not equiprobable while the ensemble averages produce the correct results . the paper is organized as follows . section [ edw ] discusses the edwards thermodynamics of the jamming transition . section [ pel - c ] describes the ensemble calculations in the potential energy landscape formalism . section [ hertz ] describes the hertzian system to be studied . section [ dos ] describes the ensemble measurements to be compared with the md measures of section [ md ] . section [ angoricity ] explains how to calculate from the data . the ergodicity test is made in section [ ergodicity ] . section [ thermo ] describes the calculation in the microcanonical ensemble where the principle of maximum entropy is verified and the coupled jamming temperature is obtained . section [ comparison ] compares our results with those of o hern _ et al . _ and section [ conclusion ] summarizes the work . appendix [ z ] includes `` de yapa '' a study of coordination number fluctuations in the edwards theory for random close packings of hard spheres .

1101.5634

the process typically referred to as the jamming transition occurs at a critical volume fraction @xmath5 where the granular system compresses into a mechanically stable configuration in response to the application of an external strain @xcite . the application of a subsequent external pressure with the concomitant particle rearrangements and compression results in a set of configurations characterized by the system volume @xmath6 ( @xmath7 is the volume fraction of @xmath8 particles of volume @xmath9 ) and applied external stress or pressure @xmath10 ( for simplicity we assume isotropic states ) . it has been long argued whether the jamming transition is a first - order transition at the discontinuity in the average coordination number , @xmath11 , or a second - order transition with the power - law scaling of the system s pressure as the system approaches jamming with @xmath12 @xcite . previous work @xcite has proposed to explain the jamming transition by a field theory in the pressure ensemble . here , we use the idea of `` thermalization '' of an ensemble of mechanically stable granular materials at a given volume and pressure to study the jamming transition from a thermodynamic viewpoint . for a fixed number of grains , there exist many jammed states @xcite confined by the external pressure @xmath10 in a volume @xmath13 . in an effort to describe the nature of this nonequilibrium system from a statistical mechanics perspective , a statistical ensemble @xcite was introduced for jammed matter . in the canonical ensemble of pressure and volume , the probability of a state is given by @xmath14 $ ] , where @xmath15 is the entropy of the system , @xmath16 is the volume function measuring the volume of the system as a function of the particle coordinates and @xmath17 is the boundary stress ( or internal virial ) @xcite of the system . just as @xmath18 is the temperature in equilibrium system , the temperature - like variables in jammed systems are the compactivity @xcite @xmath19 and the angoricity @xcite , @xmath20 in a recent series of papers @xcite the compactivity was used to describe frictional and frictionless hard spheres in the volume ensemble . here , we test the validity of the statistical approach in the combined pressure - volume ensemble to describe deformable , frictionless particles , such as emulsion systems jammed under osmotic pressure near the jamming transition @xcite . in general , if the density of states @xmath21 in the space of jammed configurations ( defined as the probability of finding a jammed state at a given @xmath22 at @xmath23 ) is known , then calculations of macroscopic observables , like pressure @xmath10 and average coordination number @xmath11 as a function of @xmath7 , can be performed by the canonical ensemble average @xcite at a given volume : @xmath24 and @xmath25 where the canonical partition function is @xmath26 and the density of states is normalized as @xmath27 . the inverse angoricity is defined as @xmath28 at the jamming transition the system reaches isostatic equilibrium , such that the stresses are exactly balanced in the resulting configuration , and there exists a unique solution to the interparticle force equations satisfying mechanical equilibrium . it is well known that observables present power - law scaling @xcite : @xmath29 @xmath30 where @xmath31 and @xmath32 for hertzian spheres and @xmath33 is the coordination number at the frictionless isostatic point ( j - point ) @xcite . the average @xmath34 indicates that these quantities are obtained by averaging over packings generated dynamically in either simulations or experiments as opposed to the ensemble average over configurations @xmath35 of eqs . ( [ eq_edp])([eq_edz ] ) . comparing the ensemble calculations , eqs . ( [ eq_edp])([eq_edz ] ) , with the direct dynamical measurements , eqs . ( [ eq_mdp])([eq_mdz ] ) , provides a basic test of the ergodic hypothesis for the statistical ensemble . our approach is the following : we first perform an exhaustive enumeration of configurations to calculate @xmath21 and obtain @xmath36 as a function of @xmath37 for a given @xmath7 using eq . ( [ eq_edp ] ) . then , we obtain the angoricity by comparing the pressure in the ensemble average with the one obtained following the dynamical evolution with molecular dynamics ( md ) simulations . by setting @xmath38 , we obtain the angoricity as a function of @xmath7 . by virtue of obtaining @xmath39 , all the other observables can be calculated in the ensemble formulation . the ultimate test of ergodicity is realized by comparing the remaining ensemble observables with the corresponding direct dynamical measures . an appealing approach for understanding out - of - equilibrium systems is to study the properties of the system s `` potential energy landscape '' ( pel ) @xcite , described by the @xmath40-coordinates of all particles in the multi - dimensional configuration space , or landscape , of the potential energy of the system ( @xmath8 is the number of particles ) . characterizing such potential energy landscapes has become an important approach to study the behaviour of out - of - equilibrium systems . for example , this approach has provided important new insights into the origin of the unusual properties of supercooled liquids , such as the distinction between `` strong '' and `` fragile '' liquids @xcite . in frictionless granular matter , the potential energy is well - defined and each jammed configuration corresponds to one local minimum in the pel . for small systems ( @xmath41 ) , it is possible to find all the minima with current computational power @xcite . for somewhat larger systems @xmath42 , it is possible to obtain a representative ensemble , without exhaustively sampling all the states . based on these stationary points , we test the combined volume - stress ensemble . the following work is only valid for frictionless systems where the potential energy of interaction is well defined . frictional grains are path dependent due to coulomb friction between particles and therefore not amenable to a pel study since there is no well defined energy of interaction . the formalism introduced by goldstein @xcite consists of partitioning the potential energy surface into a set of basins as illustrated in fig . [ pel : pes ] . the dynamics on the potential energy surface can be separated into two types : the vibrational motion inside each basin and the transitional motion between the local minima . stillinger and coworkers @xcite developed the method of inherent structure to characterize the pel . in this method , a local minimum in the pel is located by following the steepest - descent pathway from any point surrounding the minimum . the inherent structure formalism simplifies the energy landscape into local minima and ignores the vibrational motion around them . the dynamics between the inherent structures is introduced with the transition states identified with the saddle points in the pel . the transition states are stationary points like the local minima but they have at least one maximum eigendirection . for the simplest system of @xmath8 structureless frictionless particles possessing no internal orientational and vibrational degrees of freedom , the potential energy function of this n - body system is @xmath43 , where the vectors @xmath44 comprise position coordinates . as mentioned above , the most interesting points of a potential energy surface are the stationary points , where the gradient vanishes . here we explain how to locate these stationary points . the algorithm follows well established methods in computational chemistry @xcite . the procedure is analogous to finding the inherent structures @xcite of glassy systems . the algorithm employed , lbfgs algorithm , is also similar to the conjugate gradient method employed by o hern @xcite , differing in the fact that it does not require the calculation of the hessian matrix at every time step . we make the source code in c@xmath45 available at http://www.jamlab.org and free to use together with all the packings generated in this study . the algorithm has been used in the short version of this article @xcite and in a study of the pel in lennard - jones glasses to reconstruct a network of stationary states and apply a percolation picture of the glass transition @xcite . consider the taylor expansion of the potential energy , @xmath46 , around a general point in configuration space , @xmath47 , @xmath48 where @xmath49 is the gradient , @xmath50 , @xmath51 is the hessian matrix , @xmath52 , and @xmath53 is a small step vector that gives the displacement away from @xmath47 . by eq . ( [ taylor ] ) , the calculation of energy difference for a given step @xmath53 from the initial point @xmath47 is complicated . by selecting the eigenvectors of the hessian matrix @xmath54 as our local coordinates , we can simplify the taylor expansion of eq . ( [ taylor ] ) as : @xmath55 where @xmath56 , @xmath57 , @xmath58 , and @xmath59 is the eigenvalue of the hessian matrix for component @xmath37 . from eq . ( [ changes ] ) , it is easy to see that the total change of energy could simply be the sum of the changes in each directions . this may help us to raise the energy in some directions and reduce the energy at others , and finally reach a stationary point . the length of each step components can be selected as the maximum change of energy : @xmath60 as shown in fig.[pel : upanddown ] . the sign @xmath61 in this formula depends on the choice of uphill or downhill direction . in fact , for @xmath62 , it is possible to choose another step for the uphill case , since @xmath63 increases as @xmath64 , but for large steps , the taylor expansion eq . ( [ taylor ] ) may breakdown . therefore , it is important to control the step length . for @xmath65 , we reach the opposite conclusion . the stationary points can be separated into local minima and saddle points . based on the eigenvalues of the hessian matrix for the stationary point , the local minima are ordered as : @xmath66 and for a saddle point of order @xmath37 : @xmath67 generally , this algorithm searches for the nearest stationary point on the surface by following the opposite @xmath68 and along @xmath69 the various gradient directions . it is much easier to locate local minima than saddle points because , for the first , we only need to search downhill in every direction . at present one of the most efficient methods to search the local minima for large system is nocedal s limited memory broyden - fletcher - goldfarb - shanno algorithm ( lbfgs ) @xcite . the lbfgs algorithm constructs an approximate inverse hessian matrix from the gradients ( first derivatives ) which are calculated from previous points . since it is only necessary to calculate the gradients at each searching step , the lbfgs algorithm increases the computational speed of the algorithm enormously . in the newton - raphson method discussed above , the hessian matrix of second derivatives is needed to be evaluated directly . instead , the hessian matrix used in lbfgs method is approximated using updates specified by gradient evaluations . the lbfgs algorithm code can be obtained from http://www.netlib.org/opt/index.html . here we present a brief explanation of the algorithm from an initial random point @xmath70 and an approximate hessian matrix @xmath71 ( in practice , @xmath71 can be initialized with @xmath72 ) , the following steps are repeated until @xmath47 converges to the local minimum . * obtain a direction @xmath73 by solving : @xmath74 . * perform a line search to find an acceptable step size @xmath75 in the direction found in the first step , then update @xmath76 . * set @xmath77 . * set @xmath78 . * set the new hessian , @xmath79 . in the present study we do not make use of the saddle points . however , other studies using network theory to represent the pel necessitate the links between minima through the saddle points @xcite . for completeness , below we explain how to search for saddles . a particular powerful method for locating saddle points is the eigenvector following method @xcite . the eigenvector - following method , developed by cerjan , miller and others @xcite , consists of locating a saddle point from a local minimum . at each searching step towards a saddle point with @xmath37 order , the directions are separated into two types : @xmath37 uphill directions to maximization and @xmath80 downhill directions to minimization . we follow the implementation of the eigenvector - following method by grigera @xcite . we give a general description : at each searching step , a step size @xmath53 is calculated by the diagonalized hessian matrix @xcite : @xmath81 where @xmath59 are the eigenvalues of the hessian matrix and @xmath82 are the components of the gradient in the diagonal base ( @xmath83 is set to 0 for the directions where @xmath84 ) . the sign @xmath61 is chosen by the order of the saddle point . for a saddle point of order @xmath85 , the algorithm will set @xmath86 for @xmath87 and @xmath88 for @xmath89 . when @xmath90 , the step size @xmath83 converges to the newton - raphson step as eq . ( [ nrstep ] ) : @xmath91 we generate a two dimensional soft - ball system in circular boundary , which contains 31 particles of equal radius , to illustrate the method of finding stationary and saddle points in the pel . the interaction between particles ( also the interaction between particles and wall ) follows the hertzian law @xcite : @xmath92 here , @xmath93 is the interaction strength between particles @xmath94 and @xmath95 . the volume fraction is @xmath96 , which is closed to the jamming transition of @xmath97 hard disks . we first generate a random configuration , which is the initial point of the search of the minima . with the lbfgs method , we search the local minimum @xmath1 nearby this initial point . after the minimum @xmath1 is obtained , we apply the eigenvector following method to walk from the point @xmath1 on the potential energy surface to locate the transition state @xmath98 ( here the transition state is a first order saddle ) . finally , the minimum @xmath99 is located by applying lbfgs method again . figure [ pel : exp1 ] shows configurations of two local minima ( marked as red ) and the transition state ( marked as blue ) between them . the pathway from minimum a to minimum b , passing by transition state c , is shown in fig . [ pel : exp2 ] . the pathway distance is the euclidean distance , @xmath100 where @xmath101 , @xmath102 , @xmath103 is the coordinate of configuration passing along the searching method and @xmath47 is the coordinate of saddle @xmath98 . the dynamics from minimum to minimum can be represented as a walk on a network whose nodes correspond to the minima and where edges link those minima which are directly connected by a transition state . the work of doye @xcite provides an illustration of such a landscape network for a lj energy surface . to characterize the topology of the landscape network , doye @xcite study small lennard - jones clusters to locate nearly all the minima and transition states on the potential energy landscape . the inherent structure network of such a system has a scale - free and small - world properties . in a companion study @xcite we repeated the main results as doye studied . the numbers of minima and transition states are expected to increase roughly as @xmath104 and @xmath105 respectively , where @xmath8 is the number of atoms in the cluster . therefore , the largest network that we are able to consider is for a 14-atom cluster for which we have located 4158 minima and 90 738 transition states in agreement with the results of doye . in the next section we apply the above formalism to find the stationary states for a 3d granular system of hertz spheres in a periodic boundary . next we calculate the density of jammed states @xmath21 in the framework of the pel formulation for a system of hertz spheres . in the case of frictionless jammed systems , the mechanically stable configurations are defined as the local minima of the pel @xcite . the systems used for both , ensemble generation and molecular dynamic simulation , are the same . they are composed of 30 spherical particles in a periodic boundary box . the particles have same radius @xmath106 and interact via a hertz normal repulsive force without friction . the normal force interaction is defined as @xcite : @xmath107 where @xmath108 is the hertz exponent , @xmath109 is defined in terms of the shear modulus @xmath110 and the poisson s ratio @xmath111 of the material from which the grains are made . we use typical values for glass : @xmath112 gpa and @xmath113 and the density of the particles , @xmath114 kg / m@xmath115 @xcite . the interparticle potential energy is @xmath116 the hertz potential is chosen for its general applicability to granular materials . the results are expected to be independent of the form of the potential . below , we apply the lbfgs algorithms @xcite to find the local minima of the pes ( zero - order saddles ) . in this section , we first explain the method to obtain geometrically distinct minima in the pel to calculate the density of states . then we show that the density of the states , @xmath21 , does not change significantly after sufficient searching time for the configurations . in principle , if all local minima corresponding to the mechanically stable configurations of the pel are obtained , the density of states @xmath21 can be calculated . such an exhaustive enumeration of all the jammed states requires that @xmath8 not be too large due to computational limits . on the other hand , in order to obtain a precise average pressure in the md simulation , @xmath117 , @xmath8 can not be too small such that boundary effects are minimized . considering these constraints , we choose a @xmath118 particle system . in order to enumerate the jammed states at a given volume fraction @xmath7 , we start by generating initial unjammed packings ( not mechanically stable ) performing a monte carlo ( mc ) simulation at a high , fixed temperature . the mc part of the method applied to the initial packings assumes a flat exploration of the whole pel . every mc unjammed configuration is in the basin of attraction of a jammed state which is defined as a local minimum in the pes with a positive definite hessian matrix , that is a zero - order saddle . in order to find such a minimum , we apply the lbfgs algorithm provided by nocedal and liu @xcite explained above . the pel for each fixed @xmath7 likely includes millions of geometrically distinct minima by our simulation results . therefore , an exhaustive search of configurations is computationally long ; for a system of 30 particles it is impossible to find all the configurations with the current available computational power . however , we notice that it not crucial to find all the states , but rather a sufficiently accurate density of states . therefore , we check that the number of found configurations has saturated after sufficient trials and that the density of states @xmath21 has converged to a final shape under a prescribed approximation . it is also important to determine if the local minima are distinct . usually , the eigenvalues of the hessian matrix at each local minimum can be used to distinguish these mechanically stable packings . here , we follow this idea to compare minima to filter the symmetric packings . however , instead of calculating the eigenvalues of each packing , which is time consuming , we calculate a function of the distance between any two particles in the packing to improve search efficiency ( for the lbfgs algorithm , we do not need to calculate the hessian matrix ) . for each packing , we assign the function @xmath119 for each particle : @xmath120 where @xmath121 is the distance between particles @xmath94 and @xmath95 , @xmath122 is the system size and @xmath123 . we list the @xmath119 for each packing from minimum to maximum @xmath124 . since @xmath119 is a higher order nonlinear function , we can assume that two packings are the same if they have the same list . the tolerance is defined as : @xmath125 where @xmath119 and @xmath126 are the corresponding values from the lists of two packings . figure [ tolerance ] shows the distributions of the tolerance @xmath127 for packings at different volume fractions . this figure suggests that two packings can be considered the same if @xmath128 , which defines the noise level . from fig . [ dos_phi ] , we see that after one week of searching , @xmath21 does not change significantly , since the initial packings are generated by a completely random protocol . we also calculate the probability of finding new mechanically stable states for different searching days , defined as @xmath129 , where @xmath130 is the number of new configurations found on the @xmath94-th day and @xmath131 is the total number of configurations found in @xmath94 days . from fig . [ number ] , we see that , after one week searching , the probability of finding new configurations at different volume fractions seems to have converged in the linear plot . figure [ number]b shows a detail of the actual number of new configurations found and @xmath21 versus searching time in days suggesting convergence . however , the log - log plot of the inset in fig . [ number]a indicates that the algorithm is still searching for new configurations ; the power - law relation in the inset suggesting a neverending story . however , the main question is whether the observables have converged . a further test of convergence is obtained below in fig . [ checkalpha ] where the value of the inverse angoricity is measured as a function of the searching time in days . this plot suggests that enough ensemble packings have been obtained to capture the features of @xmath21 that give rise to the correct observables . we conclude that we have obtained an accurate enough density of states for this particular system size . regarding system size dependence , the presented results are still @xmath8 dependent , although they started to converge for @xmath132 and above , fig . [ dos_figure_d ] . more accurate calculations for large values of @xmath8 remain computationally impossible , but in our treatment the exact choice of @xmath8 is not as important as the consistency of the results between ensemble and md , for a given @xmath8 value . figure [ dos_figure ] shows @xmath21 versus @xmath133 for different volume fractions . ( a ) ( b ) converges as early as @xmath134 particles . the distribution @xmath21 ( inset ) has not fully converged yet but its shape has converged after @xmath135 and the first moment does not change as indicated by the average @xmath10 . ] as a function of internal virial @xmath133 for different volume fraction , @xmath7 , ranging from 0.610 to 0.670 . the inset shows the logarithmic distribution of @xmath21 . at low volume fraction ( @xmath136 ) , the distributions are sharp and the tails of the distributions are exponential . at high volume fraction ( @xmath137 ) , the distributions are much broader and the tails are gaussian . ] in order to analyze numerical results , we perform md simulations to obtain @xmath138 and @xmath139 , which are herein considered real dynamics . the algorithm is described in detail in @xcite . here , a general description is given : a gas of non - interacting particles at an initial volume fraction is generated in a periodically repeated cubic box . then , an extremely slow isotropic compression is applied to the system . the compression rate is @xmath140 , where the time is in units of @xmath141 . after obtaining a state for which the pressure @xmath10 is a slightly higher than the prefixed pressure we choose , the compression is stopped and the system is allowed to relax to mechanical equilibrium following newton s equations . then the system is compressed and relaxed repeatedly until the system can be mechanically stable at the predetermined pressure . to obtain the statical average of @xmath138 and @xmath139 , we repeat the simulation to get enough packing samples having statistically independent random initial particle positions . here , 250 independent packings are obtained for each fixed pressure ( see fig . [ all_md ] ) . @xmath142 and @xmath143 are flat averages of these @xmath144 packings by @xmath145 and @xmath146 ( a ) ( b ) from previous studies , it has been observed the pressure @xmath10 vanishes as power - law of @xmath7 when approaching the jamming transition as seen in eq . ( [ eq_mdp ] ) @xcite . we obtain ( fig . [ ango_a ] ) @xmath147 where @xmath148 is the volume fraction corresponding to the isostatic point j @xcite following eq . ( [ eq_mdz ] ) and @xmath149 . this critical value @xmath150 and the exponent , @xmath151 , are slightly different from the values obtained for larger systems ( @xmath152 ) @xcite . however , our purpose is to use the same system in the dynamical calculation and the exact enumeration for a proper comparison . since we obtain @xmath21 and @xmath117 for each volume fraction @xmath7 , we can calculate the inverse angoricity @xmath37 by eq . ( [ eq_edp ] ) . the pressure @xmath153 for a given @xmath7 is a function depending on @xmath37 as : @xmath154 [ h ] figure [ alpha_phi ] shows the result of the numerical integration of eq . ( [ eq_sedp ] ) for a particular @xmath155 as a function of @xmath37 using the numerically obtained @xmath21 from fig . [ dos_figure ] . to obtain the value of @xmath37 for this @xmath7 , we input the corresponding measure of the pressure obtained dynamically @xmath156 and obtain the value of @xmath37 as schematically depicted in fig . [ alpha_phi ] . the same procedure is followed for every @xmath7 ( see fig . [ p_vs_alpha ] ) and the dependence @xmath39 is obtained . [ h ] [ h ] we also check the inverse angoricity @xmath39 using @xmath21 for different searching days . to ensure the accuracy and convergence to the proper value . from fig . [ checkalpha ] , we can see that , after 10 days searching , @xmath39 is stable due to the fact that the density of state , @xmath21 , does not change significantly . for each @xmath7 we use @xmath21 to calculate @xmath157 by eq . ( [ eq_edp ] ) . then , we obtain @xmath39 by setting @xmath158 for every @xmath7 . the resulting equation of state @xmath39 is plotted in fig . [ ango_b ] and shows that the angoricity follows a power - law , near @xmath150 , of the form : @xmath159 with @xmath160 . the result is consistent with @xmath161 , suggesting that @xmath162 . for volume fraction much larger than @xmath150 , the system s input pressure @xmath156 reaches the plateau at low @xmath37 of the function @xmath163 ( see fig . [ p_vs_alpha ] ) and the corresponding @xmath39 becomes much smaller ( the angoricity @xmath164 becomes much larger ) , leading to large errors in the value of @xmath1 as @xmath7 becomes large . this might explain the plateau found in @xmath1 when @xmath165 as shown in fig . [ ango_b ] . angoricity is a measure of the number of ways the stress can be distributed in a given volume . since the stresses have a unique solution for a given configuration at the isostatic point , @xmath150 , the corresponding angoricity vanishes . at higher pressure , the system is determined by multiple degrees of freedom satisfying mechanical equilibrium , leading to a higher stress temperature , @xmath1 . the angoricity can also be viewed as a scale of stability for the system at different volume fractions . systems jammed at larger volume fractions require higher angoricity ( higher driving force ) to rearrange . ( a ) ( b ) in principle , using the inverse angoricity , @xmath37 , from eq . ( [ inverse ] ) we can calculate any macroscopic statistical observable @xmath166 at a given volume by performing the ensemble average @xcite : @xmath167 we test the ergodic hypothesis in the edwards s ensemble by comparing eq . ( [ eq_edab ] ) with the corresponding value obtained with md simulations averaged over ( @xmath144 ) sample packings , @xmath168 , generated dynamically : @xmath169 the comparison is realized by measuring the average coordination number , @xmath170 , the average force and the distribution of interparticle forces . we calculate @xmath171 by eq . ( [ eq_edz ] ) and @xmath172 as in eq . ( [ eq_mdab ] ) . using @xmath39 for each volume fraction , we calculate @xmath171 by : @xmath173 the average force @xmath174 is given by : @xmath175 where @xmath176 is the average force for each ensemble packing . finally , the force distribution @xmath177 is given by : @xmath178 equations ( [ eq_sedz])([eq_sedpro ] ) are then compared with the dynamical measures for a test of ergodicity in figs . [ predict_a ] and [ predict_b ] . [ h ] ( a ) ( b ) figure [ predict_a]a and [ predict_a]b show that the two independent estimations of the coordination number agree very well : @xmath179 . the average inter - particle force @xmath180 for a jammed packing is proportional to the pressure of the packing . we calculate @xmath174 and @xmath181 and find that they coincide very closely ( see fig . [ predict_b]a ) . the full distribution of inter - particle forces for jammed systems is also an important observable which has been extensively studied in previous works @xcite . the force distribution is calculated in the ensemble @xmath177 by averaging the force distribution for every configuration in the pes . figure [ predict_b]b shows the distribution functions . the peak of the distribution shown in fig . [ predict_b]b indicates that the systems are jammed @xcite . besides the exact shape of the distribution , the similarity between the ensemble and the dynamical calculations shown in fig . [ predict_b]b is significant . the study of @xmath170 , @xmath182 and @xmath183 reveals that the statistical ensemble can predict the macroscopic observables obtained in md . we conclude that the idea of `` thermalization '' at an angoricity is able to describe the jamming system very well . [ h ] ( a ) ( b ) the md simulations performed so far are at a predetermined pressure @xmath10 . for this case there is no difference between the force distribution @xmath184 and @xmath185 @xcite . on the other hand , a md simulation at a given fixed volume fraction @xmath7 , gives rise to different distributions . for each system with fixed @xmath7 , the packings can have various pressure . this suggests that the force distribution for each packing scaled by the average force over all packings , @xmath186 , should be different from the force distribution scaled by the average force of that particular packing @xmath184 @xcite . we now proceed to investigate a constant volume md , vmd simulation . [ h ] [ h ] the force distribution for vmd ensemble , @xmath187 is shown in fig . [ pofff]a . from fig . [ pofff]a , we find that the force distribution @xmath188 as a function of different volume fraction @xmath7 no longer collapse . at @xmath7 close to @xmath189 , the average system force @xmath176 for each packing changes dramatically . while at @xmath7 is much above @xmath189 , the fluctuations of the average system force @xmath190 decrease , then the force distribution @xmath187 changes continuously . we can also calculate the force distribution @xmath191 in the ensemble average : @xmath192 where @xmath174 is the overall average @xmath180 of the ensemble . from fig . [ pofff]b , we find the same tendency as obtained in md simulation . furthermore , we check the distribution of force @xmath193 for our vmd system ( see fig . [ poffv ] ) . we see that @xmath193 for different volume fraction @xmath7 collapses very well similarly to those obtained from the predetermined pressure system . this result suggests that @xmath194 is a global quantity that can be used to verify if the system is jammed or not @xcite . so far we have considered how the angoricity determines the pressure fluctuations in a jammed packing at a fixed @xmath7 . the role of the compactivity in the jamming transition can be analyzed in terms of the entropy which is easily calculated in the microcanonical ensemble from the density of states . figure [ entropydis ] shows the entropy of the system as a function of @xmath195 in phase space : @xmath196 here @xmath197 is the number of states which is the unnormalized version of @xmath21 . it is important to note that fig . [ entropydis ] shows the non - equilibrium entropy , in the edwards sense . at the edwards equilibrium , the entropy is maximum respect to changes in @xmath7 and @xmath133 . we will now see how the jammed system verifies the principle of maximum entropy . we analyze the entropy surface @xmath198 plotted versus @xmath199 in fig . [ entropydis ] . when we plot superimposed the md - obtained curve @xmath200 we see that the md values pass along the maximum of the entropy surface constrained by the coupling between @xmath10 and @xmath7 , eq . ( [ eq_mdz ] ) ( such a curve is superimposed to the entropy surface in fig . [ entropydis ] ) . due to the coupling through the contact force law , the maximization of entropy is not on @xmath10 or @xmath7 alone but on a combination of both . the entropy @xmath15 reaches a maximum at the point @xmath201 when we move along the direction perpendicular to the jamming curve @xmath202 ( see the maximization direction in fig . [ entropydis ] ) . this is a direct verification of the second - law of thermodynamics : the dynamical measures maximize the entropy of the system . we can use this result to obtain a relation between angoricity and compactivity and show how a new `` jamming temperature '' @xmath4 and the corresponding jamming `` heat '' capacity @xmath203 can describe the jamming transition . from the power - law relation @xmath204 , we have : @xmath205 where @xmath206 is the constant depending on the system . figure [ entropydis ] indicates that the jammed system always remain at the positions of maximal entropy , @xmath207 in the direction ( @xmath208,@xmath209 ) , perpendicular to the jamming power - law curve and the slope @xmath210 in order to further analyze this result , we plot the entropy distribution along the direction ( @xmath211,@xmath209 ) in fig . [ distentropy ] . we see that the entropy of the corresponding jammed states remains at the peak of the distributions along ( @xmath208,@xmath209 ) . this is clear when we plot the value of @xmath195 from md simulations in the plot of @xmath15 in fig . [ distentropy ] , blue dot . except for volumes very close to jamming , the md coincides with the maximum of @xmath15 when taken along ( @xmath211,@xmath209 ) . we notice that the maximization is quite accurate for large volume fractions . for @xmath7 close to jamming deviations are seen . we can not rule out that these deviations are finite size effects . the deviations for small @xmath7 ( fig . [ distentropy ] ) remains to be studied . they could be due to finite size effects or due to the fact that the value of @xmath150 is different for the md results and the microcanonical ensemble @xmath15 due to the small size of the system . in general , this plot verifies the maximum entropy principle in this particular direction . an analogous plot where the entropy is shown as a function of @xmath7 but along the horizontal direction ( or along the vertical direction , @xmath133 ) shows that the md entropy is not maximal along these two directions . thus , the maximization of entropy is not on @xmath133 or @xmath13 alone , but on a combination of both . this means that the entropy @xmath212 is maximum along the direction of ( @xmath211,@xmath209 ) and the slope for the entropy of the jamming power - law curve along this direction ( @xmath211,@xmath209 ) is @xmath213 ( see fig . [ slope ] ) , that is , @xmath214 [ h ] [ h ] by the definition of angoricity @xmath215 and compactivity @xmath216 , we have : @xmath217 @xmath218 where @xmath219 , @xmath220 and @xmath221 . by eq . ( [ eq_powerlog2 ] ) and eq . ( [ eq_powerlog3 ] ) , we can simplify eq . ( [ eq_powerlog1 ] ) : @xmath222 the relation between @xmath0 and @xmath1 can be obtained then ( fig . [ slope ] ) : @xmath223 from eq . ( [ eq_powerlog4 ] ) we obtain that : @xmath224 and near @xmath189 : @xmath225 we notice that the compactivity is negative near the jamming transition . a negative temperature is a general property of systems with bounded energy like spins @xcite : the system attains the larger volume ( or energy in spins ) at @xmath150 when @xmath3 and not @xmath226 [ the bounds @xmath227 imply that the jamming point at @xmath3 is `` hotter '' than @xmath226 . at the same time @xmath2 since the pressure vanishes ] . we conclude that , @xmath1 and @xmath0 alone can not play the role of temperature , but a combination of both determined by entropy maximization satisfying the coupling between stress and strain . instead , there is an actual `` jamming temperature '' @xmath4 that determines the direction @xmath228 in the @xmath229 plot of fig . [ entropydis ] along the jamming equation of state ( see fig . [ slope ] ) . by maximizing the entropy along this direction we obtain the `` jamming temperature '' @xmath4 as a function of @xmath1 and @xmath0 : @xmath230 that is : @xmath231 thus , the temperature vanishes at the jamming transition . furthermore , the `` jamming energy '' @xmath232 , corresponding to the `` jamming temperature '' @xmath4 in eq . ( [ eq_powerlog5 ] ) , has the relation as below : @xmath233 that is , @xmath234 and @xmath235 by the definition of `` heat '' capacity , we obtain two jamming capacities as the response to changes in @xmath1 and @xmath0 : @xmath236 the jamming capacity @xmath203 can be obtained as : @xmath237 finally , with eq . ( [ eq_powerlog1])([eq_powerlog3 ] ) , the capacity @xmath203 can be calculated : @xmath238 since @xmath239 and @xmath240 , we obtain @xmath241 from eq . ( [ eq_cx ] ) , the jamming capacities diverge at the jamming transition as @xmath242 and @xmath243 . however , this result does not imply that the transition is critical since from fluctuation theory of pressure and volume @xcite we obtain : @xmath244 thus , the pressure and volume fluctuations near the jamming transition do not diverge , but instead vanish when @xmath245 and @xmath246 . from a thermodynamical point of view , the transition is not of second order due to the lack of critical fluctuations . as a consequence , no diverging static correlation length from a correlation function can be found at the jamming point . however , other correlation lengths of dynamic origin may still exist in the response of the jammed system to perturbations , such as those imposed by a shear strain or in vibrating modes @xcite . such a dynamic correlation length would not appear in a purely thermodynamic static treatment as developed here . we note that static anisotropic packings can be treated in the present formalism by allowing the inverse angoricity to be tensorial @xcite . the intensive jamming temperature eq . ( [ eq_powerlog7 ] ) gives use to a jamming effective energy @xmath232 as the extensive variable satisfying @xmath247 and a full jamming capacity @xmath248 , which also diverges at jamming . however , the fluctuations of @xmath232 defined as @xmath249 has the same behavior as the fluctuations of volume and pressure , vanishing at the jamming transition @xmath250 [ @xmath251 in eq . ( [ eq_powerlog7 ] ) ] . the results so far show a general agreement between md and the ensemble average . these include the maximum entropy principle and ergodicity . we now turn to a comparison with similar simulations done by ohern _ these studies perform an exhaustive search of all configurations in the pel of frictionless particles similarly as in the present paper . however , they find that the microstates are not equiprobable , i.e. , microstates with the same pressure and volume fraction ( pressure is fixed at zero since only hard sphere states are of interest ) do not have the same probability when sampled by a given algorithm . furthermore , experimental studies of equilibration between two systems @xcite , suggests that a hidden variable is necessary to describe the microstates , further supporting the results of @xcite . the applicability of the microcanonical ensemble is based on the fact that the microstates are defined by @xmath22 . thus , the fact that the states are not equiprobable implies that there must be an extra variable needed to describe their probabilities . therefore , ergodicity and the maximum entropy principle , which are downstream from equiprobability , are not supposed to hold , in disagreement with the results shown in the present paper . to investigate this situation , we repeat the same calculations as in @xcite with our algorithms . we first rule out subtleties related to algorithmic dependent results in sampling the space of configurations . we use our 30 particles system and use @xmath252 very close to jamming and @xmath253 to look for the hard sphere packings . we search for the jammed configurations as above . we recall that the sampling of the space of configurations is not complete due to the relatively large system size but represent a good sampling as discussed above . ref . @xcite uses a different system of 14 particles in 2d for which 248,900 configurations are found exhaustively sampling the phase space ( which is estimated to have @xmath254 states ) . these simulations correspond to a system with periodic boundary conditions for which a larger space is expected than the close boundary - system of section [ example ] . however , these differences do not affect the conclusions below . we start by measuring @xmath255 which is the probability to find a given microstate @xmath256 as defined by @xcite : each packing can be obtained many times during a search and therefore @xmath255 measures the probability for which each packing occurs . the main result of @xcite is that @xmath255 differs by many orders of magnitude for states with fixed @xmath22 . indeed , even configurations which are visually very similar can be @xmath257 more frequent , see fig . 1 of @xcite . figure [ kun1 ] shows @xmath255 sorted as a function of @xmath256 , the rank , as in @xcite . this plot reproduces the results of @xcite in our system . for a fixed pressure and volume there are many states with a large difference in their probability . the least probable states are 10@xmath258 less probable than the most probable state showing a breakdown of equiprobability . the question is how to interpret the results of ergodicity in the light of the failure of equiprobability and whether there is a need for an extra variable to describe the microstates . we first mention the issue of the small system size . it is quite possible that the low probability states will completely disappear in the thermodynamic limit and the ones remaining are the most probable ones with equal probability . indeed , the flat average assumption is only valid in the thermodynamic limit and simply says that even if there exists less probable states ( @xmath259 less probable ) then they will be irrelevant in the ensemble average , thus only the most probable and flat states are important . ( a ) ( b ) we have done simulations with @xmath26014 particles and found that the least probable states are @xmath261 less probable than the most probable states . comparing with the factor @xmath259 for @xmath123 , may indicate that the system size may take care of the non - equiprobability problem . however , calculations for larger system to fully test this assertion are out of the range of current and near future computational power . second , we notice that the coordination number is also important to define the jammed states . figure [ kun2 ] plots the same states as fig . [ kun1 ] but as a function of @xmath262 , the coordination number of microstate @xmath256 . the most probable states satisfy : @xmath263 furthermore , if we sum up all the states for a given @xmath262 and plot @xmath264 vs @xmath262 we obtain eq . ( [ fk ] ) as seen in fig . this result does not mean that @xmath262 is the hidden variable but rather eq . ( [ fk ] ) provides the density of states proposed in @xcite in the thermodynamics calculation of the random close packing of spheres . indeed , we have predicted that the density of states @xmath265 , with @xmath266 playing the role of a planck constant defining the minimum size in the volume landscape . according to eq . ( [ fk ] ) , this prediction is satisfied in average with @xmath267 which is a small number as expected . this result indicates that some variability in the probabilities of the microstates is expected from the fluctuations in the coordination number of each microstate . in appendix [ z ] we elaborate an extension of the framework of @xcite to incorporate fluctuations in @xmath11 that are neglected in @xcite . the purpose is to test whether the rcp and jamming transition are affected by these fluctuations . we find that the results are consistent with those found in @xcite . we notice that for a fix @xmath262 there are still many marginal states with very small probabilities as seen in fig . if these states do not completely disappear in the thermodynamic limit , then they need to be explained . we end this discussion by providing a possible explanation for the existence of these states . the numerical breakdown of equiprobability might be related to the fact that the found packings are not indistinguishable . indeed , we ignore the rotation and translation symmetries of the packing in order to make the numerical search possible . however , for the edwards flat hypothesis , these packings should be assumed different . once we breakdown the rotational symmetry , there would be many similar packings . the high degeneracy of the high symmetric packings may be responsible for the uneven distribution , which would be , in this case , simply artificial . for instance , consider two packings with 4 particles : ( a ) a square packing with each particle on the corner and ( b ) a triangle with each particle in each corner plus one in the center . for both packings there are 4 ! = 24 different permutations , which should be considered as 24 different packings , in principle . however , since we can rotate the square packing by 90 degree and obtain the same one , there are only 24/4 = 6 distinguishable packings . similarly , for the triangle , there are 24/3 = 8 distinguishable packings . the probability between ( a ) and ( b ) is uneven ( 6:8 ) if we assume that each distinguishable packing is equal - probable . therefore , different symmetries of the packings may contribute to the unequal probabilities that we measure in the algorithms . therefore , if the edwards assumption is correct , @xmath255 should be proportional to @xmath268 , where @xmath268 is the order of the symmetry group ( point group ) of the packing @xmath256 , since there are @xmath268 degenerations ( same packing if particles are identical ) . this conjecture needs extra evaluation of the symmetry of each packing . for instance , the translation invariance is important , and for cubic periodic boundary , it is also important to include the symmetry of cubic point group c@xmath269 . we do not investigate this conjecture but rather provide the codes and packings in http://jamlab.org to do that . since the 3d case is complicated , one might try the 2d system first to easily visualize different packings . a simple question is : given two packings with different frequencies , how do they look like @xcite ? would be the high symmetric one visited more , or inversely ? we have demonstrated that the concept of `` thermalization '' at a compactivity and angoricity in jammed systems is reasonable by the direct test of ergodicity . the numerical results indicate that the full canonical ensemble of pressure and volume describes the observables near the jamming transition quite well . from a static thermodynamic viewpoint , the jamming phase transition does not present critical fluctuations characteristic of second - order transitions since the fluctuations of several observables vanish approaching jamming . the lack of critical fluctuations is respect to the angoricity and compactivity in the jammed phase @xmath270 , which does not preclude the existence of critical fluctuations when accounting for the full range of fluctuations in the liquid to jammed transition below @xmath150 . thus , a critical diverging length scale might still appear as @xmath271 @xcite , which has been recently observed by experiment @xcite . in conclusion , our results suggest an ensemble treatment of the jamming transition . one possible analytical route to use this formalism would be to incorporate the coupling between volume and coordination number at the particle level found in @xcite together with similar dependence for the stress to solve the partition function . this treatment would allow analytical solutions for the observables with the goal of characterizing the scaling laws near the jamming transition . acknowledgements : we thank nsf - cmmt and doe - geosciences division for financial support and l. gallos for discussions . 99 a. coniglio , a. fierro , h. j. herrmann and m. nicodemi , _ unifying concepts in granular media and glasses _ ( elsevier , amsterdam , 2004 ) . r. p. behringer and j. t. jenkins , _ powders and grains _ , vol . * 97 * ( balkema , rotterdam , 1997 ) . s. f. edwards and r. b. s. oakeshott , physica a * 157 * , 1080 ( 1989 ) . a. j. liu , and s. r. nagel , nature * 396 * , 21 ( 1998 ) . c. s. ohern , s. a. langer , a. j. liu , and s. r. nagel , phys . lett . * 88 * , 075507 ( 2002 ) . h. a. makse , d. l. johnson , and l. m. schwartz , phys . lett . * 84 * , 4160 ( 2000 ) . w. g. ellenbroek , e. somfai , m. van hecke , and w. van saarloos , phys . lett . * 97 * , 258001 ( 2006 ) . s. f. edwards , physica a * 353 * , 114 ( 2005 ) . s. f. edwards and d. v. grinev , phys . lett * 82 * , 5397 ( 1999 ) . r. c. ball and r. blumenfeld , phys . lett . * 88 * , 115505 ( 2002 ) . s. henkes and b. chakraborty , phys . lett . * 95 * , 198002 ( 2005 ) . e. r. nowak , j. b. knight , e. ben - naim , h. m. jaeger and s. r. nagel , phys . e * 57 * , 1971 ( 1998 ) . p. philippe and d. bideau , europhys . lett . * 60 * , 677 ( 2002 ) . m. schroter , d. i. goldman , h. l. swinney , phys . e * 71 * , 030301(r ) ( 2005 ) . j. bruji , p. wang , c. song , d. l. johnson , o. sindt , and h. a. makse , phys . lett . * 95 * , 128001 ( 2005 ) . g. d. anna and g. gremaud , nature * 413 * , 407 ( 2001 ) . c. song , p. wang , and h. a. makse , proc . sci . * 102 * , 2299 ( 2005 ) . p. wang , c. song , c. briscoe and h. a. makse , phys . e * 77 * , 061309 ( 2008 ) . m. nicodemi , phys . * 82 * , 3734 ( 1999 ) . j. brey , a. prados , and b. sanchez - rey , physica a * 275 * , 310 ( 2000 ) . a. barrat , j. kurchan , v. loreto , m. sellitto , phys . lett . * 85 * , 5034 ( 2000 ) . d. s. dean and a. lefevre , phys . lett . * 86 * , 5639 ( 2001 ) . h. a. makse and j. kurchan , nature * 415 * , 614 ( 2002 ) . m. pica ciamarra , a. coniglio and m. nicodemi , phys . * 97 * , 158001 ( 2006 ) . gao , j. blawzdziewicz and c. s. ohern , phys . e * 74 * , 061304 ( 2006 ) . n. xu , j. blawzdziewicz and c. s. ohern , phys . e * 71 * , 061306 ( 2005 ) . gao , j. blawzdziewicz , c. s. ohern , and m. shattuck , phys . e * 80 * , 061304 ( 2009 ) . f. lechenault and k. e. daniels , soft matter * 6 * , 3074 ( 2010 ) . c. song , p. wang and h. a. makse , nature * 453 * , 629 ( 2008 ) . c. briscoe , c. song , p. wang and h. a. makse , phys . * 101 * , 188001 ( 2008 ) . m. danisch , y. jin and h. a. makse , phys . e * 81 * , 051303 ( 2010 ) . s. meyer , c. song , y. jin , h. a. makse , physica a * 389 , * 5137 ( 2010 ) . y. jin , p. charbonneau , s. meyer , c. song , and f. zamponi , phys . e * 82 * , 051126 ( 2010 ) . k. wang , c. song , p. wang and h. a. makse , europhys . lett . * 91 * , 68001 ( 2010 ) . h. a. makse , n. gland , d. l. johnson and l. m. schwartz , phys . e * 70 * , 061302 ( 2004 ) . s. henkes , c. s. ohern and b. chakraborty , phys . * 99 * , 038002 ( 2007 ) . s. henkes and b. chakraborty , phys . e * 79 * , 061301 ( 2009 ) . j. bruji@xmath272 , s. f. edwards , i. hopkinson , and h. a. makse , physica a * 327 * , 201 ( 2003 ) . j. bruji@xmath272 , c. song , p. wang , c. briscoe , g. marty , and h. a. makse , phys . lett . * 98 * , 248001 ( 2007 ) . it is worth the clarification . in the study of @xcite a different exponent @xmath273 in frictionless packings was reported . the reason might be due to the protocol employed which uses a servo mechanism @xcite to constantly adjust the strain to achieved a predetermined pressure . we think that at unrealistic low pressures ( below 100 kpa ) the protocol may not equilibrate the packings properly for the parameters used . in fact , there is an excess of small forces ( noticeable in the force distribution which shows a full exponential rather than the plateau at low forces ) probably arising from some particles which were not fully equilibrated . in a later study @xcite a different protocol was used , called split algorithm in @xcite , with the result @xmath32 . d. j. wales , _ energy landscapes _ , ( cambridge university press , cambridge , 2003 ) . p. g. debenedetti and f. h. stillinger , nature * 410 * , 259 ( 2001 ) . m. goldstein , j. chem . phys . * 51 * , 3728 ( 1969 ) . f. h. stillinger and t. a. weber , science * 225 * , 983 ( 1984 ) . j. p. k. doye and d. j. wales , j. chem . phys . * 116 * , 3777 ( 1994 ) . s. carmi , s. havlin , c. song , k. wang , h. a. makse , j. phys . a : math . theor . * 42 * , 105101 ( 2009 ) . d. c. liu and j. nocedal , mathematical programming b * 45 * , 503 ( 1989 ) . c. j. cerjan and w. h. miller , j. chem . phys . * 75 * , 2800 ( 1981 ) . t. s. grigera , http://arxiv.org/abs/cond-mat/0509301 . d. j. wales and j. p. k. doye , j. chem . 119 * , 12409 ( 2003 ) . d. j. wales , j. chem . phys . * 101 * , 3750 ( 1994 ) . d. j. wales and t. r. walsh , j. chem . phys . * 105 * , 6957 ( 1996 ) . k. doye , phys . lett . * 88 * , 238701 ( 2002 ) . l. d. landau and e. m. lifshitz , _ theory of elasticity _ ( pergamon , ny , 1970 ) . h. p. zhang and h. a. makse , phys . e * 72 * , 011301 ( 2005 ) . c. s. ohern , s. a. langer , a. j. liu and s. r. nagel , phys . lett . * 86 * , 111 ( 2001 ) . j. h. snoeijer , t. j. h. vlugt , m. van hecke and w. van saarloos , phys . lett . * 92 * , 054302 ( 2004 ) . l. d. landau and e. m. lifshitz , _ statistical physics _ ( pergamon , new york , 3rd edition 1980 ) . m. wyart , s. r. nagel and t. a. witten , europhys . lett . * 72 * , 486 ( 2005 ) . m. pica ciamarra and a. coniglio , phys . * 103 * , 235701 ( 2009 ) o. dauchot , g. marty and g. biroli , phys . * 95 * , 265701 ( 2005 ) . c. song , p. wang , y. jin , h. a. makse , physica a * 389 * , 4497 ( 2010 ) . wang , c. song , y. jin , h. a. makse , physica a * 390 * , 427 ( 2010 ) . c. briscoe , c. song , p. wang , h. a. makse , physica a * 389 * , 3978 ( 2010 ) . p. wang , c. song , y. jin , h. a. makse , j. stat . p12005 ( 2010 ) . here , we develop a @xmath11-ensemble for hard spheres in the limit of zero angoricity . in the main test we found that fluctuations in @xmath11 may account for certain variability in the probability of microstates . here we investigate whether this variability affect the existence of rcp and the jamming point . we develop a partition function in edwards ensemble to study the dependence of rcp on this type of fluctuations . the partition function is @xmath274 where @xmath275 and @xmath276 , @xmath277 , and @xmath278 . we follow the notation and concepts from @xcite . we define @xmath279 , thus : @xmath280 where @xmath281 where @xmath282 according to fig . we consider the inverse fourier transform of @xmath283 : @xmath284\equiv \int_{-\infty}^{\infty } e^{2\pi i f x } p(x ) dx = \int_{0}^{\infty}\ldots\int_{0}^{\infty}\prod_i e^{-(z_i / z^ * + \beta\kappa / z_i ) } e^{2\pi i f \sum z_i / n}dz_i = \\ = \left[\int_{0}^{\infty}e^{-(z / z^ * + \beta\kappa / z ) } e^{2\pi i f z / n}dz\right]^n = \left\{\int_{0}^{\infty } \left [ 1+\left(\frac{2\pi ifz}{n}\right)+\frac{1}{2}\left(\frac{2\pi ifz}{n}\right)^2 + \ldots \right]e^{-(z / z^ * + \beta\kappa / z ) } \right\}^n . \end{split}\ ] ] thus , @xmath289 & = ( 2z^*)^n\left[b^{1/2}k_0(2 b^{1/2 } ) + \left(\frac{2\pi ifz^ * } { n}\right ) b k_1(2 b^{1/2 } ) + \frac{1}{2}\left(\frac{2\pi ifz^ * } { n}\right)^2 b^{3/2 } k_2(2 b^{1/2})+ o(n^{-3})\right]^n\\ & = ( 2z^*)^n\exp \left\ { n\ln\left[b^{1/2}k_0(2 b^{1/2 } ) + \left(\frac{2\pi ifz^ * } { n}\right ) b k_1(2 b^{1/2 } ) + \frac{1}{2 } \left(\frac{2\pi ifz^ * } { n}\right)^2 b^{3/2 } k_2(2 b^{1/2})+ o(n^{-3})\right]\right\}\\ & = ( 2z^ * b^{1/2}k_0(2 b^{1/2}))^n\exp \left\ { n\ln\left[1 + \left(\frac{2\pi ifz^ * } { n}\right ) \frac{k_1(2 b^{1/2})}{k_0(2 b^{1/2})}b^{1/2 } + \frac{1}{2}\left(\frac{2\pi ifz^ * } { n}\right)^2 \frac{k_2(2 b^{1/2})}{k_0(2 b^{1/2})}b+ o(n^{-3})\right]\right\ } \end{split}\ ] ] now , we expand @xmath290 @xmath291\right\ } \\ = & \exp \left\ { n\left[\left(\frac{2\pi ifz^ * } { n}\right ) \frac{k_1(2 b^{1/2})}{k_0(2 b^{1/2})}b^{1/2 } + \frac{1}{2}\left(\frac{2\pi ifz^ * } { n}\right)^2 \frac{k_2(2 b^{1/2})}{k_0(2 b^{1/2})}b- \frac{1}{2}\left(\frac{2\pi ifz^ * } { n}\frac{k_1(2 b^{1/2})}{k_0(2 b^{1/2})}\right)^2 b+ o(n^{-3})\right]\right\}\\ \approx & \exp\left [ 2\pi if \left(z^*b^{1/2}\frac{k_1(2 b^{1/2})}{k_0(2 b^{1/2})}\right ) - \frac{(2\pi f)^2}{2n } { z^ * } ^{2 } b\left(\frac{k_2(2 b^{1/2})}{k_0(2 b^{1/2 } ) } - \frac{k_1(2 b^{1/2})^2}{k_0(2 b^{1/2})^2}\right)\right ] , \end{split}\ ] ] then , @xmath304 \theta(\mu - z_{max } ) = \left[\frac{1}{2}\left(\frac{l(b)(l(b)-z_{max})}{b+l(b)-l(b)^2}\right)^2-\frac{(b - z_{max } l(b))(l(b)-z_{max})}{b+l(b)-l(b)^2 } \right]\theta(l(b)-z_{max}),\ ] ] thus , @xmath306\theta(l(b)-z_{max } ) \\ + \left[\frac{1}{2}\left(\frac{l(b)(l(b)-z_{min})}{b+l(b)-l(b)^2}\right)^2 -\frac{(b - z_{min } l(b))(l(b)-z_{min})}{b+l(b)-l(b)^2 } \right]\theta(z_{min}-l(b ) ) , \end{split}\ ] ] there are two phase transitions at @xmath308 and @xmath309 . for the jammed phase @xmath310 , we have @xmath311 . if @xmath312 is a small value , @xmath282 from fig . [ kun2 ] , then @xmath99 is relatively large . thus , @xmath313 and @xmath314 . similarly , @xmath315 , which is consistent with the boundaries of the phase diagram obtained in @xcite . furthermore , @xmath316 and @xmath317 , where @xmath318 . or , @xmath319 . thus , we have verified that the inclusion of fluctuations in the coordination number does not change the shape of the jamming phase diagram obtained in @xcite . these fluctuations may affect the probability of the microstates according to the density of states proposed in @xcite . a further application of this generalized @xmath11-ensemble is developed in @xcite to calculate the probability of coordination numbers in packings , with good agreement with the numerical results for different packings in the phase diagram .

we analyze data from simulations of two- and three - dimensional ( 2d and 3d ) glass - forming liquids using a correlation function defined in terms of a memory function with a negative inverse power - law tail . the self - intermediate function and the autocorrelation functions of pressure and shear stress are analyzed ; the obtained fits are very good , at least as good as with a stretched exponential . in contrast to the stretched exponential , the key shape parameter the exponent of the power - law tail seems to be the same for all three correlation functions . it decreases from a value around 2 at high temperature to a value close to 1.58 ( 2d ) , 1.50 ( 3d ) at low temperatures . at the same time the amplitude of the tail increases towards towards a limiting value corresponding to a diverging relaxation time , which is related to anomalous diffusion . on the other hand , careful analysis of the long time behavior in the case of the intermediate scattering function suggests that the memory function is cut - off exponentially , which avoids the divergence of the relaxation time . repeating the fits with an exponential cut - off included indicates that the power - law exponent is in fact independent of temperature and close to 1.58/1.50 over the whole range . instead of the divergence , a fragile - to - strong crossover in the dynamics , estimated to occur around for the 3d kob - andersen system . another key parameter of the fitting procedure may be interpreted as a short - time rate , the amount of decorrelation that occurs in a fixed , relatively short time interval ( compared to the alpha time ) . this quantity is observed to have a near - arrhenius temperature dependence , while its wavenumber dependence seems to be diffusive ( ) over a wider range than that of the relaxation itself , a further indication that this `` bare relaxation rate '' is simpler than the full dynamics .

0904.2480

time - dependent correlation functions are a standard tool for quantifying the dynamics of physical systems.@xcite for the most basic analyses , knowledge of a single quantity the relaxation time and its dependence on external parameters such as temperature , is sufficient . but often more detailed information , in particular concerning the shape of the correlation function , is required . this might be , for example , because competing theoretical explanations make different predictions about the shape , for example . in the case of highly viscous glass - forming liquids@xcite some functions typically investigated include the autocorrelation functions of energy , pressure , shear - stress , and dielectric moment . these are related through the fluctuation - dissipation theorem@xcite to the corresponding dynamic response functions : frequency dependent specific - heat , bulk modulus , shear - modulus ( or equivalently viscosity ) and dielectric constant , respectively . the key features of viscous - liquid dynamics are ( 1 ) the non - arrhenius temperature dependence of the main , or alpha , relaxation time and ( 2 ) the non - exponentiality of the corresponding correlation functions . any theory of glass - forming liquids has to be able to explain these `` nons''violations of what is generally expected in relaxing systems ( arrhenius temperature dependence and exponential , or debye , relaxation ) . this paper presents a method for fitting relaxation data in the time domain , by parameterizing not the correlation itself but its associated memory function.@xcite the latter is a unique function related to the correlation function . the motivation comes from the hope that a relatively simple description of the memory function could be attained , and this hope turns out to be justified very good fits can be achieved by taking the memory function to have the form of an inverse power law for non - zero times . moreover the exponent of the power law seems to have a common value , approaching at low temperatures a value around 1.6 for the 2d system and 1.5 for the 3d system , for the different correlation functions , something that is not true of the `` stretching parameter '' @xmath2 when fitting using a stretched exponential . in fact when the memory function parameterization is generalized to include an exponential cut - off , the exponent is temperature independent . in the next section we discuss some general ideas for describing relaxation and review the concept of the memory function associated with a correlation function . following that is a brief description of our simulations , while section [ dataanalysis ] presents the details of our analysis method as applied to time - domain data from the simulations . results from the fitting procedure are presented in section [ results ] . section [ discussion ] discusses possible ways of interpreting the power - law description . in this work we consider normalized correlation functions @xmath3 , typically ( but not exclusively ) defined in terms of a dynamical variable @xmath4 as @xmath5 for example @xmath6 could be the potential energy , pressure , or shear stress . here @xmath7 indicates deviation from the thermal average . debye relaxation corresponds to a simple exponential decay with time constant @xmath8 : @xmath9 there are many ways to define a relaxation time @xmath10 which agrees with @xmath8 in the case of debye relaxation . for our purposes the most convenient is the time integral of @xmath3 : @xmath11 one advantage of this definition is that it uses all of the information contained in @xmath3 , unlike , for example , a definition based on the time at which @xmath3 attains a certain value . it is also convenient when working with laplace transforms , since this is simply the the @xmath12 value of the laplace transform of @xmath3 . the function most commonly used to fit time - domain data is the stretched exponential or kohlrausch - watt - williams ( kww)@xcite function , @xmath13 where @xmath14 sets the time scale and @xmath15 is a number between 0 and 1 , known as the stretching parameter . exponential behavior is recovered for @xmath16 . by integrating we find that @xmath17 . fits to eq . ( [ kww ] ) are often very good , but on the other hand experiments@xcite suggest that the relaxation crosses over to exponential at the longest times ( a decade or more beyond @xmath14 ) , corresponding to the last few percent of relaxation . this has not been investigated much , if at all , in simulations , probably because it is difficult to get high quality data at the longest times . in the case of single - particle diffusion , however , the crossover to exponential behavior should coincide with the onset of fickian diffusion and the convergence of the van hove correlation function to a gaussian , which have been studied by various authors.@xcite the memory function concept was introduced by the work of zwanzig@xcite and mori,@xcite which provided a theoretical formalism for the calculation of correlation functions of many - body systems , based on projector - operator techniques . one of the historically most important uses for the memory function has been to elucidate the short - time structure of correlation functions . for example , pure exponential decay , represented by a delta - like memory function , is unphysical at short times , since the correlation , being even and smooth , should have zero slope at @xmath18 . replacing the delta function with a less singular function ( for example a step function , a gaussian , or an exponential ) gives more physical short - time behavior . in this work , however , we are concerned with the long time behavior , and in our analysis will ignore the short time behavior , which is mainly associated with vibrational motion and irrelevant for structural relaxation . the memory function , @xmath19 , is defined@xcite for a normalized autocorrelation function @xmath3 by @xmath20 the relation between @xmath3 and @xmath19 is completely invertible , which can be seen by considering the laplace transformation of this equation : @xmath21 which implies @xmath22 an invertible relation between @xmath23 and @xmath24 . thus we can think of the memory function as a kind of transform of the correlation function . setting @xmath12 in eq . ( [ psi_s_k_s ] ) gives an expression for @xmath10 : @xmath25 this will be useful below . although though of our analysis will involve a pure power - law decay of @xmath19 , it should be noted that if @xmath3 is known to decay exponentially at the longest times , the same must be true of the memory function : if @xmath3 decays as an exponentially @xmath26 at long times ( with @xmath27 is real and positive ) , this implies that the least negative singularity of @xmath28 is at @xmath29 . in particular @xmath30 is analytic , as well as real and positive , in a finite region about @xmath12 ( recall @xmath31 ) . the same is true of @xmath32 and therefore @xmath24 ( = @xmath33 ) . being finite and analytic at @xmath12 is incompatible with a power - law decay for @xmath19 , which would correspond to a power - law for @xmath24 with ( in general ) non - integer exponent as @xmath34 . if an exponential cut - off is included in @xmath19 , however , this moves the singular point a finite distance to the left along the negative real @xmath35-axis . a useful starting point for considering the meaning of @xmath19 is the case @xmath36 , which gives exponential relaxation @xmath37 , i.e. , @xmath38 , consistent with eq . ( [ tau_alpha_from_k ] ) . this is the memory - less , or markov , case . note that the ` amplitude ' @xmath39 gives the rate of the exponential decay . consider integrating eq . ( [ memoryfunctiondef ] ) as if it were a physical equation of motion : in the memoryless case , as time goes on , the delta function means that the only contribution to the integral on the right - hand side is from the current time , @xmath40 . thus , the rate of change of @xmath3 is proportional to its current value and independent of its previous values . this of course gives exponential relaxation . if we generalize slightly to the case @xmath41 , a constant , for @xmath42 , and zero otherwise , then for times much greater than @xmath43 we again expect exponential decay with relaxation time @xmath44 . is sufficiently large for a given value of @xmath45 , there will also be an oscillatory factor . ] when @xmath19 is non - zero for all @xmath46 , but @xmath47 converges sufficiently quickly ( e.g. exponentially ) to a finite positive value , we can expect exponential decay in @xmath3 for long times . for times @xmath48 when the memory function is still changing , @xmath47 gives an estimate of the instantaneous decay rate of @xmath3 . what happens when @xmath19 , which must be positive at time zero ( otherwise @xmath3 will diverge ) , becomes negative ? at that point @xmath49 starts to decrease , implying that the decay rate decreases . as long as @xmath50 , @xmath3 will exhibit slower - than - exponential relaxation the instantaneous decay rate keeps decreasing . therefore we expect that for viscous liquids the memory function is negative and significant for times up to @xmath10 , around which time @xmath47 has more or less converged to the limiting value and exponential relaxation takes over . when we consider simulation data we will be not interested in short - time vibrational contributions to @xmath3 or @xmath19 . a standard way of removing the effects of vibrations is to consider the so - called inherent dynamics,@xcite obtained by minimizing configurations along the actual simulated trajectory to the corresponding local minimum of the potential energy the inherent state . the correspondence is defined by steepest - descent minimization and is , in principle , unique for almost all configurations . it has been shown that the resulting trajectory of inherent states preserves the long time features of the dynamics . @xcite correlation functions are essentially unchanged except that the usual initial decay due to vibrations on the picosecond time scale is missing . a technical problem is introduced by this procedure , however : the inherent correlation functions tend to have non - zero , and therefore discontinuous , slopes at @xmath18 . this is not so surprising , given that the inherent trajectory is itself necessarily discontinuous , but it does pose problems for defining the short - time behavior of the memory function . to avoid these problems , and given that simulation data is generally available at regular , discrete times , we write a discrete version of eq . ( [ memoryfunctiondef ] ) : @xmath51 here time has been discretized in units of @xmath52 , so @xmath53 . equation ( [ memoryfunctiondiscrete ] ) can be straightforwardly used to calculate the discrete values of the autocorrelation function , @xmath54 , given the discrete values of the memory function , @xmath55 : start by setting @xmath56 and then for each @xmath57 in turn , calculate the sum on the right hand side , which only involves @xmath58 for @xmath59 and the known @xmath60 , to get the next unknown value , @xmath61 . it is almost as straightforward to see that the reverse transformation is possible , i.e. , given @xmath54 , to determine @xmath62 . for example , from eq . ( [ memoryfunctiondiscrete ] ) for @xmath63 one finds @xmath64 ( since we assume the normalization @xmath56 ) . knowing now @xmath65 , writing eq . ( [ memoryfunctiondiscrete ] ) for @xmath66 gives again an equation with only one unknown , namely @xmath67 . in this way the values of @xmath68 can be solved for one by one . this process may be formalized using the idea of a generating function.@xcite we consider the series @xmath69 as the coefficients of a power series in a variable @xmath70 , which ( formally ) defines the so - called generating function @xmath71 , similarly @xmath72 . we may treat generating functions as quantities which can be added , multiplied and divided ( providing the first coefficient is non - zero ) , and thus can note that the right hand of eq . ( [ memoryfunctiondiscrete ] ) is in fact the @xmath73th coefficient in the product series @xmath74 ( this is like the convolution - product correspondence in laplace transforms ) . thus the equation may be re - written as @xmath75 where the difference function @xmath76 is defined as the series @xmath77 , @xmath78 . @xmath79 is then given by @xmath80 which may be evaluated using standard algorithms for multiplying and dividing power series.@xcite mathematically this procedure is equivalent to the inversion procedure mentioned above ; thinking in terms of generating functions simply makes the implementation more straightforward . we use this transformation in the data analysis to relate correlation functions to their corresponding memory functions . in the following , we use @xmath19 to mean @xmath81 , where @xmath82 . we have simulated both two- and three - dimensional ( 2d and 3d ) binary lennard - jones ( blj ) fluids . the parameters for the blj potential for both 2d and 3d are ( where l and s stand for large and small particles , and @xmath83 and @xmath84 the energy and length scales , respectively ) @xmath85 , @xmath86 , @xmath87 , @xmath88 , @xmath89 , @xmath90 . all particles have the same mass @xmath91 . these parameters are identical to those of the 3d blj introduced by kob and andersen @xcite . the potential was truncated using an interpolating polynomial between 2.4 @xmath92 and 2.7 @xmath92 ( @xmath93 ) . in all simulations periodic boundary conditions with constant area / volume were used . in 2d the total number of particles @xmath94=700 of which 60% were of type l , while in 3d , @xmath94=1372 of which 80% were of type l. the particle density was 1.2@xmath95 in both cases ( where @xmath96 is the dimension ) . constant energy dynamics were simulated using the verlet algorithm with a time step of 0.01 @xmath97 from now on , all quantities will be reported in the units defined by @xmath98 , @xmath99 and @xmath100 . if one converts to physical units by assuming that the parameters for large particles are chosen to model argon ( i.e. , @xmath101 nm , @xmath102 kj / mol , @xmath103 u ) , then the time unit corresponds roughly to 2 ps . initial configurations consisted of fcc lattices with particles randomly assigned to be of type l or s such that the appropriate fraction was achieved . all simulations were at constant nve ; in particular the energy was controlled rather than the temperature @xmath43 , but we quote the temperature defined in terms of the mean kinetic energy per particle instead as it is more meaningful . starting at at @xmath104 ( where the liquid is not particularly viscous ) ten independent runs ( differing by the random placement of the particles on the initial lattice ) were made . the initial lattices melted immediately and equilibrated in the liquid state . for each state point , an equilibration run was carried out followed by a production run of the same length . equilibration was checked by examining the self - intermediate scattering function @xmath105 for the larger particles ; @xmath106 is 6.5 in 2d and 7.3 in 3d ( corresponding roughly to the peak of the structure factor for large particles ) . @xmath105 was required to be essentially the same for all ten runs , and the run - length to be of order 100 times the structural relaxation time ( @xmath107 ) . if the condition was not satisfied , the production run was considered to be part of the equilibration and a new production run was initiated from the final configurations . steps in temperature were made by changing the total energy according to the estimated specific heat ; a new kinetic energy was chosen such that when summed with the current potential energy the correct total energy would result . then the particles velocities were randomized using a gaussian distribution with the appropriate width and then rescaling to give the exact desired kinetic energy . during the equilibration process it was checked that the new temperature ( mean kinetic energy ) was the desired one and velocities were rescaled if necessary ( generally a small change ) . during the simulations , collective quantities such as potential and kinetic energy and the different components of stress were computed and written at regular intervals ( typically every 100 or 500 time steps , though some shorter runs were made with more frequent output in order to determine the short time behavior of the correlation functions ) . configurations were saved at `` logarithmic intervals '' to allow calculation of @xmath108 for a broad range of times @xmath48 without writing configurations as frequently as the collective quantities ( which would require a huge amount of disk storage ) . because @xmath108 involves an average over particles , it has significantly smaller statistical error and therefore has low noise out to quite long times beyond the relaxation time . it should be noted , however , that it is technically not the autocorrelation function of a dynamical variable . autocorrelation functions of collective quantities were calculated using standard fourier transform techniques . both these and @xmath108 were averaged over the ten runs . for the main part of the analysis the autocorrelation functions were also `` logarithmically averaged '' in time . this weights different time scales equally and involves grouping the data points into equal - sized bins of @xmath109 , and averaging within each bin . some of the analysis was carried on inherent - state configurations . the method used to minimize the energy was a combination of `` molecular dynamics minimization''@xcite and conjugate gradient.@xcite this combination has been found to reduce the possibility of finding the `` wrong '' minimum , which can happen if conjugate gradient alone is used.@xcite configurations were minimized every 10 , 100 or 1000 configurations more frequent minimization enables resolution of shorter time scales , but longer time scales can not be accurately probed . we start by applying the transformation described above to the simulation data . fig . [ shearcorrmem2d0.34 ] shows the 2d , t=0.34 shear - stress autocorrelation function for both true and inherent dynamics . the agreement between the two at long times is clear , although there is more noise in the inherent case due to the cost of minimization , not as long times can be simulated . the right panel of fig . [ shearcorrmem2d0.34 ] shows the result of the discrete memory transform . recall that this returns the exact memory function @xmath19 for the data . this is positive at @xmath18 , but negative for all @xmath46 ( note that @xmath48 really means @xmath110 for an integer @xmath73 ; since minimization was carried out every ten time steps , @xmath111 ) . as shown in the inset , the data suggest an inverse power - law for the negative tail , although the noise is too great to justify this for @xmath112 , a very short time compared to the relaxation time @xmath113 . nevertheless let us attempt a fit to a simple power - law in the hope that this is in fact the correct form of @xmath19 for @xmath46 . the fit yields the expression @xmath114 for @xmath19 . testing whether this expression is in fact a good representation of the true memory function is simple : we simply use eq . ( [ memoryfunctiondiscrete ] ) to construct the correlation function corresponding to the fitted @xmath19 . the value of @xmath65 is not fit ; rather the exact value from the inversion is used . the result , shown as the dashed line in the left panel of fig . [ shearcorrmem2d0.34 ] , is a disaster : after reasonable agreement up to @xmath115 , the transformed @xmath3 stops decaying and in fact begins to increase and eventually diverges . the problem can actually be foreseen by considering the time integral of @xmath19 , which in the discrete case is the sum @xmath116 . the value of @xmath117 here is 0.0286 , so that its contribution to the sum is 0.286 . the sum over the power - law part requires the evaluation of the riemann zeta function @xmath118 at @xmath119 , which is estimated numerically to be 5.97 . including the appropriate factors of @xmath52 and the prefactor yields the contribution from the power law to be -0.469 , making the whole sum negative . moreover the sum first becomes negative around where the inverted correlation function begins to increase at this point the effective decay rate has become negative , and so the correlation function grows rather than decays . clearly , direct fitting of the memory function is problematic . the fit which applies at least up to @xmath120 clearly is wrong for @xmath121 . because the noise in the exact memory function is so much larger than the true value , standard fitting procedures can not distinguish it from zero in this region . on the other hand , we can directly check that it is indeed significantly different from zero for quite long times : taking the exact memory function , but setting it to zero for times greater than @xmath122 , and inverting , we find the correlation function indicated by the dotted line in fig . [ shearcorrmem2d0.34 ] . after @xmath123 it drops below the true correlation function in a more or less exponential fashion , as expected when the memory function is only non - zero for a finite range of time . thus the apparent power - law behavior at short times must cross over to something more rapidly decaying , but still significantly different from zero . an approach other than direct fitting of the exact @xmath19 is clearly needed . since the noise in the correlation function is small compared to the function itself , comparing this to a trial function in a fitting procedure will not suffer the same problem . this suggests a practical method for determining the memory function : we choose a functional form which is quite flexible , containing many parameters . for any given set of parameter values we can generate the corresponding trial correlation function which can be compared in a least - squares sense to the actual correlation function . then we vary the parameters to optimize the match . the value of @xmath65 is fixed from the start to @xmath124 . given that the exact @xmath19 seems to have an initial inverse power - law decay for @xmath46 , we start with such a function . to provide flexibility we multiply by a piecewise linear function ; such a function will have discontinuous changes in slope but is simple and general and hopefully the dominant behavior has been captured by the power - law factor . thus this general fitting form @xmath125 is @xmath126 where @xmath127 is a piecewise linear function defined by a set of nodes @xmath128 and values @xmath129 for @xmath130 . the first node is at @xmath18 , i.e. , @xmath131 and @xmath132 is fixed at unity . we choose @xmath133 to provide sufficient flexibility ; a typical set of node locations @xmath128 is 0 , 1 , 2 , 4 , 5 , 6 , 8 , 10 , 20 , 30 , 40 , 50 , 60 , 80 , 100 , 200 , 500 , 750 , 1000 , 3000 , 5000 . the adjustable parameters are then the overall coefficient @xmath134 , the exponent @xmath135 , and the coefficients @xmath129 , @xmath136 . at this stage we do not seek the true functional form of @xmath19 ; rather , we wish to have a fairly accurate representation which is relatively free of noise . to obtain the optimal parameter values in @xmath125 , a conjugate - gradient procedure is used . some details of this are given in the appendix . [ fittedalltime2d_t0.34 ] shows the result of this procedure . while the discreteness due to the piecewise linear function is clearly visible , the overall form is clear . in particular , there is a clear cross - over at around @xmath137 from the power law we initially identified to a more rapidly decaying power law , with exponent @xmath1381.7 . this cross - over means that the time integral converges to a positive value and the effective decay rate never becomes negative . the observation that @xmath19 is well approximated by a single inverse power law for times greater than about 5 immediately suggests a simplification : if we choose to not resolve times shorter than this , a fit using only a single power law for the non - zero part of @xmath19 may be possible . [ singleplfit2d_t0.34 ] shows that this is indeed true . here times greater than 10 were included , corresponding to sampling at an interval @xmath139 . this time scale is also significant for another reason ( not necessarily unrelated ) : it corresponds to the end of the vibrational contribution to the stress autocorrelation function of the true dynamics ( see left panel of fig . [ shearcorrmem2d0.34 ] ) . this suggests that we can avoid the time consuming energy minimization process altogether and use only the data from the true dynamics . there is one problem , though : we can not get the @xmath18 value of the ( inherent ) autocorrelation function from the true dynamics . to get around this we leave it as a fitting parameter ; this gives us a procedure for fitting autocorrelation functions with three parameters : @xmath140 , @xmath134 and @xmath135 . note that @xmath141 is directly related to @xmath65 , so one could equivalently think of @xmath65 as the parameter . this fitting function involving a single inverse power law has been used in most of the analysis . it is convenient to represent that amplitude of the power - law in terms of a `` tail - weight fraction '' @xmath142 , defined such that @xmath143 corresponds to the limiting case where the sum over the whole negative ( @xmath144 ) part of @xmath68 exactly cancels the positive contribution from @xmath65 . thus we represent @xmath68 for @xmath144 as @xmath145 where @xmath146 is the riemann zeta - function . then the relaxation time is given simply by @xmath147 this form makes explicit a few things : for near zero @xmath142 , the relaxation rate is essentially the inverse of @xmath65 . as @xmath142 increases the negative tail `` cancels out '' part of the latter , increasing the relaxation time , and the limit @xmath148 gives a diverging relaxation time . examples of fits to the three functions , @xmath105 and the shear stress and pressure autocorrelation functions , are shown in the main part of fig . [ demonstratefitsisfshearpress ] for one temperature in the 3d system . here and in the all the analysis presented below , the discretization interval was @xmath52=20 . as mentioned in section [ relaxation ] , experiments@xcite suggest that autocorrelation functions switch over to exponential decay at very long times . this is incompatible with the fitting function we have identified as discussed in section [ relaxation ] , power - law decay of @xmath19 corresponds to a power - law singularity in @xmath149 as @xmath34 , implying non - analyticity of @xmath23 in the same limit , which excludes exponential behavior of @xmath19 at long times . thus we should in principle at least allow the possibility of including an exponential cut - off in the fitting function for @xmath19 . the question is whether the available data requires it to get a satisfactory fit . only in the case of the intermediate scattering function is the noise sufficiently low to allow investigation of the long time behavior ( when the correlation function has decayed to less than 1% of its original value ) . [ compareisf_3d_0p46withk ] compares the three - parameter fit using eq . [ power_law_km ] with a 4-parameter fit involving an exponential cut - off : @xmath150 where @xmath151 is the characteristic time of the exponential cut - off . the memory functions are shown in the main part of the figure , while the correlation functions are shown in the inset . for comparison , a stretched - exponential fit is also included . the three - parameter ( pure power - law ) memory - function fit noticeably underestimates the decay rate at long times . including the cut - off gives a much better fit , although some improvement is of course expected due to the extra parameter ) . notice that the stretched exponential function fits also better than the pure power - law fit at long times , as well as the cut - off memory function . for the shear stress and pressure autocorrelation functions the long time data is not good enough to warrant including the exponential cut - off as a fourth parameter , although we shall see that it can make sense to do so when there is reason to fix the exponent @xmath135 to a particular value . in the following , `` four - parameter '' and `` three - parameter '' fits refer respectively to whether the exponential cut - off was included or not , regardless of whether one or more of those parameters was constrained in the fitting process . we start by presenting the relaxation times @xmath10 for the different correlation functions . as stated earlier , we define @xmath10 as the integral of the normalized correlation function . with simulation data there is a contribution at short times due to vibrations . the relative height of this varies among the different correlation functions , making the choice of normalization problematic . in principle this can be removed by considering correlation functions of inherent quantities , but the time taken for quenches means that determining the correlation function accurately at long times is difficult . we avoid these problems by using the fitted correlation functions , normalizing by the zero - time value that emerges from the fitting process . to be consistent with the discrete - time formalism for dealing with the memory function we sum the correlation functions and multiply by @xmath52 , rather than integrating them . this makes a positive difference of order @xmath152 , negligible except when the relaxation time is of order @xmath52 . this is the case at the highest temperatures , particularly for shear stress ( note the upwards bend in the relaxation time curve at the extreme left of the plot ) . the value of @xmath10 determined from the fit varies only slightly according to whether the 3- or 4- parameter fitting function for @xmath19 is used . [ showallrelaxationtimes ] shows arrhenius plots for both 2d and 3d systems of the three correlation functions investigated . some curvature corresponding to so - called `` fragility '' of the viscous liquid is evident , although less so in the 2d data . in both 2d and 3d the shear relaxation time is noticeably shorter than those of pressure and structural relaxation ( @xmath105 ) . the pressure relaxation time tracks closely the structural one in 3d , but exceeds it noticeably in 2d . in fact the difference between pressure and shear stress relaxation in 2d is of order a factor of ten ; in 3d it is closer to a factor of four . note that at higher temperatures ( 0.50.6 ) the relaxation time is of the same order as the discretization interval @xmath153 , so a large part of the relaxation actually takes place within the first interval . unsurprisingly , this limits the ability of the fitting procedure to accurately determine the zero - time value which leads to errors in normalization and hence in @xmath10 , particularly for the collective correlation functions of pressure and shear stress where the noise is high . in these cases a fit to a stretched exponential was made first , which was then used for the memory function fit . we note finally that is apparently nothing special happening around the mode - coupling temperature @xmath154 previously identified for the 3d kob - andersen system ( albeit with a slightly different cut - off in the potential ) . we now investigate the parameters determined by the fitting procedure starting with the temperature dependence of the parameter @xmath65 , related to the short - time relaxation rate . more precisely , we consider @xmath156 , which is an effective time scale quantifying the amount of relaxation that occurs over the interval @xmath52 , a `` short - time relaxation time '' . this is plotted figure [ isf_timeparameters ] along with @xmath10 for @xmath105 . results of three different fitting schemes are shown : three - parameter fit , four - parameter fit ( i.e. , including the exponential cut - off ) and a three - parameter fit where the exponent @xmath135 is constrained to be equal for the three correlation functions at a given temperature . the motivation for the latter will become clear later on . the variation between the different fits gives an estimation of the error bars on this quantity . @xmath155 is less than @xmath10 , which is necessarily the case if @xmath157 for @xmath144 . the main point is that @xmath10 is more or less equal to the @xmath155 at high temperatures something necessarily true for exponential relaxation but increases relative to it as temperature decreases . the temperature dependence of the @xmath155 is activated , which is to say it is at least arrhenius . the data in the figure are not precise enough to determine whether the rate is super - arrhenius , but certainly the effective barrier ( the slope in the figure ) is lower than for @xmath10 itself . figure [ indivmultik0tdep ] shows comparison of @xmath155 for the three different quantities . the upper panels represent independent fits , while the lower ones represent constrained fits where @xmath135 is the same for the three quantities . figure [ showallalpha ] shows the temperature dependence of the power - law exponent @xmath135 for the two different systems , the three different correlation functions , and for different fitting schemes . considering first the top two panels , the interesting feature is that the exponent values seem to converge as the temperature is lowered , towards @xmath1381.6 in 2d and @xmath1381.5 in 3d . at higher temperatures there is significant scatter , and , for the 3d case at least , generally higher values up to around 2 . the apparent convergence at low temperatures suggests that perhaps the true values of the exponents are in fact equal for the three functions at each temperature ; the scatter in the unconstrained fits would then be understood to be due to noise in the data . we therefore attempted a fit where @xmath135 is constrained to be the same for all three correlation functions . in most cases the quality of the fit is indistinguishable by eye from the unconstrained fit ( not shown ) . the exceptions are some fits to @xmath105 in 2d where the constrained fit tends to reduce the value of @xmath135 compared to the unconstrained fit , which results in a somewhat slower decay at long times ( this could be probably compensated for by including the exponential cut - off , as explained in the following paragraph ; we have not done this ) . the temperature dependence of the constrained-@xmath135 is more or less similar to the constrained case for the 3d data , while noticeably reduced in the 2d case . the effect on the constrained fit on the values of @xmath155 can be seen in the lower panels of fig . [ indivmultik0tdep ] . there is a tendency towards smoother temperature dependence , which supports the idea that a constrained fit can yield more accurate results because the tendency to fit noise in any one of the curves is limited by the constraint . on the other hand the temperature dependence of @xmath155 for the intermediate scattering function is actually less smooth in the constrained fit it has been `` infected '' by the greater noise in the collective functions . in principle this could be compensated for assigning a greater weight to the intermediate scattering function due to the smaller noise . this has not been attempted ; it would require an unbiased estimate of the errors on the correlation functions . also shown in the lower panels of fig . [ showallalpha ] are values of @xmath135 for unconstrained 4-parameter fits to @xmath105 . what is noteworthy here is that while values at higher temperatures still show appreciable scatter , there seems to be faster convergence at low temperatures , to values close to 1.55 and 1.50 in 2d and 3d respectively ( see in particular the four lowest temperature points in the 3d case ) . this suggests the interesting possibility that the exponent could be in fact be independent of temperature as well as of which function is considered , and independent of which of the two systems . the apparently higher values at higher temperatures could be due to not including the exponential cut - off , which , like a higher value of @xmath135 , would induce faster relaxation than a pure @xmath158 power law . an attempt to find a temperature - independent @xmath135 will be described below ; first we consider the results for the parameter @xmath142 . figure [ showallf ] shows the temperature dependence of the parameter @xmath142 , the weight of the negative tail in the memory function relative to @xmath65 ( with no cut - off included ) . as @xmath43 decreases , @xmath142 increases from a low value , approaching a value close to unity at the lowest temperatures . this increase accounts for that of the overall relaxation times @xmath10 though the factor @xmath159 . the value unity can not be reached while remaining in equilibrium because this implies a diverging relaxation time unless the power - law behavior is cut off at the longest times . thus there are two possibilities for the low - temperature behavior of @xmath142 . the first is that it bends over and never reaches unity at finite temperature . in this case we can avoid including a cut - off , but the exact temperature dependence of @xmath142 determines that of @xmath10 , given that @xmath135 seems to become temperature - independent at such temperatures . the second possibility is that @xmath142 becomes unity at a finite temperature , for example @xmath1600.3 in 2d or @xmath1600.4 in 3d . if this happens a cut - off must be included to keep @xmath10 finite . when considering the full temperature range there does seem to be a bend over towards smaller slopes at lower temperatures , but the data in fig . [ showallf ] are not clean enough to draw a firm conclusion about the limiting @xmath43-dependence of @xmath142 . ) , as this would imply arrhenius dependence of @xmath10 ] it is interesting to compare with a more common characterization of the shape , the stretching parameter @xmath15 in the kww function . as explained in sec . [ fequalonelimit ] , this may be compared to @xmath162 , shown in fig . [ comparebetakww_2minusalpha ] . the values of @xmath15 decrease as @xmath43 decreases , while @xmath162 increases . in a broad sense it seems that all are converging to a value around 0.5 , but the degree of convergence for @xmath15 is much weaker . considering the 3d data in particular , the values of @xmath162 are within the range 0.470.52 , while those of @xmath15 are spread over the range 0.55 - 0.7 . the same trend can be seen , although less clearly , in the 2d data . it is possible that the @xmath15 values will converge at longer time scales , but the fact that @xmath135 ( or @xmath162 ) seems to converge more rapidly to a common value suggests that this representation may be more physically meaningful . the value @xmath1630.5 , corresponding to @xmath1641.5 , is interesting , because it has been argued theoretically and experimentally that this value is generic for structural `` alpha '' relaxation in viscous liquids , see refs . and references therein . given the apparent convergence to a common value of @xmath135 , and the hint in the 4-parameter fits to the isf in 3d ( lower - right panel of fig . [ showallalpha ] ) that this parameter may in fact be temperature dependent , we have tried to fit the high quality isf data with a common value of @xmath135 over the whole temperature range , with the exponential cut - off included . the results are shown in fig . [ isftimescales_fixed_alpha ] . the common values of @xmath135 that emerge from the fit are 1.58 for the 2d system and 1.50 for the 3d system . the quality of the `` temperature - constrained '' fits is is virtually indistinguishable by eye from that of the unconstrained 4-parameter fits ( not shown ) . the latter , are of course , numerically superior ( since they are unconstrained ) , but this seems to be mostly due to better matching of the noise in the tail . this result strengthens the case for @xmath135 having a true value of 1.5 in 3d , independent of temperature . the other interesting feature of this constrained-@xmath135 fit is that the temperature dependence of the time - scale @xmath151 of the exponential is much smoother , apparently tracking that of the relaxation time . what about the shear and pressure data ? the data was of too low quality for an unconstrained four - parameter fit to work , but by fixing @xmath135 to be equal to 1.5 we can include the exponential cut - off , making it effectively a three - parameter fit . the goodness of fit is consistently better than the original three - parameter fit ( not shown ) , and now much smoother @xmath155 values , as well as @xmath151 values are available for shear and pressure . these are shown in fig . [ alltimeparamsalpha1.5 ] . this works better for the 3d data than for the 2d data , which we suggest is due to the larger difference between @xmath10 and @xmath151 in the latter case ; the exponential cut - off has a significant effect only at longer times where the signal to noise ratio in the correlation function is smaller , and the fit thus gives erratic values . in 3d the ratio @xmath165 is in the range 23 ; in 2d the range is 89 . whether the cut - off time actually determines @xmath10 , or vice versa , or whether they determine each other in a self - consistent way , can not be answered without a theoretical understanding and/or an explicit model for the relaxation . nor do we know what may be the cause of the different ratios in 2d and 3d . to get a clearer picture more systems should be analyzed , and the possibility that @xmath151 is size dependent can not be excluded . a feature of our fitting procedure is that we leave the zero - time value of the correlation function as a fitting parameter , since the actual zero - time value of the true correlation function includes contributions from vibrational motion which rapidly decays . it is worth asking whether the value returned by the fit coincides with the value obtained by a careful determination of the inherent correlation function . since we focus on time intervals of order @xmath166 , we can run relatively long simulations , and obtain good statistics for inherent quantities , by minimizing at similar intervals . for example minimizing once every 1000 time steps ( 100 times less frequently than before ) corresponds to an interval @xmath167 , and means that the simulation time is not dominated as much by the cost of minimizing . this was done in both the 2d and 3d systems for two temperatures : 0.34 and 0.40 for 2d and 0.44 and 0.52 for 3d . in addition runs with minimization every 100 time steps were carried out for 2d , @xmath168 . .[shorttimedecorrelation]value of correlation functions ( normalized to equal unity at time zero ) at time @xmath153 ; comparison between actual inherent value ( `` i '' ) determined from the autocorrelation function of inherent quantities and values inferred from fitting the correlation function at longer times . values for three fits are shown : ( `` 1.5 '' ) the fit with @xmath135 fixed at be 1.5 , including the exponential cut - off ; ( `` @xmath169 '' ) the 3-parameter fit with no cut - off ; and ( `` se '' ) the stretched exponential fit . the correlation at @xmath52 implied by the fits is higher than the actual correlation ; alternatively put , the amount of decorrelation that takes place on the time scale @xmath52 is less than would be expected from the fits . [ cols="^,^,^,^,^,^,^",options="header " , ] table [ shorttimedecorrelation ] shows values of the normalized correlation function after one time interval @xmath153 , for two selected temperatures in each system , determined directly from the inherent state dynamics and inferred from different fits to the data for @xmath170 . although there are some exceptions ( particularly for 2d , @xmath0 ) , the trend is that the value of the normalized inherent correlation implied by the fits is higher than the actual value . for example for 2d at @xmath168 , considering the intermediate scattering function , the actual correlation is 0.86 , while the fits all imply a value between 0.930.94 . another way of putting this is that the amount of _ decorrelation _ that takes place between time zero and time @xmath52 is greater than is implied by the fits of the correlation at longer times ( note that @xmath171 ) . this can be thought of as somewhat analogous to the extra relaxation associated with the vibrational motion which has been removed , but this is part of the inherent relaxation . schrder et al . also noticed a short - time component of the inherent structural relaxation ( intermediate scattering function ) which was not accounted for by stretched exponential fits.@xcite this was particularly noticeable at high temperatures and appeared to disappear in the activated regime . the comparison above suggests that this extra relaxation does not vanish , even at quite low temperatures . this discrepancy , between the very short - time relaxation and that for later times ( @xmath172 ) , presumably corresponds to the change of apparent power - law exponent in the memory function around this time scale ( section [ matchingcorr ] ) . one could speculate that the existence of this extra component of the inherent relaxation could be associated with very small energy barriers , much smaller than the temperature . such barriers would separate distinct inherent states , and thus `` transitions '' would show up in the inherent correlation functions , but these transitions would not be activated . the author has determined that the energy barriers in these systems are exponentially distributed , thus there are always some small barriers.@xcite for simplicity , in studying the intermediate scattering above , we have restricted the analysis to a particular wavenumber , @xmath106 , and to large particles . given that a common value of the exponent @xmath135 seems to be valid for the three correlation functions we have focused on so far , it makes sense to ask whether this applies for the intermediate scattering function at other @xmath173-values . in particular , what happens at low @xmath173 , where we know that the relaxation becomes more exponential ? [ qdepfits3dt0.48 ] shows data and fits for 3d at @xmath174 . here 4-parameter fits with fixed @xmath158 were used . figure [ qdep3dt0.48timesandf ] shows the @xmath173-dependence of the relaxation time @xmath10 , inverse of the short time rate , @xmath155 and the cut - off time of the memory function @xmath155 . the inset shows the @xmath173-dependence of @xmath142 . while the power - law exponent is fixed at 1.5 for all @xmath173-values , @xmath142 varies strongly , approaching zero at small @xmath173 . this corresponds to relaxation that becomes exponential in this limit . consistent with this , @xmath155 and @xmath10 approach each other . @xmath151 seems to follow @xmath10 . in normal diffusion one expects the relaxation time to be proportional to @xmath1 , indicated by the dashed line . the relaxation time seems to approach this dependence in the low-@xmath173 limit ( as is expected , given that diffusion should be normal at long length and time scales ) . more interesting is that @xmath155 seems to show an inverse square dependence over most of its range , corresponding to normal - diffusive behavior being observed at _ short _ times . this will be discussed below . what does a negative inverse power - law memory function mean ? one interpretation derives from the formalism of the generalized langevin equation ( gle ) , which offers a framework for interpreting memory functions . the gle for a dynamical variable @xmath6 is a stochastic differential equation @xmath175 the first term on the right hand side involves an effective friction , non - local in time , where the memory kernel @xmath176 connects the friction at current time @xmath48 with the value of @xmath134 at a ( previous ) time @xmath177 . the last term @xmath178 is the so - called random force , or noise term . according to the second fluctuation - dissipation theorem , the memory kernel is the autocorrelation function of the noise , or @xmath179 it may also be shown that this memory kernel @xmath19 is the memory function associated with the autocorrelation of @xmath6the same memory function used in the analysis of this paper . therefore one can interpret the obtained memory functions as the autocorrelation functions of the random force of the corresponding gle . by `` force '' is meant the stochastic increment to the quantity @xmath134 ; in the original version of the langevin equation @xmath134 was the velocity of a particle , whose change was indeed a force ( divided by mass ) . negative @xmath19 for @xmath46 implies that the changes , which are presumably related to the so - called `` flow - events '' which define the long - time dynamics , are anti - correlated . the simplest interpretation of this would be forward - backward correlation of events that is , an event is more likely to be the reverse of the previous one than to be something else . the parameter @xmath142 is then a direct measure of this anti - correlation . as it approaches unity , this tendency becomes so strong that the relaxation time diverges ( see the next subsection ) . an alternative interpretation involves uncorrelated events but with a non - exponential waiting - time distribution . this can be understood within the continuous - time random walk formalism of montroll and weiss.@xcite the montroll - weiss equation relates the laplace transform of @xmath108 to the laplace and fourier transforms of the waiting - time and jump - size distributions , @xmath180 and @xmath181 , respectively . it is straightforward to use it to derive a relation between the waiting - time distribution and the memory function . skipping the details , this yields @xmath182 this relation can be considered a formal equivalence of the two pictures anti - correlated random forces and non - exponential waiting - time distribution . the question of which picture is _ physically _ more relevant can only be determined by more detailed analysis of the dynamics in simulations . the existence of forward - backward correlations has indeed been noted by heuer and coworkers ( see ref . for a review of this work ) , but they argue that these are trivial , operating only on relatively short time scales . they have proposed a procedure for coarse - graining the dynamics by grouping the basins of attraction of inherent states together to form `` meta - basins '' ( mbs ) . the definition of an mb is based on the requirement that once the system has escaped from an mb , it is unlikely to return there . they have shown that the dynamics on long time scales is essentially a random walk among mbs , but with non - trivial waiting - time distribution.@xcite if we accept this interpretation , the memory function analysis presented here could be used as a way to find the waiting - time distribution from the correlation function . for example , in the @xmath183 limit with no cut - off ( see next section ) one finds that the waiting time distribution has a power - law tail with the exponent @xmath135 as the memory function itself . on the other hand , is it possible that the equivalence between the waiting time distribution point of view and the forward - backward point of view is in fact more than formal ? the non - exponential distribution of waiting times could seen as a consequence of forward - backward correlations . the size of an mb is not a fixed property of the landscape,@xcite instead being defined in terms of dynamical behavior : a given mb can increase in size as temperature decreases , reflecting the increased tendency for returns ( the system needs to get further away on average before its chance of escaping for good exceeds 50% ) . correspondingly , the `` anti - correlation tendency '' @xmath142 increases towards unity ; as long as it is less than unity , however , a diffusive regime is reached the system will eventually forget where it started . in the `` ctrw picture '' , as long as the waiting - time distribution has a finite mean , upon sufficient coarse - graining it will become exponential and therefore the dynamics will be diffusive in the normal sense . the point of heuer and co - workers is that there is a coarse - graining time scale at which events are uncorrelated but the waiting - time distribution is non - trivial ( non - exponential ) . it is also true in the `` gle picture '' involving anti - correlated events that , as long as @xmath184 or there is an exponential cut - off , a normal diffusive regime must also eventually be reached under coarse - graining in time . but it is not clear yet whether coarse - graining in the gle picture can produce a dynamics that corresponds to the ctrw picture . this is a matter for future investigations . in this section we discuss briefly the limit @xmath148 , in the case where there is no exponential cut - off . as mentioned , in this limit the relaxation time diverges , but the dynamics defined by this memory function are still meaningful , and have been studied in the context of anomalous diffusion , fractional brownian motion and related processes . @xcite applications include protein dynamics@xcite and dielectric response of glassy media . @xcite a noise term in a gle with this kind of correlation function is termed fractional gaussian noise ; when it acts on an unbound degree of freedom , the latter undergoes fractional brownian motion . correlation functions generally involve a function known as the mittag - leffler ( ml ) function : for a time scale @xmath185 ; so the correlation function should really be called a `` stretched mittag - leffler function '' , but it seems to be standard to refer to the correlation function also as a mittag - leffler function , which is not strictly correct . ] they behave at intermediate times like a stretched exponential , with stretching parameter @xmath186 , before switching to a power - law at long times , such that the integral is infinite . this is why it made sense to compare @xmath162 with the fitted @xmath15 in fig . [ comparebetakww_2minusalpha ] . there are two broad classes corresponding to positively or negatively correlated noise . being respectively greater than @xmath187 , which also corresponds to @xmath188 or less than @xmath187 , corresponding to @xmath189 . at @xmath190 the exponent is formally equal to unity but its coefficient vanishes thus white ( uncorrelated ) noise is recovered . ] when comparing to the literature on fractional diffusion and related processes it is important to realize that the langevin equation is typically formulated with the dynamical variable @xmath6 as the velocity of a particle . for a free particle , if the noise is fractional gaussian with negative correlation , then the velocity autocorrelation function is an ml function . its integral diverges , therefore the position dynamics is actually super - diffusive . applications more typically involve sub - diffusive motion of a harmonically bound particle.@xcite the noise acting on the velocity is positively correlated and it is the position autocorrelation function that decays as an ml function . the memory function corresponding to this limit is @xmath191 , which is clearly scale invariant : rescaling , or coarse - graining in time produces the same dynamics . this corresponds to a fixed point in the renormalization group sense . for negatively correlated noise , the fixed point is unstable ( see the chapter by qian in ref . ) if we are not quite in this limit , then rescaling will eventually takes us away from it and back to the normal diffusive regime , as discussed in the previous subsection . the results of the analysis presented in this paper could be therefore interpreted as saying that glassy dynamics approaches sub - diffusive dynamics with a particular exponent @xmath135 , which ( in 3d ) is very close to 1.5 . as the limit @xmath143 is approached the relaxation becomes close to a mittag - leffler function , resembling a stretched exponential with @xmath192 . as we have seen , even at the lowest temperatures , actually fitting with a stretched exponential yields a different exponent , so it is hard to see the convergence to 0.5 , while this is clearer in the memory function based approach . the theoretical challenge is , of course , to understand where this exponent comes from , although there are models for viscous liquid dynamics which predict it ( refs . and references within ) . does this limit ever get reached ? although @xmath142 clearly approaches unity , it has not reached it yet at the temperatures simulated so far . moreover , there is always an exponential cut - off in the memory function . thus regardless of whether @xmath142 actually reaches unity or not , it will at some point be the exponential cut - off which will determine the relaxation time . looking at fig . [ alltimeparamsalpha1.5 ] , for example , suggests that this will happen around @xmath0 in 3d . simple - minded linear or quadratic extrapolation of @xmath142 suggests this also . if we set @xmath143 for temperatures below say @xmath0 and assume that the exponential cut - off , @xmath151 is a fixed multiple of @xmath10 ( actually the ratio seems to increase ) , then one can make an extrapolation of @xmath10 , which gives a sharp crossover to an almost arrhenius form ( not shown ) . if @xmath142 approaches unity more smoothly , then the crossover will not be so abrupt , but in any case we predict that a change in the temperature dependence will occur around @xmath0 . the parameter @xmath193 is a measure of the dynamics at short times . by fixing the time scale , it measures how much activity happens in a given time , as opposed to standard measures of relaxation , which concern how much time is required for a given amount of relaxation to occur . for exponential relaxation there is no difference , for relatively small intervals ( or long relaxation time @xmath10 ) : if the interval is @xmath52 , then the amount of correlation at this time is @xmath194 which gives @xmath195 . thus a difference between these two quantities is equivalent to non - exponential relaxation . when @xmath52 is much smaller than @xmath10 , the expected number of activated events that take place in a small volume in a time interval of this length is much smaller than unity , thus we can not expect to see any correlation effects . we suggest therefore that @xmath155 is like a `` bare '' relaxation time , measuring the dynamics before correlation effects play a role . recall ( fig . [ compareshorttimedecorrelation2dt0.34 ] ) that there is a noticeable amount of ( inherent ) relaxation in this time interval which is not accounted for by the fitted @xmath155 . we assume this to be non - activated relaxation , and not relevant for what takes place at longer time scales ( in this sense it resembles the vibrational part of the relaxation ) . the evidence supporting the idea of @xmath155 as a measure of the bare , uncorrelated relaxation rate is the near - arrhenius temperature dependence ( see fig . [ alltimeparamsalpha1.5 ] ) , and the near @xmath1 wavenumber dependence ( fig . [ qdep3dt0.48timesandf ] ) the use of a fixed , relatively short time interval to extract an apparent `` underlying arrhenius behavior '' in viscous liquid dynamics was recently demonstrated by de souza and wales @xcite . they studied the diffusion constant determined by the mean squared displacement in a fixed time interval of 25 lj units . they were also able to explicitly relate the deviations from exponential relaxation to anti - correlation of events in subsequent time windows . the present results are essentially equivalent , but obtained in the more general context of studying an arbitrary autocorrelation function . we have presented a method for fitting the kind of slowly decaying autocorrelation functions typical of the dynamics of glass - forming liquids . an explicit functional form is postulated , not for the autocorrelation function , but for its associated memory function : a positive zero - time parameter @xmath65 , and a negative inverse power law for non - zero time . an long - time exponential cut - off may optionally be included . this involves another parameter , although we have shown that good fits may be obtained by fixing the exponent to a common value , reducing the number of parameters per data set essentially to three . while it is possible to obtain the exact memory function from the autocorrelation function , it is not useful to fit the former directly . rather we match the autocorrelation function by optimizing the parameters defining the memory function . using a quite general functional form for @xmath19 we noted a crossover from one apparent power - law to another at relatively short times , which allowed a simplification by considering data separated at times longer than the cross - over , namely the simple power - law form ( with possible exponential cut - off ) . in particular we chose @xmath153 in lennard - jones units for all of the analysis . while the fits are about as good as fits with a stretched exponential , the parameters obtained have arguably greater physical significance than those of the latter . in particular the exponent @xmath135 seems to be more or less independent of the relaxing quantity , and indeed of temperature ( as long as the exponential cut - off is included ) , which is not the case for the stretching parameter @xmath15 . moreover the parameter @xmath65 may be interpreted as a short time rate ; its temperature dependence is , interesting , significantly weaker than that of the alpha relaxation times for the different autocorrelation functions . as temperature decreases , the amplitude of the power tends to approach a limiting value associated with the mathematical description of anomalous diffusion / relaxation characterized mathematically by the mittag - leffler function . we hypothesize that this value is reached at finite temperature but the associated divergence of relaxation time is removed by the presence of an exponential cut - off . including the latter in the fits for the self - intermediate scattering function allowed the identification of a third time scale , longer than the alpha time . it is hypothesized that the strong non - arrhenius temperature dependence is due to a crossover from the time associated with the short - time rate to that associated with the long - time cut - off of the memory function , and that in particular at temperatures somewhat lower than those simulated , the non - arrhenius behavior will weaken noticeably . the centre for viscous liquid dynamics `` glass and time '' is sponsored by the danish national research foundation ( dnrf ) . here we describe in a little more detail how we optimize the fit . we start by describing the procedure using the general fitting form which includes the piecewise linear function @xmath127 . suppose first we have a correlation function based on inherent dynamics , so that there is no vibrational decay at short times , and in particular the @xmath18 value is known . then we may normalize by the latter to get the normalized autocorrelation function @xmath196 where @xmath197 @xmath73 is the integer corresponding to @xmath48 . we write the fitting form as @xmath198 , where @xmath199 are the parameters of the fitting form . the node values @xmath129 are not used directly as parameters ; rather their logarithms are . this constrains them to be positive and minimizes numerical problems associated with the broad range of values that they turn out to have . for given values of the @xmath200 , we can easily compute the associated correlation function @xmath201 and then the objective function for the main data analysis we switched to using simpler functional forms without the piecewise linear function . also significant is that we used only the true dynamical correlation functions ( as opposed to those of the inherent dynamics ) which are more accurate and smooth ; this removed the need to do expensive minimizations . by considering times at intervals of order 20 , we could avoid the vibrational decay , except that now the correct initial value of correlation function had to be treated as an unknown parameter . since this affects the normalization , it is important to handle this carefully . suppose we have a non - normalized correlation function at discrete times , not including time zero : @xmath204 . since we are going to normalize anyway , we choose as the parameter not @xmath205 , but the ratio @xmath206 . and since this is required be greater than unity , it is represented internally ( i.e. , within the fitting algorithm ) in terms of a logarithm @xmath207 : @xmath208 , where @xmath207 is the actual parameter varied by the algorithm . note that @xmath141 is directly related to @xmath65 : so that one may equivalently consider @xmath65 as the fitting parameter . an important technicality must be mentioned here . if the objective function is defined as above , there is a problem due to the fact that the normalization factor is now an adjustable parameter . this means that the objective function could be reduced by making @xmath141 , and hence @xmath205 , larger . this was found to be a problem for data sets from higher temperatures where the relaxation times were relatively short and therefore the data was not sufficient to constrain @xmath141 to reasonable values . the problem was solved by normalizing by the fixed quantity @xmath210 instead ( but only for the purposes of defining the objective function ) . for @xmath57 ( times greater than @xmath153 ) we used in some cases the single power law of eq . [ power_law_km ] . the tail - weight fraction @xmath142 is constrained to lie between 0 and 1 by expressing it internally as the @xmath211 where @xmath212 is unconstrained . the exponent @xmath135 was represented as itself because it tended to vary within a relatively small range ( 1.52.2 ) , with no risk of assuming values which would cause numerical problems . in other cases an exponential cut - off factor was included , eq . [ exp_power_lawkm ] , which involves the timescale @xmath213 . because this can vary over several orders of magnitude , to aid convergence it was expressed internally in terms of its logarithm , @xmath214 , where the factor 10 was found to improve convergence ( presumably by making the typical values and range of @xmath215 comparable to those of other parameters ) .

. cartan proved that conformally flat hypersurfaces in for have at most two distinct principal curvatures and locally envelop a one - parameter family of-spheres . we prove that the gauss - codazzi equation for conformally flat hypersurfaces in is a soliton equation , and use a dressing action from soliton theory to construct geometric ribaucour transforms of these hypersurfaces . we describe the moduli of these hypersurfaces in and their loop group symmetries . we also generalise these results to conformally flat-immersions in-spheres with flat and non - degenerate normal bundle . research supported in part by nsf advance grant ] research supported in part by nsf grant dms-0707132 ]

0803.2754

an immersion @xmath6 is _ conformally flat _ if there exists a flat metric in the conformal class of the induced metric @xmath7 : that is there exists a smooth function @xmath8 such that @xmath9 is flat . this condition is equivalent to the weyl tensor of @xmath10 being zero when @xmath11 , and to the schouten tensor @xmath12 being a codazzi tensor when @xmath13 ( i.e. , @xmath14 is a symmetric @xmath15-tensor ) . the history of conformally flat immersions is long , the search for conformally flat submanifolds being a natural task in conformal geometry . the study of conformally flat _ hypersurfaces _ in @xmath0 dates back to cartan @xcite who , demonstrated that the only such hypersurfaces for @xmath11 are the channel hypersurfaces : envelopes of a 1-parameter family of @xmath2-spheres . in particular these have ( at most ) two distinct principal curvatures . for @xmath16 the problem is uninteresting as every surface is conformally flat . for @xmath13 , however , there are more varied conformally flat hypersurfaces : not only are there the channel examples , there also exist hypersurfaces with 3 distinct curvatures . these were first discussed by hertrich jeromin @xcite , who moreover described the link between conformally flat hypersurfaces in @xmath3 , curved flats in the space of circles in @xmath3 , triply orthogonal systems , and guichard nets . the classification of these hypersurfaces , however , remained unknown . given a conformally flat immersion @xmath17 with flat normal bundle , we embed @xmath18 naturally in the _ light - cone _ @xmath19 of isotropic vectors in a lorentzian @xmath20 , and construct a _ flat lift _ @xmath21 : this @xmath22 is immersed , has flat induced metric , and flat normal bundle . since both are flat , the tangent and normal bundle decomposition of the trivial @xmath23-bundle is a _ curved flat _ @xcite in the pseudo - riemannian symmetric space @xmath24 ( the grassmannian of space - like @xmath4-planes in @xmath20 ) . we are thus immediately in the realm of integrable systems . curved flats in @xmath25 with a _ good _ co - ordinate system give rise to terng s @xmath25-system ( @xcite ) , which is constructed as follows : suppose that @xmath26 is the involution of the lie group @xmath27 defined by @xmath28 , where @xmath29 and @xmath30 is the @xmath31 identity matrix . then @xmath32 is the fixed point set of @xmath26 in @xmath27 . let @xmath33 denote the @xmath34-eigenspace decomposition of @xmath35 . @xmath32 then acts on @xmath36 by conjugation . let @xmath37 be a maximal abelian subalgebra in @xmath36 , and @xmath38 a basis of @xmath37 . _ the @xmath25-system defined by @xmath37 _ is the following system for @xmath39 : @xmath40-[a_j,\xi_{x_i}]-[[a_i,\xi],[a_j,\xi]]=0,\quad i\neq j,\ ] ] where @xmath41 is the orthogonal complement of @xmath37 with respect to the killing form . unlike in the riemannian symmetric case , not all maximal abelian subalgebras in @xmath36 are conjugate under @xmath32 ; there are both semi - simple and non - semisimple such subalgebras . we note that two maximal abelian subalgebras in @xmath36 conjugate under @xmath32 give rise to equivalent @xmath25-systems , while two non conjugate maximal abelian subalgebras in @xmath36 give non - equivalent @xmath25 systems . the normal bundle of an immersion is termed _ non - degenerate _ if the dimension of the space of shape operators at each point is equal to the co - dimension . an immersion has _ uniform multiplicity one _ if it has flat normal bundle and distinct curvature normals ( equivalently all curvature distributions have rank one ) . it follows from the definition that an @xmath4-dimensional submanifold in @xmath23 with flat and non - degenerate normal bundle has uniform multiplicity one . we prove that a conformally flat @xmath4-immersion into @xmath18 with uniform multiplicity one gives rise to a flat @xmath4-immersion in the light - cone @xmath19 with flat non - degenerate normal bundle , and that the converse is also true . to study conformally flat @xmath4-immersions in @xmath18 with uniform multiplicity one is thus the same as to study flat @xmath4-immersions in @xmath19 with flat non - degenerate normal bundle . we show that there exist line of curvature co - ordinate systems for these flat immersions , and that their gauss - codazzi equations amount to the @xmath24-system defined by a semi - simple maximal abelian subalgebra @xmath37 . conversely , given a solution to the @xmath25-system defined by @xmath37 , and a null vector @xmath42 we obtain a a conformally flat immersion in @xmath18 with uniform multiplicity one . motivated by definitions in classical differential geometry , we call a diffeomorphism @xmath43 between @xmath4-immersions in space forms with flat normal bundle a _ combescure transform _ if @xmath44 maps principal directions of @xmath45 to those of @xmath46 and they are parallel . a combescure transform @xmath44 is _ christoffel _ if it is orientation reversing . given a solution to the @xmath25-system defined by @xmath37 and a null vector @xmath47 with @xmath48 , we construct a flat @xmath4-immersion @xmath49 in @xmath19 with flat normal and non - degenerate bundle . hence @xmath49 projects to a conformally flat immersion in @xmath18 with uniform multiplicity one . moreover , if @xmath50 and @xmath51 are null vectors with euclidean length @xmath52 , then @xmath53 is a combescure transform . because of the correspondence between solutions of the @xmath25-system and conformally flat @xmath4 immersions in @xmath18 with flat and non - degenerate normal bundle , all the machinery of soliton theory applies : loop - group dressing of solutions to obtain new conformal flats or simply dressing vacuum solutions to obtain more complex explicit conformally flat immersions ; existence results such as cartan - khler and inverse scattering , etc . in particular : 1 . we may dress solutions by special , simple elements , whose action may be calculated explicitly by residues . the action of such elements is seen to be by _ ribaucour transforms _ on conformal flats : corresponding immersions envelop ( have first - order contact with ) a congruence of @xmath4-spheres in such a way that principal curvature directions on the envelopes correspond under the congruence . local analytic conformally flat @xmath4-immersions in @xmath18 are determined by @xmath54 functions of one variable . the cauchy problem for the @xmath25-system with rapidly decaying initial data on a regular line can be solved globally . is _ regular _ if @xmath55 is a injective . ] although the resulting @xmath4 dimensional submanifolds may have cusp singularities , the frame is globally defined and smooth . the moduli space of such immersions has a loop group symmetry . if the normal bundle is degenerate and the curvature distributions @xmath56 ( common eigenspaces of the shape operators ) have constant ranks , then we show that all but one of the @xmath56s have rank one . such submanifolds are thus envelopes of @xmath57-parameter families of @xmath58-spheres . if , in addition , these immersions are assumed to have line of curvature co - ordinates , then the gauss - codazzi equations are the @xmath25-system defined by a non - semisimple maximal abelian subalgebra @xmath37 in @xmath36 . conversely , given a solution of the @xmath25-system defined by @xmath37 , we obtain an @xmath59-parameter family of flat lifts , each of which gives rise to a conformally flat immersion in @xmath18 with flat normal bundle , but not with uniform multiplicity one . when @xmath13 , these give _ channel immersions_. loop group dressing still works , and we can construct channel immersions from any germ of an @xmath60-valued holomorphic map at @xmath61 that satisfies the reality condition associated to @xmath25 . most of the results for conformally flat @xmath4-immersions in @xmath18 with uniform multiplicity one hold for conformally flat @xmath4-immersions in @xmath62 with flat normal bundle , @xmath4 curvature normals such that the orthogonal complement of the subbundle spanned by @xmath4 curvature normals is flat . there exist line of curvature co - ordinates and a correspondence between such immersions and solutions of the @xmath63-system . the paper is organised as follows . in section [ sec : flatlift ] , we generalise hertrich jeromin s work on conformally flat immersions of hypersurfaces in @xmath3 to @xmath4-submanifolds in @xmath18 : we outline the light - cone model and how conformal flat immersions in the sphere correspond to genuine flat immersions in the light - cone , then consider the curvature distributions of corresponding maps , and how their fundamental forms compare . the link between conformal flats and the @xmath25-system is detailed in section [ sec : uk ] , and we explain its generalisation to conformally flat @xmath4-immersions in @xmath62 in section [ sec : codim ] . we give a discussion of the dressing transformations of a negative loop on the space of solutions to the @xmath25-system and their associated conformally flat immersions ; certain dressing transforms are shown to give rise to geometric ribaucour transforms in section [ sec : rib ] . in the final section , we show that solutions of the @xmath25-system defined by a non - semisimple maximal abelian subalgebra give rise to the channel immersions . in this section we give definitions and explain , via the light - cone model , the correspondence between conformally flat @xmath4-dimensional immersions in @xmath18 and flat @xmath4-dimensional immersions in @xmath19 . we also show the existence of line of curvature co - ordinates for conformally flat @xmath4-immersions in @xmath18 with uniform multiplicity one . the light - cone model of the conformal @xmath64-sphere is now well - understood , its prime advantage being that it _ linearises _ conformal geometry in @xmath65 : the fundamental objects of the theory , subspheres @xmath66 and their intersections , become the geometry of the grassmannians @xmath67 of definite signature planes . hertrich - jeromin s book @xcite contains an excellent introduction to this , as does burstall s discussion of isothermic surfaces @xcite . let @xmath68 denote the diagonal @xmath69 matrix @xmath70 and @xmath71 the lorentzian bilinear form on @xmath72 . @xmath73 is the _ light - cone _ of isotropic vectors in @xmath72 . fix a choice of unit time - like vector @xmath74 . the restriction of @xmath75 to @xmath76 is positive definite , and hence @xmath76 is isometric to the euclidean @xmath77 . let @xmath65 denote the set of unit vectors in @xmath76 , then the map for the span of a collection of vectors , usually dropping the brackets when referring to the perpendicular space to the span of a vector . ] @xmath78 is clearly an isometry which , since each isotropic line @xmath79 intersects the plane @xmath80 exactly once , diffeomorphically puts a metric on the projective light - cone @xmath81 . however , any other choice of unit time - like @xmath82 gives a different diffeomorphism and induces a different metric . indeed the following compound map is seen to be a conformal diffeomorphism from one @xmath64-sphere to another : @xmath83 & & \ip{x+t_0 } \ar@{|->}[rr ] & & \mathrlap{-\frac{x+t_0}{(x+t_0,t'_0)}-t'_0}\\ \mathllap{\d x^2 } \ar@{|->}[rrrr ] & & & & \mathrlap{\frac{1}{(x+t_0,t'_0)^2}\d x^2 } } \text{\phantom{$\frac{x+t_0}{(x+t_0,t'_0)}-t'_0$}}\ ] ] for this reason the projective light - cone is known as the _ conformal @xmath64-sphere_. all geometric properties of @xmath65 that are genuinely conformal are detectable directly in the light - cone and do not depend on the choice of @xmath74 , as can be seen by the following theorem . the action of @xmath84 on the light - cone , and thus on any choice of @xmath64-sphere @xmath85 , is said to be by _ mbius transformations_. for @xmath86 _ all _ ( even local ! ) conformal diffeomorphisms of @xmath65 are mbius . liouville s theorem allows us to treat submanifolds of the conformal sphere similarly to those in metric geometry : the existence of submanifolds _ up to isometry _ is replaced by _ up to mbius transforms _ ; in the light - cone picture these really are isometries . in particular , starting from submanifolds in @xmath87 , we need not worry about specific choices of @xmath74 , since examining the submanifold via any other choice merely amounts to a mbius transform . we will , however , tend to assume that a fixed choice has been made , if only so that we may anchor discussions in a genuine @xmath65 . given a map @xmath88 , a _ lift _ of @xmath89 is any map @xmath90 such that @xmath91 . flat lift _ is a lift such that the induced metric @xmath92 on @xmath45 is flat . to facilitate computations , we set up notations for moving frames of flat lifts @xmath93 . suppose that @xmath94 is an @xmath95 frame on @xmath45 such that @xmath96 are tangent to @xmath97 . let @xmath98 be the 1-forms dual to @xmath96 . then @xmath99 . write @xmath100 then @xmath101 and @xmath102 the shape operator is @xmath103 , the tangential component . the two fundamental forms and the normal connection are : @xmath104 we may now state the main correspondence of the paper ( due to hertrich - jeromin @xcite when @xmath13 ) . a conformally flat immersion @xmath17 with flat normal bundle has a flat lift @xmath21 with flat normal bundle . + conversely , any immersed @xmath21 with a flat , definite signature , metric @xmath92 and flat normal bundle is a flat lift of a conformally flat immersion @xmath17 with flat normal bundle . for reference we collect some important relations . if @xmath105 is flat , we will define , in the proof , a flat lift @xmath22 as below ; similarly , given a flat lift @xmath22 and a choice of @xmath74 , we recover a conformally flat @xmath89 . everything is related by the following notation : @xmath106 since @xmath107 project to the same @xmath89 , we can always assume that @xmath108 is positive . the locations of the normal bundles viewed as subbundles of the trivial bundle are as follows , where both decompositions are orthogonal : @xmath109 it is important to note that @xmath110 . + the metric @xmath92 will almost always be definite : the only time this does nt happen is if some tangent direction is parallel to @xmath22 itself ( anything else forces a contradiction by requiring either a 2-dimensional isotropic or time - like subspace of @xmath19 ) . it follows that , for us , the _ metric _ on the normal bundle @xmath111 is always non - degenerate and of signature @xmath112 . 1 . suppose that @xmath105 is flat and define @xmath113 . then @xmath114 , and thus @xmath22 is an immersion . moreover @xmath115 is flat , hence @xmath22 is a flat lift , with positive definite tangent bundle . as is well - known , the bundle @xmath116 is then flat ( the induced connection is @xmath22-related to the levi - civita of @xmath117 on @xmath45 ) . + now let @xmath118 be parallel orthonormal sections of @xmath116 and @xmath119 parallel orthonormal sections of the normal bundle @xmath120 . moreover let @xmath121 be the unique isotropic section of @xmath111 such that @xmath122 and @xmath123 . by the flatness of @xmath117 and @xmath120 , there exist 1-forms such that the moving frame equations are @xmath124 here @xmath125 . the maurer cartan equations quickly yield @xmath126 and @xmath127 , from which we conclude that @xmath128 since the @xmath129 form a base of @xmath130 it follows that the @xmath131 are closed and thus the normal bundle @xmath111 is flat . 2 . suppose that an immersed @xmath22 has flat tangent and normal bundles . then @xmath132 is easily seen to be immersed . now let @xmath133 be any space - like parallel orthonormal sections of @xmath134 . defining @xmath135 it is easy to see that the @xmath136 are a parallel orthonormal frame of @xmath120 and hence @xmath89 has flat normal bundle . we have now reduced the study of conformally flat @xmath4-immersions in @xmath18 with flat normal bundle to the study of flat @xmath4-immersions in @xmath19 with flat normal bundle . suppose that @xmath22 has flat normal bundle and that the _ curvature distributions _ of @xmath22 have constant rank : that is the tangent bundle of @xmath45 decomposes orthogonally and smoothly as @xmath137 where each @xmath138 is an eigenspace of all the shape operators @xmath139 , and the ranks of the @xmath56 are constants . we say that @xmath22 has _ _ multiplicity _ _ thus has _ uniform multiplicity one _ if @xmath140 and @xmath141 for all @xmath142 . ] @xmath143 if each @xmath144 . to each curvature distribution there corresponds a _ curvature normal _ @xmath145 such that @xmath146 ( see e.g. @xcite ) . suppose that @xmath147 is an @xmath148 tangent frame consisting of principal curvature directions and @xmath149 a parallel @xmath150 normal frame . moreover let @xmath151 for @xmath152 and @xmath153 . then @xmath154 [ lemm : fcurv ] if @xmath22 is an immersed flat lift whose principal curvatures along any parallel normal fields have constant multiplicities , then its curvature normals are mutually orthogonal . if the curvature distribution @xmath56 has rank @xmath155 , then the corresponding curvature normal @xmath156 is isotropic . since @xmath22 is flat , the gauss equation implies that @xmath157 which yields @xmath158 . distinct curvature normals are therefore orthogonal , and repeated curvature normals are necessarily isotropic . it is now an ( almost ) obvious corollary that there can only be one @xmath159 distribution . a full discussion will wait until section [ sec : channel ] . note that a flat immersion @xmath22 with flat normal bundle is non - degenerate iff all curvature distributions have rank 1 ( uniform multiplicity one ) _ and _ all curvature normals are non - zero : theorem [ thm : ffcurvs ] will show that this second condition is vacuous . the following theorem describes how immersions with non - degenerate normal bundle come equipped with line of curvature co - ordinates . [ thm : coords ] let @xmath21 be flat with flat non - degenerate normal bundle and curvature normals @xmath160 . then there exist curvature line co - ordinates @xmath161 on @xmath45 . moreover @xmath22 and @xmath117 can be written entirely in terms of the curvature normals as @xmath162 first recall from lemma [ lemm : fcurv ] that the curvature normals are non - isotropic . since @xmath163 it follows that @xmath164 since the @xmath165 are a frame of @xmath111 , there exist functions @xmath166 such that @xmath167 . however @xmath168 . putting this together quickly gives @xmath169 . assume @xmath170 . let @xmath171 be defined by @xmath172 now @xmath173 . by , and the fact that @xmath174 and @xmath175 are unit length , it is immediate that . ] @xmath176 the @xmath129 are independent , thus we have co - ordinates @xmath177 . since the @xmath129 diagonalise the fundamental forms of @xmath22 , it follows that the @xmath178 do similarly and are thus curvature line co - ordinates . the expression for @xmath117 is then immediate . since @xmath179 and @xmath180 , it follows that if @xmath136 are parallel sections of @xmath120 , then @xmath181 are a parallel frame for the euclidean normal bundle @xmath182 . @xmath89 therefore has flat normal bundle when viewed as an immersion into @xmath183 . it also therefore has an orthogonal curvature distribution ; moreover @xmath18 will inherit a curvature distribution from @xmath184 . _ a priori _ we now have _ three _ curvature distributions on @xmath45 : the next theorem tidies things up . we also have curvature normals for @xmath22 and @xmath89 which we would like to relate . [ thm : ffcurvs ] let @xmath17 be a conformally flat immersion with flat normal bundle , and @xmath185 flat lift . then 1 . the curvature directions ( on @xmath186 ) induced by @xmath22 and @xmath89 are identical . if @xmath187 are corresponding curvature normals of @xmath22 and @xmath89 ( the latter as a map into @xmath18 and @xmath184 respectively ) , and @xmath188 the corresponding curvature directions , then we have the relations @xmath189 3 . all curvature normals of @xmath22 are non - zero . 4 . @xmath22 has non - degenerate normal bundle iff @xmath89 has uniform multiplicity one . let @xmath174 be orthonormal curvature directions for @xmath22 and define @xmath190 as in the theorem . the @xmath190 are orthonormal , and we calculate that @xmath191 hence the @xmath190 frame @xmath192 . moreover @xmath193 hence the @xmath190 are curvature directions for @xmath89 . it follows that the curvature normals for @xmath89 as a map into @xmath18 are @xmath194 . we thus have ( 1 ) and ( 2 ) . for ( 3 ) and ( 4 ) we calculate explicitly : @xmath195 supposing that @xmath196 , we readily obtain @xmath197 however taking the norm squared of both sides results in the contradiction @xmath198 it follows that @xmath156 can not be zero , yielding ( 3 ) . by we have that @xmath199 if the @xmath200 are distinct , then the @xmath156 are distinct and , by lemma [ lemm : fcurv ] , orthogonal . rearranging and taking the norm squared of yields @xmath201 , hence no pair of @xmath156 are parallel isotropic . the @xmath156 are thus all non - isotropic and @xmath111 is non - degenerate . conversely , if @xmath202 , then @xmath203 , so that either @xmath204 is a repeated curvature of @xmath22 , or @xmath205 . the latter case yields a contradiction , for it would require that @xmath206 only . this establishes ( 4 ) . observe that if @xmath89 has distinct curvature normals , then @xmath120 is automatically non - degenerate : @xmath207 . indeed the proof shows that @xmath208 . + all curvature normals of @xmath22 are non - zero , hence the only way @xmath111 can be degenerate is if @xmath22 has a repeated non - zero curvature , necessarily isotropic . thus @xmath22 is non - degenerate iff it has uniform multiplicity one . by contrast , @xmath89 may have a zero curvature normal i.e. the curvature line is a geodesic whether @xmath22 is degenerate or not . the multiplicities of @xmath89 are however identical to those of any flat lift . + when @xmath22 is non - degenerate we may obtain further relations from and : @xmath209 we will say more on the curvature distributions of conformal flats with repeated curvature normals in section [ sec : channel ] . we prove that the gauss - codazzi equation for flat @xmath4-immersions in @xmath19 with flat normal bundle and `` good '' co - ordinates is the @xmath210-system . the @xmath25-system @xcite is a general construction common to any symmetric space . we give its construction for the case @xmath211 and @xmath212 : these will be viewed as matrices where @xmath213 . @xmath32 is then represented by the block @xmath31 diagonal matrices . the lie algebras of @xmath27 and @xmath32 will be written @xmath214 and @xmath215 , while the killing perp of @xmath215 will be denoted by @xmath36 . as is well - known , @xmath25 is a pseudo - riemannian symmetric space defined by the involution @xmath216 with @xmath34-eigenspaces @xmath217 moreover , @xmath218,\ [ \fp,\fp]\subset\fk,\qquad [ \fk,\fp]\subset\fp.\ ] ] the conjugation of @xmath32 on @xmath36 is @xmath219 let @xmath37 be a maximal abelian subalgebra in @xmath36 , and @xmath220 a basis of @xmath37 . the @xmath25-system defined by @xmath37 is the pde for @xmath221 . it is known and can be easily checked that the following statements are equivalent : 1 . @xmath222 is a solution of the @xmath25-system , 2 . the following one - parameter family of @xmath223 connection 1-forms is flat for all @xmath224 : @xmath225 ) \d x_i,\ ] ] 3 . @xmath226)\d x_i$ ] is flat for some @xmath227 . we call @xmath228 the _ lax pair _ of the @xmath25-system , and a solution @xmath229 to @xmath230 satisfying the _ @xmath25-reality condition _ , @xmath231 an _ extended flat frame _ for the solution @xmath222 of the @xmath25-system . observe that : 1 . we could normalise the extended flat frame at a base point , e.g. @xmath232 , which will be called the _ normalised extended flat frame_. choosing a different extended flat frame merely affects submanifold geometry by rigid motions ( equivalently mbius transforms of @xmath18 ) . 2 . if @xmath233 is an extended flat frame , then @xmath234 for all @xmath235 and @xmath236 . there are both semisimple and non - semisimple maximal abelian subalgebras in @xmath36 . these comprise @xmath4 conjugacy classes , each of which give rise to non - equivalent @xmath25-systems . we will show in this section that the @xmath25-system defined by a semisimple abelian subalgebra in @xmath36 is the gauss - codazzi equation for flat lifts of conformally flat immersions with uniform multiplicity one . in section [ sec : channel ] we will see that solutions of a @xmath25-system defined by a non - semisimple maximal abelian subalgebra give rise to conformally flat immersions with one multiplicity greater than one . [ thm : ndimconf ] suppose that @xmath237 is a flat immersion with flat non - degenerate normal bundle , induced metric @xmath238 , and that @xmath239 is a line of curvature co - ordinate system . let @xmath240 be an @xmath241-frame such that the @xmath174 are principal curvature directions and the @xmath242 are parallel to the curvature normals of @xmath22 . let @xmath243 , @xmath33 its cartan decomposition , and @xmath244 a cartan subalgebra in @xmath36 , where @xmath245 . set @xmath246 , where @xmath247 then @xmath248 is a solution of the @xmath25-system defined by @xmath37 , @xmath249 -[a_j , \xi_{x_i } ] + [ [ a_i , \xi ] , [ a_j , \xi]]=0 , \quad i\not = j,\ ] ] and @xmath250 ) \ , \d x_i,\ ] ] where @xmath251 for @xmath152 . moreover , let @xmath252 in @xmath32 be a solution of @xmath253 \d x_i$ ] , then there exists a constant null vector @xmath254 such that @xmath255 . let @xmath256 be the curvature normals for the flat lift @xmath21 . we may assume that @xmath257 . define ( as in ) @xmath258 @xmath259 is an @xmath150 normal frame . let @xmath147 be an @xmath148 tangent frame consisting of corresponding principal curvature directions . it is clear that @xmath260 where @xmath245 , @xmath261 and the lie algebra valued 1-forms @xmath262 are the connection forms for the induced flat connections on @xmath116 and @xmath111 . the flatness @xmath263 reads @xmath264 by considering where the @xmath265 terms appear in the third equation , it is not hard to see that @xmath266 and that there therefore exists a unique map @xmath267 such that @xmath268 the first two equations of constitute a system of pdes in the entries of @xmath269 . indeed writing @xmath270 and @xmath271 , we see that @xmath272=\diag(a , b)$ ] . it is straightforward to see that @xmath222 is a map @xmath273 $ ] , and is thus a solution of the @xmath25-system . since @xmath111 is flat , and the normal connection 1-form defined by the normal frame @xmath259 is @xmath274 , there exists @xmath275 such that @xmath276 and @xmath277 is a parallel normal frame . since @xmath22 is a null parallel normal section , there exists a constant null vector @xmath278 such that @xmath279 [ thm : ndimconf2 ] let @xmath222 be a solution to the @xmath25-system , and @xmath233 an extended flat frame for the lax pair @xmath280 ) \d x_i$ ] . then : 1 . @xmath281 is of the form @xmath282 , and if @xmath254 is a constant non - zero null vector , then @xmath283 is ( locally ) a flat lift of a conformally flat immersion into @xmath18 . 2 . @xmath49 , its fundamental forms and curvature normals @xmath156 are given explicitly by @xmath284 where @xmath285 and @xmath286 . if @xmath51 is another non - zero null vector in @xmath287 , then @xmath53 is a combescure transform and is a christoffel transform if @xmath288 and @xmath289 have opposite sign , where @xmath290 . such an @xmath49 is clearly a flat lift . . then @xmath292 . it follows from @xmath293 and @xmath292 that @xmath294 write @xmath295 and @xmath296 . then @xmath297 statements ( 2 ) and ( 3 ) follow . we recall below three known approaches to constructing solutions to the @xmath25-system ( see e.g. @xcite ) . dressing : : given a solution we may apply dressing actions to find new solutions to . we shall pursue this in section [ sec : rib ] . khler : : @xcite the @xmath25-system is unchanged with respect to the substitution @xmath298 where @xmath299 for any constant invertible @xmath300 , where @xmath301 . almost any choice of @xmath300 will result in @xmath302 being _ regular _ , i.e. @xmath303 is injective . the corresponding exterior differential system is involutive , hence we may apply the cartan khler theorem : given local analytic initial data @xmath304 , there is a unique local analytic solution @xmath222 to which satisfies @xmath305 . note that @xmath306 . inverse scattering : : @xcite if @xmath307 is regular , then given rapidly decreasing initial data @xmath308 with small @xmath309 norm , there exists a unique global smooth solution @xmath222 to the @xmath25-system such that @xmath310 . we will compute the dressing action of the simplest type rational loops in section [ sec : rib ] the discussion on constructions of solutions of the @xmath25-system given above and in theorem [ thm : ndimconf2 ] imply that local flat @xmath4-immersions in @xmath19 with flat , non - degenerate normal bundle and uniform multiplicity one are determined by @xmath54 functions of one variable ( the restriction of @xmath269 to a regular line ) . we state this more precisely . fix a regular element @xmath311 in @xmath37 . given a real analytic map @xmath312 such that @xmath313 for all @xmath152 , there is a unique solution @xmath222 of the @xmath25-system defined on an open subset of the origin of @xmath314 such that @xmath315 . the same statement holds if we replace @xmath316 by a rapidly decaying smooth map on @xmath317 whose @xmath309 norm is less than @xmath318 . for each solution of the @xmath25-system , theorem [ thm : ndimconf2 ] gives a family of flat @xmath4-immersions in @xmath19 parameterised by the light - cone of @xmath287 . an immersion @xmath320 with flat normal bundle is said to have _ uniform multiplicity one _ if it has constant multiplicities and each curvature distribution has rank one . it is easy to see that the proofs of the existence of flat lifts and line of curvature co - ordinates still work for these immersions . let @xmath321 be the curvature normals for a flat lift @xmath22 , @xmath322 the _ curvature normal bundle _ , and @xmath323 the orthogonal complement of @xmath324 in @xmath111 . note that theorem [ thm : ndimconf ] holds if we replace the @xmath25-system with the @xmath325-system and assume that @xmath323 is flat . we state the analogous results below . let @xmath320 be a conformally flat immersion with flat normal bundle and uniform multiplicity one . then : 1 . there is a flat lift @xmath326 for @xmath89 with flat normal bundle and uniform multiplicity one , 2 . there exist line of curvature co - ordinates for @xmath22 . the symmetric space @xmath325 is defined by the involution @xmath327 , and the @xmath328 eigenspaces of @xmath26 are @xmath329 , and @xmath330 set @xmath331 where @xmath332 for @xmath333 and @xmath334 . then both @xmath335 are maximal abelian subalgebras in @xmath36 , and @xmath336 let @xmath326 be a flat lift of @xmath89 with flat normal bundle and uniform multiplicity one , @xmath321 the curvature normals for @xmath22 , and @xmath337 as above . suppose that the induced connection on the subbundle @xmath323 of @xmath111 is flat , where @xmath338 . \(i ) if @xmath339 for all @xmath152 , then there exists a solution @xmath340 of the @xmath325-system defined by @xmath341 and an @xmath342 frame @xmath343 such that 1 . @xmath174 are principal curvature directions , @xmath344 for @xmath152 , and @xmath345 if @xmath346 , 2 . @xmath347)\d x_i$ ] . \(ii ) if @xmath348 , then there exists a solution @xmath349 of the @xmath350-system defined by @xmath351 and an @xmath342 frame @xmath343 such that 1 . @xmath174 are principal curvature directions , @xmath352 for @xmath152 , and @xmath345 if @xmath353 , 2 . @xmath354)\d x_i$ ] . moreover , there is a constant null vector @xmath355 such that @xmath356 , where @xmath357\d x_i = g^{-1}\d g$ ] and @xmath358 . let @xmath222 be a solution of the @xmath350-system defined by @xmath341 ( @xmath351 resp . ) , @xmath233 an extended flat frame for the lax pair of @xmath222 and @xmath355 a null vector . then @xmath359 , and @xmath360 is flat with flat normal bundle , uniform multiplicity one , and the subbundle @xmath323 is flat . + a different choice of null vector @xmath51 yields a combescure transform @xmath361 of @xmath22 . we first give a brief review of the dressing action ( cf . @xcite ) for the @xmath25-system . then we construct simple elements in the rational loop group with two poles that lie in the negative loop group , and give explicit formulae for the dressing action of simple elements on solutions of the @xmath25-system and the corresponding action on the flat lifts of conformally flat @xmath4-immersions in @xmath18 . we prove in the end of the section that the dressing action of simple elements on flat lifts are ribaucour transformations enveloping @xmath4-sphere or @xmath4-hyperboloid congruences which moreover project down to ribaucour transforms enveloping @xmath4-sphere congruences of conformally flat @xmath4-immersions in @xmath18 . throughout , @xmath362 where @xmath363 will denote conjugation across the real form @xmath364 . @xmath26 will refer both to the symmetric involution of @xmath365 which defines the symmetric space @xmath25 _ and _ its derivative at the identity , whose eigenspaces are @xmath366 . fix @xmath367 . let @xmath368 , and @xmath369 the group of holomorphic maps from @xmath370 to @xmath371 that satisfy the @xmath25 reality condition . let @xmath372 denote the subgroup of @xmath373 that is the restriction of a holomorphic map on @xmath374 , and @xmath375 the subgroup of @xmath376 that is the restriction of a holomorphic map defined on @xmath377 such that @xmath378 . the birkhoff factorisation theorem implies that the multiplication maps @xmath379 and @xmath380 defined by @xmath381 and @xmath382 are injective and the images are open dense . hence the _ big cell _ @xmath383 is open and dense . the local dressing action of @xmath375 on @xmath372 is defined as follows . since the big cell @xmath384 is open and dense , given @xmath385 , there generically exist @xmath386 such that @xmath387 . then @xmath388 defines a local action of @xmath375 on @xmath372 : this is the _ dressing action_. it is proved in @xcite that the dressing action of @xmath375 induces an action of @xmath375 on the space of solutions of the @xmath25-system : @xcite let @xmath233 be the normalised flat frame of the lax pair of a solution @xmath222 of the @xmath25-system , and @xmath389 . then 1 . there exists an open subset @xmath274 of the origin in @xmath314 such that the dressing action of @xmath390 at @xmath391 , @xmath392 , is defined for all @xmath393 , in other words , we can factor @xmath394 with @xmath395 and @xmath396 for all @xmath393 , 2 . @xmath397 is the normalised frame for a solution of the @xmath25-system , which will be denoted by @xmath398 . in general it is difficult to calculate the dressing action of a given @xmath399 on @xmath372 explicitly . it is , however , now standard theory that if @xmath389 is rational @xcite , then the dressing action of @xmath390 on @xmath372 can be computed explicitly via residue calculus . moreover , the action of a rational @xmath389 with minimal number of poles ( a so - called _ simple _ element ) often corresponds to known classical transforms ( e.g. bcklund transforms of pseudospherical surfaces , darboux transforms of isothermic surfaces , etc . ) . let @xmath26 be conjugation by @xmath400 , choose a scalar @xmath401 , and an isotropic line @xmath402 such that either @xmath403 let @xmath404 and suppose in addition that @xmath405 ( equivalently @xmath406 ) . let @xmath407 denote projection onto @xmath408 away from @xmath409 . in fact , if @xmath410 with @xmath411 and @xmath412 , then @xmath413 define the _ simple element _ @xmath414 by @xmath415 then @xmath416 it is easily checked that @xmath417 is an element of @xmath375 . let @xmath39 be a solution to the @xmath25-system . an extended frame @xmath418 of the lax pair @xmath228 of @xmath222 is called the _ normalised extended frame _ if @xmath419 . since the solution of an ode depending on a holomorphic parameter is holomorphic in that parameter , the map @xmath420 lies in @xmath372 . using the dressing action of @xmath421 computed in @xcite , or alternatively from @xcite , we get the following : [ thm : dress ] let @xmath422 be the normalised extended frame for a solution of the @xmath423-system , and @xmath417 the simple element defined by . then there is an open subset @xmath274 of the origin in @xmath314 such that @xmath424 , the dressing action of @xmath425 on @xmath391 , is defined for all @xmath393 . moreover , 1 . @xmath426 and @xmath427 satisfy , 2 . @xmath428 is the normalised extended frame of a new solution of the @xmath25-system , 3 . @xmath429 is well - defined if @xmath430 . if @xmath431 , then @xmath432 is real and in @xmath95 . so @xmath433 satisfy . if @xmath434 for some real @xmath435 , then @xmath436 so if we write @xmath437 , then @xmath438 are real and @xmath439 are pure imaginary matrices . a computation shows that @xmath433 satisfy . in other words , @xmath440 satisfies the @xmath25-reality condition . to prove that @xmath441 lies in @xmath372 , we expand @xmath442 in a power series about @xmath443 and checking that it is in fact holomorphic and invertible there . the twisting condition ensures that the same is true at @xmath444 , hence @xmath442 is a map into @xmath372 ; unique factorisation finishes things off . let @xmath445 . then @xmath446 expand the above equality at @xmath61 , noting that @xmath228 is a degree one polynomial in @xmath447 ; we see that @xmath448 is degree one is bounded . ] in @xmath447 and is thus the lax pair of a new solution of the @xmath25-system . we now write down explicit formulae for the action of @xmath425 on @xmath22 and the frame of @xmath22 . recalling section [ sec : uk ] , we note that a flat lift @xmath22 can be written @xmath449 where @xmath233 is an extended frame and @xmath359 . set @xmath450 then @xmath451 and @xmath96 are principal curvature directions , @xmath452 is parallel to the curvature normal @xmath165 of @xmath22 , and @xmath453 . assume that @xmath454 . we cancel the factor of @xmath455 from the definition of the dressed frame @xmath456 and write @xmath457 ( note that the factor @xmath458 is an isometry of @xmath23 , hence @xmath459 is equal to @xmath460 up to an isometry . ) introduce the notation @xmath461 , where choices are normalised such that @xmath462 : it is easy to check that @xmath463 and @xmath464 . we have @xmath465 from which we write @xmath466 write @xmath467 and set @xmath468 then @xmath469 where @xmath470 are written with respect to the standard bases of @xmath471 . the above formulae imply that @xmath472 more is true , for @xmath22 and @xmath473 envelop of a congruence of @xmath4-spheres or @xmath4-hyperbolae and are , in fact , _ ribaucour transforms _ of each other ( defined next ) . a _ congruence _ of @xmath4-spheres ( @xmath4-hyperbolae resp . ) is a map into the space of @xmath4-spheres in a spaceform . an enveloping submanifold of a congruence is a submanifold which has first - order contact with the congruence , i.e. each sphere ( hyperbola resp . ) is tangent to the submanifold it touches . an @xmath4-sphere in @xmath23 can be written as @xmath474 for some space - like @xmath475-dimensional linear subspace @xmath476 , @xmath477 , and a constant @xmath478 . this @xmath4-sphere lies in the light - cone @xmath19 if and only if @xmath479 . an @xmath4-hyperbola can be written as @xmath480 for some lorentzian @xmath475-dimensional linear subspace @xmath476 , @xmath477 , and a constant @xmath478 . this @xmath4-hyperbola lies in @xmath19 if and only if @xmath481 . note that the projections of both @xmath4-spheres and @xmath4-hyperbolas in @xmath19 to @xmath18 are @xmath4-spheres .. ] it can be easily seen that a generic @xmath4-dimensional congruence of @xmath4-spheres ( or @xmath4-hyperbolae ) has exactly two enveloping submanifolds @xmath482 and a map @xmath483 , so that for each @xmath484 there is a @xmath4-sphere ( or @xmath4-hyperbola ) @xmath485 in the congruence such that @xmath486 and @xmath487 are tangent to @xmath485 at @xmath488 and @xmath489 respectively . we will also call the map @xmath44 a congruence . the congruence @xmath44 is said to be _ ribaucour _ if the lines of curvature on @xmath486 map to lines of curvature on @xmath487 . otherwise said , the lines of curvature correspond and the tangent line through @xmath488 in the direction @xmath490 meets the tangent line through @xmath489 in the direction @xmath491 at equal distance . this is the definition given of ribaucour transform in @xcite for isothermic surfaces . [ bh ] let @xmath492 and @xmath408 an isotropic line in @xmath493 satisfying , and @xmath417 be the simple element in @xmath375 defined by . then the dressing action @xmath494 defined by is a ribaucour @xmath4-hyperbola congruence in @xmath19 if @xmath427 is real , and a ribaucour @xmath4-sphere congruence in @xmath19 if @xmath427 is pure imaginary . moreover , 1 . @xmath495 and @xmath496 is parallel to @xmath497 for all @xmath152 , where the @xmath174 and @xmath498 are principal curvature directions for @xmath22 and @xmath459 respectively , 2 . if @xmath22 is a flat lift of a conformally flat immersion @xmath89 in @xmath18 , then @xmath459 is a flat lift of another conformally flat immersion @xmath499 in @xmath18 , and the transform @xmath500 is a ribaucour transform in @xmath18 . note that @xmath22 and @xmath459 envelop an @xmath4-sphere ( or a @xmath4-hyperbola ) congruence if there exist vector fields @xmath269 normal to @xmath22 and @xmath501 normal to @xmath459 satisfying the following conditions : 1 . @xmath502 and @xmath503 , 2 . @xmath504 , which will be denoted by @xmath505 , 3 . @xmath506 . the above conditions imply that both @xmath22 and @xmath459 are tangent to the quadrics in the affine space @xmath507 , @xmath508 at @xmath509 and @xmath510 , where the centre @xmath511 . if @xmath505 is space - like then this quadric is an @xmath4-sphere , and if @xmath505 is lorentzian then this quadric is an @xmath4-hyperbola . since @xmath509 is null , there exists a time - like @xmath512 such that the quadric lies in the affine space @xmath513 . we have seen that @xmath514 where @xmath515 and is normalised so that @xmath516 and @xmath517 . it follows from the formulae for @xmath518 and @xmath519 that @xmath520 for each @xmath142 . equate the coefficients of the @xmath174 and @xmath242 in @xmath521 to get @xmath522 where @xmath523 and @xmath524 . case 1 : @xmath525 . then @xmath526 is real . use the condition that @xmath527 lies in @xmath505 to see that @xmath269 has to be parallel to @xmath528 , hence @xmath529 for some real function @xmath89 . it follows from @xmath530 and that @xmath531 . so @xmath532 since @xmath269 is time - like , it follows that @xmath505 is lorentzian . this shows that @xmath22 and @xmath459 envelop a congruence of @xmath4-hyperbolae and @xmath533 case 2 : @xmath534 . then @xmath535 for some @xmath536 . we can use the above argument to obtain @xmath537 since @xmath538 is space like , so is @xmath505 . in other words , @xmath22 and @xmath459 envelop a congruence of @xmath4-spheres . it is easy to check that both the @xmath4-spheres and the @xmath4-hyperbolae lie in @xmath19 . as a consequence of theorem [ thm : dress ] , @xmath459 is a flat lift of a new conformally flat @xmath499 in @xmath18 with identical line of curvature co - ordinates for @xmath459 . [ rem : confinv ] the discussion of congruences in @xmath18 is particularly beautiful in the light - cone picture ( e.g. @xcite ) . a congruence of @xmath4-spheres may be viewed as a map @xmath539 into the grassmannian of positive definite @xmath59-planes ; @xmath540 . the condition that @xmath541 envelops @xmath542 then becomes very simple : @xmath543 . calderbank @xcite generalise the notion of ribaucour in this setting by demanding simply that a general codimension congruence with two enveloping submanifolds @xmath544 is ribaucour iff the bundle @xmath545 is flat . we may restate the above theorem in a more invariant manner , that views the hyperbola and sphere congruences as sub - quadrics of the quadric @xmath546 . each @xmath547 is a signature @xmath548-plane , hence @xmath539 is an @xmath4-sphere congruence in the conformal @xmath18 , enveloped by @xmath549 . we may moreover calculate the flatness of the bundle @xmath550 to see that the enveloped congruence is indeed ribaucour in the sense of burstall calderbank @xcite . since the notion of enveloped sphere congruence is conformally invariant , the theorem is true in any riemannian @xmath551 we choose . specifically : let @xmath552 be conformally flat with uniform multiplicity one , @xmath22 a flat lift and @xmath553 the projection of the transform @xmath554 by a simple element ; then @xmath555 envelop a congruence of @xmath4-spheres and have corresponding curvature directions . to summarise , we have the following theorem : simple elements act by ribaucour transforms on conformal flats with uniform multiplicity one . returning to flat lifts , we may also rephrase the construction of ribaucour transforms in terms of a system of first order pde : in particular , given a flat lift @xmath22 , @xmath556 , and @xmath557 , such that @xmath558 satisfy , the ribaucour transform of @xmath22 by @xmath559 may be constructed . let @xmath560 be as in theorem [ thm : ndimconf ] @xmath561 , @xmath562 , @xmath563 , and @xmath517 and @xmath564)\d x_i$ ] the lax pair of the solution @xmath222 of the @xmath25-system . given @xmath401 and @xmath565 satisfying . then : 1 . the following system for @xmath566-valued maps @xmath567 has a unique solution : @xmath568 2 . @xmath427 and @xmath569 satisfy , where @xmath567 is the solution to . choose @xmath570 so that @xmath571 with @xmath516 . then @xmath572 is the ribaucour transform given in theorem [ bh ] by @xmath573 . system is solvable because @xmath574 is flat . the rest follows . if we apply the dressing action of simple elements to the vacuum solution @xmath575 repeatedly , then we can construct infinitely many families of explicit conformally flat @xmath4-immersions in @xmath18 with uniform multiplicity one . we may easily obtain a permutability theorem for ribaucour transforms , or at least those that arise via simple element dressing . by theorem [ thm : dress ] , combined with linear fractional transforms @xmath576 we see that @xmath577 are pole - free and invertible at @xmath578 respectively . putting these together and applying liouville s theorem ( holomorphic on @xmath579 constant ) we see that in fact @xmath580 applied to our discussion , we see that given two ribaucour transforms via simple elements , there exists a common fourth immersion which is simultaneously a ribaucour transform of the first two ( and is not the original immersion ) . in this section , we consider conformally flat @xmath4-dimensional immersions into @xmath18 with some multiplicity greater than one . the curvature distributions of such immersions have constant ranks and are smooth . their flat lifts into @xmath19 also have constant multiplicity . we show that in fact all but one curvature distribution has rank 1 . such submanifolds envelop a @xmath488-dimensional family of @xmath581 dimensional spheres ; they are the analogues of the channel hypersurfaces in @xmath3 , and hence will be called _ channel immersions_. unlike the uniform multiplicity one case , we do not know whether line of curvature co - ordinates exist for such immersions . if line of curvature co - ordinates do exist , then the gauss - codazzi equations for such an immersion is the @xmath25-system defined by a non - semisimple maximal abelian algebra in @xmath36 . conversely , solutions to these @xmath25-systems give rise to conformally flat immersions with one multiplicity @xmath155 . [ thm : channel1 ] suppose that @xmath89 is conformally flat with flat normal bundle and constant multiplicities , with at least one multiplicity @xmath582 . then the curvature distributions are smooth and there is precisely _ one _ curvature distribution with @xmath159 so that @xmath89 has multiplicity @xmath583 . we may therefore write @xmath584 where @xmath585 , @xmath586 . let @xmath22 be any flat lift of @xmath89 and let @xmath587 be the curvature normals of @xmath22 . the @xmath156 are space - like and orthogonal , @xmath588 is isotropic and orthogonal to the @xmath156 , and all are non - zero . any flat lift @xmath22 has degenerate normal bundle and the formulae of theorem [ thm : ffcurvs ] relating curvature normals of @xmath89 and @xmath22 still hold . moreover the distribution @xmath589 is integrable , and the leaf of @xmath589 through any point is contained in a copy of @xmath590 . indeed the repeated curvature normal @xmath591 of @xmath89 is a parallel section of @xmath592 over @xmath589 and the @xmath581-sphere in question has ( euclidean ) radius @xmath593 . first recall , from theorem [ thm : ffcurvs ] , that @xmath89 and any flat lift @xmath22 share the same curvature distributions , and that any distribution of rank @xmath155 has an isotropic , non - zero curvature normal for @xmath22 . if there are two such then they must be scalar multiples , since two orthogonal non - zero isotropic vectors contradict the fact that maximal isotropic subspaces of @xmath23 are lines . the part of flat differentiation @xmath594 that maps @xmath595 between tangent and normal bundle is well - known to be a @xmath596-valued 1-form @xmath597 such that @xmath598 . since @xmath22 is parallel in @xmath111 , it follows that @xmath599 . applying this to the supposition that there are two isotropic curvature normals which are non - trivial multiples of each other gives a contradiction . for the remainder , we appeal to a theorem of terng @xcite which states that the curvature distributions of @xmath89 are integrable and that the leaf of @xmath589 through any point is an open subset of a @xmath581-plane or an @xmath581-sphere . since , for us , the leaf must lie in @xmath18 , we necessarily have ( part of an ) @xmath581-sphere . indeed one may see that @xmath600 is constant on any leaf of @xmath589 , hence the @xmath581-sphere has radius @xmath601 : since @xmath591 is parallel , this radius is independent of @xmath589 . [ thm : channel ] let @xmath602 be as in theorem [ thm : channel1 ] , @xmath22 be a flat lift of @xmath89 , and @xmath603 , @xmath588 the corresponding curvature normals of @xmath22 . suppose that @xmath22 is parameterised by line of curvature co - ordinates @xmath604 . then : 1 . there exists an @xmath95 frame @xmath605 with @xmath606 principal curvature directions , and @xmath607 for @xmath608 , and @xmath609 , 2 . @xmath610 , where @xmath611 . 3 . set @xmath612 for @xmath613 , and @xmath614 for @xmath615 . then @xmath616 is a non - semisimple maximal abelian subalgebra in @xmath36 and @xmath617 . 4 . there exists a map @xmath618 such that @xmath619\d x_i$ ] . in other words , @xmath222 is a solution of the @xmath25-system defined by @xmath620 . there exists a constant null vector @xmath254 such that @xmath621 . set @xmath622 . . 1 . @xmath359 . 2 . write @xmath625 and @xmath626 . then @xmath627 is a flat immersion with degenerate flat normal bundle and constant multiplicities , @xmath22 is parameterised by line of curvature co - ordinates , @xmath628 and the curvature normals are @xmath629 for @xmath608 and @xmath630 . the discussion of dressing and ribaucour transforms goes through exactly as in section [ sec : rib ] for channel immersions that have line of curvature co - ordinates . since , by , logarithmic derivatives of dressed frames have the same @xmath36-part , and thus similar second fundamental forms , it is clear that dressing a channel hypersurface yields another . similarly , by the correspondence of theorem [ thm : channel ] , we also get dressing and ribaucour transforms of solutions to the @xmath25-system defined by non - cartan maximal subalgebra @xmath620 . we may also repeatedly apply the dressing action of @xmath425 to the vacuum solution @xmath575 to construct infinitely many families of conformally flat channel immersions . these immersions are given by explicit formulae because the extended frame for the vacuum solution is @xmath631 . burstall f.e . : _ isothermic surfaces : conformal geometry , clifford algebras and integrable systems _ ; in _ integrable systems , geometry , and topology _ ; volume 36 of _ ams / ip stud . adv . math . _ ; pp . 182 ; amer . soc . , providence , ri ( 2006 ) . terng c.l . , wang e. : _ curved flats , exterior differential systems , and conservation laws _ ; in _ complex , contact and symmetric manifolds _ ; volume 234 of _ progr . math . _ ; pp . 235254 ; birkhuser boston , boston , ma ( 2005 ) .

we present an optimal control approach to the problem of model calibration for lvy processes based on a non parametric estimation procedure . the calibration problem is of considerable interest in mathematical finance and beyond . calibration of lvy processes is particularly challenging as the jump distribution is given by an arbitrary lvy measure , which form a infinite dimensional space . in this work , we follow an approach which is related to the maximum likelihood theory of sieves . the sampling of the lvy process is modelled as independent observations of the stochastic process at some terminal time . we use a generic spline discretization of the lvy jump measure and select an adequate size of the spline basis using the akaike information criterion ( aic ) . the numerical solution of the lvy calibration problem requires efficient optimization of the log likelihood functional in high dimensional parameter spaces . we provide this by the optimal control of kolmogorov s forward equation for the probability density function ( fokker - planck equation ) . the first order optimality conditions are derived based on the lagrange multiplier technique in a functional space . the resulting partial integral - differential equations ( pide ) are discretized , numerically solved and controlled using scheme a composed of chang - cooper , bdf2 and direct quadrature methods . for the numerical solver of the kolmogorov s forward equation we prove conditions for non - negativity and stability in the norm of the discrete solution . to set boundary conditions , we argue that any lvy process on the real line can be projected to a torus , where it again is a lvy process . if the torus is sufficiently large , the loss of information is negligible . * msc ( 2010 ) : * 93e10 ( primary ) 49k20 , 60g51 , 62g05 ( secondary ) * key words : * optimal control of pide , kolmogorov equations , fokker - planck equation , lvy processes , non - parametric maximum likelihood method , akaike information criterion , financial data .

1506.08439

lvy processes play a large rle in contemporary mathematical finance @xcite , but also in many areas of physics , see e.g. @xcite . a real valued lvy process is a stochastic process @xmath2 that has increments @xmath3 , @xmath4 , that are independent of the past . the increments are also stationary in the sense that the probability distribution of the increment only depends on the time difference @xmath5 . furthermore , @xmath6 and a stochastic continuity condition for @xmath7 holds , see e.g. @xcite . under the given conditions , the characteristic function of @xmath2 is given by the lvy khinchine representation @xmath8=e^{t\psi(k)}\ ] ] @xmath9 $ ] stands for the expected value . @xmath10 is a conditionally positive definite function @xcite that has the following representation in terms of the canonical triplet @xmath11 : @xmath12 @xmath13 are constants , @xmath14 , and the lvy measure @xmath15 is a positive measure on @xmath16 such that @xmath17 in eq . ( [ eqa : levyformula ] ) @xmath18 is the characteristic function of the set @xmath19 which takes the value @xmath20 on this set and @xmath21 otherwise . the calibration problem for lvy processes consists of the estimation of the canonical triplet @xmath11 given the observation @xmath22 of the process trajectory @xmath2 at some prescribed times @xmath23 , @xmath24 . for instance , @xmath2 could be the process of log - returns of some asset and @xmath23 could be the closing time of the @xmath25-th trading day ( historic low frequency data ) . as the @xmath25-th increment of the process @xmath26 has the same distribution as @xmath27 , if @xmath28 , from a statistical point of view this is equivalent to the @xmath29-fold independent observation of the terminal values @xmath27 at time @xmath0 . @xmath2 can also be understood as the solution to the stochastic differential equation ( sde ) of jump - diffusion type @xmath30 here @xmath31 is a standard brownian motion and @xmath32,a)\sim{\rm po}(\nu(a)\delta t)$ ] is the random counting measure of jumps of height in the set @xmath33 in the time interval @xmath34 $ ] . @xmath35 stands for the poisson distribution with intensity @xmath36 and @xmath37,a)=n((t , t+\delta t],a)-\nu(a)\delta t$ ] is the compensated or martingale jump measure for small jumps , where we require @xmath38 for some @xmath39 , see @xcite for further details . the calibration problem for lvy processes , respectively the solution of ( [ eqa : sde ] ) , unfortunately is ill posed : the collection of all lvy measures @xmath40 is infinite dimensional , while only @xmath29 observations are available . direct application of the maximum likelihood principle in this situation leads to severe over - fitting issues @xcite . in many applications , one chooses families of lvy measures @xmath41 that depend only on a finite dimensional parameter vector @xmath42 , see e.g. @xcite . furthermore , one often restricts to such parametrizations , where the density @xmath43 of the probability distribution of @xmath2 can be calculated explicitly or at low numerical cost . one then assumes that the true distribution of @xmath2 is inside the prescribed set and uses the maximum likelihood approach for calibration @xcite . this assumption might however not be justified and give rise to modelling errors . as a non - parametric alternative , one can use generic parametrizations for the density of the lvy measure @xmath15 that can be refined depending on the amount of data available . this gives rise to a hierarchy or sieve @xcite of maximum likelihood problems with a finite number of parameters . if a suitable finite parametrization has been chosen , it remains to solve the maximum likelihood estimate at a given level of parametrization . one also has to determine this level based on the quality and also stability of the fits obtained . the resulting densities can no longer be calculated analytically . also , solution of the maximum likelihood problem gives rise to high dimensional optimization problems . the maximum likelihood method requires a parametric representation of the probability density functions ( pdf ) . the pdf can however be obtained as a solution to the kolmogorov forward equation ( fokker planck equation ) . the parameters @xmath42 then enter in this equation via coefficients in the generator of the semigroup @xcite . if the lvy measure @xmath15 is not zero , the generator of both these equations does not only contain a 2nd order partial differential operator , but also an integral operator of convolution type . this places the model calibration problem in the framework of optimal control problems with partial integral differential equations ( pide ) constraints . indeed , we know that the kolmogorov forward equation is representative of a stochastic process described in terms of sdes such as that one of eq . ( [ eqa : sde ] ) , where the set of parametrization for the approximation of the pdf , would correspond to a set of controls of the stochastic dynamic equation , so that , jointly to the maximum likelihood problem , it corresponds to a stochastic optimal control problem . the classical way to deal with the optimal control of stochastic process is by the dynamic programming principle and the related hamilton - jacobi - bellman equation for stochastic processes @xcite . however , this problem has been recently framed as a constrained pde optimization problem , where the pde is the fokker - planck , i.e. kolmogorov forward , equation @xcite . following this framework , the solution of the maximum likelihood problem , i.e. the stochastic optimization , is found by solving the first order optimality conditions in a functional space , that is the optimality system consisting of two pides , named forward and backward ( or adjoint ) equations , plus an optimality condition . this optimality system can be numerically solved by a gradient - based iterative algorithm as follows . the kolmogorov forward equation has a set of control parameters in order to maximize the log - likelihood functional for its terminal value . these controls involve the kolmogorov backward equation ( adjoint equation ) with suitable terminal condition , corresponding to the log - likelihood functional . hence , given an initial approximation of the unknown parametrization , first solve the forward equation , then set up the terminal condition and solve the adjoint one . with both the forward and adjoint solutions , by using the optimality condition equation the gradient is computed . then with a descending gradient technique , such as a non linear conjugate gradient method , found a better approximation of the control parameters and repeat until the satisfying accuracy for the parametrization is found . since this maximum likelihood problem could have an high dimensional space ad a huge number @xmath29 of observations , a fast , stable and enough accurate numerical solver for the pide is required . in our case the kolmogorov forward equation is a pide of parabolic differential operator type . such kind of pides are , e.g. , investigated in the option pricing models as a generalization of the black - scholes equation . the first difficulty to numerically solve this equation is the integral . in fact , in the case of using a fully implicit method , it would lead to solve a dense system of equation , for this reason implicit - explicit ( imex ) or operator splitting methods can be applied to bypass this problem ( see @xcite ) . the solution of the kolmogorov forward equation is a probability density function that is non negative with constant integral over the domain . such properties must be owned from the discrete solution too . the chang - cooper ( cc ) is a non - negative and conservative numerical method that has been used to solve the classical fokker - planck equation @xcite . here , we use a numerical method that can be classified as imex , since we use the cc method with an implicit time difference scheme for the differential operators of our pide , and evaluate the integral operator at the previous time step solution , i.e. in an explicit way . we prove for the resulting numerical solver : conservativeness , non - negativity and stability in the @xmath1-norm . the numerical solver for the adjoint equation is obtained directly from the solver for the forward equation by using the `` discretize then optimize '' approach to the optimization problem . finally , we quote , that for related work with vanishing lvy measure see e.g. @xcite , and for estimation procedures based on non parametric approximations of the empirical characteristic function see e.g. @xcite . an approach based on the method of moments and asymptotic expansions of lvy densities can be found in @xcite . the article is organised as follows : in the following section [ sec : reg ] we describe the hierarchy , of estimation problems . we shall show that the estimation problems that can actually be solved numerically can come arbitrarily close , at increasing computational cost , to the fully general lvy estimation problem . we also show that the use of periodic boundary conditions in the kolmogorov equations can be understood in terms of mapping the original lvy process on the real line to a derived lvy process on the torus . in section [ sec : kol ] , we set up the maximum likelihood estimation problem for a given parametrization and derive kolmogorov s forward ( fokker - planck ) equation and its adjoint ( kolmogorov backward ) equation with terminal conditions set by the log - likelihood objective functional . this maximum likelihood estimation problem is solved in the framework of the fokker - planck optimal control of stochastic processes , as a constrained pde optimal control problem . in section [ sec : num ] the discretization for kolmogorov s equations and the optimal control scheme is derived following a chang - cooper and imex approach . in particular we prove the structural properties of the numerical solution , i.e. conservativeness , non - negativity and stability . section [ sec : cons ] gives numerical tests for the consistency of the proposed procedure based on simulated data . we propose to use akaike s information criterion ( aic ) @xcite to choose an adequate parametrization from the hierarchy of spline parametrizations for density of the lvy measure . three different tests are performed : we first fit data that are simulated from a given distribution within our hierarchy of lvy distributions . the fits obtained are shown to be of very good quality and aic - selection criterion reproduces almost the original parametrization . as a second test we fit simulated data from a bi - directional gamma process , i.e. the difference of two independent gamma processes @xcite , which is not inside one of the parametrizations of degree @xmath44 , but can be approximated by those . the bi - directional gamma process is augmented by a small diffusive component and projected to the torus . the aic selection criterion and the fitting results again reproduce the final distribution of this process rather adequately . as a final test , we select financial data from the german stock exchange dax in a period between april 1998 and march 2002 and consider daily log - returns over 1000 trading days . this period represents a rather stable period for the dax with an almost constant level of the volatility . after projection , the aic based method selects a six - parameter spline approximation of the lvy measure density . the resulting fits again give a decent reproduction of the empirical distribution . our conclusions and an outlook are given in the final section [ sec : out ] . the estimation problem for the lvy measure @xmath15 is plagued by several issues . here we take a step by step approach towards the derivation of a hierarchy of estimation problems that approximate the original one . let @xmath45 , @xmath46 , be a family of probability density functions with variable dimension @xmath44 of the parameter set @xmath42 , and let @xmath47 be the ( unknown ) probability density of @xmath48 . for @xmath44 fixed , we apply the maximum likelihood method to select an estimated value @xmath49 based on increasing sample size . it is known from the general theory of maximum likelihood @xcite that the estimated @xmath50 converges almost surely to the true value @xmath42 , provided @xmath51 holds . this leaves the question open , which parametrization or which value for @xmath44 one should choose . we solve this problem by maximizing the akaike information criterion ( aic ) . maximizing the aic corresponds to minimizing ( asymptotically ) the expected kulback - leibler distance , or relative entropy , between the true distribution @xmath47 and its parametric estimate @xmath52 . see @xcite[chapter 6 ] for a detailed derivation . let us now define the parametrizations with @xmath44 parameters . we intend to show that , for sufficiently large @xmath53 , we can approximate the original lvy distribution to an arbitrary precision . this requires several steps of approximation : [ [ truncating - small - and - large - jumps . ] ] truncating small and large jumps . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the total mass of the lvy measure , @xmath54 , can be infinite . this quantity defines the average number of jumps per unit time of the associated lvy process @xcite . the easiest way to deal with this is to truncate small jumps by setting @xmath55 which is a finite measure by equation ( [ eqa : levycondition ] ) . using ( [ eqa : levycondition ] ) and dominated convergence , one can moreover prove that @xmath56 for @xmath57 as @xmath58 . here @xmath59 is given by ( [ eqa : levyformula ] ) with @xmath15 replaced by @xmath60 . by the continuity theorem of paul lvy , see e.g. ( * ? ? ? * theorem 23.8 ) , this then implies ( weak ) convergence in law of the respective probability distributions . at the same time , a finite lvy measure @xmath15 permits one to re - parametrize @xmath10 of eq . ( [ eqa : levyformula ] ) via @xmath61 with @xmath62 . in the following we assume @xmath15 to be finite and use parametrization ( [ eqa : levyformulaii ] ) . the last term in ( [ eqa : levyformulaii ] ) now has the structure of a compound poisson distribution , i.e. @xmath2 can be represented as @xmath63 where @xmath64 is normally distributed , @xmath65 is poisson distributed with intensity @xmath66 and @xmath67 are i.i.d . random variables with distribution given by the normalized lvy measure , @xmath68 . also @xmath69 and @xmath67 are all stochastically independent . with a similar argument , we can cut off large jumps by replacing @xmath15 with @xmath70 . also in this case , in the limit @xmath71 , the truncated lvy distributions converge in law to the non truncated one . in the following we thus assume that the support of @xmath15 is contained in some finite interval @xmath72 $ ] . the appropriate size of this region can be estimated e.g. by the chebyshev s inequality using empirical mean and variance from the data . [ [ regularizing - the - lvy - measure . ] ] regularizing the lvy measure . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + given a non negative , continuously differentiable function @xmath73 with compact support such that @xmath74 , and setting @xmath75 , we define @xmath76 . we consider the regularised measures @xmath77 . inserting this measure in ( [ eqa : levyformulaii ] ) , using fubini s theorem and dominated convergence , one easily shows that @xmath78 . this again implies convergence of the related probability distributions in law . [ [ spline - approximation - of - the - densities . ] ] spline approximation of the densities . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let thus @xmath79 with @xmath80 non negative , continuously differentiable and with compact support . let @xmath81 , @xmath82 , be a collection of uniform grid points in @xmath83 such that the support of @xmath84 is covered and @xmath85 . define @xmath86 by the linear interpolation between points @xmath87 . again , one easily sees that @xmath86 converges to @xmath88 as @xmath89 . also , for small @xmath90 , the functions @xmath86 have support in a fixed compact interval and are uniformly bounded by the maximum value of @xmath88 . if we insert measures @xmath91 into ( [ eqa : levyformulaii ] ) , this expression converges to the related one with @xmath79 . this suffices to prove that an approximation of the probability distributions ( in law ) is feasible with 1st order spline densities . [ [ fixing - drift - and - diffusion - terms . ] ] fixing drift and diffusion terms . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + one might or might not like to include the drift and diffusion term determined by @xmath92 and @xmath93 into the estimation problem . although , in general , these quantities have to be estimated , here we keep a small fixed value for @xmath93 for reasons of numerical stability of the kolmogorov equations . as long as this value underestimates the true diffusion , this corresponds to a splitting of the lvy process @xmath94 into stochastically independent components where @xmath95 is determined by the purely gaussian lvy triplet @xmath96 . @xmath97 will be a lvy process that contains drift , the excess diffusion and jumps . however , the distribution of @xmath97 at time @xmath0 can be approximated in the sense of convergence in law by compound poisson distributions without drift and lvy terms . the explicit construction can e.g. be found in @xcite . we can thus approximate our estimation problem to arbitrary accuracy with a problem where drift and diffusion take fixed values . we also note that a non vanishing diffusion implies the existence of a probability density function @xmath98 for the distribution of @xmath2 . [ [ periodic - boundary - conditions - from - the - projection - to - a - torus . ] ] periodic boundary conditions from the projection to a torus . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + another problem with the kolmogorov forward equation is the issue of boundary conditions . we have already shown that we can approximate the estimation problem by one where the lvy measure @xmath15 has support inside a large interval @xmath72 $ ] . let @xmath99 be the torus @xmath72 $ ] with the end points of the interval identified . let @xmath100 , then define the operator @xmath101 as @xmath102 a group operation on @xmath99 , with @xmath103 the modulus operation . let furthermore @xmath104 be the group homomorphism defined by @xmath105 . let @xmath106 be a stochastic process on @xmath99 . if @xmath2 is a lvy process on the group @xmath107 , the same applies to @xmath108 with respect to the group @xmath109 . note that in the definition of lvy processes only the group structure of @xmath107 enters . lvy processes are naturally defined on locally compact abelian groups like @xmath107 or also @xmath109 , see @xcite . let us now consider the characteristic function of the @xmath99-valued process @xmath108 at time @xmath110 . by the periodicity of @xmath99 , only @xmath111 values from @xmath112 are needed . to derive a lvy - kinchine formula ( [ eqa : levyformula ] ) for @xmath108 on @xmath99 from that of @xmath2 on @xmath83 , we consider for such values of @xmath111 @xmath113&=&\mathbb{e}\left[e^{i\phi(y(t))k } \right]\nonumber\\ & = & \mathbb{e}\left[e^{iy(t)k } \right ] = e^{t\psi(k)}.\end{aligned}\ ] ] inserting ( [ eqa : levyformulaii ] ) , we obtain @xmath114 with @xmath115 the image measure on @xmath15 under @xmath116 . note that under the hypothesis that @xmath15 has support in @xmath72 $ ] , @xmath15 can be reconstructed from @xmath115 as @xmath117 , where @xmath118 is the natural embedding . using this , we identify @xmath15 and @xmath115 in the following . summing up , we consider the maximum likelihood parameter estimation problem with fixed @xmath119 , a piecewise linear density function with @xmath44 grid points for the finite lvy measure and periodic boundary conditions on @xmath99 . by refining the grid for the linear interpolation , enlarging the size of the torus and letting the fixed diffusion go to zero , we can approximate the distribution of any lvy process with one of our candidate processes in the topology set by weak convergence in law . this constitutes the hierarchy of maximum likelihood estimation procedures . in this section we formulate the optimization problem for the maximum likelihood parameters estimation . the maximum likelihood estimator as an optimizing objective functional is given together to the kolmogorov forwards pide as constraints . the optimality system is written by using the lagrange multipliers method in a functional space , by also including the karush - kuhn - tucker conditions for the non negativity of the optimizing parameters . [ [ objective - functional - and - forward - equation . ] ] objective functional and forward equation . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let the @xmath29 independent sample values @xmath120 be given , and @xmath121 , where @xmath122 $ ] . these values can e.g. be obtained as @xmath123 , where @xmath124 is the group homomorphism defined in eq . ( [ eq : homo ] ) and @xmath125 is the lvy process on @xmath83 . we deal with the problem to find the pdf of @xmath126 such that it best fits with the sample values . for this purpose we consider the maximum likelihood problem in the framework of pide - constrained optimization : we have to find the maximum likelihood estimator @xmath127 with respect to the parametrization of the measure given by @xmath42 , where @xmath128 is the ( normalized ) log - likelihood with the constraint given by the following kolmogorov forward ( fokker - planck ) equation for the lvy process @xmath108 on the torus @xmath99 with lvy data @xmath129 using the parametrization ( [ eqa : levyformulaii ] ) and @xmath130 : @xmath131 where @xmath98 represent the pdf of the process at time @xmath110 . this pide is defined in the interval of time @xmath132 $ ] , and with periodic boundary conditions on @xmath133 . here @xmath134 is a set of triangular shaped basis for the set of continuous functions that are linear on @xmath135 , see the preceding section , @xmath136 \nonumber \\ { \theta}_j(s)=1-(s-\theta_j)/{\delta}\qquad s\in [ \theta_j , \theta_j+{\delta } ] , \nonumber\end{aligned}\ ] ] where @xmath137 for @xmath138 , are the points of a discrete uniform mesh of step size @xmath139 defined on the domain . the periodicity @xmath140 is assumed . the existence and uniqueness of the solution of the fokker - planck equation ( [ eq : fp ] ) is well established , also for initial conditions belonging to the class of measures @xcite . [ [ the - optimality - system . ] ] the optimality system . + + + + + + + + + + + + + + + + + + + + + + if we write the mapping @xmath141 between the maximization parameters and the pdf , then we introduce the so - called reduced cost functional @xmath142 , so that the maximization problem becomes @xmath143 a local maxima @xmath144 for @xmath145 can be found by solving the optimality system obtained by vanishing the variations of the following lagrangian functional @xmath146 p(x , t ) dx\,dt-\sum_{j=1}^{n_\theta}\pi_j\alpha_j,\end{aligned}\ ] ] where @xmath147 and @xmath148 fulfil the usual karush - kuhn - tucker ( kkt ) conditions @xmath149 and @xmath150 . these are important to include the non - negativity constraints for the control variables . note that if the condition @xmath151 is violated for some @xmath152 , the density of the measure @xmath153 is negative in a neighbourhood of @xmath81 and thus is not a lvy measure any more . the sum @xmath154 should be extended only on the active constraints , i.e. when @xmath155 . for those values of @xmath25 on the maximum where @xmath156 we have @xmath157 . first we calculate the variation @xmath158 for the adjoint equation . in the following the variations are calculated separately for each addend of the r.h.s . we get @xmath159 for the term @xmath160 we apply the substitution @xmath161 , then exchange @xmath162 , so that it recasts to @xmath163 . then , again , we substitute @xmath164 and , by inserting also the former term , we get @xmath165 { \delta}f(x , t ) dx\,dt.\ ] ] for the variation of the time derivative , integrating by parts , one obtains @xmath166 the variation @xmath167 holds because of the cauchy initial condition , while the variation in @xmath0 can be defined in some points @xmath168 . next , we integrate by parts the term with the first order derivative in @xmath169 and obtain @xmath170 . due to periodicity in the first term , @xmath171 has to be zero , hence @xmath172 . from the diffusive term we get @xmath173_{{\omega}_a}^{{\omega}_b } dt + { \int_{\omega}}{\partial}_x^2 p(x , t ) dx dt.\ ] ] the first boundary term is zero because of the periodic condition of the variation of the derivative of @xmath174 at the boundaries , and because of the previous periodic condition on @xmath175 . the second is analogous and has to vanish , we therefore get the continuity condition @xmath176 . by collecting all the terms under double integral , we get the adjoint equation . the remaining boundary term @xmath177 will be considered below . to calculate the variation on @xmath174 in the functional @xmath178 we perform an additional integration in space , so that @xmath179 where @xmath180 is the @xmath181dirac measure , then variate @xmath182 , hence @xmath183 so that the first order terms plus the remaining boundary , give @xmath184 this expression have to be zero for each @xmath185 . it represents the terminal condition for the adjoint equation : that is @xmath186 , and @xmath187 if @xmath188 . in case of multiplicity of @xmath168 the condition becomes @xmath189 , with @xmath190 running on the multiplicity value . summarizing , the adjoint equation ( kolmogorov s backward equation ) is as follows : @xmath191 we note that by reverting the sign of the time we get the same pide as the forward equation ( up to a reflection of the drift and jump direction ) , hence this equation has a unique solution , also for the non regular final value problem @xcite . second , we variate in eq.([eq : lagr ] ) the fitting parameters @xmath192 , from which we found the optimality equations : @xmath193 where @xmath194 runs on the set of values where @xmath195 . note that the active @xmath196 do not change the gradient , but simply balance non - zero gradient components that point to the directions where the inequality constraint @xmath197 is violated . as in our case we deal with simple box - constraints on the @xmath198 themselves , we can set those components of the negative gradient equal to zero that correspond to an active index @xmath194 and are negative , when determining the update . this then accounts for the effect of the @xmath196 , see e.g. @xcite . the @xmath199 order necessary optimality system consists of the eqs . ( [ eq : fp ] ) , ( [ eq : adj ] ) and ( [ eq : opt ] ) . its solution gives values @xmath200 that are candidates for maximizing the functional ( [ eq : functional ] ) . note that maximum likelihood fits in most cases do not correspond to convex optimization problems and one always has to account for the perils of local minima that are sub - optimal globally . [ [ forward - equation - in - flux - form . ] ] forward equation in flux form . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the forward equation ( [ eq : fp ] ) , can be written in flux form : @xmath201 , where @xmath202 is the flux defined as @xmath203 by using @xmath204 , it is easy to verify that eq . ( [ eq : flux ] ) is equivalent to eq . ( [ eq : fp ] ) . further , from the conservation of the total probability , it follows that the flux has the periodic boundary condition @xmath205 . from this we immediately get the periodic condition on the first derivative @xmath206 . the numerical solution of the optimality system is found by a non linear gradient conjugate iterative procedure @xcite . at each iteration the solution of two pides , the forward and the adjoint one , must be found . in particular the structural properties of the pdf solution must be satisfied , as well as a stability condition of the pides numerical scheme solver . for the numerical discretization of the kolmogorov forward equation we use the chang - cooper scheme ( cc ) @xcite , joint to a 2nd order backward differentiation formula ( bdf2 ) for the discrete time operator . the cc method was proposed for a fokker - planck resp . kolmogorov equation @xcite without the integral term . it is stable , second - order accurate , non - negative , and conservative numerical scheme @xcite . the cc method is used for the differential operators , the integral term is treated separately according to an imex methodology . we denote the following @xmath207 and @xmath208 , then the kolmogorov forward equation reads as follows @xmath209 where @xmath210 consider a uniform grid of size @xmath211 on the space domain @xmath212 given by @xmath213 and a uniform grid on the time domain @xmath214 : t_m = m\,\delta t , m=0,\ldots , n_t , { \delta}t = t / n_t\}$ ] . let @xmath215 denote the approximated values of the continuous solution of the fpe . we employ the following discretization of ( [ eq : flux ] ) @xmath216 where @xmath217 is the bdf2 operator . @xmath218 is the sum of the integrals of eq . ( [ eq : flux ] ) calculated with the mid - point scheme @xmath219 where @xmath220 , @xmath221 represents the translated @xmath222 by the value @xmath223 and continued by periodicity , @xmath224 , @xmath225 note also that the summation starts from @xmath226 , because the point @xmath227 is the same of that @xmath228 . therefore , the solution at a new time step is calculated by solving the following equation for the unknown @xmath229 @xmath230 with the initial condition @xmath231 this scheme is based on the fluxes at @xmath232 cell boundaries . the partial flux at the position @xmath233 is computed as follows @xmath234 \ , f^{m+1}_{i+1 } - \left(\frac{1}{h } c - \delta b \right)\ , f^{m+1}_{i } .\ ] ] this formula results from the following linear convex combination of @xmath174 at the points @xmath235 and @xmath236 : @xmath237.\ ] ] the idea of implementing this combination was proposed by chang and cooper in @xcite and it was motivated with the need to guarantee positive solutions that preserve the equilibrium configuration . indeed , the cc method is related to exponential fitting methods , such as that one proposed by allen and southwell @xcite , and by the scharfetter - gummel discretization scheme @xcite . the value of the parameter @xmath238 is @xmath239 , where @xmath240 , which can be shown to be monotonically decreasing from @xmath20 to @xmath21 as @xmath241 goes from @xmath242 to @xmath243 . notice that with the choice of @xmath238 given above , the numerical scheme shares the same properties of the continuous fp equation that guarantee positiveness and conservativeness . this is a special case of the cc scheme because in the general one , the functions @xmath244 and @xmath245 may depends on @xmath246 , hence also @xmath238 may depend on @xmath246 , too . both the cc scheme @xcite and the mid - point are second order accurate , then a second order numerical scheme results . let @xmath247 be the discrete solution at the time @xmath248 , with @xmath249 omitted due to periodicity , and @xmath250 . the action of the finite difference operator for @xmath251 in eq . ( [ eq : flux_eq ] ) reads as matrix @xmath252 whose elements are defined by @xmath253 where @xmath254 , @xmath255 . hence , @xmath256 , and then the eq . ( [ eq : disc_fp ] ) can be written in matrix form , as follows @xmath257 is the matrix coefficients related to eqs . ( [ eq : disc_fp ] ) and ( [ eq : ccflux ] ) . we note that this method needs of a second starting point , that can be calculated by using a first order euler scheme with a smaller time step size than @xmath258 . the implicit euler scheme for the eqs . ( [ eq : forw ] ) and ( [ eq : ccflux ] ) is @xmath259 these two numerical schemes own some properties that can be easily proved , but we list here as remarks . [ rem : eul_cons ] the euler - cc scheme ( [ eq : euler ] ) to eqs . ( [ eq : forw ] ) and ( [ eq : ccflux ] ) , defined in the periodic domain @xmath260 , is conservative . in fact , @xmath261 , and @xmath262 because the set of values of @xmath263 are the same as @xmath264 , being the last only translated by @xmath111 . hence , @xmath265 . [ rem : cons_bdf ] provided that @xmath265 , then the bdf2-cc scheme ( [ eq : discfp_matrix ] ) to eqs . ( [ eq : forw ] ) and ( [ eq : ccflux ] ) , defined on the periodic domain @xmath260 , is conservative . in fact for the same constraints on @xmath252 and @xmath266 as above , we get the identity @xmath267 . the positivity of the numerical scheme is proved by using the theorem for the class of @xmath268-matrix @xcite . given a positive matrix @xmath269 , @xmath270 , we say that @xmath271 is a non singular @xmath268-matrix if @xmath272 , where @xmath273 is the spectral radius of @xmath269 . a non singular @xmath268-matrix has the important property @xmath274 [ th : pos_eul ] let @xmath275 , with @xmath276 defined in ( [ eq : const_a ] ) , then the euler scheme ( [ eq : euler ] ) to eq . ( [ eq : forw ] ) , defined in the periodic domain @xmath260 , is positive preserving . the argument is as follows : let @xmath277 the matrix operator such that + @xmath278 . such a matrix is non negative because @xmath198 and @xmath279 are . the numerical scheme ( [ eq : euler ] ) can be recast as @xmath280 where @xmath281 is a positive matrix . provided that @xmath282 and @xmath283 the r.h.s . is a non negative vector . we observe that the matrix on the l.h.s is always diagonal dominant , hence it has a convergent regular splitting and consequently is an @xmath268-matrix @xcite . therefore , @xmath284 is non negative and @xmath285 will be too . in order to prove the positivity of the bdf2 numerical scheme ( [ eq : discfp_matrix ] ) , we need of the following lemma that gives a lower bound to the velocity of decreasing of the solution . [ lem : eul ] let the vector @xmath286 be given non negative . take a number @xmath287 , then the solution @xmath288 calculated with the euler scheme of eq . ( [ eq : euler ] ) after @xmath289 time steps satisfies the following inequality @xmath290 provided that @xmath291 , with parameters defined in eq . ( [ eq : diff_flux_matrix ] ) . a proof is given for a particular case in @xcite ( see also refs . therein ) . here we prove it as follows . given @xmath285 and @xmath292 calculated with ( [ eq : euler ] ) , let define @xmath293 . by applying the operator @xmath294 , we get @xmath295 , i.e. @xmath296 where @xmath281 is a positive matrix . now provided the bound for @xmath297 , then the r.h.s . is positive and from th . [ th : pos_eul ] we get that @xmath298 . by iterating that inequality @xmath289 times , we get the thesis . [ rem : bound_dt_eul ] the upper bound on @xmath297 in lemma [ lem : eul ] results to be @xmath299 for @xmath287 , hence the condition on the lemma is stricter than those on non negativity of thm . [ th : pos_eul ] . now we show a lemma similar to lemma [ lem : eul ] valid for the bdf2 scheme . [ lem : bdf2 ] let @xmath300 and @xmath301 be the time step size of the numerical scheme of eq . ( [ eq : discfp_matrix ] ) that generates the sequence of vectors @xmath285 for @xmath302 from the starting vectors @xmath303 . if there exists @xmath304 such that @xmath305 and @xmath306 , then @xmath307 for all @xmath308 . we apply the operator @xmath309 to @xmath310 , @xmath311 and use eq . ( [ eq : discfp_matrix ] ) to the first term on the r.h.s . to get @xmath312f^{m+1 } -\xi f^m + \dt(2\tilde a + \xi r)f^{m+1},\ ] ] where @xmath281 is a positive matrix . we know that @xmath309 is an @xmath268-matrix and its inverse is always non - negative . also @xmath313 is non negative . hence , we can prove non negativity of @xmath314 , provided that @xmath315 for a value @xmath316 , because of the hypothesis @xmath305 , that also states that @xmath317 . the last inequality is just the bound on @xmath297 in the assertion that gives a positive value for @xmath297 only when @xmath318 . indeed , this lemma proves positivity of the numerical solution of eq . ( [ eq : discfp_matrix ] ) , provided that @xmath319 , and @xmath320 . @xmath321 is the second starting value of the numerical scheme , that can be calculated with the euler scheme ( [ eq : euler ] ) . [ th : pos_bdf2 ] let @xmath319 the discrete initial condition ( [ eq : disc_init ] ) , and let @xmath322 the second starting value calculated with the euler scheme ( [ eq : euler ] ) with an appropriate time step , such that @xmath323 , for @xmath318 . then , the bdf2 scheme ( [ eq : discfp_matrix ] ) to eq . ( [ eq : forw ] ) , defined in the periodic domain @xmath260 , is positive preserving for the solution @xmath285 , with @xmath324 . the proof is an application of the lemmas [ lem : eul ] and [ lem : bdf2 ] . in order to establish the stability of the discrete numerical schemes of eqs . ( [ eq : discfp_matrix ] ) and ( [ eq : euler ] ) , we need inequalities of the form @xmath325 evaluated in a suitable norm with @xmath326 possibly less or equal than @xmath20 . we prove that it realizes for the @xmath20-norm with @xmath327 . [ th : stab_eul ] let the positivity condition of theorem [ th : pos_eul ] be fulfilled , i.e. @xmath328 . then , the euler scheme ( [ eq : euler ] ) is stable in the @xmath20-norm , that is @xmath329 for all @xmath330 . let @xmath331 and invert the matrix operator at l.h.s . , then eq . ( [ eq : eul_proof ] ) reads as @xmath332.\ ] ] now we observe that @xmath333 hence , @xmath334 since @xmath328 , all the components of the vectors inside the norm at the r.h.s . are positive , so that the modulus for the evaluation of the @xmath20-norm can be removed . using @xmath262 as in rem . [ rem : eul_cons ] , we get the statement of the theorem . now we can prove the stability of the numerical scheme with bdf2 integration of eq . ( [ eq : discfp_matrix ] ) . [ th : stab_bdf2 ] let the positivity condition of the theorem [ th : pos_bdf2 ] be fulfilled , i.e. let @xmath297 be the time step size of the numerical scheme of eq . ( [ eq : discfp_matrix ] ) , @xmath319 the discrete initial condition ( [ eq : disc_init ] ) and @xmath322 the second starting value evaluated at the time @xmath297 . if there exists a real number @xmath335 such that @xmath336 with @xmath318 and @xmath337 , then the bdf2 scheme ( [ eq : discfp_matrix ] ) is stable in the @xmath20-norm , that is @xmath329 for all @xmath330 . the numerical scheme ( [ eq : discfp_matrix ] ) can be written as @xmath338 where @xmath277 is defined as in thm . [ th : pos_eul ] . we apply @xmath339 and evaluate the @xmath20-norm to both sides . following the same calculations as in thm . [ th : stab_eul ] , we get that @xmath340 . from the bound on @xmath297 , we note that @xmath341 this means that for all @xmath342 in the interval @xmath343 , it is @xmath344 with @xmath345 . now we have that @xmath346 by virtue of the positivity condition , @xmath347 by our assumptions , and @xmath348 , hence is guaranteed that the sum in the r.h.s . is a non negative vector and the modulus in the calculation of the @xmath20-norm can be removed . using the property given in rem . [ rem : eul_cons ] , we conclude that @xmath349 . [ rem : stability ] indeed , in the stability theorem [ th : stab_bdf2 ] the equality @xmath350 holds . in fact , because of the conservativeness from rem . [ rem : cons_bdf ] we have @xmath351 , and under the non negativity condition of theorem [ th : pos_bdf2 ] all the components of the vectors @xmath352 are non negative , so that the previous conservativeness identity corresponds to the 1-norm equivalence . further , we can state that for these numerical schemes the conservativeness and the non negativity imply the stability of the discrete operator . we can finally conclude from the lax equivalence theorem , that for regular solutions of the kolmogorov forward equation @xmath353 $ ] , provided that the hypothesis of thm . [ th : stab_bdf2 ] , then the numerical scheme of eq . ( [ eq : discfp_matrix ] ) yields discrete solutions that are second order convergent in time and space . the non negativity conditions of for the euler scheme of lemma [ lem : eul ] and bdf2 of lemma [ lem : bdf2 ] , can be correspondingly written as @xmath354 and @xmath355 we note that for @xmath356 or @xmath357 the upper bound for @xmath297 scales as @xmath358 . for @xmath359 it scales as @xmath360 . [ [ adjoint - equation . ] ] adjoint equation . + + + + + + + + + + + + + + + + + the discrete adjoint equation can be found by discretizing the lagrangian function of eq . ( [ eq : lagr ] ) and then performing the variations on the discrete variables . this is know as the _ discretize - then - optimize _ approach ( see ref . @xcite for details ) . this technique yields the following discrete adjoint equation @xmath361 where @xmath362 is the transpose of @xmath268 , and @xmath363 , with @xmath364 . the numerical stability is given by the same condition for the forward equation , since the transpose of the operator @xmath268 has the same eigenvalues , but in this case the non negativity and conservativeness property are not required . care has to be taken for the discrete terminal condition , since it can not be defined through the eq . ( [ eq : adj ] ) for the presence of the @xmath238-dirac measure . for this purpose we discretize the term ( [ eq : int_func ] ) as follows @xmath365 where @xmath366 are the points of the integral average theorem . then we use the approximation @xmath367 , so that , by performing the variation @xmath368 on this discrete functional , we get the discrete terminal condition @xmath369 according to eq . ( [ eq : adj ] ) , it completes the formulation of the discrete adjoint problem . [ [ discrete - gradient . ] ] discrete gradient . + + + + + + + + + + + + + + + + + + the discrete of the reduced gradient related to the optimality condition of eq . ( [ eq : opt ] ) is calculated with the mid - point quadrature formula . each component @xmath25 is given by @xmath370 where @xmath371 . [ [ non - linear - conjugate - gradient - method . ] ] non linear conjugate gradient method . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the availability of the discrete gradient allows us to implement a non linear conjugate gradient scheme ( nlcg ) in order to solve the optimization problem ( [ eq : max_reduced ] ) . nlcg represents an extension of the linear conjugate gradient method to non - quadratic problems @xcite . the optimality system is solved by implementing the gradient given by the following algorithm : [ evaluate_gradient ] 1 . solve the discrete fp equation ( [ eq : disc_fp ] ) with given initial condition ( [ eq : disc_init ] ) ; 2 . solve the discrete adjoint fp equation ( [ eq : discadj_matrix ] ) with terminal condition ( [ eq : disc_term ] ) compute the approximated discrete gradient @xmath372 by using ( [ eq : disc_grad ] ) ; 4 . end . in a nlcg scheme . the search directions are recursively as @xmath373 where @xmath374 in this paragraph stands for the iteration index , @xmath375 is the numerical gradient , with @xmath376 . let @xmath377 an estimation of the best rates at the iteration @xmath111 , the next one for a minimum point are given by @xmath378 where @xmath379 is a steplength obtained with a line - search that satisfies the armijo condition of sufficient decrease of @xmath145 s value as follows @xmath380 where @xmath381 ; see @xcite . notice that we use the inner product of the @xmath382 space . for the formula of @xmath383 we use the formulation due to dai and yuan @xcite @xmath384 where @xmath385 . summarizing , the nlcg scheme is implemented as follows [ alg : nlcg ] * input : initial approx . @xmath386 , @xmath387 , index @xmath227 , maximum @xmath388 , tolerance @xmath389 . 1 . while ( @xmath390 & & @xmath391 ) do 2 . search the steplength @xmath392 , by sequentially shrinking , along @xmath393 satisfying ( [ armijo ] ) ; 3 . set @xmath394 . i.e. eq . ( [ ncgupdate ] ) , according to the kkt condition , the eventually negative components of @xmath395 are set to @xmath21 . 4 . compute @xmath396 using algorithm [ evaluate_gradient ] ; 5 . compute @xmath397 given by ( [ betadaiyuan ] ) ; 6 . let @xmath398 , i.e. eq ( [ ncgdir ] ) 7 . set @xmath399 ; 8 . end while [ ncg ] [ [ correction - factor - for - the - logarithm - in - the - objective . ] ] correction factor for the logarithm in the objective . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the numerical evaluation of the functional of eq . ( [ eq : functional ] ) has the problem of the logarithm in the points @xmath168 where the pdf at the final time has vanishing values . hence , the functional is replaced as follows @xmath400 with @xmath401 . [ [ nearest - grid - point - for - sample - values ] ] nearest grid point for sample values + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the discrete pdf is defined on the mesh grid @xmath260 , the sample values @xmath168 used for the evaluation of the pdf are approximated to the nearest values of the space mesh grid @xmath260 . this approximation affects both the value of the functional and the terminal condition for the adjoint equation . [ [ von - mises - distribution . ] ] von mises distribution . + + + + + + + + + + + + + + + + + + + + + + + the initial distribution @xmath402 of eq.([eq : fp ] ) is set as the following von mises distribution @xmath403 where @xmath404 is the modified bessel function of order @xmath21 , and @xmath405 is the concentration parameter that should be taken large in order to approximate the dirac delta function in zero as initial data for the forward pide . in this section we perform the non parametric estimation of lvy density distribution function , that is to find the value @xmath406 such that best fits with the given data . we present two validation test cases and one application case to finance . [ [ testing - for - consistency . ] ] testing for consistency . + + + + + + + + + + + + + + + + + + + + + + + + we perform a numerical test on the consistency of our estimation procedure . according to the maximum likelihood technique , consistency here means that , if we fix a parametrization @xmath44 and the parameter values , we can ( approximately ) reconstruct these values form our estimation procedure and maximization of the aic , provides that a sufficiently large sample from the true distribution is given . we simulate such a sample using pseudo random realizations for the lvy process @xmath108 . details on the simulation methods can be found e.g. in @xcite . however note that in our case , cyclic boundary conditions have to be taken into account . the data setting for our test case is as follows : the space domain @xmath407 , the final time @xmath408 , the initial von mises distribution has center @xmath409 and wideness @xmath410 , the drift of the stochastic process is @xmath411 and the gaussian volatility is @xmath412 . the setting for the numerical solution is : space grid size @xmath413 , time grid size @xmath414 . the setting for the optimization is critical , we found the following parameters by the experience : initial approximation of the parameter rates @xmath415 , constant of the armijo condition @xmath416 , initial step - length of point 2 . of algorithm [ alg : nlcg ] is set to @xmath417 and shrink by a factor @xmath418 , @xmath419 . as a first test , we perform a fit for a set of @xmath420 values generated by a monte carlo algorithm for a simulated lvy process on the circle , with the following five values of the jump rates : @xmath421 . we solve the fitting problem , i.e. calculating the estimates to @xmath422 , for different numbers of interpolatory functions : @xmath423 . the center @xmath424 of the basis functions @xmath425 are equally spaced in the domain @xmath426 at the places @xmath427 , @xmath428 , @xmath429 , this means the basis functions do not cover all the domain @xmath430 . in the following table the calculated value of @xmath431 for each problem are reported versus @xmath432 l|ccccc & @xmath433 & @xmath434 & @xmath435 & @xmath436 & @xmath437 + + @xmath438 & 3.4502 & 3.1771 & 2.9746 & 2.8580 & 2.8452 + @xmath439 & 1.1089 & 1.5577 & 1.8100 & 1.9922 & 2.1003 + @xmath440 & 0.4505 & 0.6576 & 1.0198 & 1.3025 & 1.5083 + @xmath441 & & 0.3362 & 0.4951 & 0.7607 & 1.0077 + @xmath442 & & & 0.2490 & 0.3946 & 0.6137 + @xmath443 & & & & 0.2042 & 0.3428 + @xmath444 & & & & & 0.1847 we see the good match for @xmath435 with the original rates @xmath50 . in figs . [ fig : data_fit3],[fig : data_fit5 ] and [ fig : data_fit6 ] we can also appreciate the good data fitting between the calculated pdf and the histograms of the simulated monte carlo data , for the proposed optimization problem with @xmath445 . rates . histograms : experimental lvy data collected in @xmath446 bins . solid line calculated pdf . right . dashed line calculated pdf with the original @xmath447 rates.,title="fig:",scaledwidth=45.0% ] rates . histograms : experimental lvy data collected in @xmath446 bins . solid line calculated pdf . right . dashed line calculated pdf with the original @xmath447 rates.,title="fig:",scaledwidth=45.0% ] rates . histograms : experimental lvy data collected in @xmath446 bins . solid line calculated pdf . right . dashed line calculated pdf with the original @xmath447 rates.,title="fig:",scaledwidth=45.0% ] rates . histograms : experimental lvy data collected in @xmath446 bins . solid line calculated pdf . right . dashed line calculated pdf with the original @xmath447 rates.,title="fig:",scaledwidth=45.0% ] rates . histograms : experimental lvy data collected in @xmath446 bins . solid line calculated pdf . right . dashed line calculated pdf with the original @xmath447 rates.,title="fig:",scaledwidth=45.0% ] rates . histograms : experimental lvy data collected in @xmath446 bins . solid line calculated pdf . right . dashed line calculated pdf with the original @xmath447 rates.,title="fig:",scaledwidth=45.0% ] another interesting problem is the selection of the number of parameters @xmath432 and the corresponding basis functions @xmath448 for the best data fit . in fig . [ fig : aic_test ] we depict the result of the akaike s information criterion ( aic ) @xcite , given by @xmath449 a common choice in statistics is to pick that parametrization that maximises the aic . we can see that criterion gives the value @xmath450 , while the correct value is @xmath447 . the difference in the aic is however rather small for @xmath432 between 5 and 7 . [ [ fitting - data - from - a - bi - directional - gamma - process . ] ] fitting data from a bi - directional gamma process . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in the second test we fit the final position at @xmath408 of @xmath451 samples of a stochastic process with the jumps distributed according a bi - directional gamma process with lvy measure @xmath15 on @xmath452 given by the density @xcite @xmath453 here @xmath454 is the so - called _ shape _ parameter and @xmath455 is the _ rate _ note that this is not a finite measure , so we are out of the compound poisson class , and the trajectory of the bidirectional gamma process as infinitely many ( small ) jumps . in @xcite only the unidirectional gamma process is described . let @xmath456 be such a unidirectional gamma process , then the lvy measure is @xmath457 let thus @xmath456 and @xmath458 be two independent copies of the gamma process , then @xmath459 is our bi - directional gamma process , which is the jump part of our lvy process that also includes diffusion as in the first experiment . if we project @xmath2 to the torus @xmath460 $ ] , the effect on the projected levy measure @xmath115 , see ( [ eqa : wrappedlevymeasure ] ) , of the projected lvy process @xmath106 is @xmath461 using @xmath462 with @xmath463 the lerch transcendent , for @xmath464 and @xmath465 , we get @xmath466 as the lvy measure on the torus for the projected bi - directional gamma process . in the simulations data are generated by scaling the rate parameter such that @xmath467 , and setting the shape to @xmath468 . finally , the data @xmath2 are projected to the torus @xmath469 . in fig . [ gamma_fit ] we report the result of the aic test the fit with @xmath470 basis functions centered to @xmath471 , whose the calculated rates are @xmath472 . we conclude that our procedure results in high quality fits , even for lvy distributions that are not part of our hierarchy of parametrizations , but can only be approximated by these . basis functions.,title="fig:",scaledwidth=45.0% ] basis functions.,title="fig:",scaledwidth=45.0% ] [ [ financial - data . ] ] financial data . + + + + + + + + + + + + + + + as an example from a real world problem we report the result of fitting of the german stock exchange ( dax ) index , which is publicly available , e.g. from the yahoo finance , see figure [ dax_part ] left panel . within the data of all closing quotes between april 1998 and march 2015 , there are several periods of elevated volatility , so called volatility bursts . obviously , this contradicts the description of the market with an exponential levy model @xmath473 @xcite , as the statistical law of @xmath3 does not only depend on @xmath5 . in order to avoid the pitfalls of time dependent ( or stochastic @xcite ) volatility , we identify a period of comparatively stable volatility of 1000 trading days between april 1998 and february 2002 , see the right panel of figure [ dax_part ] . this data set has a small drift value @xmath474 which corresponds to the increased value of the stocks of 67% nominal interest rate in 1000 trading days ( followed by severe losses in the subsequent period ) . the empirical daily volatility ( i.e. standard deviation of daily log - returns ) in this period of time is @xmath475 . the obvious absence of axial symmetry prohibits a gaussian ( black - scholes ) market model from the outset . our goal is to find a suitable description of this sample with an exponential lvy market model from our hierarchy of parametrizations . the data has been mapped to the @xmath460 $ ] torus , by rescaling and wrapping the daily log - returns below / above 3% , i.e. @xmath72=[-0.03,0.03]$ ] . three data sets , all of them negative , were situated outside this band.=[-0.05,0.05]$ ] , see section [ sec : reg ] . at the present stage , it is however not the intention of this work to provide a ready to use basis for risk estimates for financial applications . ] the drift value is adapted to a re - scaled torus @xmath469 . thus the drift on the torus of length @xmath476 is @xmath477 . one fourth of the total empirical variance @xmath478 is attributed for the fixed diffusion , compare section [ sec : reg ] , which yields a coefficient for the laplace operator equals to @xmath479 on the torus rescaled to @xmath469 . with this setting we calculated the fitting of the distribution , with equally spaced basis functions in the interval @xmath469 . in fig . [ dax_fit ] ( left panel ) we report the result of the aic test and the fit with @xmath480 ( right panel ) . the selected basis functions are centred at @xmath481 , whose the calculated rates are @xmath482 . although only three parameters are different from zero , the aic is maximized at @xmath436 . that the aic at @xmath433 is lower is explained by the fact , that the more localized basis functions in the @xmath436-basis are more adequate to fit the data . it is also a misinterpretation that the chosen parametrization misses an effective description with three parameters , since the position of the grid points are additional parameters . note that the zero entries of the 1st , 2nd and 6th slot @xmath148 actually correspond to small positive values and only represented as zero when rounded to the 3rd digit . basis functions , corresponding to the maximal aic.,title="fig:",scaledwidth=45.0% ] basis functions , corresponding to the maximal aic.,title="fig:",scaledwidth=45.0% ] in the present article , we demonstrate the use of optimal control for pide for the non parametric estimation of lvy processes . here the pide is given by kolmogorov s forward equation which allows one to calculate the terminal distribution of the lvy process at time @xmath0 . the objective functional is the log - likelihood evaluated on a sample of terminal values of the lvy process . based on the study of lvy distributions , we set up approximate estimation problems that can be tackled by maximum likelihood estimation along with model selection based on akaike s information criterion ( aic ) . as the density of lvy probability distributions in most cases can not be determined analytically , numerical solutions of the kolmogorov forward equation ( fokker - planck equation ) and its backward ( adjoint ) analogue are needed for the efficient maximization of the log - likelihood functional . for the numerical solution of the optimality system we used the chang - cooper method with a mid - point quadrature rule and second order backward time differentiation formula . this numerical scheme is second order accurate and conservative , and we found conditions for stability and positivity of the numerical solution . we use a non linear conjugate gradient method to find the optimality condition . we have shown that this method works for spline discretizations of the density of the lvy measure with symmetric boundary conditions for up to 11 parameters . the results consistently fit simulated data from the family of discretizations itself . the same turns out to be true from lvy processes that only can be approximated by such discretizations , if the number of parameters goes to infinity , like the gamma process . here the aic provides an effective mechanism to choose an adequate discretization at a given sample size . finally , we have demonstrated that also real - world , financial data can be effectively fitted using our strategy . the future potential of this solution lies in the fact that , unlike fft / spectral based calibration procedures that are widely used in financial engineering @xcite , the present approach naturally generalizes to processes that originate as the solution of stochastic differential equations ( sde ) with state dependent coefficients . such local volatility models are frequently used in contemporary financial engineering . in this work , we used historic and low frequency data for non parametric model calibration . high frequency historical data and implicit volatility data @xcite are natural candidates to set up new objective functionals for related control problems that go beyond the control of the terminal distribution . another relevant problem is the notorious occurrence of local minima in the maximum likelihood estimation . we expect this to be more severe , when the number of parameters significantly increases . an interesting hybrid approach would combine the robustness of non - parametric spectral calibration methods as a sort of pre - conditioner with the highly efficient maximum likelihood estimation .

scattering problems for periodic structures have been studied a lot in the past few years . a main idea for numerical solution methods is to reduce such problems to one periodicity cell . in contrast to periodic settings , scattering from locally perturbed periodic surfaces is way more challenging . in this paper , we introduce and analyze a new numerical method to simulate scattering from locally perturbed periodic structures based on the bloch transform . as this transform is applied only in periodic domains , we firstly rewrite the scattering problem artificially in a periodic domain . with the help of the bloch transform , we secondly transform this problem into a coupled family of quasiperiodic problems posed in the periodicity cell . a numerical scheme then approximates the family of quasiperiodic solutions ( we rely on the finite element method ) and backtransformation provides the solution to the original scattering problem . in this paper , we give convergence analysis and error bounds for a galerkin discretization in the spatial and the quasiperiodicity s unit cells . we also provide a simple and efficient way of implementation that does not require numerical integration in the quasiperiodicity , together with numerical examples for scattering from locally perturbed periodic surfaces computed by this scheme .

1611.06360

in this paper , we present a numerical method for solving scattering problems from locally perturbed periodic surfaces . scattering problems for periodic or quasiperiodic incident fields from periodic structures have been well studied over at least 25 years . the common way of solving is reduction to one periodicity cell , which avoids the need for computing numerical solutions in unbounded domains . however , if such reduction fails due to non - periodicity of the incident field or the surface , one needs to seek for new approaches . the approach we present in this paper is based on the floquet - bloch transform . it builds up a relationship between a non - periodic problem and a family of quasiperiodic problems reduced to one single period . with this transform , the scattering problems from periodic surfaces and non - periodic incident fields have been discussed in @xcite and @xcite . based on these theoretic results , a numerical scheme has been developed to solve these kinds of scattering problems in @xcite . following this type of technique , we introduce in this paper an algorithm for solving scattering problems from local perturbations of periodic surfaces that is pretty close to the one from the recent paper @xcite . our convergence analysis is for various reasons different , as @xcite for instance strongly relies on integral equations in the spatial variable . the source of inspiration for all these techniques seems to be the paper @xcite on wave propagation in full - space periodic media . to briefly present our numerical approach in some detail , we firstly rely on the floquet - bloch transform , defined on functions living in periodic domains , and hence reformulate the locally perturbed problem by a suitable diffeomorphism between the locally perturbed and the purely periodic domain . applying the bloch transform to the new problem yields a family of quasiperiodic scattering problems posed in one single periodicity cell . we state the classic error analysis for finite element discretizations using low - order approximation in @xmath0 . the interesting feature of this discretization is that all integrals in the quasiperiodicity parameters can be computed by hand , such that standard solvers become attractive to tackle the full problem . by finite element discretizations for the spatial parts of the problem and , roughly , the trapezoidal rule discretizing the inverse bloch transform , one gets a large but sparse block - linear system to solve . to this end , we use the gmres iteration with a specially designed incomplete @xmath1-decomposition as pre - conditioner for the numerical solution of the linear system . for dirichlet scattering problems on perturbed periodic surfaces one can , at least in two dimensions , of course exploit the corresponding numerical convergence theory for boundary integral equation approximations from rough surface scattering theory , see , e.g. , @xcite . there are , however , few methods specifically designed for such locally perturbed periodic scatterers . in @xcite and @xcite , the authors give a method that approximates the dirichlet - to - neumann map on the transparent edges of a periodic waveguide modeled by the helmholtz equation . another method that uses the so - called recursive doubling procedure constructs the sommerfeld - to sommerfeld maps at artificial boundaries of such a waveguide , see @xcite and @xcite . both of these methods are motivated by the infinite half - guide and inspired by the limiting absorption principle . this paper is organized as follows . in section [ se : scatter ] , we describe the direct scattering problem corresponding to a locally perturbed periodic surface . in sections [ se : persca ] and [ se : qpscattperturb ] , we use the bloch transform to obtain an equivalent family of quasiperiodic problems . in section [ se : invbloch ] we give a discrete inverse bloch transform and estimate the finite element method applied to the individual quasiperiodic scattering problems . the numerical implementation for the bloch transform based method is illustrated in section [ se : nummet ] . in the last section [ se : numeg ] , several numerical examples indicate the efficiency of that method . appendix [ se : bloch ] briefly introduces the bloch transform and some of its mapping properties . _ notation : _ we denote quasiperiodic sobolev spaces with regularity @xmath2 and quasiperiodicity @xmath0 by @xmath3 , such that @xmath4 denotes a _ periodic _ sobolev space and _ not _ a space of functions that vanish on some boundary . despite functions in sobolev spaces are merely defined almost everywhere , we usually omit to write this down . moreover , @xmath5 is a generic constant with value that might change from one appearance to the other . in this section , we model scattering from a local perturbation @xmath6 of a periodic surface @xmath7 . suppose @xmath8 is defined by a @xmath9-periodic lipschitz continuous function @xmath10 , i.e. , @xmath11 for any @xmath12 and itself defines the periodic domain @xmath13 above the graph of @xmath14 . the locally perturbed periodic surface @xmath6 is then defined via a second lipschitz continuous function @xmath15 that satisfies @xmath16 \text { and defines } \quad \gamma_{{\mathrm{p}}}:=\{{(y_1,\,\zeta_{{\mathrm{p}}}(y_1))}:\ , y_1\in{\ensuremath{\mathbb{r}}}\}.\ ] ] we further assume without loss of generality that there is @xmath17 such that both lipschitz surfaces @xmath6 and @xmath14 are included in @xmath18 and introduce the perturbed periodic domain @xmath19 . then @xmath20 and @xmath21 as well . for some positive number @xmath22 we further introduce truncated domains @xmath23 note that we have for simplicity assumed that @xmath24 only in @xmath25 ! the scattering problem we consider is described by the helmholtz equation with dirichlet boundary condition for the total wave field @xmath26 , @xmath27 where @xmath28 is the wavenumber and @xmath29 is the incident field . moreover , the scattered field @xmath30 satisfies the so - called angular spectrum representation , @xmath31 here , @xmath32 is the square root extended to the complex plane by a branch cut at the negative imaginary axis ( such that its real part and imaginary part are non - negative for numbers in the upper complex half - plane ) , and @xmath33 is the fourier transform of @xmath34 , i.e. , @xmath35 and extended by density to functions in @xmath36 . thus , we can define the exterior dirichlet - to - neumann map @xmath37 , @xmath38 recall from appendix [ se : bloch ] the spaces @xmath39 and @xmath40 with its subspace @xmath41 of functions that vanish on @xmath14 . the operator @xmath37 is bounded from @xmath42 to @xmath43 for all @xmath44 , see @xcite , and the the variational formulation of - is to find @xmath45 ( that is , more precisely , the restriction of the total wave field to @xmath46 , but we omit this fact from now on ) such that @xmath47 dx-\int_{\gamma_h}t^+(u|_{\gamma_h})\overline{v } \ , ds=\int_{\gamma_h}\left[\frac{\partial u^i}{\partial x_2}-t^+(u^i|_{\gamma_h})\right]\overline{v } \ , ds\ ] ] for all @xmath48 with compact support in @xmath49 . due to @xcite we know that this variational problem is uniquely solvable for all @xmath28 and all bounded anti - linear right - hand sides . [ eq : chand2010 ] for @xmath44 and any incident field @xmath50 , the variational problem possesses a unique solution @xmath51 . the ( floquet-)bloch transform @xmath52 reduces differential equations involving periodicity to , roughly speaking , quasiperiodic scattering problems from the unit cell of the periodic structure . before exploiting this reduction , we need to recall some results on non - perturbed periodic scattering . in this section , all proofs are omitted and we refer to appendix [ se : bloch ] and the references therein . for an incident solution @xmath29 to the helmholtz equation @xmath53 in @xmath54 , the dirichlet scattering problem from the periodic surface @xmath14 defined in the last section is described for the total wave field @xmath55 as in by @xmath56 subject to the radiation condition for the restriction of @xmath57 to @xmath58 for some @xmath59 . the variational formulation of this problem is hence to find a solution @xmath60 to with @xmath61 replaced by @xmath46 , that is , @xmath62 { \,\mathrm{d}x \ , } - \int_{\gamma_h } t^+ [ u ] \big|_{\gamma_h } \overline{v } \ , ds = \int_{\gamma_h } \left [ \frac{\partial u^i}{\partial x_2 } - t^+ [ u^i]\big|_{\gamma_h } \right ] \overline{v } \ , ds\ ] ] for all @xmath63 with compact support in @xmath49 . the bloch transform of a solution @xmath64 to the surface scattering problem involving the periodic surface @xmath14 solves a corresponding quasiperiodic scattering problem . to introduce the corresponding variational formulation , we recall the wigner - seitz cell @xmath65 $ ] of periodicity @xmath9 and the periodic sets @xmath66 as well as sobolev spaces @xmath67 , @xmath68 , and @xmath69 for quasiperiodicity @xmath70 : = ( -\pi/\lambda , \pi/\lambda]$ ] from appendix [ se : bloch ] ; @xmath71 $ ] is the so - called brillouin zone . we first rely on a well - known periodic dirichlet - to - neumann operator @xmath72 on @xmath73 that is continuous from @xmath74 into @xmath68 for all @xmath75 , @xmath76\bigg|_{\gamma_h^\lambda } \quad \text{for } \varphi = \sum_{j\in{\ensuremath{\mathbb{z } } } } \hat{\varphi}(j ) e^{{\mathrm{i}}(\lambda^\ast j - \alpha ) x_1}.\ ] ] obviously , @xmath72 is a periodic version of the operator @xmath37 from . second , we introduce a bounded sesqui - linear form @xmath77 on @xmath78 corresponding to the helmholtz equation with dirichlet boundary condition on @xmath14 , @xmath79 { \,\mathrm{d}x \ , } - \int_{\gamma_h^\lambda } t_\alpha^+ \left ( w|_{\gamma_h^\lambda } \right ) \ , \overline{v}|_{\gamma_h^\lambda } \\ , ds,\ ] ] and state an equivalence result that can be shown along the lines of theorem 9 in @xcite . [ th : equiscalarperio ] suppose @xmath29 belongs to @xmath40 for some @xmath80 . then a function @xmath81 solves if and only if @xmath82 solves @xmath83\ ] ] for all @xmath84 and almost every @xmath85 . the last theorem s assumption that @xmath29 belongs to @xmath40 for some @xmath86 is in two dimensions satisfied , e.g. , for ( non - periodic ) point sources or herglotz wave functions with , roughly speaking , vanishing horizontal part , see @xcite . the periodic scattering problem is always uniquely solvable if , e.g. , @xmath14 is graph of a lipschitz function , see @xcite . [ th : uniquealpha ] if @xmath14 is graph of a lipschitz continuous function , then is solvable for all @xmath87 and the solution operators @xmath88 are uniformly bounded in @xmath85 . the solution @xmath89 does generally not belong to @xmath90 , which follows actually already from @xcite . [ th : exisolscalperio ] assume that @xmath14 is graph of a lipschitz continuous function . if @xmath50 for @xmath91 , then the solution @xmath89 to belongs to @xmath69 and the solution @xmath92 to belongs to @xmath93 . if @xmath94 , then @xmath95 is continuous from @xmath96 into @xmath67 . we merely show the continuity result : reference @xcite states that the solution @xmath64 to belongs to @xmath97 if the incident field decays as indicated for @xmath80 ( even for @xmath98 ) . the transformed solution @xmath99 hence belongs to @xmath69 . if @xmath94 , such functions are continuous in @xmath0 due to sobolev s embedding theorem ( or morrey s estimate ) in one dimension ( see @xcite ) , such that @xmath95 is continuous from @xmath96 into @xmath100 . ( the norm in @xmath67 is simply the norm of @xmath101 ! ) in particular , the evaluations @xmath102 in @xmath67 depend continuously on @xmath0 . now we start to analyze scattering problems from locally perturbed periodic surfaces based on the bloch transform from appendix [ se : bloch ] and our knowledge on quasiperiodic scattering from section [ se : persca ] . as the variational formulation of the locally perturbed periodic surface scattering problem is set in the non - periodic space @xmath103 , we have to transform it into a problem formulated in the periodic domain @xmath46 . to this end , we use the diffeomorphism @xmath104 from @xmath46 into @xmath61 , defined by @xmath105 the support of @xmath106 is contained in @xmath107 as the support of @xmath108 is by assumption included in @xmath109 $ ] , too , see . the transformed total field @xmath110 then satisfies by the transformation theorem the following variational problem in @xmath46 , @xmath111 dx - \int_{\gamma_h } \hspace*{-1 mm } t^+(u_{{\mathrm{t}}}|_{\gamma_h})\overline{v_{{\mathrm{t } } } } \ , ds = \int_{\gamma_h } \hspace*{-1 mm } \left[\frac{\partial u^i}{\partial x_2}-t^+(u^i|_{\gamma_h})\right]\overline{v_{{\mathrm{t } } } } \ , ds\ ] ] for all @xmath112 with compact support in @xmath49 and coefficients @xmath113 \in l^\infty(\omega_h,{\ensuremath{\mathbb{r}}}^{2\times 2}),\\ c_{{\mathrm{p}}}(x)&:=\big|\det\nabla\phi_{{\mathrm{p}}}(x)\big|\in l^\infty(\omega_h).\end{aligned}\ ] ] we reformulate by applying the inverse bloch transform composed with the bloch transform to the weak solution @xmath114 . as @xmath115 outside @xmath61 there holds that @xmath116 and @xmath117 are both supported in @xmath107 , and an explicit computation shows that the bloch transform of @xmath118 equals to @xmath119 in the space @xmath120 , and the bloch transform of @xmath121 is @xmath122 in @xmath123 . ( despite , both functions have compact support in @xmath107 . ) if we assume that the incident field @xmath29 belongs to @xmath40 for some @xmath80 , then the bloch transform @xmath124 belongs to @xmath125 and satisfies for all test functions @xmath126 that @xmath127^{1/2 } \int_{\omega_h^\lambda } ( a_{{\mathrm{p}}}-i)\nabla \big({\mathcal{j}}_\omega^{-1 } w_{{\mathrm{b}}}\big ) \cdot \nabla \big(\overline{{\mathcal{j}}_\omega^{-1 } v_{{\mathrm{b } } } } \big)\ , dx \\ & - k^2 \left[\frac{\lambda}{2\pi}\right]^{1/2}\int_{\omega_h^\lambda } ( c_{{\mathrm{p}}}-1 ) { \mathcal{j}}_\omega^{-1 } w_{{\mathrm{b}}}\ , \overline{{\mathcal{j}}_\omega^{-1 } v_{{\mathrm{b } } } } \ , dx = \int_{{w_{\hspace*{-1pt}\lambda^\ast}}}\int_{\gamma_h^\lambda } f(\alpha,\cdot ) \ , \overline{v_{{\mathrm{b}}}}(\alpha,\cdot ) \ , ds \ , d\alpha , \end{split}\ ] ] for the right - hand side @xmath128 with @xmath129 given by @xmath130.\ ] ] the corresponding coupled strong formulation is @xmath131^{1/2}\big[\nabla\cdot\left[(a_{{\mathrm{p}}}-i)\nabla \big({\mathcal{j}}_\omega^{-1 } w_{{\mathrm{b}}}\big ) \right]+k^2(c_{{\mathrm{p}}}-1 ) \big({\mathcal{j}}_\omega^{-1 } w_{{\mathrm{b}}}\big ) \big]\ ] ] with boundary conditions @xmath132 on @xmath133 and @xmath134 = \partial ( { \mathcal{j}}_\omega u^i ( \alpha,\cdot))/\partial x_2 - t^+_{\alpha}\big[({\mathcal{j}}_\omega u^i)(\alpha,\cdot)|_{\gamma^\lambda_h}\big]$ ] on @xmath135 . [ th : equivalent ] assume that the incident field @xmath29 belongs to @xmath40 for some @xmath91 . then @xmath136 satisfies if and only if @xmath137 satisfies . from the arguments before , it is easy to see that if @xmath114 satisfies , then @xmath138 solves . if @xmath139 satisfies , then the property @xmath140 implies that @xmath141 satisfies . we next consider unique solvability of . [ th : regquasi ] if @xmath6 is graph of a lipschitz continuous function , then is uniquely solvable in @xmath69 for all incident fields @xmath50 for @xmath80 . if @xmath6 is graph of a lipschitz continuous function , then @xcite implies that both variational formulations and , equivalently , are uniquely solvable in @xmath97 for incident fields in @xmath50 , @xmath98 . theorem [ th : equivalent ] now implies that is uniquely solvable , too . before we study error estimates for a discretization of the variational formulation of @xmath138 in the next section , we need to show an auxiliary result on the regularity of this solution . [ th : regulh2 ] assume that the restrictions of @xmath142 and @xmath29 to @xmath58 belong to @xmath42 and to @xmath143 for @xmath80 , and that @xmath144 and @xmath145 . then the quasiperiodic solutions @xmath146 to do all belong to @xmath147 and the inverse bloch transformation @xmath148 belongs to @xmath149 . from @xcite we know that the variational problem possesses a unique solution that is bounded in @xmath150 by the norm of @xmath29 in @xmath151 . from the regularity of @xmath144 and @xmath152 , we deduce that @xmath6 is @xmath153-smooth such that elliptic regularity results , see , e.g. , @xcite , imply that both @xmath154 and the restriction of @xmath29 to @xmath58 itself belong actually to @xmath155 . in turn , these regularity results further imply by localization that @xmath156 ( see , e.g. , @xcite ) . thus , @xmath157 belongs to @xmath149 by the @xmath153-smoothness of @xmath104 defined via @xmath152 and @xmath144 in , and its bloch transform belongs to @xmath158 by the mapping properties of @xmath52 : @xmath159 . for almost every @xmath0 , the solution @xmath138 to hence belongs to @xmath147 we finally state an equivalent way of writing if the incident field @xmath29 belongs to @xmath40 for some @xmath160 . [ th : testfuncs ] if @xmath6 is graph of a lipschitz continuous function and if @xmath50 for @xmath160 , then the solution @xmath161 equivalently satisfies for all @xmath162 and all @xmath163 that @xmath164^{1/2 } \int_{\omega_h^\lambda } ( a_{{\mathrm{p}}}-i)\nabla \big({\mathcal{j}}_\omega^{-1 } w_{{\mathrm{b}}}\big ) \cdot \nabla \overline{v_\alpha } \ , dx \\ & - \left[\frac{\lambda}{2\pi}\right]^{1/2 } k^2 \int_{\omega_h^\lambda } ( c_{{\mathrm{p}}}-1 ) { \mathcal{j}}_\omega^{-1 } w_{{\mathrm{b}}}\ , \overline{v_\alpha } \ , dx = \int_{\gamma_h^\lambda } f(\alpha,\cdot ) \ , \overline{v_\alpha } \ , ds . \nonumber\end{aligned}\ ] ] reference @xcite states that the solution @xmath64 to belongs to @xmath40 if the incident field decays as indicated for @xmath80 ( even for @xmath165 ) . as @xmath104 merely modifies @xmath64 in a bounded region , the transformed field @xmath157 decays with the same rate as @xmath166 and @xmath124 hence belongs to @xmath167 for @xmath168 . such functions are continuous in @xmath0 with values in @xmath101 due to sobolev embeddings ( or morrey s estimate ) in one dimension ( see @xcite ) . thus , we can test equation by dirac distributions in @xmath169 , multiplied by functions @xmath170 , to get that @xmath138 solves for each @xmath171 . in turn , if @xmath138 satisfies the latter ( infinite number of ) equations , constructing a complete countable family of test functions in @xmath172 shows that @xmath138 solves as well . ( to this end , one takes an orthonormal family of the separable hilbert space @xmath172 ; separability can be shown by first considering functions that are piecewise constant in @xmath0 and take values in the periodic sobolev functions @xmath173 , and , second , multiplication of these functions by @xmath174 . ) in this section , we discuss a galerkin discretization of the variational formulation of @xmath138 together with an error estimate for the solution to the discretized problem . of course , this makes it necessary to introduce a suitable finite element space first . we actually chose the simplest type of ( nodal ) elements , which is not crucial but avoids technicalities . ( for instance , when using periodic boundary integral equations instead , we would need to take care of exceptional wave numbers where uniqueness of solution fails . ) we assume hence to know a family of regular and quasi - uniform meshes @xmath175 , @xmath176 , of the domain @xmath107 such that for each mesh width @xmath177 the nodes on the right and left boundary of @xmath107 have the same height . this in particular ensures that piecewise linear and globally continuous functions on that mesh can be extended to periodic functions on a regular and quasi - uniform mesh of @xmath46 . to construct such periodic functions we omit now all nodal points on the left boundary of @xmath107 , denote the piecewise linear and globally continuous nodal functions equal to one at exactly one of the remaining nodes and zero at all others by @xmath178 , and denote the discrete subspace spanned by these functions by @xmath179 . it is well - known ( see , e.g. , @xcite ) that for functions @xmath180 there holds @xmath181 to introduce our finite element space , we introduce uniformly distributed grid points @xmath182 and consider a basis @xmath183 of the space of functions that are piecewise constant on each interval @xmath184 $ ] for @xmath185 such that @xmath186 equals one on the @xmath187th interval and zero else . the finite element space @xmath188 we consider is spanned by products of these two bases , multiplied by @xmath189 , @xmath190 it is easy to see that functions in @xmath188 indeed belong to @xmath172 : without the exponential factor in they are clearly periodic in @xmath0 and in @xmath191 as functions defined in @xmath192 ; further , multiplication by @xmath189 implies for @xmath193 and @xmath194 that @xmath195 such that @xmath196 is @xmath0-quasiperiodic and hence belongs to @xmath67 . introduce now , abstractly , the sesqui - linear form @xmath197^{1/2}\int_{\omega_h^\lambda}\left[(a_{{\mathrm{p}}}-i)\nabla { w}\cdot\overline{\nabla v}-k^2(c_{{\mathrm{p}}}-1 ) \ , w \overline{v}\right ] dx \quad \text{on } h^1(\omega_h^\lambda ) \times h^1(\omega_h^\lambda).\ ] ] for the boundary term @xmath198 in @xmath199 from , we now seek a finite element solution @xmath200 to the finite - dimensional problem @xmath201 for all @xmath202 . as functions in @xmath188 are for fixed @xmath203 piecewise exponential in @xmath0 on each interval @xmath204 $ ] , the inverse bloch transform in the latter problem can be explicitly computed : @xmath205^{1/2 } \sum_{j=1}^n \int_{\alpha_n^{(j)}-\pi/(n\lambda)}^{\alpha_n^{(j)}+\pi/(n\lambda ) } w_{n , h}(\alpha , x ) { \,\mathrm{d}\alpha \ , } \nonumber \\ & = \left [ \frac{\lambda}{2\pi } \right]^{1/2 } \sum_{j=1}^n \sum_{\ell=1}^m v_{n , h}^{(j,\ell ) } { \varphi}_{m}^{(\ell)}(x ) \int_{\alpha_n^{(j)}-\pi/(n\lambda)}^{\alpha_n^{(j)}+\pi/(n\lambda ) } e^{-{\mathrm{i}}\alpha x_1 } { \,\mathrm{d}\alpha \ , } \nonumber \\ & = \left [ \frac{\lambda}{2\pi } \right]^{1/2 } \sum_{j=1}^n g_n^{(j ) } ( x_1 ) \sum_{\ell=1}^m v_{n , h}^{(j,\ell ) } { \varphi}_{m}^{(\ell)}(x ) = : { \mathcal{j}}_{\omega , n}^{-1 } \big ( \ { w_{n , h } ( \alpha_n^{(j)},\cdot)\}_{j=1}^n \big ) \label{eq : aux462}\end{aligned}\ ] ] where @xmath206 / x_1 \quad \text { if $ x_1 \not = 0$,}\ ] ] and @xmath207 . note that hence defines a numerical approximation @xmath208 to the inverse bloch transform that we rely on in our numerical examples later on . [ th : numinvbloch ] assume that @xmath209 for @xmath210 and that @xmath144 and @xmath152 are @xmath153 diffeomorphisms . then the linear system is uniquely solvable in @xmath188 for any right - hand side @xmath211 if @xmath212 is large enough and @xmath213 is small enough . the solution @xmath214 satisfies the error estimate @xmath215 \(a ) despite we have explicitly introduced the finite dimensional space @xmath188 via piecewise linear and globally continuous functions on a mesh of @xmath107 , theorem [ eq : erroru ] holds for any family of finite - dimensional spaces that satisfies . of course , is also uniquely solvable for any other continuous linear form on @xmath69 as right - hand side . + ( b ) the assumption of theorem [ th : numinvbloch ] for @xmath29 applies for instance if @xmath29 is the dirichlet green s function of the half space for all @xmath216 , see @xcite . the proof exploits the regularity result in theorem [ th : regulh2 ] stating that @xmath217 . the latter function is continuous in @xmath0 by theorem [ th : testfuncs ] . solvability of the given discretized sesqui - linear problem that features a continuous sesqui - linear form that satisfies a grding s inequality as well as an injectivity condition is due to basic finite elements theory , see , e.g. , @xcite . the indicated error bound follows from the corresponding standard convergence estimate @xmath218 to prove the additional @xmath219-estimate , we need to consider the adjoint problem to find @xmath220 such that , for some given @xmath221 , there holds @xmath222 for all @xmath223 . conjugating the entire latter equation obviously yields a fredholm problem such that is suffices to show uniqueness of solution to deduce existence of solution . if @xmath224 annihilates the latter sesqui - linear form , @xmath225 we deduce that @xmath226 solves as well the homogeneous primal problem and hence vanishes by uniqueness of the primal problem ( see theorem [ th : regquasi ] ) . the arguments proving the @xmath227-regularity estimate for the solution @xmath138 from theorem [ th : regulh2 ] directly transfer to the solution @xmath228 to the adjoint problem , such that there is @xmath229 with @xmath230 . recall that the difference @xmath231 of the solutions to the continuous and discretized primal problem satisfy galerkin orthogonality , @xmath232 for all elements @xmath202 of the discretization space . this shows that @xmath233 for all @xmath202 . if we choose @xmath234 as the orthogonal projection of @xmath228 onto @xmath235 , then firstly @xmath236 together with , this estimate secondly implies that @xmath237 holds for all @xmath221 with norm equal to one . in consequence , theorem [ th : regulh2 ] and imply that @xmath238 in this section , we describe the numerical implementation of the variational problem in detail . for convenience , we solve for the scattered field instead of for the total field and further periodize all quasiperiodic functions , such that the sesqui - linear forms will become @xmath0-dependent instead of the function spaces . recall that the scattered field @xmath239 belongs to @xmath240 and satisfies the variational problem @xmath241^{1/2 } \int_{\omega_h^\lambda } ( a_{{\mathrm{p}}}-i)\nabla \big({\mathcal{j}}_\omega^{-1 } w^s \big ) \cdot \nabla \big(\overline{{\mathcal{j}}_\omega^{-1 } v_{{\mathrm{b } } } } \big)\ , dx \\ & - k^2 \left[\frac{\lambda}{2\pi}\right]^{1/2}\int_{\omega_h^\lambda } ( c_{{\mathrm{p}}}-1 ) { \mathcal{j}}_\omega^{-1 } w^s \ , \overline{{\mathcal{j}}_\omega^{-1 } v_{{\mathrm{b } } } } \ , dx = 0 \end{split}\ ] ] for all @xmath242 , together with the variationally formulated boundary conditions @xmath243 \overline{t}(\alpha,\cdot ) \ , dx \ , d\alpha = 0 \quad \text { for all } t \in l^2({{w_{\hspace*{-1pt}\lambda^\ast}}};\ , h^{-1/2}_\alpha(\gamma)).\ ] ] we next periodize all functions involved in the latter formulation , that is , we introduce @xmath244 such that @xmath245 and @xmath246 in @xmath247 are for fixed @xmath0 two @xmath9-periodic functions in @xmath191 . further , @xmath248 belongs to @xmath249 . as the gradient @xmath250 transforms to @xmath251 , the variational problem for @xmath252 equivalently reformulates for @xmath245 as @xmath253 dx - \int_{\gamma^\lambda_h}{\widetilde}{t}_\alpha^+(w_{0}|_{\gamma^\lambda_h})\overline{v_{0 } } \ , ds \right ] d\alpha \\ & \qquad + \left[\frac{\lambda}{2\pi}\right]^{1/2 } \int_{\omega_h^\lambda } ( a_{{\mathrm{p}}}-i)\nabla { \mathcal{j}}_\omega^{-1 } ( \exp(-{\mathrm{i}}\alpha \ , \cdot ) w_{0 } ) \cdot \overline{{\mathcal{j}}_\omega^{-1 } \nabla(\exp(-{\mathrm{i}}\alpha \ , \cdot ) v_{0 } ) } \ , dx \\ & \qquad \qquad - k^2 \left[\frac{\lambda}{2\pi}\right]^{1/2 } \int_{\omega_h^\lambda } ( c_{{\mathrm{p}}}-1 ) { \mathcal{j}}_\omega^{-1 } ( \exp(-{\mathrm{i}}\alpha \ , \cdot ) w_{0 } ) \ , \overline{{\mathcal{j}}_\omega^{-1 } ( -\exp({\mathrm{i}}\alpha \ , \cdot ) v_{0 } ) } \,dx = 0 \end{aligned}\ ] ] for all @xmath254 and @xmath255 \overline{t_{0}}(\alpha , x)\ , ds \ , d\alpha = 0\ ] ] for all @xmath256 . in , the modified dirichlet - to - neumann map @xmath257 is defined on periodic functions @xmath258 in @xmath259 by @xmath260\bigg|_{\gamma_h^\lambda}.\ ] ] for simplicity , we introduce short - hand notation for , writing @xmath261^{1/2 } \int_{\omega_h^\lambda } \left[(a_{{\mathrm{p}}}-i)\nabla { \mathcal{j}}_\omega^{-1 } ( \exp(-{\mathrm{i}}\alpha \ , \cdot)w ) \cdot \nabla \overline{{\mathcal{j}}_\omega^{-1}(\exp(-{\mathrm{i}}\alpha \ , \cdot ) v ) } \right . \\ & \qquad\qquad\qquad - \left . k^2(c_{{\mathrm{p}}}-1 ) { \mathcal{j}}_\omega^{-1 } ( \exp(-{\mathrm{i}}\alpha \ , \cdot)w ) \ , \overline{{\mathcal{j}}_\omega^{-1}(\exp(-{\mathrm{i}}\alpha \ , \cdot ) v ) } \right ] dx \nonumber\end{aligned}\ ] ] for arbitrary @xmath262 and abbreviate the term inside the @xmath0-integral in the first line in by @xmath263 . then reads @xmath264 let us emphasize that @xmath265 implements the coupling due to the perturbation of the periodic surface between the different quasiperiodic components of the bloch transformed solution . we next discretize the latter family of problems by finite elements and recall from the definition of the finite - dimensional approximation space @xmath188 in the piecewise constant set of functions @xmath266 in @xmath0 , as well as the piecewise linear and globally continuous nodal basis @xmath267 of the finite - dimensional approximation space @xmath179 of basis functions that vanish on @xmath268 . we now introduce a larger approximation space @xmath269 with @xmath270 spanned by _ all _ basis functions defined on the mesh , i.e. , also those that do not vanish on @xmath268 , such that @xmath271 for @xmath272 and @xmath273 if we denote the nodes of the mesh defining @xmath274 by @xmath275 , then @xmath276 is piecewise linear on each triangle of the mesh and satisfies @xmath277 for @xmath278 . in particular , the mesh functions @xmath279 are linked to the nodes @xmath280 that are contained by @xmath268 ; these basis functions hence yield the boundary values of a function in @xmath274 . this finite element space @xmath274 then defines @xmath281 as well as subspaces @xmath282 of functions that are constant in @xmath283 $ ] , @xmath284 for @xmath185 , and a corresponding subspace of functions that vanish on @xmath268 , @xmath285 thus , setting @xmath286 we note that the solution @xmath287 to can be represented by a unique element @xmath288 in @xmath289 as @xmath290 . ( here , @xmath291 denotes the first component of the function s argument . ) the inverse bloch transform @xmath292 applied to @xmath293 that implicitly appears in hence equals the numerical inverse bloch transform @xmath208 from , applied to @xmath294 , @xmath295 the latter equation actually shows via how we implement the bloch transforms in the form @xmath265 from in our numerical examples . the galerkin discretization can now be reformulated via the tuple @xmath296 just introduced , @xmath297 for all tuples @xmath298 and @xmath185 . the boundary conditions on @xmath268 can be implemented directly at the nodal points of the finite elements : to this end , consider all nodal points @xmath280 on @xmath268 and denote the corresponding basis function of @xmath274 by @xmath276 , @xmath299 . the first coordinate of these points hence is @xmath300 , such that we impose that @xmath301 for notational simplicity , let us now identify the function @xmath302 , which is by construction supported in @xmath303 \times \overline{\omega_h^\lambda}$ ] , with the element @xmath304 in @xmath305 , and further set @xmath306 as well as @xmath307 then we get the following discrete variational problem for @xmath308 that is equivalent to , @xmath309 and all @xmath310 , together with the constraints @xmath311 for @xmath185 and @xmath312 . the last equation is due to our choice of the finite element space @xmath274 equivalent to . numerically , we actually solve a slightly restructured linear system that relies on a further unknown @xmath313 that equals the ( discrete ) inverse bloch transform of @xmath314 . in our discretization , the constraint @xmath315^{1/2 } \sum_{j=1}^n g_n^{(j ) } ( x_1 ) \ , e^{-{\mathrm{i}}\alpha_n^{(j ) } x_1 } w_{0}^{(j)}(x)\ ] ] is added to the linear system as a constraint for @xmath316 . ( the weight functions @xmath317 have been explicitly defined in . ) to this end , recall the @xmath318 nodal points @xmath319 of the mesh defining the periodic finite element space @xmath274 that are not on @xmath268 , as well as the @xmath320 nodes @xmath280 on @xmath268 . also recall the basis functions @xmath321 of @xmath274 linked to these nodal points by the conditions @xmath322 . abbreviating @xmath323 , the discrete solution @xmath324 with @xmath325 solves the linear system @xmath326^{1/2 } \sum_{j=1}^n g_n^{(j)}\big((x_{m'}^{(m)})_1 \big ) \ , e^{-{\mathrm{i}}\alpha_n^{(j ) } ( x_{m'}^{(m)})_1 } w_{n , h}^{(j , m ) } & = 0 \quad \text{for } m=1 , \dots , m ' . \label{eq : sys}\end{aligned}\ ] ] if we introduce vectors @xmath327 , @xmath328 , and @xmath329 where @xmath330 for @xmath331 and @xmath332 for @xmath333 , then is equivalent to the quadratic matrix - vector equation @xmath334 with complex @xmath335-matrices @xmath336 and @xmath337 defined by @xmath338 for @xmath339 and @xmath340 else , as well as @xmath341 for @xmath342 and @xmath343 else . further , @xmath344^{1/2}\frac{1}{n}\ , \mathrm{diag}\bigg [ g_n^{(j)}\big ( \big ( x_{m'}^{(1 ) } \big)_1 \big ) e^{-{\mathrm{i}}\alpha_n^{(j ) } \big ( x_{m'}^{(1 ) } \big)_1 } , \ , \dots , \ , g_n^{(j ) } \big(\big ( x_{m'}^{(m ' ) } \big)_1\big ) e^{-{\mathrm{i}}\alpha_n^{(j ) } \big ( x_{m'}^{(m ' ) } \big)_1}\bigg]\ ] ] for @xmath345 . of course , after solving this linear system , we have already computed the discrete inverse bloch transform @xmath346 of the individual solutions @xmath347 via @xmath348 . multiplying @xmath349 by @xmath189 yields an approximation in @xmath107 to the transformed solution @xmath157 to the surface scattering problem from the locally perturbed periodic surface @xmath6 we considered originally . consequently , @xmath64 itself can on @xmath107 be approximated via the formula @xmath350 . when solving the large linear system of size @xmath351 , one needs to employ an iterative method , as direct solvers become inefficient ( at least without parallelization ) . we chose the gmres iteration as solution method and pre - condition it in two steps : 1 . construct an incomplete lu decomposition @xmath352 for each @xmath335-matrix @xmath336 , for @xmath353 . define the lower triangular matrix @xmath354 and the upper triangular matrix @xmath355 . 2 . use the gmres iteration procedure with @xmath356 be the pre - conditioner with a tolerance @xmath357 that we typically choose to be @xmath358 . this choice certainly is somewhat preliminary as we did not perform large - scale tests agains other preconditioners , and in particular not against parallelized solvers , to tackle . in this section , we give some numerical results for the above - presented bloch transform based method , discretized in , together with error estimates and computation times to indicate efficiency . we always choose the incident field as the half - space dirichlet s green s function @xmath359 , \quad x \not = y \in { \ensuremath{\mathbb{r}}}^2_+,\ ] ] where @xmath360 is the mirror point of the source point @xmath361 . the source point @xmath362 is in all experiments located below both the periodic and the locally perturbed surfaces on the one hand , and the @xmath191-axis on the other hand . this artificial scattering problem then possesses @xmath363 as an explicit solution , which makes the explicit computation of the error of the resulting solution very simple . recall that @xmath364 is the smooth function defined by @xmath365 when @xmath366 and @xmath367 , fix the period @xmath9 as @xmath368 , set @xmath369 and @xmath370 such that the incident field has the form @xmath371 we give the numerical results for two different periodic surfaces given by @xmath372 for each surface , two perturbations are considered : @xmath373 + 1\right ) \qquad \text{for } -2\leq t\leq 0 \text { and $ 0 $ else , and } \\ g_2(t ) & = \exp\left(\frac{1}{(t+1)(t-1)}\right)\sin\left[\pi(t+1)\right ] \qquad \text{for } -1\leq t\leq 1 \text { and $ 0 $ else.}\end{aligned}\ ] ] the surfaces @xmath374 , @xmath375 , @xmath376 and @xmath377 are four locally perturbed periodic surfaces defined by @xmath378 [ cols="^,^ " , ] for each surface , we chose @xmath379 and evaluated the numerical solutions @xmath380 on @xmath381 for two different source points , i.e. , @xmath382 and @xmath383 and two wave numbers @xmath384 and @xmath385 by numerically solving the linear system by the described preconditioned gmres algorithm . as mentioned , the exact scattered field equals ( minus ) the incident field , which allows to compute relative errors @xmath386 . table [ surf1p1k1 ] and table [ surf1p1k10 ] show the relative errors for the numerical solutions for the surface @xmath374 , table [ surf1p2k1 ] and table [ surf1p2k10 ] shows the results for @xmath375 , table [ surf2p1k1 ] and table [ surf2p1k10 ] show the results for @xmath376 , and the results of @xmath377 are in table [ surf2p2k1 ] and table [ surf2p2k10 ] . for each example , the results are computed for mesh sizes @xmath387 and @xmath388 . @xmath389 & @xmath390 & @xmath391 & @xmath392 + @xmath393&@xmath394e@xmath395&@xmath396e@xmath395&@xmath397e@xmath395&@xmath398e@xmath395&@xmath399e@xmath395 + @xmath400&@xmath401e@xmath395&@xmath402e@xmath395&@xmath403e@xmath395&@xmath404e@xmath395&@xmath405e@xmath395 + @xmath406&@xmath407e@xmath395&@xmath408e@xmath395&@xmath409e@xmath395&@xmath410e@xmath395&@xmath411e@xmath412 + @xmath413&@xmath414e@xmath395&@xmath415e@xmath412&@xmath416e@xmath412&@xmath417e@xmath412&@xmath418e@xmath412 + @xmath419&@xmath420e@xmath395&@xmath421e@xmath412&@xmath422e@xmath412&@xmath423e@xmath412&@xmath424e@xmath412 + @xmath390 & @xmath391 & @xmath392 + @xmath393&@xmath425e@xmath426&@xmath427e@xmath426&@xmath428e@xmath429&@xmath430e@xmath429 + @xmath400&@xmath431e@xmath426&@xmath432e@xmath426&@xmath433e@xmath429&@xmath434e@xmath395 + @xmath406&@xmath435e@xmath426&@xmath432e@xmath426&@xmath436e@xmath429&@xmath437e@xmath395 + @xmath413&@xmath435e@xmath426&@xmath432e@xmath426&@xmath438e@xmath429&@xmath439e@xmath395 + @xmath419&@xmath435e@xmath426&@xmath432e@xmath426&@xmath424e@xmath429&@xmath440e@xmath395 + @xmath389 & @xmath390 & @xmath391 & @xmath392 + @xmath393&@xmath441e@xmath395&@xmath442e@xmath395&@xmath443e@xmath395&@xmath443e@xmath395&@xmath443e@xmath395 + @xmath400&@xmath444e@xmath395&@xmath445e@xmath395&@xmath403e@xmath395&@xmath446e@xmath395&@xmath403e@xmath395 + @xmath406&@xmath447e@xmath395&@xmath448e@xmath395&@xmath409e@xmath395&@xmath449e@xmath412&@xmath450e@xmath412 + @xmath413&@xmath451e@xmath395&@xmath452e@xmath395&@xmath453e@xmath412&@xmath454e@xmath412&@xmath455e@xmath412 + @xmath419&@xmath451e@xmath395&@xmath456e@xmath395&@xmath457e@xmath412&@xmath458e@xmath412&@xmath459e@xmath412 + @xmath390 & @xmath391 & @xmath392 + @xmath393&@xmath460e@xmath426&@xmath409e@xmath426&@xmath461e@xmath429&@xmath462e@xmath429 + @xmath400&@xmath463e@xmath426&@xmath409e@xmath426&@xmath404e@xmath429&@xmath434e@xmath395 + @xmath406&@xmath460e@xmath426&@xmath409e@xmath426&@xmath405e@xmath429&@xmath464e@xmath395 + @xmath413&@xmath460e@xmath426&@xmath409e@xmath426&@xmath436e@xmath429&@xmath465e@xmath395 + @xmath419&@xmath460e@xmath426&@xmath409e@xmath426&@xmath436e@xmath429&@xmath466e@xmath395 + @xmath389 & @xmath390 & @xmath391 & @xmath392 + @xmath393&@xmath467e@xmath395&@xmath468e@xmath395&@xmath469e@xmath395&@xmath470e@xmath395&@xmath470e@xmath395 + @xmath400&@xmath471e@xmath395&@xmath472e@xmath395&@xmath473e@xmath395&@xmath474e@xmath395&@xmath474e@xmath395 + @xmath406&@xmath475e@xmath395&@xmath476e@xmath395&@xmath477e@xmath395&@xmath478e@xmath395&@xmath479e@xmath395 + @xmath413&@xmath480e@xmath395&@xmath481e@xmath395&@xmath482e@xmath412&@xmath483e@xmath412&@xmath484e@xmath412 + @xmath419&@xmath435e@xmath395&@xmath485e@xmath412&@xmath486e@xmath412&@xmath487e@xmath412&@xmath488e@xmath412 + @xmath390 & @xmath391 & @xmath392 + @xmath393&@xmath489e@xmath426&@xmath481e@xmath426&@xmath490e@xmath429&@xmath491e@xmath429 + @xmath400&@xmath492e@xmath426&@xmath481e@xmath426&@xmath461e@xmath429&@xmath493e@xmath429 + @xmath406&@xmath492e@xmath426&@xmath481e@xmath426&@xmath422e@xmath429&@xmath477e@xmath429 + @xmath413&@xmath492e@xmath426&@xmath481e@xmath426&@xmath422e@xmath429&@xmath494e@xmath429 + @xmath419&@xmath492e@xmath426&@xmath481e@xmath426&@xmath424e@xmath429&@xmath494e@xmath429 + @xmath389 & @xmath390 & @xmath391 & @xmath392 + @xmath393&@xmath495e@xmath395&@xmath496e@xmath395&@xmath497e@xmath395&@xmath497e@xmath395&@xmath497e@xmath395 + @xmath400&@xmath498e@xmath395&@xmath499e@xmath395&@xmath500e@xmath395&@xmath473e@xmath395&@xmath500e@xmath395 + @xmath406&@xmath501e@xmath395&@xmath448e@xmath395&@xmath494e@xmath395&@xmath479e@xmath395&@xmath479e@xmath395 + @xmath413&@xmath502e@xmath395&@xmath503e@xmath395&@xmath504e@xmath412&@xmath505e@xmath412&@xmath454e@xmath412 + @xmath419&@xmath502e@xmath395&@xmath506e@xmath395&@xmath507e@xmath412&@xmath458e@xmath412&@xmath508e@xmath412 + @xmath390 & @xmath391 & @xmath392 + @xmath393&@xmath509e@xmath426&@xmath510e@xmath426&@xmath511e@xmath429&@xmath512e@xmath429 + @xmath400&@xmath509e@xmath426&@xmath510e@xmath426&@xmath513e@xmath429&@xmath514e@xmath429 + @xmath406&@xmath515e@xmath426&@xmath510e@xmath426&@xmath516e@xmath429&@xmath517e@xmath429 + @xmath413&@xmath515e@xmath426&@xmath510e@xmath426&@xmath518e@xmath429&@xmath519e@xmath429 + @xmath419&@xmath515e@xmath426&@xmath510e@xmath426&@xmath436e@xmath429&@xmath519e@xmath429 + as is shown in tables 1 - 8 , the relative error decreases in @xmath520 and @xmath177 up to error stagnation . for wave number @xmath384 , the error caused by @xmath520 is the dominant one , such that the error decrease as @xmath177 gets small is sometimes comparatively small , see tables 1 , 3 , 5 , and 7 . for @xmath385 , this is the exact opposite , see tables 2 , 4 , 6 , and 8 . when @xmath177 is small enough ( see the results for @xmath392 in figures 1 , 3 , 5 , 7 ) , the relative error decreases faster than the rate @xmath521 than proved theoretically in theorem [ th : numinvbloch ] . in tables [ surf2p1k1 t ] and [ surf2p1k10 t ] , we also show the computation times of our serial code imlemented in matlab for examples 5 and example 6 computed on a workstation with an intel i7 - 4790 processor ( 8 cores at 3.60 ghz ) and 32 gb ram . this data excludes the smallest mesh size and the largest discretization of the brillouin zone @xmath96 that we merely treated on a comparatively slow workstation with significantly larger memory of 264 gb . finally , we balance the two error terms in the @xmath522-estimate @xmath523 $ ] from by choosing @xmath524 for @xmath525 and @xmath520 equal to @xmath526 . the @xmath527 pairs are hence @xmath528 , @xmath529 and @xmath530 . figure [ fig : rate ] shows plots in logarithmic scale of the relative @xmath522-errors for the eight examples from above . the slopes for examples 1 , 3 , 5 , and 7 is roughly about @xmath531 , for example 2 , 4 , 6 , and 8 they are roughly about @xmath532 . this means that the numerical results converges at the rate of @xmath533 or even faster than shown in theorem [ th : numinvbloch ] . -errors for the eight considered examples with @xmath534 plotted in logarithmic scale over @xmath520.,scaledwidth=60.0% ] @xmath389 & @xmath390 & @xmath391 + @xmath393 & 0.74 & 3.3 & 25 & 277 + @xmath400 & 1.5 & 7.4 & 55 & 580 + @xmath406 & 3.7 & 19 & 126 & 1242 + @xmath413 & 9.3 & 55 & 316 & 2740 + @xmath390 & @xmath391 + @xmath393 & 8.4 & 43 & 341 + @xmath400 & 20 & 95 & 691 + @xmath406 & 46 & 217 & 1500 + @xmath413 & 120 & 531 & 6524 + we briefly recall mapping properties of the ( floquet-)bloch transform @xmath52 ; standard references on this topic are @xcite or @xcite , but see also ( * ? ? ? * annexe b ) . we define that transform on smooth functions @xmath535 with compact support in @xmath536 by @xmath537^{1/2 } \sum_{j\in { \ensuremath{\mathbb{z } } } } u\left ( \begin{smallmatrix } x_1 + \lambda j \\ x_2 \end{smallmatrix } \right ) e^{{\mathrm{i}}\ , \lambda j \ , \alpha } , \qquad x = \left ( \begin{smallmatrix } x_1 \\ x_2 \end{smallmatrix } \right ) \in \overline\omega , \ , \alpha \in { \ensuremath{\mathbb{r}}}.\ ] ] ( the same transform for functions defined in @xmath46 is denoted by @xmath52 as well . ) one easily computes that @xmath538 is @xmath0-quasiperiodic with respect to @xmath9 , @xmath539^{1/2 } \sum_{j\in { \ensuremath{\mathbb{z } } } } u \big ( \begin{smallmatrix } x_1 + \lambda(j+1 ) \\ x_2 \end{smallmatrix } \big ) \ , e^{-{\mathrm{i}}\alpha \cdot \lambda j } = e^{-{\mathrm{i}}\lambda \ , \alpha } { \mathcal{j}}_\omega u(\alpha , x)\ ] ] for @xmath540 . further , @xmath541 is for fixed @xmath203 a fourier series in @xmath0 with basis functions @xmath542 that are @xmath543 periodic . thus , introducing the unit cell and the brillouin zone as @xmath544 \quad \text{and } \quad { { w_{\hspace*{-1pt}\lambda^\ast}}}= \bigg ( -\frac{\lambda^\ast}{2 } , \frac{\lambda^\ast}{2}\bigg ] = \bigg ( -\frac{\pi}{\lambda } , \frac{\pi}{\lambda}\bigg]\ ] ] implies that knowledge of @xmath545 in @xmath546 ( or @xmath547 ) defines the bloch transform @xmath548 everywhere in @xmath192 ( or @xmath549 ) . this observation reflects in mapping properties of the bloch transform . to this end , recall the bessel potential spaces @xmath550 for @xmath75 , together with their weighted analogues @xmath551 for @xmath552 , equipped with their natural norms . the spaces @xmath553 are defined analogously and @xmath554 contains all @xmath555 such that @xmath556 . for @xmath557 we further rely on the subspace of quasiperiodic functions @xmath558 of @xmath559 ( see for a definition of @xmath0-quasiperiodicity ) . this space contains all @xmath0-quasiperiodic distributions @xmath560 with finite norm @xmath561 for @xmath85 , where @xmath562 is the @xmath187th fourier coefficient of @xmath560 ( the dual evaluation between @xmath560 and @xmath563 ) . functions in these spaces can be represented by their fourier series , @xmath564 the analogous spaces for functions defined in @xmath46 are @xmath565 with the usual @xmath566-norm on @xmath46 . next , we consider all distributions in @xmath567 that are @xmath568-periodic in their first variable @xmath0 and quasiperiodic in the first component @xmath191 of their second variable @xmath203 , the quasiperiodicity being equal to the first variable . for integers @xmath569 and @xmath75 , these distributions define norms @xmath570 and hilbert space @xmath571 as set of those distributions with finite @xmath572-norm . interpolation in @xmath573 and a duality argument allow to define these spaces for all @xmath574 . for a regularity result , we actually also require the family of spaces @xmath575 for @xmath576 , that are defined by replacing the @xmath577-norm in @xmath0 by a @xmath578-norm ; the @xmath579th power of the norm of these spaces equals @xmath580 { \,\mathrm{d}\alpha \ , } , \quad 1 \leq p < \infty.\ ] ] for all spaces introduced so far involving @xmath46 , @xmath581 , @xmath54 , or @xmath582 it is convenient to define the closure of smooth functions that vanish in a neighborhood of @xmath14 or @xmath268 in the norms defined above ; the corresponding subspaces are then denoted by @xmath583 , @xmath584 , and so on . [ th : blochomega ] the bloch transform @xmath52 extends to an isomorphism between @xmath554 and @xmath585 as well as between @xmath586 and @xmath587 for all @xmath552 . further , @xmath52 is an isometry for @xmath588 with inverse @xmath589^{1/2 } \int_{{w_{\hspace*{-1pt}\lambda^\ast}}}w(\alpha , x ) e^{i \alpha \ , \lambda j } { \,\mathrm{d}\alpha \ , } \qquad \text{for } x \in \omega_h^\lambda.\ ] ] we actually merely consider the inverse bloch transform in @xmath107 , where the exponential factor in the latter integral can be omitted . finally , we introduce sobolev spaces on @xmath58 and @xmath590 by identifying @xmath58 with @xmath591 and @xmath592 with @xmath593 . the resulting spaces are then denoted by @xmath594 , @xmath595 , and @xmath596 for @xmath597 . for @xmath598 it is well - known that these are natural trace spaces of volumetric @xmath566-spaces , see @xcite . the second author was supported by the university of bremen and the european union fp7 cofund under grant agreement n@xmath599600411 . a. lechleiter and r. zhang . a convergent numerical scheme for scattering of aperiodic waves from periodic surfaces based on the floquet - bloch transform . technical report , center for industrial mathematics , university of bremen , ` http://arxiv.org/abs/1603.07860 ` , 2016 . a. meier , t. arens , s. n. chandler - wilde , and a. kirsch . a nystrm method for a class of integral equations on the real line with applications to scattering by diffraction gratings and rough surfaces . , 12:281321 , 2000 .

we describe a construction that maps any connected graph on three or more vertices into a larger graph , , whose independence number is strictly smaller than its lovsz number which is equal to its fractional packing number . the vertices of represent all possible events consistent with the stabilizer group of the graph state associated to , and exclusive events are adjacent . mathematically , the graph corresponds to the orbit of under local complementation . physically , the construction translates into graph - theoretic terms the connection between a graph state and a bell inequality maximally violated by quantum mechanics . in the context of zero - error information theory , the construction suggests a protocol achieving the maximum rate of entanglement - assisted capacity , a quantum mechanical analogue of the shannon capacity , for each . the violation of the bell inequality is expressed by the one - shot version of this capacity being strictly larger than the independence number . finally , given the correspondence between graphs and exclusivity structures , we are able to compute the independence number for certain infinite families of graphs with the use of quantum non - locality , therefore highlighting an application of quantum theory in the proof of a purely combinatorial statement .

1211.4250

partitioning a phase space into orbits is a central step in the study of any physical or formal dynamics . immediately after the introduction of graph ( /stabilizer ) states in quantum coding theory @xcite and in the context of measurement - based quantum computation @xcite , it was evident that the related task , when considering the dynamics at the subsystems level , requires approaches of combinatorial flavour . while substantial attention has been given to orbits obtained by the application of local unitaries ( with a clear motivation coming from the classification of multipartite entanglement @xcite ) , the question to decide whether two graph states are equivalent under the action of the local clifford group has been settled , by showing @xcite that the equivalence classes are in one - to - one correspondence with local complementation orbits ( also called _ kotzig orbits _ @xcite ) . in other words , the existence of a sequence of local complementations relating the associated graphs guarantees equivalence under local clifford operations and _ viz_. even though this link does not embrace full local unitary equivalence , having now counterexamples to the lu - lc conjecture ji10 , sarvepalli10 , it unveils a rich interface between the structure of useful multi - qubit systems and a number of mathematical ideas . indeed , local complementation ( or , equivalently , @xmath2-transformation ) is a fundamental operation for studying circle graphs @xcite . this notion has been instrumental for unifying certain properties of eulerian tours and matroids via isotropic systems @xcite , constructs associated to vector spaces over @xmath3 ; and it appears in string reconstruction problems ( related to dna sequencing ) and graph polynomials @xcite . given an equivalence class induced by local complementation , in the present work we shall describe a method for constructing a larger graph associated to the equivalence class . the method makes use of the stabilizer group of an arbitrary graph state from the class . each of these graphs is identified with an exclusivity structure and a related _ non - contextuality inequality _ ( for short , nc inequality ) . such an inequality is an upper bound on the sum of probabilities of a set of events , with some exclusivity constraints ( a technical discussion about events and exclusivity will be given in section [ con ] ) . nc inequalities are satisfied by any non - contextual hidden variable theory , _ i.e. _ , any physical theory for which the probability of seeing an event is independent of the choice of measurements . quantum mechanics or more general theories can violate such inequalities . for more details , see @xcite . in this reference , the graph ( and more generally an hypergraph ) , whose vertices are events , is employed to characterize the correlations for classical and general probabilistic theories satisfying that the sum of probabilities of pairwise exclusive events can not be larger than 1 . the maximum values for the three physical theories : classical , general , and quantum , were computed through the three well - known combinatorial parameters : the independence number , the fractional packing number , and the lovsz number , respectively . as a consequence , it becomes evident that quantum and general probabilistic correlations satisfying that the sum of probabilities of pairwise exclusive events can not be larger than 1 have semidefinite and linear characterizations , respectively . quantum mechanics is _ sandwiched _ between the other two theories . the framework introduced in @xcite permits to quantitatively discuss classical , quantum , and more general theories through the analysis of a single mathematical object and to have a general technique to single out quantum correlations with _ ad hoc _ degree of contextuality . for example , a generic graph with independence number strictly smaller than the lovsz number is associated to a nc inequality violated by quantum mechanics . if , in addition , the graph has equal lovsz and fractional packing number , then it can be associated to a nc inequality that is maximally violated by quantum mechanics , meaning that no general probabilistic theory satisfying that the sum of probabilities of pairwise exclusive events can not be larger than 1 can achieve a larger value . ( of course , there are graphs for which all theories coincide , as , for example , perfect graphs . ) in the present paper , we propose a construction that translates into the combinatorial language developed in @xcite the connection between every graph state of three or more qubits and a bell inequality maximally violated by quantum mechanics found in @xcite . namely , we describe a construction that maps any graph on three or more vertices @xmath0 into a larger graph , @xmath1 , such that its independence number is strictly smaller than its lovsz number which is equal to its fractional packing number . the vertices of @xmath1 represent all possible events consistent with the stabilizer group of the graph state associated to @xmath0 and exclusive events are adjacent . the construction has also applications in zero - error information theory . it leads to a straightforward protocol achieving the maximum rate of zero - error entanglement - assisted capacity @xcite . we conjecture that this quantity for a graph @xmath1 is always strictly larger than its shannon capacity . a proof of this statement would possibly require a rank bound _ a la haemers_. while it is difficult to compute this bound in general , it may be easier in our case , since @xmath1 has a very particular structure because of the connection with the stabilizer group . the violation of the bell inequality is here expressed by the one - shot version of this capacity being strictly larger than the independence number . the correspondence between graphs and exclusivity structures allows us compute the independence number of the graphs @xmath4 , where @xmath5 is the complete graph , by taking advantage of well - known techniques used in quantum non - locality . since two graphs yield the same ( up to isomorphism ) graph in our construction if and only if they are equivalent under local complementation , the construction can be interpreted as a method to represent local complementation orbits . somehow this is in analogy with the notion of a two - graph , a well - studied mathematical object which represents equivalence classes under the operation of switching ( see ch . 11 of @xcite ) . our work is innovative with respect to @xcite and @xcite in many ways . first , we present a characterization of local complementation orbits , a result of pure combinatorial nature . our representative graphs are obtained via a construction inspired by quantum mechanics . we find tools to analyze the properties of such graphs in @xcite . second , by using the results in @xcite , we discover that local complementation orbits are naturally associated to bell inequalities . we improve the mathematical representation of such bell inequalities by writing down an explicit operational form , namely , a _ pseudo - telepathy game_. the form that we have introduced is often easier to use in both theoretical purposes and the design of laboratory experiments . third , we introduce a novel connection between the results presented in @xcite and @xcite , and zero - error information theory . such a connection is also interesting on its own , since it provides a way to construct families of channels with a separation between classical and quantum capacities starting from _ any _ graph . the remainder of the work is organized as follows . the next section introduces the required terminology and notions : the language of graph theory , non - locality , and channel capacities . the construction is described in section 3 . section 4 discusses the relevant graph - theoretic parameters . section 5 contains examples . we highlight that physical arguments can be useful to consider difficult tasks such as computing the independence number . section 6 is devoted to zero - error capacities . we show that our construction produces infinite families of graphs for which the use of entanglement gives the maximum possible zero - error capacity . section 7 classifies graphs ( or local complementation orbits ) according to the objects obtained via the construction . we point out a link with boolean functions and propose a conjecture about connectedness of the graph representatives . a ( simple ) _ graph _ @xmath6 is an ordered pair : @xmath7 is a set whose elements are called _ vertices _ ; @xmath8 is a set whose elements are called _ edges_. the set @xmath9 does not contain an edge of the form @xmath10 , for every @xmath11 . the vertices forming an edge are said to be _ we denote by @xmath12 the _ neighbourhood _ of the vertex @xmath13 , _ i.e. _ , @xmath14 . an _ independent set _ in a graph @xmath0 is a set of mutually non - adjacent vertices . the _ independence number _ of a graph @xmath0 , denoted by @xmath15 , is the size of the largest independent set of @xmath0 . @xmath16 of a graph @xmath6 , is a graph such that @xmath17 and @xmath18 . an _ induced subgraph _ @xmath16 of a graph @xmath6 with respect to @xmath17 is a graph with vertex set @xmath19 and edge set @xmath20 . a _ clique _ in a graph @xmath0 is a subgraph whose vertices are all adjacent to each other . an _ orthogonal representation _ of @xmath0 is a map from @xmath7 to @xmath21 for some @xmath22 , such that adjacent vertices are mapped to orthogonal vectors . an orthogonal representation is _ faithful _ when vertices @xmath23 and @xmath24 are mapped to orthogonal vectors if and only if @xmath25 . the _ lovsz number _ @xcite is defined as follows : @xmath26where the maximum is taken over all unit vectors @xmath27 and all orthogonal representations , @xmath28 , of @xmath0 . the _ fractional packing number _ is defined by the following linear program : @xmath29where the maximum is taken over all @xmath30 under the restriction @xmath31 @xcite , and for all cliques @xmath32 , where @xmath33 is the set of all cliques of @xmath0 . in this paper , by fractional packing number , we mean @xmath34 and denote it as @xmath35 . given a graph @xmath6 , the _ graph state _ @xmath36 ( see , for example , @xcite ) associated to @xmath0 is the unique @xmath37-qubit state such that @xmath38where @xmath39 is the generator labeled by a vertex @xmath40 of the stabilizer group of @xmath36 . a generator for @xmath11 is defined as @xmath41where @xmath42 , @xmath43 , and @xmath44 denote the pauli matrices ( sometimes denoted as @xmath45 , @xmath46 , and @xmath47 ) acting on the @xmath13-th qubit . therefore , @xmath39 can be obtained directly and univocally from @xmath0 . the stabilizer group of the state @xmath36 is the set @xmath48 of the stabilizing operators @xmath49 of @xmath36 defined by the product of any number of generators @xmath39 . note that , for convenience , we shall remove the identity element from @xmath48 . therefore , the set @xmath48 contains @xmath50 elements . given a graph @xmath51 , the operation of _ local complementation _ ( lc ) on @xmath40 transforms @xmath0 into a graph @xmath52 on the same set of vertices . to obtain @xmath52 , we replace the induced subgraph of @xmath0 on @xmath53 by its complement . it is easy to verify that @xmath54 . the set of graphs is partitioned into _ lc orbits _ ( also known as _ kotzig orbits _ ) by the repeated action of local complementation on each graph @xcite . the lc orbits are then equivalence classes . we assume familiarity with the basics of quantum information theory . the reader can find a good introduction in ( * ? ? ? * chapter 2 ) . non - local game _ is an experimental setup between a referee and two players , alice and bob . ( it can also be defined with more players , but we do not consider this case here . ) the game is not adversarial , but the players collaborate with each other . they are allowed to arrange a strategy beforehand , but they are not allowed to communicate during the game . the referee sends alice an input @xmath55 and sends bob an input @xmath56 , according to a fixed and known probability distribution @xmath57 on @xmath58 . alice and bob answer with @xmath59 and @xmath60 respectively , and the referee declares the outcome of the game according to a verification function @xmath61win @xmath62 , lose @xmath63 . so , the non - local game is completely specified by the sets @xmath64 , the distribution @xmath57 , and the verification function @xmath65 . a _ classical strategy _ is _ w.l.o.g . _ a pair of functions @xmath66 and @xmath67 for alice and bob , respectively . a _ quantum strategy _ consists of a shared bipartite entangled state @xmath68 and povms @xmath69 for every @xmath55 for alice and @xmath70 for every @xmath56 for bob . on input @xmath71 , alice uses the positive operator valued measurement ( povm ) @xmath72 to measure her part of the entangled state and bob does similarly on his input @xmath73 . alice ( resp . bob ) answers with @xmath74 ( resp . @xmath75 ) corresponding to the obtained measurement outcome . therefore , the probability to output @xmath76 given @xmath77 is @xmath78 . the classical and quantum values ( or winning probabilities ) for the game are : @xmath79 a _ bell inequality _ for a non - local game is a statement of the form @xmath80 for @xmath81 $ ] . it is violated by quantum mechanics if @xmath82 . a non - local game is called a _ pseudo - telepathy _ game if @xmath83 , _ i.e. _ quantum players win with certainty , while classical players have nonzero probability to lose . non - local games are a special form of bell experiment . in general , a _ bell operator _ @xmath84 is a linear combination of observables and a bell inequality is a statement of the form @xmath85 where the maximum runs over classical states . a quantum state is said to _ violate _ the bell inequality if @xmath86 . zero - error information theory was initiated in @xcite ; a review is @xcite . a _ classical channel _ @xmath87 with input set @xmath88 and output set @xmath89 is specified by a conditional probability distribution @xmath90 , the probability to produce output @xmath73 upon input @xmath71 . ( precisely this is a discrete , memoryless , stationary channel . ) two inputs @xmath91 are _ confusable _ if there exists @xmath56 such that @xmath92 and @xmath93 we then define the _ confusability graph of channel @xmath87 _ , @xmath94 , as the graph with vertex set @xmath88 and edge set @xmath95 . the _ one - shot zero - error capacity _ of @xmath87 , @xmath96 , is the size of a largest set of non - confusable inputs . this is just the independence number @xmath97 of the confusability graph . in the entanglement - assisted setting , the sender ( alice ) and receiver ( bob ) share an entangled state @xmath98 and can perform local quantum measurements on their part of @xmath98 . the general form of an entanglement - assisted protocol used by alice to send one out of @xmath99 messages to bob with a single use of the classical channel @xmath87 can be described as follows ( also see @xcite ) . for each message @xmath100 $ ] , alice has a povm @xmath101 with @xmath102 outputs . to send message @xmath103 , she measures her subsystem using @xmath104 and sends through the channel the observed @xmath55 . bob receives some @xmath56 with @xmath92 . if the right condition holds ( as we will explain below ) , bob can recover @xmath103 with certainty using a projective measurement on his subsystem . it is not hard to state a necessary and sufficient condition for the success of the protocol . if alice gets outcome @xmath55 upon measuring @xmath104 , bob s part of the entangled state collapses to @xmath105 . given the channel s output @xmath73 , bob can recover @xmath103 if and only if @xmath106 bob can recover the message with a projective measurement on the mutually orthogonal supports of @xmath107 for all messages @xmath103 . in such a case we say that , assisted by the entangled state @xmath98 , alice can use the povms @xmath108 as her strategy for sending one out of @xmath99 messages with a single use of @xmath87 . the _ entanglement - assisted one - shot zero - error channel capacity _ of @xmath87 , @xmath109 , is the maximum integer @xmath99 such that there exists a protocol for which condition holds . we are now ready to outline the setting where alice and bob share a maximally entangled state in the above protocol . we will refer to this particular case later in section [ sec : capacity ] . let the ( canonical ) maximally entangled state of local dimension @xmath37 be defined as follows : @xmath110 } { |i\rangle}{|i\rangle},\ ] ] where @xmath111}$ ] is the standard basis of @xmath112 . when alice and bob share a maximally entangled state and alice performs a projective measurement observing @xmath113 , bob s part of the state collapses to @xmath114 . this implies that bob can distinguish between @xmath113 and @xmath115 perfectly if and only if @xmath116 . therefore , if alice uses projective measurement @xmath117 for message @xmath103 and players share a maximally entangled state , then condition is true if and only if @xmath118 considering more than a single use of the channel , one can define the asymptotic zero - error channel capacity @xmath119 and the asymptotic entanglement - assisted zero - error channel capacity @xmath120 by @xmath121 and @xmath122 since @xmath109 depends solely on the confusability graph @xmath94 @xcite , we can talk about @xmath123 for a graph @xmath0 , meaning the entanglement - assisted one - shot zero - error capacity of a channel with confusability graph @xmath0 . similarly we can talk about quantities @xmath124 and @xmath125 . let @xmath0 be a graph on @xmath37 vertices and consider the @xmath37-qubit graph state @xmath36 . let @xmath48 be the stabilizer group of @xmath0 . for each @xmath126 , with @xmath127 , let @xmath128 be the _ weight of _ @xmath129 . let @xmath130 be the set of the _ events _ of @xmath129 , _ i.e. _ the measurement outcomes that occur with non - zero probability when the system is in state @xmath131 and the stabilizing operators @xmath49 are measured with single - qubit measurements . the set of all events is @xmath132 . two events are _ exclusive _ if there exists a @xmath133 for which the same single - qubit measurement gives a different outcome . a graph representing a kotzig orbit can be naturally defined as follows : [ def : h ] let @xmath0 be a graph . let @xmath48 be the stabilizer group of the graph state of @xmath0 . we denote by @xmath1 the graph whose vertices are the events in @xmath48 and the edges are all the pairs of exclusive events . we give an example for events and exclusiveness . let @xmath134 and @xmath135 ( we omit the superscripts for simplicity ) . this means that @xmath136 , _ i.e. _ , if the system is prepared in @xmath36 and @xmath137 is measured by measuring @xmath138 on the first qubit ( with possible results @xmath139 or @xmath140 ) , @xmath88 on the second qubit , and @xmath138 on the third qubit , then the product of the three results must be @xmath140 . therefore , @xmath141 , where hereafter @xmath142 denotes the event the result 1 is obtained when @xmath138 is measured on qubit 1 , the result @xmath139 is obtained when @xmath88 is measured on qubit 2 , and the result @xmath139 is obtained when @xmath138 is measured on qubit 3 . as another example : if @xmath143 and @xmath144 , then @xmath145 . we now give an example of a graph representing a kotzig orbit . let us consider @xmath146 , the path on three vertices . we construct @xmath147 . the stabilizer group @xmath48 ( minus the identity ) has the following elements : @xmath148 , @xmath149 , @xmath150 , @xmath151 , @xmath152 , @xmath153 , and @xmath154 . for all @xmath155 , obtain all possible events ( _ i.e. _ , those which can happen with non - zero probability ) when three qubits are prepared in the state @xmath36 and three parties measure the observables corresponding to @xmath49 . for instance , when @xmath156 , alice measures @xmath157 , bob measures @xmath158 , and charlie does not perform any measurement . since the three qubits are in state @xmath36 , there are only two possible outcomes : alice obtains @xmath159 and bob obtains @xmath160 , denoted as @xmath161 ; or alice obtains @xmath162 and bob obtains @xmath163 , denoted as @xmath164 . for @xmath165 , the only events that can occur are @xmath166 , @xmath142 , @xmath167 , and @xmath168 . the other events for the remaining @xmath169 s are obtained in a similar way . now , let us construct the graph @xmath147 : the vertices represent possible events ; two vertices are adjacent if and only events are exclusive ( _ e.g. _ , @xmath161 and @xmath170 ) . notice that each @xmath49 of weight @xmath171 generates @xmath172 vertices . a drawing of @xmath147 is in fig . associated to the path on three vertices , @xmath173 , consists of two connected components : the upper component in the drawing is @xmath174 , the complement of the shrikhande graph @xcite ; the lower component is @xmath175 , the 6-vertex ( 1,3)-circulant graph . we have @xmath176 , @xmath177 , @xmath178=\protect\vartheta[ci_8(1,3)]=\protect\alpha^ * [ ci_8(1,3)]=3 $ ] . therefore , @xmath179 , while @xmath180.,width=249 ] each @xmath1 can be interpreted as in @xcite . every graph is in fact associated to an nc inequality and is constructed by expressing the linear combination of joint probabilities of events in the nc inequality as a _ sum _ @xmath48 . for a graph in @xcite , an event in @xmath48 is represented by a vertex and exclusive events are represented by edges . constructing such a graph from the inequality is straightforward , when the absolute values of the coefficients in the linear combination are natural numbers ( which , to our knowledge , is always the case for all relevant nc inequalities ) . as already mentioned in the introduction , this graph - theoretic framework can be used to single out games with _ ad hoc _ quantum advantage and quantum correlations with _ ad hoc _ degree of contextuality ( see @xcite , respectively ) . if we apply the method to graphs @xmath0 and @xmath181 in the same orbit under local complementation @xcite then _ we obtain the same graph _ @xmath182 . the reason is that the graph states @xmath131 and @xmath183 share the same set of perfect correlations ( up to relabeling ) , so also share the same graph in which all possible exclusive events are adjacent . this paper constructs @xmath1 from @xmath0 , as described earlier , where each of the @xmath184 operators , @xmath129 , generated by @xmath0 , in turn generates a clique of size @xmath172 in @xmath182 , where @xmath171 is the weight of operator @xmath129 . in this section , we present a classification of all @xmath1 from all graphs @xmath0 for @xmath185 . this classification is greatly simplified by the fact that if two graphs , @xmath0 and @xmath181 , are in the same local complementation ( lc ) orbit , then @xmath186 . so we need only classify for one representative from each orbit . a choice of representatives for @xmath187 , for connected graphs only , are listed in the second column of table [ bigclass ] . the action of local complementation on vertex @xmath24 of graph @xmath0 to yield graph @xmath181 can be realised , in the context of graph states , by a specific local unitary action : @xmath188 where @xmath189 , @xmath190 , and @xmath191 . here , @xmath192 is the stabilizer group associated with the graph @xmath0 , where we omit @xmath193 for convenience . similarly , let @xmath194 be the stabilizer group associated with @xmath195 . we show how to obtain @xmath196 from @xmath48 . let @xmath197 , where @xmath198 , and @xmath199 . define the mapping @xmath200 as follows : @xmath201 moreover , define @xmath202 . in words , @xmath203 is the total number of @xmath89 matrices at tensor positions @xmath204 of @xmath129 . [ lcperm ] the action of lc at vertex @xmath24 of @xmath0 maps @xmath0 to @xmath205 and @xmath48 to @xmath196 , where @xmath206 this action is a permutation , @xmath207 , of the pauli matrices at tensor position @xmath24 of each @xmath129 and a permutation , @xmath208 , of the pauli matrices at tensor positions in @xmath209 of each @xmath129 , followed by a global multiplication by @xmath210 . for example , consider the graph @xmath0 with two edges @xmath211 and @xmath212 . then @xmath213 is the graph with edges @xmath214 ( so the star @xmath215 and the complete graph @xmath216 are in the same kotzig orbit ) . we have that @xmath217 and that @xmath218 . for example @xmath219 is mapped to @xmath220 , where @xmath221 , @xmath222 , and @xmath223 as @xmath89 occurs at tensor positions @xmath140 and @xmath224 of @xmath129 , where @xmath225 . + we use ( [ lumap ] ) . for vertex @xmath24 we replace @xmath226 with @xmath227 , for each of @xmath228 to obtain @xmath229 . similarly , for vertices in @xmath209 we replace @xmath226 with @xmath230 , for each of @xmath228 to obtain @xmath231 . [ lcinvariance]let @xmath232 be the kotzig orbit of graphs generated by the action of successive local complementation on @xmath0 . then @xmath233 every vertex in @xmath1 represents a measurement , @xmath129 , of @xmath131 , combined with a certain measurement result , as specified by the bars under @xmath71 , @xmath73 , and @xmath234 , as appropriate . this measurement is equivalent to a measurement , @xmath235 of @xmath236 , where @xmath235 , @xmath236 , and the new measurement results are obtained from @xmath129 and @xmath131 by the same local unitary transform , namely the transform in ( [ lumap ] ) . since the two measurement scenarios are equivalent , then the edge relationship between vertices in @xmath1 is preserved in @xmath237 , _ i.e. _ , @xmath238 . the theorem is then extended to any two @xmath239 as @xmath240 can be obtained from @xmath181 by a series of local complementations . proposition [ lcinvariance ] implies that we only have to classify for one representative member , @xmath0 , ( arbitrarily chosen ) of each kotzig orbit , @xmath241 , of @xmath37-vertex graphs . table [ bigclass ] classifies , computationally , graphs @xmath242 for @xmath243 from @xmath51 , where @xmath244 is the size of the kotzig orbit of @xmath0 up to re - labeling ( graph isomorphism ) . [ thm : hparams]let @xmath182 be a graph representing a kotzig orbit . then , for @xmath245 , @xmath246 we firstly give an intuition of the statement , explaining how the theorem can be seen as a consequence of the results in @xcite . a formal and stand - alone proof will follow later in this section . let @xmath36 be the graph state with corresponding graph @xmath0 . it was shown in @xcite that the sum of the elements of the stabilizer group of @xmath36 , @xmath247 is a bell operator such that @xmath248 and @xmath249 when restricting to classical states ( where @xmath48 is the stabilizer group defined earlier ) . in words , the graph state @xmath36 violates the corresponding bell inequality up to its algebraic maximum this fact together with ( * ? ? ? * equation 6 ) enforces that @xmath250 and @xmath251 . the construction in definition [ def : h ] simply transforms the bell operator , originally written as a sum of mean values , into a sum of probabilities of events , in order to construct the graph associated with the exclusivity structure . the statement , therefore , combines known facts from quantum information in a novel way in order to prove a purely graph theoretical result . the proof of theorem [ thm : hparams ] requires the following definition : [ def : orth]let @xmath16 be a graph representing a kotzig orbit . let @xmath252 be the event at vertex @xmath253 , where @xmath254 , @xmath255 , and @xmath256 , for each @xmath257 . let @xmath258 be defined as follows:@xmath259here , @xmath260 is an arbitrary ray in @xmath261 and @xmath262 are the eigenvectors of the pauli matrix @xmath89 with eigenvalue @xmath263 and @xmath139 , respectively . canonical orthogonal representation _ of @xmath182 is the set of vectors @xmath264 . for example , in @xmath147 ( see fig . [ h1 ] ) , the element of the canonical orthogonal representation of the vertex labeled by @xmath265 is @xmath266 . notice that if @xmath260 is chosen to be non - orthogonal to any of the vectors @xmath267 then the representation is faithful . let @xmath268 be a graph representing a kotzig orbit of a graph @xmath0 . the proof is structured in three parts : _ ( 1 ) _ we prove that @xmath269 ; _ ( 2 ) _ we prove that @xmath270 ; _ ( 3 ) _ finally , we prove that @xmath271 . the first two parts together prove that @xmath251 , since @xmath272 , for any graph @xmath0 ( see , _ e.g. _ , cabello10 ) . we begin with the first part : _ ( 1 ) _ it follows directly from eq . that @xmath273 . we know that the eigenvectors with eigenvalue @xmath263 of each operator @xmath274 are in one - to - one correspondence with the vertices of a clique in @xmath182 : @xmath275 . these are elements of the canonical orthogonal representation of @xmath182 . from the definition of the stabilizer group , for all @xmath276 and for all eigenvectors @xmath277 ( @xmath278 ) with eigenvalue @xmath139 , we have @xmath279 , because @xmath36 is in the @xmath263 eigenspace . now , let @xmath280 be an hermitian eigendecomposition of @xmath274 . thus , @xmath281where the inequality in the last line follows because a canonical orthogonal representation of @xmath182 together with the state @xmath36 represents a feasible solution for the semidefinite formulation of the lovsz number in eq . . _ ( 2 ) _ from the linear programming formulation of the fractional packing number ( see eq . ) , it is easy to see that a partition of the set of vertices into @xmath22 cliques gives an upper bound to @xmath282 . to see this choose one vertex , say @xmath13 , per clique and set its weight @xmath283 . we get a partition of @xmath182 into @xmath50 cliques if we consider the events associated with each @xmath274 . _ _ ( 3 ) _ _ we use an argument very similar to ( * ? ? ? * lemma 1 and theorem 1 ) . if the number of vertices of @xmath0 is two then the result does not hold as @xmath284 ( by direct calculation ) . each connected graph with more than two vertices has a subgraph with three vertices . for each of those we can see ( also by direct calculation ; see table [ bigclass ] of section [ kotzig ] ) that @xmath285 . therefore , we just need to show that if @xmath181 is a subgraph of @xmath0 with @xmath286 vertices and @xmath287 , where @xmath288 is the representative of the kotzig orbit of @xmath181 , then @xmath271 for @xmath245 . notice that @xmath289 , the stabilizer group of @xmath181 , is a subset of @xmath48 . therefore , in the graph @xmath182 we find cliques associated with @xmath289 , but containing slightly different events . for each @xmath290 , the corresponding @xmath276 has the same structure , with eventually some additional @xmath138 operators . let @xmath291 be the subgraph of @xmath182 induced by the vertices in cliques associated with the elements of @xmath289 . we need to show that if in @xmath288 there is no vertex per clique to form a maximal independent set then neither are there in @xmath291 . therefore , @xmath271 . towards a contradiction , suppose there is an independent set @xmath292 of @xmath291 such that @xmath293 . we distinguish two cases : * if the events at the vertices in @xmath292 do not have any @xmath294 element then we can map them to an independent set in @xmath288 of size @xmath295 just by ignoring the additional @xmath138 operators . this contradicts the hypothesis that @xmath287 . * if the events at the vertices in @xmath292 do have @xmath294 elements then we can find another independent set @xmath296 with the same cardinality such that the events at its vertices do not have any @xmath294 element . we can find @xmath296 as follows . it is easy to check that an operator @xmath274 has the form @xmath297 if and only if it has an odd number of @xmath298 and @xmath299 , with @xmath300 . therefore , complementing @xmath301 and all occurrences of @xmath298 and @xmath299 in the events at the vertices of the independent set @xmath292 , we obtain the events in @xmath296 with the desired properties , and so we are back to the previous case . the bell inequalities described by the graph @xmath182 are exactly the same as in @xcite , but in the form of a pseudo - telepathy game . for any graph @xmath0 on @xmath37 vertices , let us define an @xmath37-player game for @xmath0 as follows . input set _ for each player is @xmath302 and the _ output set _ is @xmath303 . the set of valid inputs is the set of elements of the stabilizer group of @xmath36 . the players win on input @xmath276 if and only if the sign of the product of their outputs equals the sign of @xmath274 . the graph game for @xmath0 is a pseudo - telepathy game . it is easy to see that if the players share the graph state @xmath36 and each player performs the measurement corresponding to her input , then they always win . on the other hand , we show that a classical strategy for the game can be used to construct an independent set of @xmath268 and _ viz_. we now consider the first direction . if there exists a strategy which answers correctly to @xmath22 questions then there exists an independent set with @xmath22 elements . a classical strategy is _ w.l.o.g . _ a set of functions for each player from the input set to the output set . therefore , for all the winning inputs @xmath274 , there will be a single output @xmath304 , corresponding to a vertex of @xmath182 . it is easy to verify that there can not be an edge between any pair of these vertices . since the strategy wins on @xmath22 input pairs , the independent set has @xmath22 elements . for the other direction , we show that if there exists an independent set @xmath292 of @xmath182 having size @xmath22 , then there exists a strategy for the game on @xmath0 that answers correctly to at least @xmath22 of the @xmath50 questions . by the structure of @xmath182 , the independent set @xmath292 can not contain vertices @xmath305 such that , for the same input @xmath71 , @xmath306 for some @xmath307 . hence , we have the following strategy : on input @xmath71 , each player outputs the unique @xmath74 determined by the vertices in the independent set . the size @xmath22 of the independent set implies that the players answer correctly to at least @xmath22 input pairs . let @xmath5 be the complete graph on @xmath37 vertices . as said before , @xmath1 can be associated to the bell inequality in which the bell operator is the sum of all stabilizer operators of @xmath36 gunhe04 . the first observation is that the graph state associated to @xmath309 is the @xmath37-qubit greenberger - horne - zeilinger ( ghz ) state hein03 . the second observation is that , in that case , the bell inequality corresponding to @xmath1 is the sum of a well - known bell inequality maximally violated by the @xmath37-qubit ghz state @xcite plus a trivial bell inequality not violated by quantum mechanics @xcite . for example , for @xmath134 , the bell inequality corresponding to @xmath310 is @xmath311 , where @xmath312 recall that @xmath313 denotes the mean value of the product of the outcomes of the measurement of @xmath88 on qubit @xmath140 , @xmath138 on qubit @xmath224 , and @xmath314 on qubit @xmath315 . the inequality @xmath316 is the well - known bell inequality introduced in @xcite , while @xmath317 is a trivial inequality not violated by quantum mechanics @xcite . the sum @xmath318 has @xmath319 terms , with four terms generating cliques of size @xmath320 and the other three terms generating cliques of size @xmath321 . then , we can see that @xmath322 is equal to the maximum number of quantum predictions that a deterministic local theory can simultaneously satisfy . by quantum predictions we mean : @xmath323 , @xmath324 , @xmath325 , @xmath326 , @xmath327 , @xmath328 , and @xmath327 . in this case , the maximum number of quantum predictions that a deterministic local theory can simultaneously satisfy is @xmath329 : @xmath315 out of @xmath323 , @xmath324 , @xmath325 , @xmath326 , plus the other @xmath315 ( @xmath327 , @xmath328 , and @xmath327 ) . equivalently , @xmath330 is the maximum quantum violation , denoted by @xmath331 , minus the minimum number of quantum predictions which _ can not _ be satisfied by a deterministic local theory . since the minimum number of quantum predictions which can not be satisfied by a deterministic local hidden variable theory is @xmath331 minus the maximum value of the bell operator for a deterministic local theory , denoted by @xmath332 , and all of them divided by two , then@xmath333this expression is very useful since , for the bell inequalities for the @xmath37-qubit ghz states @xcite , @xmath334the interesting point is that the values of @xmath335 and @xmath336 are well - known @xcite , and @xmath337 ( recall that @xmath338 . ) for @xmath37 odd , @xmath339 , and @xmath340 ; for @xmath37 even , @xmath339 and @xmath341 . inserting these numbers in eq . ( [ des ] ) and then in eq . ( [ aln ] ) , we obtain the statement . in this section , we show that for every graph @xmath0 on @xmath37 vertices the graph @xmath1 has zero - error entanglement - assisted capacity @xmath50 . theorem thm : hparams states that @xmath342 . the result gives a separation between @xmath343 and @xmath344 . it is known that for all graphs , the lovsz number upper bounds the entanglement - assisted shannon capacity @xcite . therefore , @xmath343 saturates its upper bound . there are few ( and very recently discovered ) classes of graphs for which this separation is known . for example , one is based on the kochen - specker theorem @xcite and other ones are based on variations of orthogonality graphs @xcite . here , we present a new family of graphs and a construction method , which can also be interpreted as a graph theoretic technique of independent interest . the most important point is that every graph gives rise to a member of the family through our construction . this property opens directions for future studies , for example , identifying subclasses or hierarchies where the separation is large or is easy to quantify . [ thm : capacity ] let @xmath182 be a graph from a kotzig orbit . then @xmath345 . from theorem [ thm : hparams ] and ( * corollary 14 ) we obtain the upper bound @xmath346 we need to show a matching lower bound on @xmath343 . we do this by exhibiting a strategy for entangled parties to send one out of @xmath50 messages in the zero - error setting through a channel with confusability graph @xmath182 . the strategy is as follows . alice and bob share a maximally entangled state of local dimension @xmath347 . observe that @xmath182 can be partitioned into @xmath50 cliques , one for each element of the stabilizer group . the clique corresponding to @xmath276 consists of the vertices associated with the mutually exclusive events in the set @xmath348 ; we denote by @xmath348 the set of events related to @xmath274 as in section [ con ] . for each @xmath349 , alice performs a projective measurement on her part of the shared state . the outcomes of the measurement are the elements of @xmath348 . since the parties share a maximally entangled state , alice s strategy has to satisfy two properties to be correct : 1 . for each @xmath349 , the projectors associated to elements of @xmath348 form a projective measurement ( because alice needs to perform a projective measurement for each message @xmath13 to be sent ) . 2 . for each edge @xmath350 , projectors associated with @xmath23 and @xmath24 must be orthogonal ( to satisfy the zero - error constraint ) . the next step is to exhibit projectors in alice s strategy and show that both properties are satisfied . in what follows we use the notation in definition [ def : orth ] . we begin by examining the case where @xmath274 does not contain any identity operator . in this case , each projective measurement will consist of projectors of rank @xmath140 acting on @xmath351 . order the elements of @xmath348 arbitrarily . let @xmath274 be of the form @xmath352 , where @xmath353 . define for each @xmath354 the _ occurrence number _ @xmath355 based on a chosen ordering : if the same eigenvector of @xmath356 occurs in @xmath354 for the @xmath357-th time in the chosen ordering then @xmath358 . construct projectors starting from the canonical orthogonal representation and an ancillary space of dimension @xmath359 . for @xmath354 , let @xmath360 we show that property 1 is satisfied . these projectors are mutually orthogonal for all vertices @xmath361 . we need to prove that their sum is the identity . from the structure of the events in @xmath348 we observe that , for each @xmath362 , the eigenvectors with eigenvalue @xmath263 ( and @xmath139 ) occur in half of the elements of @xmath348 . therefore , in the construction of the projectors , a pair of @xmath363 eigenvectors for each @xmath362 is summed for each ancillary subspace . the sum of each subspace is the identity . hence , the total sum is the identity for the whole space . we show now that also property 2 is satisfied . if two projectors are in the same clique , orthogonality follows from the discussion above . consider now two projectors of adjacent vertices from two different cliques that project to the same ancillary subspace . since we started from an orthogonal representation , those projectors are orthogonal . now , consider the more general case where @xmath274 can contain identity operators . let @xmath274 be of the form @xmath352 , where @xmath364 . we assume that @xmath274 has weight @xmath365 . first consider the case where the first @xmath365 operators are different from identity , @xmath366 . to construct the projective measurement for @xmath348 , we initially construct the projectors for the first @xmath365 operators as in the previous case . we obtain rank-@xmath140 projectors acting on @xmath367 . choose a basis for @xmath368 and let the projectors be @xmath369 this ensures that the dimensions match and that properties @xmath140 and @xmath224 hold . to finish the proof , we need to prove the general case where identity operators are in arbitrary positions and not all at the end . in this case , split the construction into subspaces so that each subspace has all the identities at the end . obtain the projectors for the subspaces as described above and then obtain the final projectors by making tensor products of the projectors for the subspaces . we immediately have the following corollary from theorems [ thm : hparams ] and [ thm : capacity ] and the lovsz number upper bound on @xmath370 . let @xmath182 be a graph from a kotzig orbit . then , @xmath371 . let @xmath372 be the number of connected components of @xmath182 - we follow @xmath372 by the number of vertices in each connected component , _ e.g. _ , @xmath373 $ ] means that there are two connected components , one with 6 vertices and one with 16 vertices . the degree sequence of @xmath182 , @xmath374 , is written as @xmath375 , meaning that there are @xmath75 vertices of degree @xmath74 , @xmath376 vertices of degree @xmath377 , @xmath378 vertices of degree @xmath379 , etc . for each kotzig orbit , we wish to compute the independence number , @xmath380 . this is also listed in table [ bigclass ] . each stabilizing operator of @xmath48 is multiplied by a global coefficient @xmath263 or @xmath139 . for instance , for the operators of @xmath381 ( see fig [ h1 ] and associated discussion ) , there are six ` @xmath263 ' coefficients and one ` @xmath139 ' coefficient . if one selects one vertex from each of the 6 cliques in @xmath182 generated by the 6 operators of this example which have a ` @xmath263 ' coefficient , then one can be sure that they are mutually unconnected in @xmath182 . ( more specifically , one can select for each operator the event where all the measurements gave outcome @xmath263 . this event is always present if the operator has coefficient @xmath263 . ) so @xmath382 , _ i.e. _ the number of ` @xmath263 ' coefficients in @xmath48 for a given @xmath0 yields a lower bound on @xmath380 , for each @xmath0 . this idea leads to the following lemma . let @xmath383 be the number of operators in @xmath48 with a ` @xmath139 ' coefficient where , in general , @xmath383 is not an invariant of the kotzig orbit of @xmath0 . let @xmath384 and @xmath385 , where @xmath241 is the set of graphs in the kotzig orbit of @xmath0 . table [ bigclass ] lists both the computed independence number , @xmath380 , for each @xmath182 , and the range of lower bounds , @xmath387 , on @xmath380 . one observes that the lower bound is often tight but not always . for example , for the graph @xmath388 , computations show that @xmath389 and @xmath390 . so @xmath391 . in this case the bound is tight as @xmath380 is computed to be @xmath392 . @xmath393 & 3 & 0 & 1,6/ \\ \hline 3 & 12,13 & 2 & 22 & 2[6,16 ] & 6 & 1 - 1 & 3,6/9,16 \\ \hline 4 & 14,24,34 & 2 & 84 & 2[20,64 ] & 13 & 4 - 4 & 11,12/13,2/17,6/43,64 \\ & 14,23,24,34 & 4 & 76 & 1 & 13 & 2 - 4 & 15,4/25,4/27,4/29,32/35,32 \\ \hline 5 & 15,25,35,45 & 2 & 316 & 2[60,256 ] & 25 & 10 - 10 & 33,20/45,10/49,30/195,256/ \\ & 15,25,34,45 & 6 & 280 & 1 & 25 & 6 - 12 & 49,6/51,2/89,12/91,12/107,24/129,72/133,24/159,128/ \\ & 15,24,25,34,35 & 10 & 268 & 1 & 25 & 6 - 12 & 55,4/99,16/101,4/103,4/107,16/117,64/129,32/137,32/139,32/147,64/ \\ & 12,13,24,35,45 & 3 & 256 & 1 & 25 & 6 - 10 & 99,40/117,120/135,96 \\ \hline 6 & 16,26,36,46,56 & 2 & 1206 & 2[182,1024 ] & 51 & 20 - 20 & 101,30/143,30/147,90/151,2/167,30/841,1024/ \\ & 16,26,36,45,56 & 6 & 1030 & 1 & 51 & 16 - 28 & 145,12/189,2/207,2/211,6/299,24/301,24/343,4/351,28/ \\ & & & & & & & 383,32/455,32/459,96/587,256/665,512/ \\ & 16,26,35,45,46 & 16 & 976 & 1 & 49 & 14 - 30 & 169,6/207,2/349,12/351,12/357,24/383,8/411,48/449,8/ \\ & & & & & & & 453,24/471,32/473,16/477,48/481,48/483,48/533,128/557,128/ \\ & & & & & & & 565,64/567,64/611,256/ \\ & 15,26,36,45,56 & 4 & 1044 & 1 & 49 & 14 - 30 & 173,12/321,36/323,36/479,192/541,64/545,192/679,512/ \\ & 15,26,35,36,45,46 & 5 & 958 & 1 & 51 & 20 - 28 & 213,6/355,32/395,12/397,12/441,96/515,384/547,192/ \\ & & & & & & & 549,192/559,4/567,28/ \\ & 16,24,35,46,56 & 10 & 958 & 1 & 51 & 12 - 28 & 181,4/213,2/323,4/325,4/347,8/349,8/379,32/401,64/455,8/ \\ & & & & & & & 459,24/477,32/487,64/489,64/515,256/593,256/599,32/603,96/ \\ & 15,26,34,35,46,56 & 25 & 940 & 1 & 49 & 14 - 30 & 187,4/337,4/339,4/357,16/379,16/383,32/431,64/435,32/ \\ & & & & & & & 469,32/471,32/493,32/495,32/497,128/521,128/529,64/531,64/ \\ & & & & & & & 575,128/601,64/603,64/ \\ & 16,24,26,35,36,45 & 21 & 922 & 1 & 47 & 16 - 28 & 193,2/355,16/357,16/405,16/413,64/425,64/441,8/475,32/ \\ & & & & & & & 477,32/479,64/503,256/511,32/513,32/557,256/563,8/567,24/ \\ & 14,25,36,45,46,56 & 5 & 990 & 1 & 47 & 20 - 28 & 197,6/355,12/357,12/433,192/473,96/487,8/491,24/579,192/ \\ & & & & & & & 581,192/625,256/ \\ & 12,13,25,36,45,46 & 16 & 904 & 1 & 45 & 18 - 30 & 355,32/395,24/407,96/431,48/485,384/539,192/565,64/567,64/ \\ & 12,13,14,23,25,36,45,46,56 & 2 & 936 & 1 & 45 & 22 - 26 & 427,360/571,576/ \\ \hline \end{array}\ ] ] the _ symmetrical form _ of @xmath394 makes it relatively easy to prove that , for @xmath245 , @xmath395 is comprised of two disjoint subgraphs , and the results of section [ examples ] allow us to identify these two disjoint subgraphs , _ i.e. _ @xmath396 , where @xmath397 has @xmath398 vertices and @xmath399 has @xmath400 vertices - ( see the equations in ( [ aln ] ) for the case of @xmath401 ) . it is likewise easy to show that @xmath402 , and therefore we know that @xmath403 . ( sketch ) consider the subgraph of @xmath0 at vertices @xmath408 , where we assume @xmath409 . one can confirm that @xmath410 is an operator with a global coefficient of @xmath139 and we can represent this in @xmath411 by the cubic term @xmath412 . moreover this must be true for each such pair of edges in @xmath0 . likewise consider the five vertices @xmath413 in @xmath0 where we assume that @xmath414 . one can confirm that @xmath415 is an operator with a global coefficient of @xmath139 , and the operators @xmath416 , @xmath417 , and @xmath418 also have global coefficients equal to @xmath139 . we represent this situation in @xmath411 by the sum of the cubic terms @xmath419 . the lemma follows from an elaboration of this argument . specifically the global coefficient of @xmath420 is @xmath421 . lemma [ cubic ] allows us to obtain an equation for @xmath422 . we evaluate @xmath422 by computer for @xmath423 to be @xmath424 , respectively and use @xcite to find the sequence a051253 and the reference @xcite where a recurrence formula is provided for @xmath425 , namely @xmath426 the graph @xmath1 is always connected except when @xmath0 is in the kotzig orbit of the star graph , @xmath427 , in which case @xmath1 splits into 3 disjoint components for @xmath428 , and @xmath224 disjoint components for @xmath245 . the conjecture is verified , computationally , for @xmath429 . a potential way to prove it is to try to construct two connected components by adding codewords and show that this forces one to be in the @xmath427 orbit . by using code - theoretic techniques , we can lower and upper bound the size of cliques in @xmath1 . remember that @xmath48 is the stabilizer group generated by @xmath0 , comprising @xmath184 operators , @xmath129 , and that @xmath430 is the _ weight _ of operator @xmath129 . let @xmath431 and @xmath432 . we have that @xmath433 , for any @xmath48 , because , for any graph @xmath0 , then @xmath434 is always an operator of weight @xmath37 . it is also well - known that the stabilizer group for a graph state characterizes a self - dual additive code over @xmath435 of length @xmath37 whose minimum distance , @xmath436 , is given by @xmath437 @xcite . so , in subsequent discussions , we refer to @xmath437 and @xmath438 by @xmath436 and @xmath37 , respectively . let @xmath439 be the set of maximal cliques in @xmath182 , where each @xmath440 is a subset of @xmath19 , the set of vertices of @xmath182 , over which there is a clique in @xmath182 . let @xmath441 and @xmath442 be the minimum and maximum size of a maximal clique , respectively , in graph @xmath182 . so @xmath443 and @xmath444 . recall that @xmath442 denotes the clique number of @xmath182 . consider an arbitrary set of two operators , @xmath446 . then both operators have @xmath447 at their second , third , and fourth tensor positions . we say that the two operators have a set overlap of @xmath447 and this overlap is of size @xmath448 . let @xmath449 and consider an arbitrary splitting of the assignments to @xmath447 of @xmath450 , @xmath451 , @xmath452 , @xmath453 , @xmath454 for operator @xmath455 and @xmath456 , @xmath457 , @xmath458 for operator @xmath459 . then @xmath1 contains a size-11 clique over the vertices @xmath460 and @xmath461 . it is straightforward to verify that this clique can not be extended by adding another vertex from the subset of vertices in @xmath182 originating from the operators in @xmath462 . more generally , any splitting of the @xmath463 assignments to @xmath447 is possible , each giving a different clique . for our example , the operator @xmath455 contributes @xmath464 vertices to the clique and operator @xmath459 contributes @xmath465 vertices to the clique . so this clique is of size @xmath466 . it is a maximal clique if and only if there is no @xmath467 such that @xmath468 and @xmath469 . if , instead , @xmath470 , then the set overlap is reduced to @xmath471 and of size @xmath472 . as an example , consider just the partition where @xmath473 is assigned to operator @xmath455 , @xmath474 to operator @xmath459 , and @xmath475 to operator @xmath476 . then this assignment identifies the vertices @xmath477 , and @xmath478 , and @xmath479 . the operator @xmath455 contributes @xmath480 vertices to the clique , operator @xmath459 contributes @xmath481 vertices to the clique , and operator @xmath482 contributes @xmath483 vertices to the clique . so this clique is of size @xmath484 . it is a maximal clique if and only if there is no @xmath467 such that @xmath468 and @xmath469 . therefore , a strategy to find all maximal cliques in @xmath1 is to find all subsets @xmath449 where @xmath485 , and such that there is no @xmath467 where @xmath468 and @xmath469 . then for each of these subsets , @xmath462 , and for each @xmath486-wise ordered partition , @xmath487 of @xmath488 , where @xmath489 , and for each assignment of the integers in set @xmath490 according to partition @xmath491 , there is associated a maximal clique , @xmath492 , of size , @xmath493 let @xmath494 . then , for the specific @xmath495 where @xmath496 we can minimise the clique size by assigning @xmath497 . so a special case of the above equation identifies a maximal clique of minimum size @xmath498 . similarly let @xmath499 . then , for the specific @xmath495 where @xmath500 we can maximise the clique size by assigning @xmath497 . a special case of the above equation identifies a maximal clique of maximum size @xmath501 . we arrive at both @xmath502 and @xmath503 by observing that all members of @xmath48 occur in at least one @xmath462 identified . in a step towards enumerating the maximal cliques of @xmath1 , we first enumerate the number of maximal cliques of size given by ( [ maximalsize ] ) , as generated by a fixed @xmath462 and @xmath486-wise ordered partition , @xmath491 . let @xmath504 satisfy @xmath505 , and let @xmath506 . the lemma follows immediately by counting the number of ways that one can assign to the integers an @xmath486-way partition of @xmath488 . for instance , for @xmath472 and partition @xmath510 , one can assign the integers as @xmath511 , @xmath512 , @xmath513 , @xmath514 , @xmath515 , @xmath516 , @xmath517 , @xmath518 , @xmath519 , + @xmath520 , @xmath521 , @xmath522 - in total @xmath523 ways . although , in theorem [ cliqbound ] , we have obtained tight lower and upper bounds for the size of the maximal cliques of @xmath1 in the general case , in terms of @xmath436 , and @xmath37 , respectively , it remains open to obtain tight lower and upper bounds for the size of the maximal independent sets of @xmath1 in the general case . however we have obtained a lower bound , @xmath383 , on @xmath524 , although this bound is not tight in general . it also remains open to provide equations for @xmath525 in the general case . given that this paper highlights the gap between @xmath524 and @xmath526 , it is particularly desirable to develop equations for @xmath524 in the general case . of course , it would be nice to find a formula linking @xmath524 and @xmath527 - however it should be noted that @xmath527 is not an invariant of the kotzig orbit of @xmath0 . acknowledgements we would like to thank mary beth ruskai for asking during the quantum information workshop at the centro de ciencias de benasque ( july 2011 ) the question that motivated the results : is there a connection between graph states and the graphs representing exclusivity structures studied in @xcite ? we thank the associate editor and a referee for the very useful comments that improved the readability of the paper . ac is supported by the project no . fis2011 - 29400 ( spain ) . part of this work has been done while ac and ss where visiting the department of informatics at the university of bergen . the financial support of the university of bergen is gratefully acknowledged . part of this work has been done while gs was visiting university college london . gs is supported by ronald de wolf s vidi grant 639.072.803 from the netherlands organization for scientific research ( nwo ) . r. duan , s. severini , and a. winter , zero - error communication via quantum channels , non - commutative graphs and a quantum lovsz @xmath528 function , arxiv:1002.2514v2 [ quant - ph ] ; to appear in _ ieee trans . theory_. o. ghne , g. tth , p. hyllus , and h. j. briegel , bell inequalities for graph states , _ phys . rev . lett . _ * 95 * , 120405 ( 2005 ) . m. hein , j. eisert , and h. j. briegel , multiparty entanglement in graph states , _ phys . rev . a _ * 69 * , 062311 ( 2004 ) . j. krner and a. orlitsky , zero - error information theory , _ ieee trans . inf . theory _ * 44 * , 2207 ( 1998 ) . m. van den nest , j. dehaene , and b. de moor , graphical description of the action of local clifford operations on graph states , _ phys . a _ , * 69 * , 022316 ( 2004 ) . m. a. nielsen and i. l. chuang , _ quantum computation and quantum information _ ( cambridge university press , cambridge , 2000 ) . r. raussendorf , d. e. browne , and h. j. briegel , measurement - based quantum computation with cluster states , _ phys . , * 68 * , 022312 ( 2003 ) . p. sarvepalli and r. raussendorf , on local equivalence , surface code states and matroids , _ phys . a _ * 82 * , 022304 ( 2010 ) . c. e. shannon , the zero - error capacity of a noisy channel , _ ire trans . th . _ * 2 * , 8 ( 1956 ) . e. r. scheinerman and d. h. ullman , _ fractional graph theory _ ( john wiley & sons , new york , 1997 ) . s. s. shrikhande , the uniqueness of the @xmath529 association scheme , _ annals of mathematical statistics _ * 30 * , 781 ( 1959 ) . d. schlingemann , r. f. werner , quantum error - correcting codes associated with graphs , _ phys . a _ * 65 * , 012308 ( 2001 ) .

the following question was raised by tuza in 1990 and erds et al . in 1992 : if every edge of an-vertex chordal graph is contained in a clique of size at least four , does have a clique transversal , i.e. , a set of vertices meeting all non - trivial maximal cliques , of size at most ? we prove that every such graph has a clique transversal of size at most if , which is the best possible bound .

1601.05305

we investigate a problem posed by erds et al . * problem 4 ) and tuza ( * ? ? ? * problem 1 ) on clique transversals in chordal graphs . to state the problem , we require the following definitions . a graph is _ chordal _ if it contains no induced cycle of length four or more . we define a _ clique _ to be a complete subgraph of a graph ( here , we deviate from the standard definition ) and call a clique _ maximal _ if it is an inclusion - wise maximal complete subgraph . @xmath5-clique _ is a clique containing @xmath5 vertices ; moreover we call a clique non - trivial if it is a @xmath5-clique for some @xmath6 . a _ clique transversal _ of a graph @xmath1 is a set @xmath7 of vertices such that every non - trivial maximal clique of @xmath1 contains a vertex from @xmath7 . finally , a chordal graph @xmath1 is _ @xmath5-chordal _ if each edge of @xmath1 is contained in a @xmath5-clique . every @xmath8-chordal @xmath0-vertex graph has a clique transversal of size at most @xmath9 @xcite , and every @xmath10-chordal @xmath0-vertex graph has a clique transversal of size at most @xmath11 @xcite . motivated by these results , erds et al . * problem 4 ) and tuza ( * ? ? ? * problem 1 ) posed the following . [ quest ] does every @xmath12-chordal graph with @xmath0 vertices have a clique transversal of size at most @xmath2 ? flotow @xcite constructed counterexamples for some small values of @xmath0 and andreae and flotow @xcite constructed arbitrarily large @xmath0-vertex @xmath12-chordal graphs admitting no clique transversal with fewer than @xmath13 vertices , conjecturing this to be tight up to a constant factor . [ conj ] every @xmath12-chordal graph with @xmath0 vertices has a clique transversal of size at most @xmath14 . this conjecture also appears as ( * ? ? ? * problem 77 ) . andreae @xcite proved this conjecture when restricting to @xmath12-chordal graphs with every maximal clique containing at most four vertices . in this paper , we prove conjecture [ conj ] and determine the optimal function of @xmath0 for which question [ quest ] holds true , that is we prove the following . [ thm - main ] every @xmath12-chordal graph @xmath1 with @xmath4 vertices has a clique transversal of size at most @xmath15 . as shown in proposition [ prop - lower ] , the bound proven in theorem [ thm - main ] is best possible for every @xmath4 . it is well - known that every chordal graph is an intersection graph of subtrees of a tree . such intersection representations lead to tree - decompositions , which we now define . tree - decomposition _ of a graph @xmath1 is a tree @xmath16 such that * each node @xmath17 of @xmath16 is _ associated _ with a subset @xmath18 of vertices of @xmath1 , * two vertices @xmath19 and @xmath20 of @xmath1 are joined by an edge if and only if there exists a node @xmath17 such that @xmath21 , and * for every vertex @xmath19 of @xmath1 , the nodes @xmath17 with @xmath22 induce a subtree of @xmath16 . a graph @xmath1 admits a tree - decomposition @xmath16 if and only if @xmath1 is chordal . we will always refer to the vertices of a tree - decomposition as nodes in order to distinguish clearly between the vertices of graphs and tree - decompositions . the vertices associated with a node @xmath17 of a tree - decomposition of a graph @xmath1 form a clique in @xmath1 ; we will say that @xmath17 _ corresponds _ to this clique . the helly property of subtrees of a tree implies that every maximal clique corresponds to a node of a tree - decomposition . however , not all nodes of a tree - decomposition need correspond to a maximal clique . we now state the following folklore lemma that asserts it is possible to modify a tree - decomposition of a chordal graph such that each node corresponds to a maximal clique . [ lm - maximal ] every chordal graph @xmath1 has a tree - decomposition @xmath16 such that each node of @xmath16 corresponds to a maximal non - trivial clique of @xmath1 , and all the nodes of @xmath16 are associated with different subsets of vertices of @xmath1 . let @xmath16 be a tree - decomposition of @xmath1 with the minimum number of nodes . the minimality of @xmath16 implies that no node is associated with a single vertex , i.e. , all cliques associated with nodes of @xmath16 are non - trivial . let @xmath17 be a node of @xmath16 and let @xmath23 be the clique corresponding to @xmath17 . we will show that @xmath23 is a maximal clique . if @xmath17 had a neighbor @xmath24 such that every vertex associated with @xmath17 were also associated with @xmath24 , we could contract the edge @xmath25 and obtain a smaller tree - decomposition of @xmath1 , which is impossible by the choice of @xmath16 . hence , for every neighbor @xmath24 , there exists a vertex @xmath19 associated with @xmath17 but not with @xmath24 . suppose that @xmath23 is not a maximal clique , i.e. , there exists a clique @xmath26 of @xmath1 containing @xmath23 . let @xmath27 be the node that corresponds to @xmath26 and let @xmath24 be the neighbor of @xmath17 on the path from @xmath17 to @xmath27 . furthermore , let @xmath19 be a vertex associated with @xmath17 but not with @xmath24 . since the nodes associated with @xmath19 induce a subtree , the vertex @xmath19 is not associated with the node @xmath27 . consequently , @xmath26 does not contain @xmath19 . we conclude that the clique @xmath23 is maximal . in the proof of theorem [ thm - main ] , we require a particular type of tree - decomposition , which we now define . a tree - decomposition of a @xmath12-chordal graph is _ nice _ if it is rooted and satisfies the following : * no two adjacent nodes correspond to the same clique , * if a node @xmath17 corresponds to a @xmath5-clique @xmath23 , then @xmath28 and if @xmath29 , then @xmath23 is maximal , * if a node @xmath17 corresponds to a @xmath5-clique , @xmath30 and @xmath17 is not the root , then @xmath31 vertices associated with @xmath17 are also associated with the parent of @xmath17 , and * if @xmath1 contains a maximal @xmath10-clique , then the root node corresponds to a maximal @xmath10-clique . we now show that every @xmath12-chordal graph admits a nice tree - decomposition , even with the additional restriction on the clique corresponding to its root . [ prop - nice ] let @xmath1 be a @xmath12-chordal graph and let @xmath23 be a maximal @xmath10-clique of @xmath1 , if it exists ; otherwise let @xmath23 be a clique of @xmath1 of order at least four . then , @xmath1 admits a nice tree - decomposition such that its root corresponds to @xmath23 . let @xmath16 be a tree - decomposition of @xmath1 such that each node of @xmath16 corresponds to a maximal non - trivial clique , which exists by lemma [ lm - maximal ] . we construct a nice tree - decomposition of @xmath1 from @xmath16 . we first introduce a new node @xmath32 and associate it with the vertices of @xmath23 . we join @xmath32 to a node of @xmath16 that corresponds to a clique containing @xmath23 ( if @xmath23 is maximal , then this node corresponds to @xmath23 itself ) , and root the tree - decomposition at @xmath32 . observe that the second and forth properties from the definition of a nice tree - decomposition hold in this modified tree - decomposition . while the modified tree - decomposition contains a node @xmath17 associated with vertices @xmath33 , @xmath30 , such that two of these vertices , say @xmath34 and @xmath35 , are not associated with the parent @xmath24 of @xmath17 , we proceed as follows : we introduce a new node @xmath27 to be both a child of @xmath24 and the parent of @xmath17 , associating @xmath27 with @xmath36 . this process terminates since the sum of the squares of the sizes of the symmetric difference of vertex sets associated with adjacent nodes in the tree - decomposition decreases in each step . this tree - decomposition now satisfies the second , third and forth properties from the definition of a nice tree - decomposition . if the tree - decomposition contains two adjacent nodes corresponding to the same clique , we contract the edge joining them . after this process terminates , we have a nice tree - decomposition of @xmath1 rooted at a node corresponding to the clique @xmath23 . to prove our main result , we design an algorithm that constructs a small clique transversal of a given @xmath12-chordal graph @xmath1 , whose input is @xmath1 together with any one of its nice tree - decompositions . during the course of the algorithm , nodes of the input nice tree - decomposition will be removed sequentially , whilst simultaneously in @xmath1 , selected pairs and triples of vertices will become temporarily distinguished , determining which vertices are to be removed or colored red with every iteration . distinguished pairs and triples will be collectively referred to as distinguished tuples , and it will hold that : * a distinguished tuple will never contain a red vertex , * the vertices of each distinguished tuple will induce a complete subgraph , and * at least one vertex of a distinguished tuple will eventually become red . the red vertices will form the sought clique transversal . we will refer to an algorithm of this kind as a _ clique transversal algorithm_. to bound the size of the constructed clique transversal , we apply a double counting argument . initially , each vertex will be assigned two zotys and each node of the input nice tree - decomposition corresponding to a maximal clique of order three will be assigned one zoty . as each vertex of the chordal graph is removed or colored red , its assigned zotys will be reassigned in accordance with the algorithm . whenever a vertex is colored red , seven zotys will be removed from the graph these seven zotys will be referred to as _ paid _ for coloring a vertex red . it may also occur that some zotys are removed from the graph without creating a red vertex these zotys will be referred to as _ saved_. in addition , each distinguished pair will always be assigned three zotys and each distinguished triple two zotys at the instant of its creation . when any distinguished tuple ceases to exist , its zotys will also be reassigned . to aid the accessibility of our arguments , we will sequentially design three algorithms with the properties above . to establish a further notion , a clique transversal algorithm will be denoted _ cash - balanced _ if it abides by the rules presented above for reassigning zotys . [ prop - zlotys ] let @xmath1 be a @xmath12-chordal graph and let @xmath37 be the clique transversal produced by a clique transversal algorithm that is cash - balanced . if @xmath0 is the number of vertices of @xmath1 , @xmath38 is the number of maximal @xmath10-cliques of @xmath1 and @xmath39 is the number of zotys saved by the algorithm , then @xmath40 initially , @xmath41 zotys are assigned amongst the vertices of @xmath1 and @xmath38 zotys are assigned to maximal @xmath10-cliques . since @xmath39 zotys are saved during the course of the algorithm , it stands that @xmath42 zotys have been paid for red vertices . since exactly seven zotys are paid for each red vertex , the bound on the size of @xmath37 follows . we begin by presenting a clique - transversal algorithm that may fail to save any zotys . we will then modify this algorithm for the case of @xmath12-chordal graphs with maximal @xmath10-cliques and finally the case of @xmath12-chordal graphs with no maximal @xmath10-cliques . the algorithm that we design in this subsection will be referred to as the _ basic algorithm_. the basic algorithm processes the nodes of an input nice tree - decomposition of a @xmath12-chordal graph from the leaves towards the root using the rules we now describe . each time a node is processed , it is removed from the tree - decomposition ; moreover , the vertices of the graph associated with only that node are also removed . let @xmath17 be a leaf node of the tree - decomposition , let @xmath7 be the set of vertices associated with @xmath17 , and let @xmath43 be the set of vertices associated with @xmath17 but not with its parent . we analyze four cases concerning the possible sizes of @xmath7 and @xmath44 . * case 1 : the size of @xmath7 is at least four and the size of @xmath44 is one . * let @xmath19 be the single vertex contained in @xmath44 . the algorithm follows the first of the following rules that applies and then removes both the node @xmath17 from the tree - decomposition and the vertex @xmath19 from the graph . rule b1 . : : if the vertex @xmath19 is red , the algorithm performs no additional steps . : : if the vertex @xmath19 is contained in at least three distinguished tuples , it becomes red . the tuples containing @xmath19 cease to exist and their zotys are removed . during this process , seven zotys are paid for coloring @xmath19 red and at least one zoty is saved . : : if the vertex @xmath19 is contained in at least two distinguished pairs , it becomes red . the pairs containing @xmath19 cease to exist and their zotys are removed . as in the case of rule g1 , at least one zoty is saved . : : if the vertex @xmath19 is contained in a distinguished pair and a distinguished triple , it becomes red and the two tuples cease to exist . no zotys are saved . : : if the vertex @xmath19 is contained in two distinguished triples , the remaining two vertices of each triple will become a distinguished pair . each pair keeps the zotys of the original triple it was contained in and is reassigned one additional zoty from @xmath19 . : : if the vertex @xmath19 is contained in a distinguished triple , the remaining two vertices of the triple become a distinguished pair . this pair keeps the two zotys of the original triple and gets one zoty from @xmath19 . the other zoty of @xmath19 is saved . : : if the vertex @xmath19 is contained in a distinguished pair and the other vertex , call it @xmath20 , of that pair is contained in another distinguished tuple , the vertex @xmath20 becomes red . the tuples containing @xmath20 cease to exist . since we remove at least nine zotys ( the four zotys of @xmath19 and @xmath20 and at least five zotys of the tuples ) , at least two zotys are saved . : : if the vertex @xmath19 is contained in a distinguished pair , the other vertex of the pair becomes red and the pair ceases to exist . : : if the vertex @xmath19 is not in a tuple and one of the vertices of @xmath45 is red , the algorithm also performs no additional steps ; the two zotys of @xmath19 are saved . : : if the vertex @xmath19 is not in a tuple and no vertex of @xmath45 is red , an arbitrary three vertices of @xmath45 form a distinguished triple and this triple is reassigned two zotys from @xmath19 . note that at least one zoty is saved when one of rules g1g5 is applied , so we will refer to these five rules as _ good _ rules . similarly , rules b1b5 will be referred to as _ bad _ rules ( since no zotys are saved ) . * case 2 : the size of @xmath7 is at least four and the size of @xmath44 is greater than one . * we apply some of the rules from case 1 . as long as @xmath44 contains a non - red vertex that is not contained in any distinguished tuple ( and that has not been processed ) , we process it , i.e. , we use rule b5 or g5 . we then use the first applicable rule to process each remaining vertex in @xmath44 in an arbitrary order , followed by removing the node @xmath17 from the tree - decomposition . * case 3 : the size of @xmath7 is three and the size of @xmath44 is one . * let @xmath19 be the single vertex contained in @xmath44 . the algorithm follows the first of the following rules that applies and then removes the node @xmath17 from the tree - decomposition and the vertex @xmath19 from the graph . : : if the vertex @xmath19 is red , the algorithm performs no additional steps . : : if the vertex @xmath19 is contained in at least two distinguished tuples , it becomes red and all the tuples containing @xmath19 cease to exist . coloring @xmath19 red is paid using the two zotys of @xmath19 , at least four zotys from the tuples containing @xmath19 and the one zoty of node @xmath17 . : : if the vertex @xmath19 is contained in exactly one distinguished tuple , an arbitrary vertex @xmath20 of the tuple different from @xmath19 becomes red and all distinguished tuples containing @xmath20 cease to exist . note that at least seven zotys are removed with this rule : the two zotys of the vertex @xmath19 , the two zotys of the new red vertex , at least two zotys from the ceased tuples , and the one zoty of the node @xmath17 . : : if @xmath7 contains a red vertex , the algorithm performs no additional steps ; one zoty assigned to the node @xmath17 is saved . : : the other two vertices of @xmath7 become a distinguished pair , which is assigned the two zotys of @xmath19 and the one zoty of node @xmath17 . * case 4 : the size of @xmath7 is three and the size of @xmath44 is greater than one . * let @xmath19 be an arbitrary vertex of @xmath44 and follow the first applicable rule among the rules t1t5 . if the remaining two vertices of @xmath7 do not form a distinguished pair after this step , we remove the remaining vertices of @xmath44 from the graph and remove the node @xmath17 from the tree - decomposition . assume that the remaining two vertices of @xmath7 form a distinguished pair . let @xmath20 be the vertex of @xmath46 if it exists ; otherwise , let @xmath20 be an arbitrary vertex of @xmath7 different from @xmath19 . the vertex @xmath20 is colored red and all distinguished pairs containing @xmath20 cease to exist . the remaining vertices of @xmath44 and the node @xmath17 are then removed . notice that 7 zotys are removed : the 4 zotys of the two vertices @xmath19 , @xmath20 and 3 zotys from the distinguished pair . this concludes the description of the basic algorithm . its description yields that the basic algorithm is a cash - balanced clique transversal algorithm . note that if we apply the basic algorithm to a nice tree - decomposition of a @xmath12-chordal graph @xmath1 with no maximal @xmath10-cliques , we get a clique transversal of @xmath1 of size at most @xmath47 by proposition [ prop - zlotys ] . a _ branch _ is a rooted subtree of a nice tree - decomposition of a @xmath12-chordal graph such that : * its root @xmath32 corresponds to a @xmath12-clique , * the parent of @xmath32 corresponds to a clique that shares two vertices with @xmath32 , * the branch contains all descedants of @xmath32 , and * no descendant of @xmath32 corresponds to a @xmath10-clique . the following proposition easily follows from the definition of a nice tree - decomposition . [ prop - branch ] let @xmath1 be a @xmath12-chordal graph with @xmath38 maximal @xmath10-cliques . every nice tree - decomposition of @xmath1 has at least @xmath48 branches . let @xmath16 be a nice tree - decomposition of @xmath1 . suppose that @xmath17 is a node of @xmath16 corresponding to a @xmath10-clique formed by vertices @xmath49 , @xmath50 and @xmath51 . since @xmath1 is a @xmath12-chordal graph , every pair of the vertices @xmath49 , @xmath50 and @xmath51 is contained in a @xmath12-clique . in particular , for every such pair , there exists a node adjacent to @xmath17 that contains both vertices of the pair . let @xmath52 , @xmath53 and @xmath54 be these nodes for the three different pairs of the vertices @xmath49 , @xmath50 and @xmath51 . since the @xmath10-clique corresponding to @xmath17 is maximal , the nodes @xmath52 , @xmath53 and @xmath54 are different . observe that if @xmath55 , @xmath56 , is a child of @xmath17 , then the subtree rooted at @xmath55 is a branch unless it contains a node corresponding to a @xmath10-clique . let @xmath57 be a rooted tree obtained from @xmath16 as follows : the nodes of @xmath57 are the nodes of @xmath16 corresponding to @xmath10-cliques ; the root of @xmath57 is the root of @xmath16 ; and a node @xmath17 is the parent of a node @xmath24 if the path from @xmath17 to @xmath24 contains no node corresponding to a @xmath10-clique . hence , every leaf of @xmath57 will have at least two children in @xmath16 such that the subtrees rooted at them form a branch , and every node with exactly one child in @xmath57 has at least one child in @xmath16 such that the subtree rooted at it is a branch . moreover , if the root of @xmath57 has @xmath58 children in @xmath57 and @xmath59 , it has at least @xmath60 children in @xmath16 such that the subtrees rooted at them form a branch . in other words , if the degree of a node of @xmath57 is @xmath58 , this node in @xmath16 has at least @xmath60 children such that the subtrees rooted at them form a branch . since @xmath57 has @xmath38 nodes and the sum of their degrees is @xmath61 , we derive that @xmath16 must have at least @xmath62 branches . we now modify the basic algorithm to save at least one zoty while processing each branch of an input nice tree - decomposition . [ lm - branch ] there exists a cash - preserving clique transversal algorithm that saves at least one zoty when processing the nodes of each branch of an input nice tree - decomposition . consider a branch rooted at a node @xmath32 . observe that we can freely choose the order in which the nodes of the branch are processed provided that all the descendants of each node are processed before the node itself . in particular , if we find an order such that at least one of the good rules are used , we save one zoty as desired . let @xmath17 be an arbitrary leaf node of the branch and let @xmath63 be the path from @xmath17 to @xmath32 . furthermore , let @xmath64 and @xmath65 be the two vertices of the clique corresponding to @xmath32 that are also contained in the clique corresponding to the parent of @xmath32 . the node @xmath17 will be the first node of the branch to be processed . suppose first that the clique corresponding to @xmath17 has two vertices that are not contained in the clique corresponding to @xmath52 . this implies that @xmath17 corresponds to a @xmath12-clique . hence , the first two rules that apply are either rules b5 and g3 or rule g5 twice . in both cases , a good rule is applied and hence at least one zoty is saved . the remaining nodes of the branch can be processed in an arbitrary order . if the clique corresponding to @xmath17 has exactly one vertex not contained in the clique corresponding to @xmath52 , then either rule b5 or rule g5 is the first rule to apply . in the latter case , a good rule is applied , so we need only analyze the former . we process all nodes of the branch in an arbitrary order , but stipulate that the nodes @xmath66 are to be processed last . we can assume that no good rules are applied while the branch is processed otherwise , one zoty is saved . in particular , the distinguished triple created by rule b5 during the removal of the node @xmath17 can not cease to exist before processing the node @xmath52 ( without applying rule g4 ) . let @xmath5 be the largest index such that a subset of the vertices of the clique corresponding to @xmath67 form a distinguished triple prior to processing the node @xmath67 , and let @xmath68 be the set of vertices contained in the cliques corresponding to the nodes @xmath69 . by the choice of @xmath5 , either rule b2 or b3 applies to @xmath67 ; let @xmath19 be the vertex that is removed at this step . observe that the choice of @xmath5 implies that , after @xmath19 is removed , only rules b1 and b4 can apply while processing @xmath69 , i.e. , the distinguished pairs induce a matching on non - red vertices of @xmath68 with the possible exception of @xmath64 , @xmath65 or both , which may be unmatched . we modify the basic algorithm to remove the vertex @xmath19 more economically depending on which of rules b2 or b3 removed it , saving at least one zoty as required . * case 1 : the vertex @xmath19 was removed by rule b2 . * let @xmath49 and @xmath70 be the two vertices forming a distinguished triple with @xmath19 and let @xmath50 be the vertex forming a distinguished pair with @xmath19 . if @xmath71 or @xmath72 , we color @xmath50 red and remove @xmath19 , including the tuples containing it . in this way , two zotys are saved . hence , we can assume that the vertices @xmath49 , @xmath70 and @xmath50 are mutually distinct . note that at least one of these three vertices must be distinct from @xmath64 and @xmath65 . if @xmath50 is contained in a distinguished pair with another vertex , we proceed as follows . we color @xmath50 red and remove @xmath19 . we also remove the distinguished pairs containing @xmath50 and the distinguished triple @xmath73 , before creating a new distinguished pair @xmath74 . this procedure removes @xmath75 zotys ( two from each of @xmath19 and @xmath50 , three from each distinguished pair containing @xmath50 and two from the distinguished triple ) , which we reassign as follows : seven zotys are paid for coloring @xmath50 red , three zotys are reassigned to the new distinguished pair , and two zotys are saved . if @xmath50 is not contained in a distinguished pair with another vertex , @xmath50 is either @xmath64 or @xmath65 . by symmetry , we can assume that @xmath76 . consequently , at least one of the vertices @xmath49 and @xmath70 is distinct from @xmath65 . we can assume @xmath77 by symmetry . let @xmath78 be the vertex contained in a distinguished pair with @xmath49 . if @xmath79 , we swap the roles of @xmath49 and @xmath70 . in particular , we can assume that neither @xmath49 nor @xmath78 is @xmath65 . we can now proceed as follows : the vertices @xmath49 and @xmath76 are colored red while the vertex @xmath19 and all distinguished tuples containing @xmath49 or @xmath50 are removed . a total of @xmath80 zotys are removed ( two from each of the vertices @xmath19 , @xmath49 and @xmath50 ; five from the distinguished tuples containing @xmath19 ; and three from the distinguished pair containing @xmath49 ) as payment for coloring the vertices @xmath49 and @xmath50 red . we run the basic algorithm and observe that the vertex @xmath78 is removed by rule g5 ( the vertex @xmath64 is contained in all the cliques corresponding to the nodes @xmath81 ) , which results in saving two zotys . * case 2 : the vertex @xmath19 was removed by rule b3 . * if both distinguished triples containing @xmath19 also contain another vertex , say @xmath20 , we color @xmath20 red and remove the vertex @xmath19 , including the two triples containing it . in this way , we have removed eight zotys , of which seven are paid for coloring @xmath20 red and one is saved . hence , we can assume that @xmath19 is the only vertex contained in the intersection of the two distinguished triples . we proceed by coloring the vertex @xmath19 red and removing the two triples containing it . unfortunately , we are only able to pay six out of the seven zotys requisite to color @xmath19 red . we argue that we can always remove two zotys while processing the nodes @xmath82 without creating a red vertex ; one of which will pay off the one zoty debt for coloring @xmath19 red and the other will constitute the required saving in the branch . let @xmath83 be the four other vertices contained in the two distinguished triples that contained @xmath19 . let @xmath84 be the largest index such that the clique corresponding to @xmath85 contains all four of the vertices @xmath83 . by symmetry , we can assume that @xmath86 is not contained in the clique corresponding to the parent of @xmath85 . note that only rules b1 and b4 apply while processing the nodes @xmath87 . in particular , no new distinguished tuples are created and none of the vertices @xmath83 become red . if the clique corresponding to the node @xmath85 contains a red vertex , we simply apply rule g5 to @xmath86 . if the clique of @xmath85 contains no vertices in addition to @xmath83 , we may still remove @xmath86 and its two zotys since the clique is not maximal ( it is a subset of the clique corresponding to @xmath67 ) . hence , we can assume that the clique of @xmath85 contains a non - red vertex @xmath88 . we first assume that @xmath89 ( note that the case @xmath90 is symmetric ) . if no vertex from @xmath91 is @xmath65 , we color @xmath88 red and observe that all of the vertices @xmath83 are eventually removed by rule g5 . thus we have removed @xmath92 zotys : seven zotys are paid for coloring @xmath88 red , one zoty pays off the zoty debt for coloring @xmath19 red , and two zotys are saved . if one of the vertices @xmath91 , say @xmath93 , is @xmath65 , we create a distinguished pair formed from @xmath89 and @xmath94 , and when processing the vertices @xmath95 , simply remove them from the graph ( the cliques corresponding to the nodes containing @xmath95 also contain both @xmath64 and @xmath65 , so they will eventually contain a red vertex ) . in this way , a total of @xmath96 zotys are removed : one pays off the zoty debt for coloring @xmath19 red , three are assigned to the new distinguished pair , and two zotys are saved . if the vertex @xmath88 is neither @xmath64 nor @xmath65 , it must be contained in a distinguished pair with another vertex , say @xmath97 . when processing @xmath85 , we color @xmath88 red , which is paid for with the three zotys assigned to the distinguished pair containing @xmath88 and four zotys shared between the vertices @xmath86 and @xmath88 . if any of the vertices @xmath98 are removed before @xmath88 , we must have applied rule g5 . the two zoty we save from this are used to pay off the one zoty debt for coloring @xmath19 red and make the required saving in the branch . otherwise , @xmath88 is removed before the vertices @xmath98 , which implies there is a clique containing each of the vertices @xmath99 . after @xmath88 is removed , the case of the vertices @xmath100 is completely analogous to the original case of the vertices @xmath83 , and we proceed in the very same way , i.e. , we repeat the steps presented in this and the preceding two paragraphs . since the process must eventually terminate , we find one zoty to pay off the debt and also save one zoty . propositions [ prop - zlotys ] and [ prop - branch ] and lemma [ lm - branch ] yield the following . [ thm - main - a ] every @xmath12-chordal graph @xmath1 containing a maximal @xmath10-clique has a clique transversal with at most @xmath101 vertices . as identified earlier , the basic algorithm together with proposition [ prop - zlotys ] gives , in this case , a clique transversal of @xmath1 of size at most @xmath47 , which is greater than our proposed bound of @xmath101 . to attain this improvement , we need to slightly modify the algorithm depending on the structure of @xmath1 . [ thm - main - b ] every @xmath12-chordal graph @xmath1 that contains no maximal @xmath10-clique has a clique transversal with at most @xmath101 vertices unless @xmath102 . let @xmath1 be a counterexample to the statement of the theorem containing the minimum number of vertices and , subject to this , the minimum number of edges . clearly , @xmath1 is connected , in particular , @xmath1 has no isolated vertices . fix a rooted tree - decomposition @xmath16 of @xmath1 with the maximum number of leaves such that every node corresponds to a maximal clique of @xmath1 ( such a tree - decomposition exists by lemma [ lm - maximal ] ) . if @xmath1 has at most two maximal cliques , then it has a clique transversal of size one . so , we can assume that @xmath1 has at least three maximal cliques , which implies that the tree - decomposition @xmath16 has at least three nodes . in particular , the root of @xmath16 has at least two children . we now show that we may assume each leaf node of @xmath16 corresponds to a @xmath12-clique . suppose that @xmath16 has a leaf node @xmath24 that corresponds to a @xmath5-clique with @xmath30 . let @xmath44 be the set of vertices of @xmath1 that are associated with @xmath24 , and let @xmath7 be the set of vertices associated with the parent of @xmath24 in @xmath16 . if @xmath103 , then let @xmath19 be any vertex in @xmath104 and let @xmath105 be the graph obtained from @xmath1 by removing the vertex @xmath19 . if @xmath106 , then let @xmath19 be the unique vertex in @xmath104 , let @xmath20 be any vertex of @xmath107 , and let @xmath105 be the graph obtained from @xmath1 by removing the edge @xmath108 . observe that in both cases , @xmath105 is a @xmath12-chordal graph and any clique transversal of @xmath105 is also a clique - transversal of @xmath1 . however , this contradicts the choice of @xmath1 as a minimal counterexample . suppose that @xmath16 has a node @xmath17 adjacent to a leaf @xmath24 such that at least two vertices associated with @xmath24 are not associated with @xmath17 . let @xmath7 and @xmath44 be the sets of vertices of @xmath1 associated with nodes @xmath17 and @xmath24 respectively , and let @xmath109 . note that @xmath110 since @xmath1 is connected . consider a nice tree - decomposition @xmath57 such that its root corresponds to the clique induced by @xmath44 ; such a tree - decomposition @xmath57 exists by proposition [ prop - nice ] . note that the root of @xmath57 has a single child and exactly the @xmath5 vertices of @xmath104 are not associated with its only child . if @xmath111 , the subtree rooted at the only child of the root of @xmath57 is a branch and we process its nodes following lemma [ lm - branch ] , which results in saving one zoty . if one of the nodes @xmath44 is red , then we can save at least @xmath12 zotys and conclude that @xmath1 has a clique transversal of size at most @xmath112 . if no node of @xmath44 is red , we color one of the vertices of @xmath113 red , making a saving of one zoty from the eight zotys assigned to the vertices of @xmath44 ( additional zotys are saved if the two vertices of @xmath113 form a distinguished pair ) , and conclude that @xmath1 has a clique transversal of size at most @xmath114 . if @xmath29 , then exactly one vertex , say @xmath19 , of @xmath44 is also contained in @xmath7 . let @xmath20 be another vertex associated with the child of the root of @xmath57 . we process the nodes of the subtree of @xmath57 rooted at the single child of its root as in lemma [ lm - branch ] , as if the subtree were a branch sharing the vertices @xmath19 and @xmath20 with the rest of the graph . in this way , we save one zoty . if @xmath19 is red , we can save all six zotys assigned to the vertices of @xmath104 . if @xmath19 is not red , we color @xmath19 red and save one zoty out of eight zotys assigned among the vertices of @xmath44 . in total , we save at least two zotys in both cases and conclude that @xmath1 has a clique transversal of size at most @xmath101 . from now on , we assume that all but one vertex associated with each leaf node of @xmath16 are also associated with its parent in @xmath16 . we next show that the tree @xmath16 contains no node @xmath24 with a single child that is a leaf . suppose that such a node @xmath24 exists . let @xmath17 its child , let @xmath27 be its parent , and let @xmath7 , @xmath44 and @xmath115 be the vertex sets associated with the nodes @xmath17 , @xmath24 and @xmath27 , respectively . if @xmath116 , then we can modify the tree @xmath16 by replacing the edge @xmath25 with @xmath117 , which increases the number of leaves and so contradicts the choice of @xmath16 . so , we assume that @xmath118 . suppose that @xmath119 . if @xmath120 , let @xmath19 be a vertex contained in @xmath121 . otherwise , we have that @xmath122 and thus @xmath123 , and we let @xmath19 be a vertex contained in @xmath124 . remove @xmath19 from @xmath44 and let @xmath105 be the @xmath12-chordal graph corresponding to this modified tree - decomposition . any clique transversal of @xmath105 is also a clique transversal of @xmath1 , contradictory to the choice of @xmath1 as a minimum counterexample . hence , we can assume that @xmath125 . let @xmath126 and @xmath127 . by symmetry , we can assume that @xmath128 ( recall that @xmath118 and @xmath44 is a maximal clique of @xmath1 ) . we now construct another @xmath12-chordal graph @xmath129 as follows . let @xmath37 be a two - element subset of @xmath115 that contains all the vertices of @xmath130 , and let @xmath131 be the set @xmath132 , where @xmath133 and @xmath134 are two new vertices . we construct a new tree decomposition @xmath135 from @xmath16 by removing the nodes @xmath17 and @xmath24 and introducing a new node @xmath39 associated with the set @xmath136 and adjacent to the node @xmath27 . let @xmath129 be the chordal graph with the tree - decomposition @xmath135 . observe that @xmath129 is @xmath12-chordal and all maximal cliques of @xmath1 , except for those induced by @xmath7 and @xmath44 , are also maximal cliques of @xmath129 . let @xmath137 be a nice tree - decomposition of @xmath129 with its root node corresponding to the clique of @xmath129 induced by @xmath136 . observe that the subtree of @xmath137 rooted at the child of the root of @xmath137 containing a node associated with @xmath115 is a branch . hence , we can process all nodes of @xmath137 except for the root , saving one zoty by lemma [ lm - branch ] . note that neither of the new vertices @xmath133 and @xmath134 , nor their zotys , are removed during this process and that the output set intersects all maximal cliques of @xmath1 , except possibly those induced by @xmath7 and @xmath44 . one can think of this as if we had processed all nodes of @xmath16 except for @xmath17 and @xmath24 , while removing all vertices of @xmath1 except for @xmath138 , saving at least one zoty . if at least one of the vertices @xmath51 and @xmath139 is red , the set of red vertices is a clique transversal of @xmath1 and we save four zotys assigned to the vertices @xmath49 and @xmath50 . if none of the vertices @xmath51 or @xmath139 are red , we color one of them red and remove eight zotys assigned to the vertices @xmath138 ; in this way , we have saved an additional one zoty . in total , we have saved at least two zotys and so the constructed clique transversal of @xmath1 contains at most @xmath101 vertices . hence , we can assume onwards that every node adjacent to a leaf has at least two children . consider a node @xmath17 of @xmath16 that has at least two children that are leaves . let @xmath24 and @xmath27 be two such children , and let @xmath7 , @xmath44 and @xmath115 be the sets of vertices of @xmath1 associated with the nodes @xmath17 , @xmath24 and @xmath27 respectively . note that @xmath140 , @xmath141 and @xmath142 . let @xmath143 and observe that @xmath144 . we show it must hold that @xmath111 . if @xmath29 , obtain a nice tree - decomposition from @xmath16 following the proof of proposition [ prop - nice ] . note that @xmath17 has remained the parent of both @xmath24 and @xmath27 . we start processing this nice tree - decomposition by removing the two vertices contained in @xmath104 and @xmath145 , removing the nodes @xmath24 and @xmath27 , and introducing a distinguished triple formed by the vertices of @xmath146 . this results in saving two zotys . we process the rest of the graph using the basic algorithm . if @xmath147 , we fix a nice tree - decomposition @xmath57 such that its root is associated with @xmath148 with two children associated with @xmath44 and @xmath115 . such a nice tree - decomposition @xmath57 indeed exists : introduce a new node @xmath133 associated with @xmath148 , make @xmath133 a child of @xmath17 , and make both @xmath24 and @xmath27 children of @xmath133 . next , reroot the tree - decomposition at @xmath133 and subdivide edges as necessary to obtain a nice tree - decomposition . note that neither the edge @xmath149 nor the edge @xmath150 needs subdividing . we now process all the nodes of @xmath57 , except for @xmath151 , using the basic algorithm . let @xmath20 be the vertex of @xmath104 , @xmath152 the vertex of @xmath145 , @xmath153 the vertices of @xmath113 and @xmath154 the vertices of @xmath155 . note that neither @xmath20 nor @xmath152 can be contained in a distingushed tuple . if any vertex from @xmath156 is contained in either at least three distinguished triples , or a distinguished pair and another distinguished tuple , color that vertex red and remove the zotys assigned to it , including the distinguished tuples containing it . continue while such a vertex exists . note that we remove at least seven zotys each time . if two distinguished triples share at least two vertices , replace them with a single distinguished pair containing two of the shared vertices , and repeat until no such distinguished triples exist . we next analyze two cases depending on the value of @xmath5 . * * case @xmath157 : * if the vertices @xmath158 are contained in at least two distinguished tuples , then these tuples are either two disjoint distinguished pairs , a disjoint distinguished pair and triple , or two distinguished triples sharing a single vertex . in each case , it is possible to color two vertices from @xmath159 in such a way that at least one vertex from @xmath153 , at least one vertex form @xmath160 , and at least one vertex from each distinguished tuple are red . since we have removed at least 16 zotys ( @xmath75 zotys assigned among the vertices @xmath20 , @xmath152 and at least four of the vertices @xmath158 contained in the distinguished tuples , and at least four zotys assigned among the two distinguished tuples ) , we have saved at least two zotys . + so , we can assume that there is at most one distinguished tuple . if this tuple contains the vertex @xmath51 , we color @xmath51 red and remove all remaining vertices and zotys . in this way , we remove at least @xmath92 zotys ( eight zotys assigned among the vertices @xmath20 , @xmath152 and at least two of the vertices @xmath158 contained in the distinguished tuple ; and at least two zotys assigned to the distinguished tuple ) , so we have saved at least three zotys . + if the single distinguished tuple does not contain the vertex @xmath51 , we color one of its vertices red in such a way that if possible both sets @xmath161 and @xmath162 contain a red vertex . by symmetry , we can assume that we have colored @xmath139 . if none of the vertices @xmath153 are red , we color @xmath51 red ( note that @xmath163 can not be red in this case ) . we now remove all remaining vertices and zotys from the graph . if we colored only a single vertex red , we have removed at least @xmath92 zotys ( eight zotys assigned among the vertices @xmath20 , @xmath152 and at least two of the vertices @xmath158 contained in the distinguished tuple ; and at least two zotys assigned to the distinguished tuple ) , so we have saved at least three zotys . if we colored two vertices red , we have removed at least @xmath164 zotys ( @xmath80 zotys assigned among the vertices @xmath20 , @xmath152 and @xmath158 ; and at least two zotys assigned to the distinguished tuple ) , so we have saved at least two zotys . + finally , we analyze the case where there is no distinguished tuple . if none of the vertices @xmath153 are red or none of the vertices @xmath160 are red , we color the vertex @xmath51 red and remove all remaining vertices and zotys from the graph . in this way , we remove at least @xmath92 zotys ( those assigned among the vertices @xmath20 , @xmath152 and at least three of the vertices @xmath158 ) , so we have saved at least three zotys . if at least one of the vertices @xmath153 is red and if at least one of the vertices @xmath160 is red , we color no vertex red and save at least the four zotys assigned to @xmath20 and @xmath152 . * * case @xmath165 : * if the vertices @xmath166 contain at least three distinguished tuples , they must form three disjoint distinguished pairs . we color one vertex from each pair red in such a way that at least one vertex from @xmath153 and at least one vertex from @xmath167 are red . we then remove all remaining vertices and zotys . in this way , we have removed @xmath168 zotys ( @xmath164 zotys assigned among the vertices @xmath20 , @xmath152 and @xmath166 ; and nine zotys assigned among the three distinguished pairs ) , so we have saved four zotys . + if the vertices @xmath166 contain two distinguished tuples , then the distinguished tuples together cover at least four vertices and it is possible to color two of these vertices red such that at least one vertex from @xmath153 , at least one vertex from @xmath167 , and at least one vertex from each tuple are red . we then remove all remaining vertices and zotys . in this way , we have removed at least @xmath164 zotys ( at least @xmath75 zotys assigned among the vertices @xmath20 , @xmath152 , and at least four from the vertices @xmath166 ; and at least four zotys assigned among the two distinguished tuples ) , so we have saved at least two zotys . + if the vertices @xmath166 contain a single distinguished tuple , we color a vertex from this distinguished tuple red such that if possible at least one vertex from @xmath153 and at least one vertex from @xmath167 are red . by symmetry let us color the vertex @xmath139 red . if none of the vertices @xmath153 are red , we color any one of them . we then remove all remaining vertices and zotys . if the procedure colors only a single vertex red , we have removed at least @xmath92 zotys ( eight zotys assigned among the vertices @xmath20 , @xmath152 and at least two of the vertices @xmath166 contained in the distinguished tuple ; and at least two zotys assigned to the distinguished tuple ) , so we have saved at least three zotys . if two vertices are colored red , we have removed at least @xmath164 zotys ( @xmath80 zotys assigned among the vertices @xmath20 , @xmath152 and @xmath166 ; and at least two zotys assigned to the distinguished tuple ) , so we have saved at least two zotys . + finally , suppose that there is no distinguished tuple . if no vertex from @xmath153 is red , color @xmath49 red . similarly , if no vertex from @xmath167 is red , color @xmath139 red . we continue by removing all remaining vertices and zotys . we have removed four zotys assigned to the vertices @xmath20 and @xmath152 and an additional six zotys for each colored vertex . in total , we have saved at least two zotys . since we save at least two zotys in both cases , the size of the constructed clique transversal is at most @xmath101 as desired . we finish the proof by analyzing three cases based on the structure of @xmath16 . * * case : @xmath16 contains at least two different nodes such that each has at least two children that are leaves . * + let @xmath17 and @xmath169 be two such nodes ( note that one of @xmath17 and @xmath169 can be the root ) . let @xmath24 and @xmath27 be two children of @xmath17 that are leaves . let @xmath7 , @xmath44 and @xmath115 be the sets of vertices of @xmath1 associated with the nodes @xmath17 , @xmath24 and @xmath27 respectively . note that @xmath140 and @xmath170 . hence , we can assume that @xmath171 , @xmath172 and @xmath173 . in the analogous way , we define @xmath174 , @xmath175 and @xmath176 . + next , we obtain a nice tree - decomposition from @xmath16 following the proof of proposition [ prop - nice ] . observe that both @xmath24 and @xmath27 have remained as children of @xmath17 and both @xmath174 and @xmath175 have remained as children of @xmath169 . we start processing the graph @xmath1 by removing the vertices @xmath49 , @xmath177 , @xmath178 and @xmath179 , removing the nodes @xmath24 , @xmath27 , @xmath174 and @xmath175 , and introducing two distinguished pairs @xmath180 and @xmath181 . in this way , we save two zotys . the rest of the graph is then processed following the basic algorithm . * * case : the first case does not apply and the root of @xmath16 has a non - leaf child . * since the first case does not apply , the root of @xmath16 has exactly two children such that one is a leaf and the other is not . let @xmath32 be the root of @xmath16 and @xmath182 its child that is a leaf . let @xmath183 be the vertices associated with @xmath182 , and let @xmath184 be one of the vertices associated with @xmath32 but not with @xmath182 . by symmetry , we can assume that the vertices @xmath185 , @xmath186 and @xmath187 are associated with @xmath32 . furthermore , let @xmath17 be a deepest non - leaf node of @xmath16 , and @xmath24 and @xmath27 be two of its children . as in the previous case , we can assume that the node @xmath24 is associated exactly with vertices @xmath188 , the node @xmath27 is associated exactly with vertices @xmath189 , and the node @xmath17 is associated with vertices @xmath190 ( with possibly some additional vertices ) . + we now introduce a new node @xmath191 associated with the vertices @xmath192 , and make both @xmath32 and @xmath182 to be children of @xmath191 ; the node @xmath191 becomes the new root . next obtain a nice tree - decomposition from @xmath16 following the proof of proposition [ prop - nice ] . note that @xmath24 and @xmath27 have remained as children of @xmath17 , and @xmath182 has remained as a child of @xmath191 . we start processing the graph @xmath1 by removing the vertices @xmath49 and @xmath177 , removing the nodes @xmath24 and @xmath27 , and introducing a distinguished pair @xmath180 . in this way , we have saved one zoty . we then process the remaining nodes except for @xmath191 and @xmath182 . if one of the vertices @xmath185 , @xmath186 and @xmath187 is red , the vertex @xmath193 is removed using rule g5 and we save additional two zotys . otherwise , @xmath193 is removed using rule b5 and the vertices @xmath185 , @xmath186 and @xmath187 become a distinguished triple . observe that one of the good rules applies when processing the root and we save additional one zoty . in both cases , we have saved at least two zotys in total . * * case : all children of the root of @xmath16 are leaves . * observe that @xmath16 is a nice tree - decomposition . let @xmath24 and @xmath27 be two children of the root of @xmath16 . we can assume that the node @xmath24 is associated exactly with vertices @xmath188 , the node @xmath27 is associated exactly with vertices @xmath189 , and the root is associated with vertices @xmath190 ( and with possibly some additional vertices ) , but not with @xmath49 and @xmath177 . + if the root has only two children , then the vertex @xmath51 is contained in all cliques , and the statement of the theorem follows . if the root has exactly three children , then its third child is associated with exactly two of the vertices @xmath194 ; in particular , the third child is associated with @xmath51 or @xmath139 ( or both ) . by symmetry , we can assume that it is associated with @xmath51 . consequently , the vertex @xmath51 is contained in all cliques , and the statement of the theorem again follows . + we now assume that the root has at least four children , i.e. , it has two additional children @xmath174 and @xmath175 . let @xmath195 be the vertices associated with @xmath174 and let @xmath196 be the vertices associated with @xmath175 listed in such a way that the vertices @xmath197 are associated with the root ( some of the vertices @xmath190 might be among the vertices @xmath197 ) . we now process the graph @xmath1 . we start with removing the vertices @xmath49 , @xmath177 , @xmath178 and @xmath179 , removing the nodes @xmath24 , @xmath27 , @xmath174 and @xmath175 , and introducing distinguished pairs @xmath180 and @xmath181 . in this way , we save two zotys . we then process the rest of the graph using the basic algorithm . since we save at least two zotys in each of the cases , the size of the constructed clique transversal is at most @xmath101 as desired . we conclude by showing that the bound given in theorem [ thm - main ] is tight . to this end , we extend the construction presented in @xcite for @xmath198 to all values of @xmath0 . [ prop - lower ] for every @xmath4 , there exists an @xmath0-vertex @xmath12-chordal graph with no clique transversal with fewer than @xmath15 vertices . if @xmath199 , @xmath15 is equal to one and the complete graph @xmath200 shows that the assertion of the proposition is true . we next recall the construction presented in @xcite . let @xmath201 , @xmath202 , be the following graph with @xmath203 vertices ( see figure [ fig - lower1 ] ) . the vertex set of @xmath201 is formed from @xmath204 vertices @xmath205 , @xmath206 , and eight vertices @xmath207 and @xmath208 . the vertices @xmath207 form a @xmath12-clique ; each triple of vertices @xmath209 , @xmath206 , forms a @xmath10-clique ; each quadriple of vertices @xmath210 , @xmath206 , forms a @xmath12-clique ; and the vertices @xmath208 also form a @xmath12-clique . note that these @xmath211 cliques are vertex disjoint . in addition , @xmath201 contains @xmath12-cliques formed by vertices @xmath212 for @xmath206 ; vertices @xmath213 for @xmath214 ; and finally vertices @xmath215 , and @xmath216 . observe that @xmath201 is a @xmath12-chordal graph with @xmath203 vertices and @xmath211 disjoint maximal cliques . this proves the proposition for @xmath198 . for @xmath217 , we proceed as follows . let @xmath5 be the largest integer such that @xmath218 and let @xmath219 . if @xmath220 , we add @xmath221 new vertices @xmath222 to the graph @xmath201 and join each of them to all of the four vertices @xmath208 . the resulting graph is @xmath12-chordal and its clique transversal number is at least @xmath223 . if @xmath224 , we add @xmath221 new vertices @xmath222 to the graph @xmath201 such that the @xmath221 vertices are mutually adjacent , and join the vertices @xmath225 , @xmath226 to both @xmath227 and @xmath228 . the resulting graph is @xmath12-chordal . since the clique formed by the vertices @xmath222 is vertex - disdjoint from the @xmath211 disjoint maximal cliques of @xmath201 identified earlier , the clique transversal number of the resulting graph is at least @xmath229 . the first author would like to thank the undergraduate research support scheme of the university of warwick for support provided to him during the summer 2015 while carrying out this work . 9 m. aigner and t. andreae : _ vertex - sets that meet all maximal cliques of a graph _ , manuscript , 1986 . t. andreae : _ on the clique - transversal number of chordal graphs _ , discrete math . * 191 * ( 1998 ) , 311 . t. andreae and c. flotow : _ on covering all cliques of a chordal graph _ , discrete math . * 149 * ( 1996 ) , 299302 . p. erds , t. gallai , and zs . tuza : _ covering the cliques of a graph with vertices _ , discrete math . * 108 * ( 1992 ) , 279289 . c. flotow : _ obere schranken fr die clique - transversalzahl eines graphen _ , thesis , univ . hamburg , 1992 . tuza : _ covering all cliques of a graph _ , discrete math . * 86 * ( 1990 ) , 117126 . tuza : _ unsolved combinatorial problems , part i _ , brics lecture ser . ls-01 - 1 , 2001 , 30 pp .

this paper concerns what background independence itself is ( as opposed to some particular physical theory that is background independent ) . the notions presented mostly arose from a layer - by - layer analysis of the facets of the problem of time in quantum gravity . part of this coincides with two relational postulates which are thus identified as classical precursors of two of the facets of the problem of time . these are furthemore tied to the forms of each of the gr hamiltonian and momentum constraints . other aspects of background independence include the algebraic closure of these constraints , expressing physics in terms of beables , foliation independence as implemented by refoliation invariance , the reconstruction of spacetime from space . the final picture is that background independence a philosophically desirable and physically implementable feature for a theory to have has the facets of the problem of time among its consequences . thus these arise naturally and are problems to be resolved , as opposed to avoided ` by making one s physics background - dependent in order not to have these problems ' . this serves as a selection criterion that limits the use of a number of model arenas and physical theories . * edward anderson * .

1310.1524

background independence is often ( but not always ) considered in attempts to combine general relativity ( gr ) and quantum theory . standard quantum theory is _ not _ background - independent because it clings to either newton s absolute structures ( absolute time in particular ) or to the distinct absolute structures @xcite that special relativity replaced these with . i.e. minkowski spacetime as a fixed background with its other type of privileged inertial frames and its timelike killing vector . on the other hand , gr is not only a relativistic theory of gravitation but also einstein s attempt to free physics from background structures . here frames are only local and are determined by matter content , time is but a coordinate choice , and the generic gr spacetime has no killing vectors of any kind . einstein s attempt did not directly address @xcite further relational machian issues concerning time and space . however , it has since been determined that the theory of gr he arrived at in fact already manages @xcite to incorporate these . to proceed with explaining this , firstly the more standard and traditional dynamical / canonical formulation of gr wheeler s geometrodynamics @xcite is required . here gr is reformulated as a dynamics of evolving spatial 3-geometries which stack together to form the standard formulation s spacetime . this is usually described by evolving spatial 3-metrics which , unlike the description in terms of 3-geometries , have 3-coordinate ( or , better , 3-diffeomorphism ) redundancy but also the benefit of explicit computibility . secondly , if one starts out from leibniz mach relational first principles about time and space , one arrives at an equivalent variant of the geometrodynamics formulation of gr . ( a similar analysis also applies @xcite if one considers instead ashtekar variables - type dynamical formulations of gr @xcite that have dominated the canonical , but not dynamical , gr literature since the late 1980 s . ) background independence notions are often held to carry substantial physical , philosophical and conceptual content @xcite . this is for reasons that go back to the centuries - old absolute versus relative ( relational ) motion debate @xcite . it should be said that the background independence attempted from einstein to a ) ashtekar variables programs such as loop quantum gravity @xcite and b ) the present paper is at the _ metric level _ @xcite . this is as opposed to topological - level background independence , for which humankind largely remains physically and mathematically unprepared to this day ( see the conclusion ) . background independence is also a stated goal in m - theory @xcite , and furnishes one approach to supergravity ( particularly in canonical form ) @xcite . the above examples gives some idea of the opening sentence s ` often ' . on the other hand , its ` not always ' is exemplified by covariant perturbative quantization of gr and of alternative theories ( including perturbative string theory ) , and by further theories mentioned in the conclusion . some precursors for the current paper s material and presentation are as follows . most of it arose from analysing which facets of and isham s problem_s _ of time @xcite in quantum gravity already occur at the classical level @xcite . this was found to recover @xcite many of the postulates , implementations , results and insights of barbour - type relationalism @xcite . this theory of background independence can also be seen as an extension from two postulates with implementations to _ nine _ of a variant of barbour s relationalism : an extension that goes as far as covering all of the quantum aspects of the problem of time . i accredit gryb @xcite for previously considered a theory of background independence based just on ( a different variant of ) barbour s two relational postulates and implementations , with links to a small fraction of problem of time facets mentioned . however , the present work exceeds that by nine to two and by having comprehensive links to all of the known quantum and classically problem of time facets . moreover , barbour s relationalism s two postulates proved to be highly unselective among the well - known alternative theories to gr @xcite . on the other hand , the conclusion gives a number of examples of how the present paper s nine are substantially more selective in this manner . thus this paper provides a coherent enlargement of barbour s work into a theory of background independence that 1 ) is both usefully restrictive of widely proposed fundamental theories of physics . 2 ) it exposits the problem of time as a _ consequence _ of the physically , philosophically and conceptually desirable option of directly implementing ( metric ) background independence . note that 8/9ths of the possible alternative conceptual facets of the quantum problem of time already possess classical precursors @xcite . thus we start with the classical version . the outline of the current paper is as follows . after the structural considerations and toy models outlined in sec 2 , i cover temporal relationalism in sec 3 and configurational relationalism in sec 4 . these are the barbour - type postulates and implementations . i then cover constraint closure in sec 5 , expression in terms of beables in sec 6 , gr - type spacetime relationalism in sec 7 , foliation independence as attained by refoliation invariance in sec 8 , spacetime reconstruction in sec 9 . these are the 7/9ths of the aspects of background independence that bring about 7/9ths of the facets of the problem of time [ labelled exclusively by 1 ) to 9 ) in this article ] that are to be faced by _ a local _ resolution of the problem of time @xcite . the remaining ninths are globally defined specifications ( which lead to the global problem of time ) and uniqueness specifications ( which lead to the multiple choice problem of time ) . these are much less well understood in detail , as regards 1 ) the extent of the list of global sub - aspects of background independence ( and of consequent global problem of time ) . 2 ) also as regards addressing uniqueness specifications ( and of the consequent multiple choice problem ) in a manner that is both mathematically sharp and applies to a wide range of models / theories . so whilst i do give a conceptual outline of these in secs 11 and 12 , they largely remain among the frontiers of knowledge ( see the conclusion ) . sec 10 concerns the classical paths and histories counterparts of constraint closure and beables . sec 12 also summarizes the connection between this paper s background independence postulates and problem of time facets . these are labelled both with the names and conceptualizations used by kucha and isham @xcite and under my updated names for these @xcite . refer to appendix a for a mosaic of figurelets that support the main text . i adopt the position that dynamics is primary ( as opposed to e.g. spacetime being primary ) . i start with the notion of configuration @xmath0 for @xmath1 some indexing set . i.e. the generalization of the ` position ' half of ` positions and momenta ' in the case of basic point - particle mechanics . configuration space @xmath2 is then the space of all possible values of the configurations that a system can possess . one is then to build composite objects out of the configurations . these include the _ changes _ of configuration @xmath3 , actions , and the momenta @xmath4 that correspond to the @xmath5 . some simple examples of configuration spaces are as follows ; these also serve to introduce the toy models employed in the current paper . example 1 ) for the mechanics of @xmath6 particles in dimension @xmath7 in flat space the simplest ( and redundant ) configuration space is @xmath8 @xcite . this is the space of possible values of the particle positions @xmath9 . [ underline denotes @xmath7-dimensional vector and @xmath10 a particle label running from 1 to @xmath6 . ] example 2 ) background - independent formulation of mechanics * relational particle mechanics ( rpm ) * @xcite comes in scaled and pure - shape versions @xcite . these already possess 5/9 of the aspects of background independence at the classical level and 6/8 at the quantum level @xcite , rendering them useful model arenas for quite a few problem of time investigations . the configuration space for this example is @xmath11 for the scaled rpm of @xmath6 particles with the redundant translations with respect to an absolute origin trivially taken out by splitting out and discarding the centre of mass motion . this is the space of possible values of a basis of relative interparticle ( cluster ) separations , most conveniently jacobi s , @xmath12 , for label @xmath13 running from 1 to @xmath14 . in 1-@xmath7 the scaled rpm has no other redundancies , so this manages to be a relational model with very simple and well - known configuration space mathematics . ( thus it has well - understood dynamics and quantum theory . in this 1-@xmath7 case we drop the underline on the @xmath15 . ) each of the above two examples have an obvious euclidean metric on configuration space ( the @xmath16 one rather than the spatial @xmath17 one etc ) . example 3 ) for full gr , a redundant configuration space is @xmath18 @xcite . i.e. the space of all positive - definite 3-metrics @xmath19 on the fixed topological manifold @xmath20 . the latter is taken here for simplicity to be compact without boundary , and which is interpreted as the theory s notion of space . [ @xmath21 are the spatial coordinates and @xmath22 are spatial indices running from 1 to 3 . ] riem(@xmath23 ) furthermore possesses a physically - realized curved ( + + + + + ) signature per space point metric . i.e. the gr configuration space metric alias inverse dewitt supermetric , @xmath24 . example 4 ) a simpler subcase of example 3 ) is * minisuperspace * ( @xmath20 ) @xcite : the space of homogeneous positive - definite 3-metrics on @xmath20 . these are notions of space in which every point is the same . this possesses an overall ( rather than per space point ) curved ( + + + + + ) ` minisupermetric ' . nested easier subcases within this are diagonal minisuperspace [ flat ( + + ) minisupermetric ] and isotropic minisuperspace [ flat ( ) minisupermetric ] , for instance for @xmath25 with standard hyperspherical metric . minisuperspace @xcite has 7/9ths of the aspects of background independence classically and 8/9ths quantum mechanically , though quite a few of these imply _ very quickly resolved _ problem of time facets by virtue of homogeneity . while the rpm and minisuperspace cases are simple to manipulate , they miss the subtleties specifically associated with diffeomorphisms @xcite . example 5 ) a first model in which these appear nontrivially is * halliwell hawking s model * of inhomogeneous perturbations about spatially-@xmath26 minisuperspace . this models a quantum - cosmological origin for the cmb hot - spots and galaxies . rpm s and minisuperspace complementarily support by one or the other having all _ other _ background independence / problem of time aspects of this last model . here one considers the first few ( usually two ) orders of the perturbation of the metric . each of these form a simplified configuration space in place of the full riem(@xmath20 ) that are currently under investigation . this aspect of background independence adopts the leibnizian ` there is no time for the universe as a whole ' principle @xcite at the primary level for closed universes . at the classical level , this is mathematically implemented by the following . \i ) postulate _ geometrical jacobi synge type actions_. these _ happen _ to be parametrization - irrelevant ( a mathematically straightforward but conceptually - curious dual property ) , which is equivalent in turn to the earlier reparametrization - invariant conceptualization @xcite of this mathematical implementation . my argument is this way round using and being named for the dual geometrical property on the grounds that it is a conceptual advance for background - independent physics to not name one s entities or techniques after physically - irrelevant properties . ii ) additionally , such actions are not to contain any extraneous time such as newton s absolute time . nor are they to contain any extraneous time - like variables such as the adm lapse of gr ( named after the notion of ` time elapsed ' ) . the euler lagrange formulation of mechanics and the arnowitt deser misner ( adm ) formulation of gr immediately fail to obey ii ) . however , they are physically equivalent to the below formulations which do obey both i ) and ii ) . example 1 ) jacobi s action principle @xcite for spatially - absolute mechanics . example 2 ) another case of jacobi s action principle occurs for scaled 1-d rpm with translations trivially removed @xcite . example 3 ) misner s principle @xcite for minisuperspace gr . all three of these actions principles are geometrical _ geodesic principles _ with respect to the physical line element @xmath27 . alternatively , they are _ parageodesic _ ( conformal to geodesic ) principles with respect to the configuration space line element . these two alternatives can be summarized for all three examples at once as follows : s = : = s . [ pi - geom - gestalt ] here @xmath28 is the geometrical ` kinetic arc element ' built out of the configuration space metric @xmath29 . also @xmath30 is the ` potential factor ' . this takes the form @xmath31 for spatially - absolute mechanics , @xmath32 for 1-@xmath7 scaled rpm or @xmath33 for minisuperspace ( and , in sec 4 , generic , gr ) . @xmath34 is total energy and @xmath35 is potential energy . @xmath36 generically of the form @xmath37 $ ] is the ricci scalar corresponding to @xmath38 [ though for minisuperspace it is just for function dependence , @xmath39 $ ] for functional dependence and @xmath40 $ ] for mixed function ( before the ;) and functional ( after the ;) dependence . this then leaves me with @xmath41 for ordinary brackets in equations . ] @xmath42 ) ] and @xmath43 is the cosmological constant . @xmath44 itself is the physical arc element that is conformally related to the kinetic arc element by the overall conformal factor @xmath45 . finally , ` ( para)geodesic ' indeed implies ` geometrical ' . moreover , in the given examples this refers to a particularly simple notion of geometry . i.e. riemannian geometry in the first two cases and semi - riemannian indefinite - signature geometry in the last case . the geometrical arc - element for these all take the _ quadratic _ form of the square root of a sum of squares as given in the above expression for the action . synge s extension @xcite involves a wider range of physical theories being encoded by action principles built out of arc elements that encode more general notions of geometry than the above . the following consequences ensue . i ) dirac @xcite noted that primary constraints are implied by reparametrization - invariant actions . this argument clearly carries over to geometrical actions that happen to be , dually , parametrization - irrelevant actions . thus it can be recast as ` primary constraints are implied by temporally - relational actions ' . thus how the action for minisuperspace gr manages to encode the hamiltonian constraint , : = n_ijkl^ij^kl/ - r = 0 . [ ham ] this is relevant because in the more commonly used adm approach , ( [ ham ] ) arises from variation with respect to the lapse , which in turn is necessarily absent from misner s action for this to implement temporal relationalism . so there is an initial worry that the crucial constraint @xmath46 may , somehow , have gone missing . however , dirac s insight pin - points how upon formulating gr free of lapse , a distinct primary constraint route takes the place of the variation with respect to the lapse route to @xmath46 of the adm formulation . thus in the background - independent paradigm for physics , the important hamiltonian constraint of gr @xmath46 arises directly from the demand of temporal relationalism . indeed , furthermore , the precise form of the gr @xmath46 is dictated by the precise ( quadratic ) way the action is built to be temporally relational . this in turn induces @xmath46 s quadratic dependence on the momenta , which , moreover , is well - known to cause the schrdinger picture manifestation of the most well - known quantum problem of time facet : the frozen formalism problem . thus indeed temporal relationalism the background independence aspect is a deeper and already classically - present replacement for the ` frozen formalism problem ' problem of time facet . for spatially - absolute mechanics following from the jacobi action , the quadratic energy constraint : = _ i^2/2m_i + v(^i ) = e plays an analogous role to @xmath46 , and the @xmath47 , @xmath48 of this does likewise for 1-@xmath7 scaled rpm . in each case , the quadratic form of the geometrical arc - element in the action gives rise to a purely quadratic constraint ` quad ' that then results in a quantum frozen formalism problem . this is the first of a number of justifications for considering such mechanics as useful model arenas for conceptually understanding some parts of the more elaborate case of gr itself . \ii ) the above primary - level timelessness from adopting the background independence leibniz s principle of temporal relationalism is , moreover , both already classically present and easier to reconcile with time nonetheless appearing to play a prominent role in the world around us . this reconciliation involves classical time being emergent at a secondary level via _ mach s time principle _ : ` time is to be abstracted from change ' . three distinct proposals for this then involve ` any change ' ( rovelli @xcite ) , ` all change ' ( barbour @xcite ) and my _ sufficient totality of locally significant change ( stlrc ) _ @xcite . as detailed in @xcite , all three of these proposals have some sense in which they are ` democratic ' however , only the last two take into consideration that ` some clocks are better than others ' is an essential part of accurate timekeeping @xcite . additionally , only the first and the third are operationally realizable . thus overall , stlrc wins out . the time abstracted from this is a _ generalization _ of the astronomers _ ephemeris time _ @xcite that emphasizes that such a procedure is in practise _ local_. thus i term it a ` glet ' , and posit the specialization of the machian emergent time resolution to ` glet is to be abstracted from stlrc ' . for the actions in question , _ emergent jacobi time _ is such an emergent time . this time arises as a simplifier @xcite of the change - momentum relations and jacobi equation of motion . these are the temporal relationalism complying equivalents of velocity - momentum relations and euler lagrange equations . it takes the form to incorporate selection of ` calendar year zero ' , @xmath49 . ] ^ ( ) = s / . . at first sight , this is an ` all change ' resolution , but a careful look reveals that in practise it is indeed a ` stlrc ' resolution . it amounts to a relational recovery of newtonian , proper and cosmic times in the contexts in which one expects to have each such notion of time . finally , suppose that one is in the presence of a heavy light ( @xmath50@xmath51 ) split , such as for cosmology for @xmath50 scale and @xmath51 small inhomogeneities . then this scheme is only fully machian once one passes from zeroth - order emergent times of the form ^()_0 = f[h , h ] to at least - first order emergent times of the form @xcite ^()_1 = f[h , l , h , l ] . [ temj1 ] such schemes indeed provide the classical @xmath51-degrees of freedom with the opportunity to contribute to the timestandard . \i ) now as well as starting with a geometrically - simple but redundant configuration space @xmath2 , one considers also a group @xmath52 of continuous transformations acting on the @xmath2 that are taken to be physically irrelevant . this encompasses both \a ) _ spatial relationalism _ : translations and rotations relative to absolute space in mechanics , or 3-diffeomorphisms in gr . b ) _ internal relationalism _ : a reformulation of the more familiar type of gauge theories from particle physics . the meaning of configurational relationalism is that the true physics resides in the quotient space @xmath53 . for example , for scaled rpm s this is _ relational space _ @xmath54 . for pure - shape ( i.e. additionally scale - free ) rpm s , it is _ shape space _ @xmath55 . here rot(@xmath7 ) are rotations , dil are dilations and denotes semidirect product . on the other hand , for gr @xmath53 is usually taken to be wheeler s @xcite _ superspace _ , superspace(@xmath20 ) = riem(@xmath20)/diff(@xmath20 ) . moreover cases can be made to consider _ conformal superspace _ , cs(@xmath20 ) = riem(@xmath20)/conf(@xmath20 ) diff(@xmath20 ) or _ cs + v _ , \{cs + v}(@xmath20 ) = riem(@xmath20)/vpconf(@xmath20 ) diff(@xmath20 ) . here , diff(@xmath20 ) are spatial 3-diffeomorphisms and conf(@xmath20 ) are conformal transformations . v stands for ` volume ' , i.e. the spatial volume of @xmath20 a merely global - valued variable . then vpconf(@xmath20 ) are volume - preserving conformal transformations . conformal superspace is the space of conformal 3-geometries , so this case is a theory of _ conformogeometrodynamics_. \{cs + v}(@xmath20 ) adjoins to this the dynamics of a single volume degree of freedom . both of these spaces have close ties to prevalent approaches to the initial - value problem for gr @xcite . configurational relationalism can be mathematically implemented in an indirect manner in a very wide range of circumstances by the following ` @xmath52-act @xmath52-all ' method . given an object @xmath56 that corresponds to the theory with configuration space @xmath2 , one first applies a group - action of @xmath52 to this denoted @xmath57 . then one applies some operation @xmath58 that makes use of all of the @xmath59 so as to cancel out the appearance of @xmath60 in the group action , such as summing , integrating , averaging or extremizing over @xmath52 . one example of such an implementation is for @xmath56 a classical action built upon @xmath2 with application of the basic infinitesimal group - action to obtain s _ = _ ( ) _ g s , _ gs : = ||- _ g || _ . then one extremizes over @xmath52 as per the variational principle now also including variation with respect to @xmath60 .. ] this particular example is barbour s _ best matching _ the extremization produces an equation , lin@xmath61 , that , in the @xmath62 variables formulation , is to be solved for the auxiliary variables @xmath63 themselves and then substituted back into the action . this produces a final @xmath52-independent expression that _ directly _ implements configurational relationalism . note the extra @xmath64 factor in the gr case . in more detail , this method initially looks to be ` taking a step in the wrong direction ' by extending @xmath2 to the bundle @xmath65 that appends the further unphysical @xmath52-auxiliary degrees of freedom to the redundancies already present in @xmath2 . however , lin@xmath61 , thus named for its linearity in the momenta , is , unlike the preceding quadratic constraints , uncontroversially a gauge constraint . gauge constraints these use up _ two _ degrees of freedom per @xmath60 degree of freedom . thus each degree of freedom appended wipes out not only itself but also one of @xmath2 s redundancies , so that one indeed ends up on @xmath53 as required to implement configurational relationalism . note that the expression given involves formulating the actually - present auxiliary variables as @xmath63 rather than as , in the earlier literature @xcite , @xmath60 . this is necessary for these not to spoil the parametrization - irrelevance that implements temporal relationalism . moreover , then one might naively expect cyclic equations ( jacobi @xmath66 parallel of the conventional lagrangian @xmath67 ) in place of multiplier equations ( jacobi @xmath68 parallel of the conventional lagrangian @xmath69 ) . however , @xmath60 is entirely physically meaningless . thus in particular its value at the end points ( or , more generally , end nos ) of the variation are meaningless . so it is to be subjected to free end nos variation , and this fixes @xmath70 at the end nos , but @xmath71 is @xmath72-constant , so @xmath70 everywhere . thus using cyclic differentials @xmath63 in place of a lagrange multiplier implementation of the @xmath52-auxiliaries leaves the equations of one s theory unchanged . additionally , in the presence of configurational relationalism , the preceding sec s resolution of temporal relationalism picks up a complication : ^ ( ) = ( _ g s /. [ t - jbb ] jbb stands for ` jacobi barbour bertotti ' @xcite , and the complication in question can be summed up as ` more implicit via containing an extremum ' and ` formulated in terms of unphysical auxiliaries @xmath63 ' . however , if one succeeds in carrying out best matching , the @xmath63 are replaced by an extremal expression solely in terms of @xmath62 . by this , both aspects of this complication are washed away and one has an expression for @xmath73 just like the previous sec s ( but in terms of the _ reduced _ configuration space s geometry ) . example 1 ) the scaled relational mechanics case of ( [ t - jbb ] ) follows from the relational action [ see fig [ rpm - coordi].b ) ] , s _ b s : = _ ^3x _ f s _ f s : = ||_f h|| _ , ^ ( ) = ( _ f s / . ) . configurational relationalism the background independence aspect is related to one of the problem of time facets as follows . it is a threefold generalization of the thin sandwich problem facet @xcite . \i ) simply note that this is the subcase of this corresponding to @xmath74 and @xmath75 _ is _ the thin sandwich problem when cast in terms of solving for an adm or baierlein sharp wheeler ( bsw ) @xcite multiplier coordinate shift @xmath76 see fig 2.a ) and b ) for an outline of the thin sandwich prescription and its thick sandwich precursor ; note that this problem of time facet is indeed usually considered at the classical level @xcite . one can then pass to solving for the simultaneously temporal relationalism complying cyclic differential of a frame variable , @xmath77 , without altering in any @xcite way the mathematical form of the thin sandwich equation @xcite . \ii ) reconceptualize in terms of any @xmath2 and a @xmath52 that is compatible with @xmath2 , and one has the best matching implementation of configurational relationalism \iii ) reconceptualize so that the elimination is at any level rather than at the lagrangian ( @xmath78 variables ) or jacobian ( @xmath62 variables ) level , and one has the fully general ` @xmath52-act , @xmath52-all ' implementation of configurational relationalism . additionally , best matching is explicitly solvable for example 1 ) 1- and 2-@xmath7 @xcite ( as is its pure - shape counterpart ) . by use of kendall s shape theory @xcite the simplest configuration space geometries for 1- and 2-@xmath7 rpm s are @xmath79 and @xmath80 ( pure shapes ) . then by further use of the coning construction @xcite , @xmath81 and @xmath82 ( shapes and scale ) . the first three of these are very well known as geometries and as regards subsequent classical and quantum mechanics thereupon and supporting linear methods of mathematical physics . these render many qm and problem of time strategy calculations tractable and available for comparison with each other , which is a rarity in the latter field . triangleland is further aided in this way by @xmath83 and @xmath84 albeit the latter is not the usual flat geometry . it is , however , conformally flat @xcite . the ensuing mathematics s simplicity , which extends even well into the usually complex rearrangements necessary for the investigation of problem of time strategies , is a major asset in this rpm model arena . this is because it secures many computational successes beyond the usual points at which these break down for full gr / many other toy models . pure - shape rpm configuration spaces are analogous to conformal superspace ( cs ) @xcite for gr , and scaled ones to wheeler s superspace in one sense and to cs + volume @xcite in another sense . @xcite demonstrated solvability , by mixture of basic maths and interdisciplinarity with statistical theory of shape , molecular physics and a few other areas for @xmath6-a - gonlands ( particle physics , instantons ) . for use below , scaled triangleland s reduced configuration spaces possesses cartesian coordinates that are non - obviously related to the merely - relative jacobi coordinates . i.e. the ` dragt coordinates ' @xcite dra_1 = 2_1 _ 2 , dra_2 = 2\{_1 _ 2}_3 , dra_3 = _ 2 ^ 2 - _ 1 ^ 2 ( the form of which arises by what is more widely known as the ` hopf map ' ) . these are all cleanly interpretable as the product of a scale factor @xmath10 and a lucid shape quantity . respectively , these shape quantities are an anisoscelesness , four times the mass - weighted area of the triangle and the ellipticity ( difference of partial moments of inertia ) . \ii ) much as temporal relationalism came with a second statement about no extraneous times , configurational relationalism comes with a second statement of no extraneous spatial metric . [ in all cases in this paper , accompanying ` no extraneity ' statements are metric geometry extraneities , the lapse itself being interpretable as playing the role of a 1-metric of time . ] this paper now branches , firstly into canonical bracket structure in secs 5 and 6 and then into spacetime structure in secs 7 , 8 and 9 . as well as the @xmath5 and @xmath4 introduced above , one requires to endow the combined space of these with the further poisson brackets structure , @xmath85 . then the constraints have poisson brackets among themselves . for all of rpm s , gr and minisuperspace , the constraints already found close algebraically at the classical level without further issue . cases with best matching explicitly attained are aided in the matter of closure by a ) having less constraints to make brackets out of . b ) by single finite - theory classical c - number constraints always closing with themselves by symmetric entries into an antisymmetric bracket . in more detail , for gr , the constraints poisson brackets form the dirac algebroid and d@xmath86 for @xmath87 and d@xmath88 , d@xmath89 for @xmath46 . @xmath90 \ , \d a^{\tfw}(z^i)$ ] is an ` inner product ' notation for the smearing of a @xmath91-tensor density valued constraint @xmath92 by an opposite - rank differential @xmath91-tensor smearing with no density weighting , @xmath93 . ] ( m_i | l^i ) ( m_j | m^j ) = ( m_i | [ l , m ] ^i ) , [ m , m ] ( h | k ) ( m_i , l^i ) = ( _ l h | k ) , [ h , m ] ( h | j ) ( h | k ) = ( m_i | j ^i k ) , [ h , h ] where @xmath94 ( a notation familiar from qft ) . the first bracket means that the 3-diffeomorphisms on a given spatial hypersurface close in the usual way as a lie algebra . [ i.e. the right - hand side is of the form of a sum of constraints each multiplied by a _ structure constant_. ] the second bracket means that @xmath46 is a good object ( a scalar density ) under 3-diffeomorphisms . the third bracket , however , is more complicated in both form and meaning . for now , i note that the presence of @xmath95 on its right - hand side is a structure function . thus this bracket renders the overall algebraic structure of the brackets more complicated than a lie algebra . it is , rather , an algebroid ; in particular , it is the _ dirac algebroid_. this is entirely unlike the _ unsplit _ gr s 4-diffeomorphisms of spacetime @xmath96 which form a genuine lie algebra just like the 3-diffeomorphisms do . the dirac algebroid is a highly diffeomorphism - specific structure . in the case of minisuperspace , it has the last right - hand side zero by homogeneity ( @xmath97 annihilates everything ) . i also note that further interpretation of the third bracket is largely the basis of secs 8 and 9 . constraint closure is indeed a necessary check since the constraints that one readily finds can sometimes imply further conditions . dirac @xcite gave a systematic treatment for this at the classical level , in terms of poisson brackets @xmath98 on the space of the @xmath5 and @xmath4 . one s theory s constraints poisson brackets are capable of the following @xcite . \i ) producing further constraints . \ii ) or of diagnosing these or pre - existing constraints as in fact being second - class . \iii ) or of producing a further type of equation called ` specifiers ' that fix the forms of some of the theory s auxiliary quantities . further constraints thus produced may mean that the @xmath52 one intended to enforce as physically meaningless is rejected or enlarged by the physics of one s theory . second - class constraints thus diagnosed do not have the property of using up a total of two degrees of freedom per auxiliary @xmath52-degree of freedom . thus the presence of such may be sending the physical theory to other than the intended quotient . these issues are met by some of the following means . \i ) sometimes one s attempted physical theory has to be abandoned . not all theories one can write down are consistent ! dirac illustrated this with the lagrangian @xmath99 , whose euler lagrange equation reads 0 = 1 ... \ii ) sometimes the @xmath52 one initially selected has to be abandoned . for instance , attempting gr - like metrodynamics instead of geometrodynamics turns out to be impossible because @xmath87 turns out to be an integrability of @xmath46 . \iii ) one can try to deal with any second - class constraints one finds by gauge - fixing them . or by extending the phase space . or by replacing the poisson bracket with a dirac bracket that factors out the second - class constraints @xcite . having found constraints and introduced the classical poisson brackets structure , one is additionally free to ask about which objects have zero poisson brackets with the constraints . these objects , termed _ observables _ or _ beables _ , are more useful than just any @xmath5 s and @xmath4 s or functionals of these . this usefulness follows from their containing physical information only . at the very least this is required to be present in final answers to questions posed to a physical theory . my particular program emphasizes _ beables _ : quantities that just ` are ' , rather than carrying any connotations of ` observing ' . at the classical level , this concept is already more suitable from a cosmological perspective @xcite . one still has an ambiguity as to which notion of beables to use . _ dirac beables _ are required to have zero brackets with all of the constraints : @xmath100 . _ beables _ are less stringent in being required to have zero brackets only with the linear constraints : @xmath101 . note 1 ) beables are the end - product of succeeding to find a full algebra of gauge - invariant quantities . the ` poisson brackets ' in use might in fact belong to an extended phase space , or be dirac brackets @xcite constructed so as to remove second - class constraints . the gauge - invariant quantities in question correspond to the successful realization of configurational relationalism s @xmath52 as a means of passage to a less redundant quotient configuration space @xmath53 . note 2 ) immediately from the jacobi identity on two constraints and one beable , the ` poisson bracket ' of two beables is itself a beable . thus each of dirac and beables constitute a further classical algebra . an alternative choice suggested are _ rovelli partial observables _ @xcite , which have no such requirements and are not physical quantities per se , though correlations between pairs of such are physical quantities . one often imagines each as measured by a localized observer , hence this concept is usually expressed as observables rather than as beables . moreover one can conceive of beables with all the other rovellian properties , that are e.g. locally decohered by processes bereft of the trappings of measurements and observers . [ rovelli s scheme also makes use of ` total observables ' that are similar to dirac s notion of observables . ] see secs [ sr ] , [ ph ] for yet more notions of beables / observables . the connection with a problem of time facet is as follows . and especially dirac beables are hard to find , and even more especially at the quantum level . this accounts for the problem of beables ( more often called ` problem of observables ' ) facet being a problem . that it may be a forteriori a problem of time facet is tied to commutation with @xmath46 also possibly carrying connotations of constants of the motion , i.e. the conundrum that one s desirable beables might also all be frozen in time . additionally , at least directly formulated dirac beables have to contend with possible nonexistence , even at the classical level , due to nonintegrability / chaos . some benevolent cases are as follows . \i ) trivial configurational relationalism ( or resolved best matching ) readily imply possession of a full set of classical beables . e.g. explicitly for the relational triangle , = f[dra , ^dra ] , for @xmath102 the lucid scaled - up shape quantities introduced in sec 3 with conjugate variables @xmath103 . such shape quantities are also available for the general 1- and 2-@xmath7 rpm and for diagonal minisuperspace models . \ii ) in the case of trivial configurational relationalism , halliwell provided @xcite an indirect prescription for dirac beables . i.e. objects commuting with quad also , which i further promoted to the case of resolved best matching too @xcite ( constructed for the relational triangle ) . the problem of beables then consists of finding objects which brackets - commute with all the constraints ( dirac beables ) or perhaps just with the linear constraints . \iii ) i distinguish between specific and merely formal resolutions of the problem of beables . e.g. for the relational triangle , one has a specific set of shape quantities ( sec [ cr ] ) . whereas for gr one can only talk formally in terms of the spatial 3-geometries and their conjugates . \i ) now as well as considering a spacetime manifold @xmath96 , consider a @xmath104 of irrelevant transformations acting upon @xmath96 . \ii ) also do not consider any extraneous spacetime structures , in particular indefinite background spacetime metrics . for gr , @xmath105 = diff(@xmath96 ) , which , as mentioned above , _ do _ form a lie algebra , | [ ( _ |x^ ) , ( _ |y^ ) ] | = ( _ |[x , y]^ ) . there is still an issue as to what role the diff(@xmath96 ) generators @xmath106 play here . unlike with sec 4 s diff(@xmath20 ) auxiliaries , these @xmath106 are not conventionally associated with dynamical constraints . nor is the above classical realization of a lie bracket conventionally taken to be a poisson bracket . because of that , there is conventionally no complete spacetime analogue of the previous sec s notion of beables / observables . however , the notion of diff(@xmath96 ) invariant quantities given by objects @xmath107 such that | [ ( _ |x^ ) , ( s_|z^ ) ] | = 0 does remain useful and physically meaningful . the jacobi identity applying to all lie brackets , these @xmath107 are themselves also guaranteed to close as a lie algebra . moreover , the well - known einstein hilbert action of the spacetime formulation of gr is built out of a good spacetime scalar . thus there are no manifest corrections at the level of the action in this case . [ the @xmath108 s that entered sec 4 s actions did so via their split - off kinetic terms not constituting good @xmath52 objects without such corrections . ] on the other hand , corrected - derivative entities analogous to best matching do still exist for spacetime , even though these are now to have an interpretation in terms of an auxiliary rather than physically - realized 5-manifold . e.g. gr perturbation theory can be cast as an unphysical but technically - useful 5-@xmath7 stack of spacetime 4-geometries that are inter - related via lie derivative terms @xcite . with respect to , now , a _ spacetime 4-vector_. such has the status of a _ resolved _ problem at the classical level . more usually , one makes a choice to work with one of split or unsplit spacetime . however , a few approaches to background independence and quantum gravity do _ combine _ spacetime and split spacetime concepts . thus they manifest all of temporal , spatial and spacetime relationalism at once . see sec [ ph ] for examples . _ foliation dependence _ is a type of privileged coordinate dependence . this runs against the basic principles of what gr contributes to physics . thus * foliation independence * is an aspect of background independence . * refoliation invariance * is that evolving via the dashed surface and via the dotted one in fig 2.e ) give the same physical answer . this is also known as _ path independence_. this refers to paths in gr s configuration space , each of which represents a distinct sequence of spatial 3-geometries . refoliation invariance is a property manifested by the classical canonical formulation of gr that guarantees foliation independence in this context . this occurs via fig 2.e ) being guaranteed by the poisson bracket ( [ h , h ] ) closing it up modulo a mere diffeomorphism on the second hypersurface as per the third figurelet in fig 2.g ) ( an insight of teitelboim @xcite ) . it is therefore a diffeomorphism - specific resolution . by this property , gr spacetime is not just a strutting together of spaces . rather , it manages to be many such struttings at once in a physically mutually consistent manner . even the relational form of mechanics fails to manifest this property ; this is one of the key limitations of the rpm model arena . on the other hand , minisuperspace is often considered in terms of the privileged foliation by spatially homogeneous surfaces , which greatly simplifies this facet @xcite . both these limitations are to be expected since refoliation invariance is a diffeomorphism - specific issue . moreover , just as dirac pointed out that foliation of minkowski spacetime by arbitrary spatial 3-surfaces does involve diffeomorphism - specific issues , a similar treatment does also exist for minisuperspace cosmologies . finally , hojman - kucha - teitelboim @xcite obtained the form of @xmath46 from less assumptions than forming the adm split of spacetime gr . they proceeded via the deformation algebroid for a hypersurface taking the form of the already - known dirac algebroid . this does however still presuppose ( embeddability into ) spacetime . on the other hand , barbour foster murchadha began a program @xcite ( the most recent form of which is in @xcite ) deriving the form of @xmath46 without presupposing spacetime structure . they assumed temporal and configurational relationalism instead . thus their approach additionally amounts to a classical spacetime reconstruction ( from space ) . another kind of spacetime reconstruction would be from a discrete ontology rather than a continuum [ whether for spacetime or for space : fig 2.f ) ] . we shall see in sec [ qm - sec ] that spacetime reconstruction is further motivated at the quantum level , but it does already admit a classical precursor . in general , if classical spacetime is not assumed , one needs to get it back in a suitable limit . this can be hard ; in particular , the less structure is assumed , the harder a venture it is . in the case of assuming a space continuum , mere consistency imposed by the dirac procedure whittles down a general ansatz to one of four alternatives . i.e. lorentzian , galilean , carrollian relativity and cmc ( constant mean curvature ) slicing @xcite . these arise together as the different ways to kill off an obstruction term that is a product of four factors . it is novel for such an alternative to arise from principles of dynamics considerations . [ this is as opposed to the historical dichotomy between universal local galilean or lorentzian relativity . ] it is furthermore intriguing that it gives the cmc slicing option familiar from york s work @xcite on the initial value problem as an option on the same kind of footing as this dichotomy . one might guess that @xmath107 could be viewed as beables in the distinct sense of bergmann . this is because both are gauge - invariant path notions . a path in geometrodynamical phase space corresponds to sweeping out a slicing of a spacetime , encoding both the sequence of 3-metrics and the extrinsic curvature of each of these . however , bergmann and komar @xcite note that these have invariances much larger than diff(@xmath96 ) , and their position is that the largest group of invariances is to be taken . thus their @xmath109 are distinct from @xmath107 by their connection to a larger group digg(@xmath96 ) the diffeomorphism induced gauge group in place of diff(@xmath96 ) . on the other hand , in classical histories theory @xcite , classical paths in configuration space ( which represent particular slicings of spacetime ) are elevated to have a similar canonical status to ordinary configurations . i.e. they are taken to possess conjugate _ histories momenta _ and _ histories brackets _ : poisson brackets on the ensuing histories phase space . there is then a notion of histories constraints , @xmath110 and of histories dirac and kucha beables , ^t _ , k^t _ _ = 0 , ^t _ , d^t _ _ = 0 , each of which closes as an algebra by the corresponding jacobi identity . due to having this interpretation pinned upon them , histories beables are distinct from each of spacetime beables and bergmann beables . i finally comment on approaches that are concurrently covariant _ and _ canonical . such have been outlined by savvidou @xcite , treated for toy models by kouletsis and kucha @xcite and in greater generality ( but only at the classical level ) by kouletsis @xcite . these would be expected to concurrently exhibit temporal , configurational and spacetime relationalism cast in terms of histories , and both configurational and histories beables notions . this issue is investigated further in @xcite . it would be preferable if all entities and strategies used are globally are well - defined , avoidable or accountable for in terms of a meaningful physical interpretation . a few examples of consequent classical - level global problem of time subfacets impinging upon the current paper s exposition of background independence are as follows . ( i shall give a substantially more complete list of these in @xcite , alongside various classification schemes . local in space or in time itself ? a simple issue of patching together maps , or a far less simple issue of patching together partial differential equations solutions or unitary evolutions ? ) \i ) if there is a time - function , is it defined everywhere in space and in time itself ? \ii ) if there are true dynamical degrees of freedom , are these defined everywhere in space and in time ? \iii ) how does one deal with the emergent time and the best - matching procedure globally ? \iv ) do definitions of beables hold globally , and , if not , how does one move from one set to the other ? \v ) the refoliation - invariance and spacetime reconstruction demonstrations remain only local . \1 ) * frozen formalism problem*. the quadratic hamiltonian constraint of gr gives a time - independent schrdinger equation at the quantum level in a situation in which one might expect a time - dependent one . this is well - known to be a consequence of @xmath46 being quadratic and not linear in the momenta . what is hitherto less well - known is sec 3 s exposition that this in turn is a consequence of the form of relational action and thus of the * temporal relationalism * that this implements . note 1 ) the machian classical emergent time resolution breaks down at the quantum level : it does not unfreeze the gr quantum wave equation . however , this resolution can now be replaced by the equally - machian semiclassical resolution , which additionally differs in the detail of its corrections for obvious stlrc reasons . i.e. now somewhat distinct quantum @xmath51-changes have an opportunity to contribute in place of classical @xmath51-changes . for instance , in the semiclassical approach @xcite wavefunctions of born oppenheimer wkb form = ( is(h)/)|(l , h ) , @xmath111 can be interpreted as @xcite a machian emergent semiclassical time . to zeroth order , this coincides with @xmath112 , but both fail to be machian on account of allowing neither classical nor semiclassical @xmath51-change to contribute . to first order , however , ^()_1 = f[h , l , h , |(l , h ) ] , which is clearly distinct from ( [ temj1 ] ) . note 2 ) the wave equation does not suffice to obtain physical answers , which require also an inner product input . this is tied to time issues via conservation of probability / unitarity and it is in general unclear how to select an inner product with suitable properties for quantum gravity . i.e. the * inner product problem * subfacet of the frozen formalism problem . note 3 ) one s quantum formalism may include further objects needing to be cast into temporally - relational form , such as path integrals and quantum operators . \2 ) * quantum configurational relationalism*. if one succeeds in reducing out @xmath52 at the classical level , one may not need to face a @xmath52 again at the quantum level . on some occasions however [ see 3 ) below ] , the quantum theory requires a distinct @xmath113 . were that to happen , or were one unable to reduce @xmath52 out at the classical level , one could continue to use the indirect @xmath52-act , @xmath52-all method for all subsequent objects required by one s theory . for instance , for classical notions of distance , classical and quantum notions of information and quantum wave equations and operators @xcite . in the semiclassical case , were this to happen , one needs to replace not only the integrand of ( [ t - jbb ] ) with operator - ordering- and expectation - term corrections . but also consider what to replace the classical action within the accompanying extremization condition . the first of these is considered in @xcite and the second will be written up in v4 of @xcite . \3 ) * functional evolution problem*. this is @xcite whether the hamiltonian and momentum constraints are all that one needs at the quantum level by virtue of closing and of not producing anomalies . however , this name for the problem refers only to the field - theoretic case . including the finite case also involves naming it after the portmanteau of the partial and the functional derivatives @xcite . but even then it is still only a quantum - level name , whereas the term * constraint closure problem * extends it to include its classical counterpart too . at the quantum level , this involves the following further issues . \a ) the classical bracket is replaced by the quantum commutator . however , this does not always obey the same algebraic structure as the classical bracket , for starters for global reasons @xcite . \b ) additionally , the classical constraints are replaced by operator - valued quantum constraints . however , this is subject to operator - ordering ambiguities , and to possible regularization and functional - equation well - definedness issues . furthermore , how these are operator - ordered will almost always affect the form of the quantum commutator brackets , thus feeding into both a ) and c ) . \c ) _ anomalies _ can arise . these are a type of brackets algebra obstruction that specifically alters the classical symmetry group at the quantum level . thus it can be that a suggested use of @xmath52 as a group of irrelevant transformations is accepted by one s classical theory only to be rejected at the quantum level . note 4 ) a ) , b ) and c ) need not always be subfacets of the constraint closure problem because some instances of them are unrelated to time , space or frames . thus in this case these play no part in background independence or the problem of time . \4 ) the * problem of observables * @xcite has been argued to be better named and conceived of as the * problem of beables*. this is now not only as regards the cosmological setting but also as regards the non - existence of ordinary qm s notion of observers or measuring apparatuses in many relevant universe regimes . the word ` beable ' was indeed always meant @xcite to carry further realist - interpretation connotations at the quantum level . variants of such interpretations that remain alive and well include 1 ) decoherence paradigms , 2 ) many - worlds interpretations , 3 ) histories theory and 4 ) the contextual - realist approach of doering and isham @xcite . these all have variants that can be used for whole - universe quantum cosmology . the quantum problem of beables thus involves three changes as compared to its classical precursor . i.e. it now involves zero commutator brackets between operator - valued versions of both the constraints and the beables . these further assume the end - product of background independence postulate 3 ) has been attained . plus one now has to worry about , firstly , how to select a subalgebra of one s classical beables to promote to quantum beable operators . secondly , about how one is to operator - order these so as to ensure the following . \i ) the beables close among themselves . \ii ) the beables succeed in forming zero brackets with the final selection of operator - ordered and regularized quantum constraints . the quantum problem of beables is consequently that and especially dirac beables are in general even harder to find at the quantum level than they were at the classical level . \5 ) * spacetime relationalism*. this does not occur if one simply splits spacetime and works canonically . however , if an unsplit spacetime formulation is used , one has the advantages of no frozen wave equation or inner product problem or foliation issues to worry about . at the quantum level , such unsplit approaches furthermore make use of feynman s path integral technique . whilst very successful in qft , for quantum gravity these face their own set of problems . \i ) in place of an inner product problem , there is a * measure problem*. moreover , this is diffeomorphism - specific and thus is itself directly a problem of time facet resulting from the spacetime relationalism background independence aspect . \ii ) the gravitational path integral is , at the very least , problematic as regards being well - defined . ( discrete approaches can overcome this aspect ) . \iii ) the analogue of the wick rotation familiar from qft ( ` from imaginary time to real time ' ) in general goes awry in curved spacetimes . this is due to ambiguities arising as regards which contours to use @xcite . also , qft is free to rely on the canonical approach as regards computing what its feynman rules are . thus ordinary qm ` in terms of path integrals ' can sometimes in fact be a combined path - integral _ and _ canonical approaches . then in quantum gravity , such a combined scheme would pick up the problems of _ both _ approaches . finally , one might use quantum - level histories theory instead , i.e. with more structures than are usually assumed in a path integral approach . this changes some of the mathematics involved , and is at least one way of attempting combined spacetime - and - canonical schemes . however , such programs remain largely incomplete ( @xcite is classical only ) . \6 ) * foliation dependence problem*. there is no known general counterpart at the quantum level of the classical refoliation invariance guarantee of the third figurelet of fig 2.g ) , not even at the semiclassical level . [ the present state of affairs here is well outlined in e.g. @xcite . ] \7 ) the * spacetime reconstruction problem * from space and/or a discrete ontology as per fig 2.f ) is further motivated at the quantum level as per fig 2.j ) . \8 ) * global problems of time*. these are difficulties with choosing an everywhere - valid timefunction , and further global holes in other problem of time facets . a couple of particular specifically - quantum global issues with which there has been some recent progress @xcite are as follows ( see @xcite for more ) . \a ) what does it mean to patch together unitary evolutions ? \b ) are quantum - level beables / observables defined everywhere , and , if not , how does one patch between different sets of these ? \9 ) * multiple choice problem*. this is only relevant once the quantum level is under consideration . i.e. as per fig 2.l ) , canonical equivalence of classical formulations of a theory does not imply unitary equivalence of the quantizations of each @xcite . by this , different choices of timefunction can lead to inequivalent quantum theories . postulate 9 ) ( uniqueness specifications ) for background independent schemes would then be that these are to be quantum - mechanically unambiguous . [ more precisely , there should only be physically meaningful ambiguities . ] i have characterized background independence in terms of three types of relationalism ( temporal , configurational and spacetime ) , constraint closure , expression in terms of beables and some further properties of gr spacetime which include refoliation invariance . this account in no way fully covers globality and uniqueness specifications , though i have also outlined some of these . this characterization is rather more selective than merely demanding the first two types of relationalism , which is a positive feature as regards background independence being a filter of theories . e.g. geometrodynamics and loop quantum gravity pass the filter . i have also exposited the problem of time as a _ consequence _ of the physically , philosophically and conceptually desirable option of directly implementing ( metric ) background independence . [ this is to be contrasted this with pragmatic statements to avoid the problem of time by having one s theory carry some types of background dependence . nonetheless , i emphasize that such models _ do _ remain useful as examples that carry some aspects of background independence and not others . this isolation of various aspects has aided in the present paper s quest to _ conceptually and mathematically characterize _ background independence . ] example 1 ) relational mechanics is less background independent than gr due to having no notion of refoliation and its invariance . example 2 ) so are einstein - aether theory and hoava lifshitz theory ( and causal dynamical triangulation via its giving hoava litshitz theory as its classical limit @xcite ) , due to their being foliation - dependent . example 3 ) scalar gravity ( nordstrm , einstein fokker theories ) has an absolute causal structure , see e.g. p 28 of @xcite ( as well as failing the light deflection test , by which they are physically overruled ) . example 4 ) supersymmetry can not be considered at the classical level within the paradigm given here . in fact , classical fermions can not be considered . [ these are moreover well - known to be questionable entities at the classical level on the further grounds of possessing inherently quantum - mechanical properties . ] the reason that the above can not be considered is that , whilst they can be cast in leibnizian form , fail to have an ` all change ' or ` stlrc ' machian resolution of temporal relationalism . i.e. classical fermionic change can not enter the expression for the emergent classical machian time . thus the relational status of fermions ( and consequently of supergravity and m - theory ) need to be studied specifically at the quantum level , which is beyond what the current article can cover . example 5 ) perturbative strings and covariant perturbative quantization of gravity are also excluded on grounds of possessing background metric structures . thus two major future directions for this work are considering how one can relationally formulate each of a ) supersymmetry specifically at the quantum level and b ) theories possessing extended objects ( such as strings or branes ) . a third future direction concerns semiclassical and fully quantum understanding of whether and how constraint closure , refoliation invariance and spacetime reconstruction of classical gr carry over . a fourth future direction is considering the classical and quantum discrete counterpart of the current paper . e.g. @xcite has taken a number of useful steps in this direction . [ spacetime reconstruction becomes more formidable in many such approaches , and the status of diffeomorphism invariance is less well studied than in the continuum case . ] a remaining caveat is that this paper s detailed account is solely a local theory of _ metric - level _ background independence . topological background independence is a more advanced and much more highly undeveloped subject . a few tentative steps taken so far in this direction are as follows . \i ) an extension of loop quantum gravity based not on holonomies but on monodromies has been suggested @xcite . \ii ) one could instead attempt a spectral approach to geometry such as in @xcite . * acknowledgements*. i thank those close to me for support . i thank flavio mercati for some help with making figure 2 . marc lachieze - rey , malcolm maccallum , claus kiefer , don page , reza tavakol , juan valiente - kroon and jeremy butterfield for help with my career . harvey brown and simon saunders for inviting me to speak on these matters at oxford . many of the above , sean gryb and julian barbour for discussions . grant fqxi - rfp3 - 1101 from the foundational questions institute ( fqxi ) fund , administered by silicon valley community foundation , theiss research and the cnrs , hosted with marc lachieze - rey at apc . 99 a. einstein , dialog ber einwnde gegen die relativittstheorie _ die naturwissenschaften _ * 6 * 697 ( 1918 ) ; a. einstein s foreword to m. jammer , _ concepts of space . the history of theories of space in physics . third , enlarged edition . _ ( dover , new york 1993 ) . c.j . isham , in lect . notes phys . * 434 * ( 1994 ) , gr - qc/9310031 . isham , write - up of gr14 plenary lecture , gr - qc/9510063 . butterfield and c.j . isham , in _ physics meets philosophy at the planck scale _ , ed . c. callender and n. huggett , cambridge university press ( 2000 ) , gr - qc/9903072 . kucha , in _ proceedings of the 4th canadian conference on general relativity and relativistic astrophysics _ ed . g. kunstatter , d. vincent and j. williams ( world scientific , singapore 1992 ) ; c.j . isham , in _ integrable systems , quantum groups and quantum field theories _ ed . ibort and m.a . rodrguez ( kluwer , dordrecht 1993 ) , gr - qc/9210011 . i. newton , _ philosophiae naturalis principia mathematica _ ( 1686 ) : an english translation , see e.g i.b . cohen and a. whitman ( university of california press , berkeley , 1999 ) . in particular , see the scholium on absolute motion therein ; _ the leibnitz clark correspondence _ alexander ( manchester 1956 ) ; bishop g. berkeley , _ the principles of human knowledge _ ( 1710 ) ; _ concerning motion ( de motu ) _ ( 1721 ) ; e. mach , _ die mechanik in ihrer entwickelung , historisch - kritisch dargestellt _ barth , leipzig 1883 ) : an english translation is _ the science of mechanics : a critical and historical account of its development _ ( open court , la salle , ill . 1960 ) ; j.b . barbour , _ absolute or relative motion ? vol 1 : the discovery of dynamics _ ( cambridge university press , cambridge 1989 ) ; _ mach s principle : from newton s bucket to quantum gravity _ ed . barbour and h. pfister ( birkhuser , boston 1995 ) ; o. pooley , to appear in r. batterman ( ed . ) , _ the oxford handbook of philosophy of physics _ , http://philsci-archive.pitt.edu/9055/1 . halliwell , in _ the future of theoretical physics and cosmology ( stephen hawking 60th birthday festschrift volume ) _ ed . gibbons , e.p.s . shellard and s.j . rankin ( cambridge university press , cambridge 2003 ) , gr - qc/0208018 . m. gell - mann and j.b . hartle , phys . * d47 * 3345 ( 1993 ) ; j.b . hartle , in _ gravitation and quantizations : proceedings of the 1992 les houches summer school _ ed . b. julia and j. zinn - justin ( north holland , amsterdam 1995 ) , gr - qc/9304006 ; c.j . isham and n. linden , j. math . phys . * 36 * 5392 ( 1995 ) , gr - qc/9503063 . m. bojowald , p.a . hoehn and a. tsobanjan , * 28 * 035006 ( 2011 ) , arxiv:1009.5953 ; phys . * d83 * 125023 ( 2011 ) , arxiv:1011.3040 ; p.a hoehn , arxiv:1110.5631 ; p.a . hoehn , e. kubalova and a. tsobanjan , arxiv:1111.5193 .

we define and study distributions in that we callnormal . for they are really multidimensional normal , for they have densities , compact support and many properties that resemble properties of ordinary multidimensional normal distribution . we also consider some generalizations of these distributions and indicate close relationship of these distributions to askey - wilson weight function i.e. weight with respect to which askey - wilson polynomials are orthogonal and prove some properties of this weight function . in particular we prove a generalization of poisson - mehler expansion formula .

0911.0109

the aim of this paper is to define , analyze and possibly accustom new distributions in @xmath0 . they are defined with a help of two one - dimensional distributions that first appeared recently , partially in noncommutative context and are defined through infinite products . that is why it is difficult to analyze them straightforwardly using ordinary calculus . one has to refer to some extent to notations and results of so called @xmath1series theory . however the distributions we are going to define and examine have _ purely commutative , classical probabilistic meaning_. they appeared first in an excellent paper of boejko et al . @xcite as a by product of analysis of some non - commutative model . later they also appeared in purely classical context of so called one - dimensional random fields first analyzed by w. bryc at al . in @xcite and @xcite . from these papers we can deduce much information on these distributions . in particular we are able to indicate sets of polynomials that are orthogonal with respect to measures defined by these distributions . those are so called @xmath1hermite and al - salam - chihara polynomials - a generalizations of well known sets of polynomials . thus in particular we know all moments of the discussed one - dimensional distributions . what is interesting about distributions discussed in this paper is that many of their properties resemble similar properties of normal distribution . as stated in the title we consider three families of distributions , however properties of one , called multidimensional @xmath1normal , are main subject of the paper . the properties of the remaining two are in fact only sketched . all distributions considered in this paper have densities . the distributions in this paper are parametrized by several parameters . one of this parameters , called @xmath4 belongs to @xmath5 $ ] and for @xmath6 the distributions considered in this paper become ordinary normal . two out of three families of distributions defined in this paper have the property that all their marginals belong to the same class as the joint , hence one of the important properties of normal distribution . conditional distributions considered in this paper have the property that conditional expectation of a polynomial is also a polynomial of the same order - one of the basic properties of normal distributions . distributions considered in this paper satisfy gebelein inequality -property discovered first in the normal distribution context . furthermore as in the normal case lack of correlation between components of a random vectors considered in the paper lead to independence of these components . finally conditional distribution @xmath7 considered in this paper can be expanded in series of the form @xmath8 where @xmath9 is a marginal density , @xmath10 are orthogonal polynomials of @xmath9 and @xmath11 are also polynomials . in particular if @xmath12 that is when instead of conditional distribution of @xmath13 we consider only distribution of @xmath14 then @xmath15 . in this case such expansion formula it is a so called poisson - mehler formula , a generalization of a formula with @xmath16 being ordinary hermite polynomials and @xmath17 that appeared first in the normal distribution context . on the other hand one of the conditional distributions that can be obtained with the help of distributions considered in this paper is in fact a re - scaled and normalized ( that is multiplied by a constant so its integral is equal to @xmath18 ) askey - wilson weight function . hence we are able to prove some properties of this askey - wilson density . in particular we will obtain a generalization of poisson - mehler expansion formula for this density . to define briefly and swiftly these one - dimensional distributions that will be later used to construct multidimensional generalizations of normal distributions , let us define the following sets @xmath19 & if & \left\vert q\right\vert < 1 \\ \left\ { -1,1\right\ } & if & q=-1% \end{array}% \right . .\]]let us set also @xmath20 and @xmath21 if @xmath22 . sometimes to simplify notation we will use so called indicator functions @xmath23 the two one - dimensional distributions ( in fact families of distributions ) are given by their densities . the first one has density:@xmath24defined for @xmath25 @xmath26 . we will set also @xmath27for @xmath28 considered distribution does not have density , is discrete with two equal mass points at @xmath29 . since this case leads to non - continuous distributions we will not analyze it in the sequel . the fact that such definition is reasonable i.e. that distribution defined by @xmath30 tends to normal @xmath31 as @xmath32 will be justified in the sequel . the distribution defined by @xmath33 @xmath34 will be referred to as @xmath35normal distribution . the second distribution has density : @xmath36 defined for @xmath25 @xmath37 , @xmath38 @xmath39 . it will be referred to as @xmath40conditional normal , distribution . for @xmath6 we set @xmath41 ( in the sequel we will justify this fact ) . notice that we have @xmath42 for all @xmath43 . the simplest example of multidimensional density that can be constructed from these two distribution is two dimensional density @xmath44that will be referred to in the sequel as @xmath45 and two parameters . one playing similar rle to parameter @xmath46 in two - dimensional normal distribution . the other parameter @xmath47 has a different rle . in particular it is responsible for modality of the distribution and of course it defines its support . as stated above , distribution defined by @xmath30 appeared in 1997 in @xcite in basically non - commutative context . it turns out to be important both for classical and noncommutative probabilists as well as for physicists . this distribution has been accustomed i.e. equivalent form of the density and methods of simulation of i.i.d . sequences drawn from it are e.g. presented in @xcite . distribution @xmath48 although known earlier in nonprobabilistic context , appeared ( as an important probability distribution ) in the paper of w. bryc @xcite in a classical context as a conditional distribution of certain markov sequence . in the following section we will briefly recall basic properties of these distributions as well as of so called @xmath1hermite polynomials ( a generalization of ordinary hermite polynomials ) . to do this we have to refer to notation and some of the results of @xmath1series theory . the paper is organized as follows . in section 2 after recall some of the results of @xmath1series theory we present definition of multivariate @xmath1normal distribution . the following section presents main result . the last section contains lengthy proofs of the results from previous section . we will use traditional notation of @xmath1series theory i.e. @xmath49 _ { q}\allowbreak = \allowbreak 0;$ ] @xmath50 _ { q}\allowbreak = \allowbreak 1+q+\ldots + q^{n-1}\allowbreak = \allowbreak \frac{1-q^{n}}{1-q}% , $ ] @xmath50 _ { q}!\allowbreak = \allowbreak \prod_{i=1}^{n}\left [ i% \right ] _ { q},$ ] with @xmath49 _ { q}!\allowbreak = 1,\qatopd [ ] { n}{k}% _ { q}\allowbreak = \allowbreak \left\ { \begin{array}{ccc } \frac{\left [ n\right ] _ { q}!}{\left [ n - k\right ] _ { q}!\left [ k\right ] _ { q } ! } & , & n\geq k\geq 0 \\ 0 & , & otherwise% \end{array}% \right . . it will be useful to use so called @xmath1pochhammer symbol for @xmath51 with @xmath52 , @xmath53 . often @xmath54 as well as @xmath55 will be abbreviated to @xmath56 and @xmath57 if it will not cause misunderstanding . it is easy to notice that @xmath58 _ { q}!$ ] and that@xmath59 { n}{k}_{q}\allowbreak = $ ] @xmath60 . let us also introduce two functionals defined on functions @xmath61 , @xmath62and sets : @xmath63spaces @xmath64 and @xmath65 are hilbert spaces with the usual definition of scalar product . let us also define the following two sets of polynomials : -the @xmath1hermite polynomials defined by @xmath66_{q}h_{n-1}(x|q ) , \label{he}\]]for @xmath67 with @xmath68 @xmath69 and -the so called al - salam - chihara polynomials defined by the relationship for @xmath70@xmath71_{q}p_{n-1}(x|y,\rho , q ) , \label{alsc}\]]with @xmath72 @xmath73 . polynomials ( [ he ] ) satisfy the following very useful identity originally formulated for so called continuous @xmath1hermite polynomials @xmath74 ( can be found in e.g. @xcite thm . 13.1.5 ) and here below presented for polynomials @xmath75 using the relationship @xmath76@xmath77 { m}{j}_{q}\qatopd [ ] { n}{j}_{q}\left [ j\right ] _ { q}!h_{n+m-2k}\left ( x|q\right ) . \label{identity}\ ] ] it is known ( see e.g. @xcite ) that @xmath1hermite polynomials constitute an orthogonal base of @xmath78 while from @xcite one can deduce that @xmath79 constitute an orthogonal base of @xmath80 . thus in particular @xmath81 . consequently , if @xmath82 has also @xmath1normal distribution , then @xmath83 . it is known ( see e.g. @xcite formula 13.1.10 ) that@xmath84where @xmath85 { n}{i}% _ { q}. \label{wn}\ ] ] we will also use chebyshev polynomials of the second kind @xmath86 , that is @xmath87 and ordinary ( probabilistic ) hermite polynomials @xmath88 i.e. polynomials orthogonal with respect to @xmath89 . they satisfy @xmath90term recurrences : @xmath91with @xmath92 @xmath93 some immediate observations concerning @xmath47-normal and @xmath94conditional normal distributions are collected in the following proposition : [ uwaga]@xmath95 @xmath96 @xmath97 @xmath98 @xmath99 @xmath100 @xmath101 @xmath102 @xmath103 @xmath104 @xmath105 @xmath106 pointwise . @xmath107 @xmath108 @xmath109 pointwise . @xmath95 is obvious . @xmath97 follows observation that ( [ he ] ) simplifies to ( _ 0 ) and ( [ _ 1 ] ) for @xmath110 and @xmath6 respectively . @xmath100 first two assertions follow either direct observation in case of @xmath111 or comparison of ( [ alsc ] ) and ( [ _ 1 ] ) considered for @xmath112 and then multiplication of both sides by @xmath113 third assertion follows following observations : @xmath114 @xmath115 @xmath116 , @xmath117 @xmath118 for @xmath67 which is an equation ( [ _ 0 ] ) with @xmath119 replaced by @xmath120 . @xmath107 first assertions are obvious . rigorous prove of pointwise convergence of respective densities can be found in work of @xcite . to support intuition we will sketch the proof of convergence in distribution of respective distributions . to do this we apply @xmath97 and @xmath121 . and see that @xmath122 @xmath123 and @xmath124 as @xmath125 . now keeping in mind that families @xmath126 and @xmath127 are orthogonal with respect to distributions defined by respectively @xmath128 and @xmath129 we deduce that distributions defined by @xmath128 and @xmath129 tend to normal @xmath130 and @xmath131 distributions weakly as @xmath132 since both @xmath133 and @xmath131 are defined by their moments , which are defined by polynomials @xmath134 and @xmath135 before we present definition of the multidimensional @xmath1normal and related distributions , let us generalize the two discussed above one - dimensional distributions by introducing @xmath136normal distribution as the distribution with the density @xmath137 for @xmath138 @xmath139 @xmath140 $ ] . that is if @xmath141normal then @xmath142normal . similarly let us extend definition of @xmath94conditional normal by introducing for @xmath138 @xmath139 @xmath140,$ ] @xmath143 @xmath144-conditional normal distribution as the distribution whose density is equal to @xmath145 let @xmath146 and @xmath147 , @xmath140 $ ] . now we are ready to introduce a multidimensional @xmath1normal distribution @xmath148 multidimensional @xmath1normal distribution @xmath149 is the continuous distribution in @xmath150 that has density equal to @xmath151where @xmath152 @xmath153 @xmath154 @xmath155 as an immediate consequence of the definition we see that @xmath156 . one can also easily see that @xmath157 is a shift parameter and @xmath158 is a scale parameter . hence in particular @xmath159 . in the sequel we will be mostly concerned with distributions @xmath160 following assertion @xmath95 of proposition [ uwaga ] we see that distribution @xmath161 is the product distribution of @xmath162 i.i.d . @xmath1normal distributions . another words `` lack of correlation means independence '' in the case of multidimensional @xmath1normal distributions . more generally if the sequence @xmath163 contain , say , @xmath164 zeros at , say , positions @xmath165 @xmath166 then the distribution of @xmath167 is a product distribution of @xmath168 independent multidimensional @xmath1normal distributions : @xmath169 thus in the sequel all considered vectors @xmath170 will be assumed to contain only nonzero elements . let us introduce the following functions ( generating functions of the families of polynomials):@xmath171 _ { q}!}h_{i}\left ( x|q\right ) , \label{gf_h } \\ \tau \left ( x , t|y,\rho , q\right ) \allowbreak & = & \newline \allowbreak \sum_{i=0}^{\infty } \frac{t^{i}}{\left [ i\right ] _ { q}!}% p_{i}\left ( x|y,\rho , q\right ) . \label{gf_asc}\end{aligned}\ ] ] the basic properties of the discussed distributions will be collected in the following lemma that contains facts from mostly @xcite and the paper bms . [ wlasnosci]i ) for @xmath172@xmath173 _ { q } ! & when & n = m% \end{array}% \right . .\ ] ] \ii ) for @xmath174@xmath175 \iii ) for @xmath172@xmath176 _ { q } ! & when & n = m% \end{array}% \right . .\ ] ] \iv ) @xmath177 \v ) for @xmath178@xmath179convergence is absolute , where @xmath180 is defined by ( wn ) . \vi ) for @xmath181 and @xmath182@xmath183convergence ( [ gf_h ] ) is absolute in @xmath184 & @xmath119 and uniform in @xmath119 . moreover @xmath185 is positive and @xmath186 . @xmath187 @xmath188 \vii ) for @xmath189 @xmath37 and @xmath190@xmath191convergence ( [ gf_asc ] ) is absolute in @xmath184 & @xmath119 and uniform in @xmath119 . moreover @xmath192 is positive and @xmath193 . @xmath194 \viii ) for @xmath189 @xmath195@xmath196_{q}!}% h_{n}(x|q)h_{n}(y|q ) \label{p - m}\]]and convergence is absolute @xmath197 @xmath198 & @xmath119 and uniform in @xmath119 and @xmath199 \i ) it is formula 13.1.11 of @xcite with obvious modification for polynomials @xmath75 instead of @xmath74 ( compare ( [ q - cont ] ) ) and normalized weight function ( i.e. @xmath200 ii ) exercise 15.7 of @xcite also in @xcite , iii ) formula 15.1.5 of @xcite with obvious modification for polynomials @xmath201 instead of @xmath202 and normalized weight function ( i.e. @xmath203 iv ) see ( 2.6 ) of @xcite . v ) exercise 12.2(b ) and 12.2(c ) of @xcite . vi)-viii ) the exact formulae are known and are given in e.g. ia ( thm . 13.1.1 , 13.1.6 ) and @xcite ( 3.6 , 3.10 ) . absolute convergence of @xmath204 and @xmath205 follow ( [ ogr_h ] ) and v ) . positivity of @xmath206 and @xmath205 follow formulae @xmath207 and @xmath208 . values of integrals follow ( [ gf_h ] ) and ( [ gf_asc ] ) and the fact that @xmath209 and @xmath210 are orthogonal bases in spaces @xmath78 and @xmath211 [ marginal]every marginal distribution of multidimensional @xmath1normal distribution @xmath212 is multidimensional @xmath1normal . in particular every one - dimensional distribution is @xmath1normal . more precisely @xmath213th coordinate of @xmath214 vector has @xmath215 normal distribution . by considering transformation @xmath216 we reduce considerations to the case @xmath217 . first let us consider @xmath218 dimensional marginal distributions . the assertion of corollary is obviously true since we have assertion iv ) of the lemma [ wlasnosci ] . we can repeat this reasoning and deduce that all @xmath219 @xmath220 @xmath221 @xmath222 dimensional distributions are multidimensional @xmath1normal . the fact that @xmath223 dimensional marginal distributions are @xmath1normal follows the fact that @xmath224 is a one - dimensional density and integrates to @xmath18 . [ warunkowanie]if @xmath225 , then \i ) @xmath226@xmath227 thus in particular @xmath228and @xmath229 . \ii ) @xmath230@xmath231where @xmath232 @xmath233 @xmath234 . thus in particular this density depends only on @xmath235 and @xmath236 . \i ) as before , by suitable change of variables we can work with distribution @xmath237 then following assertion iii ) of the lemma [ wlasnosci ] and the fact that @xmath238 dimensional marginal , with respect to which we have to integrate is also multidimensional @xmath1normal and that the last factor in the product representing density of this distribution is @xmath239 we get i ) . \ii ) first of all notice that joint distribution of @xmath240 depends only on @xmath241 since sequence @xmath242 @xmath243 is markov . it is also obvious that the density of this distribution exist and can be found as a ratio of joint distribution of @xmath244 divided by the joint density of @xmath245 keeping in mind that @xmath246 have the same marginal @xmath128 and because of assertion iv of lemma [ wlasnosci ] we get the postulated form . having lemma [ wlasnosci ] we can present proposition concerning mutual relationship between spaces @xmath78 and @xmath247 defined at the beginning of previous section . [ przestrzenie]@xmath248 . besides @xmath249 @xmath250 @xmath251 and @xmath252 for every @xmath253 firstly observe that : @xmath254 which is elementary to prove . we will use modification of the formula ( [ p - m ] ) that is obtained from it by dividing both sides by @xmath255 that is formula:@xmath256 _ { q}!}h_{n}\left ( x|q\right ) h_{n}\left ( y|q\right ) .\end{aligned}\]]now we use ( [ ogr_h ] ) and assertion v ) of lemma [ wlasnosci ] and get @xmath257 @xmath258hence @xmath259 and @xmath260 for every @xmath253 thus @xmath261 . conversely to take a function @xmath261 . we have @xmath262now we keeping in mind that @xmath263 is a quadratic function in @xmath264 we deduce that it reaches its maximum for @xmath265 on the end points of @xmath266 . hence we have @xmath267since for @xmath268@xmath269and we see that @xmath270@xmath271 so @xmath253 notice that the assertion of proposition [ przestrzenie ] is not true for @xmath272 since then the respective densities are @xmath130 and @xmath273 . using assertion of proposition [ przestrzenie ] we can rephrase corollary [ warunkowanie ] in terms of contraction @xmath274 ( defined by ( [ kontrakcja ] ) , below ) . for @xmath275 we have @xmath276where @xmath277 is a contraction on the space @xmath278 defined by the formula ( using polynomials @xmath75 for @xmath279 : @xmath280by the way it is known that @xmath281 is not only contraction but also ultra contraction i.e. mapping @xmath282 on @xmath283 ( boejko ) . we have also the following almost obvious observation that follows , in fact , from assertion iii ) of the lemma [ wlasnosci ] . [ regresja]suppose that @xmath284 and @xmath275 . assume that for some @xmath285 and @xmath286 \i ) if @xmath287polynomial of degree at most @xmath288 of @xmath289 then function @xmath290 must be also a polynomial of degree at most @xmath291 \ii ) if additionally @xmath292 @xmath293where @xmath294 \i ) the fact that @xmath295 is a function of @xmath236 only , is obvious . since @xmath296 we can expand it in the series @xmath297 . by corollary [ warunkowanie ] we know that @xmath298 for @xmath299 . now since @xmath300 for @xmath301 and @xmath302 we deduce that @xmath303 for @xmath304 \ii ) suppose @xmath305 . we have @xmath306 . hence @xmath307 _ { q}!\leq r^{2}\sum_{i=1}^{\infty } g_{i}^{2}\left [ i\right ] _ { q}!\allowbreak = \allowbreak r^{2}\mathbb{e}g^{2}(x_{i}).$ ] as it follows from the above mentioned definition , the multidimensional @xmath1normal distribution is not a true generalization of @xmath308dimensional normal law @xmath309 . it a generalization of distribution @xmath309 with very specific matrix @xmath310 namely with entries equal to @xmath311 for @xmath312 and @xmath313 for @xmath314 where @xmath315 ; @xmath316 are some positive numbers and @xmath317 @xmath318 follows the fact that two dimensional @xmath1normal distribution of say @xmath319 has density @xmath320 if @xmath321 suppose that @xmath322 then @xmath323 form a finite markov chain with @xmath324 @xmath215normal and transition density @xmath325-conditional normal distribution following assertions vi ) and vii ) of lemma [ wlasnosci ] we deduce that for @xmath326 @xmath37 functions @xmath327 and @xmath328 are densities . hence we obtain new densities with additional parameter @xmath184 . this observation leads to the following definitions : let @xmath329,t^{2}<1/(1-q),x\in s\left ( q\right ) $ ] . a distribution with the density @xmath330 will be called _ modified _ @xmath331__normal ( briefly _ _ @xmath331mn distribution ) . we have immediate observation that follows from assertion vi ) of lemma wlasnosci . [ mn]i ) @xmath332 \ii ) let @xmath333mn . then for @xmath334 \i ) using assertions vi ) and viii ) of the lemma [ wlasnosci ] we get : @xmath335_{q}!}h_{i}\left ( y|q\right ) \sum_{j=0}^{\infty } \frac{\rho ^{j}}{[j]_{q}!}h_{j}\left ( y|q\right ) h_{j}\left ( x|q\right ) dy.\end{aligned}\]]now utilizing assertion ii ) of the same lemma we get : @xmath336_{q}!}h_{i}\left ( x|q\right ) \allowbreak = \allowbreak \varphi \left ( x , t\rho |q\right ) f_{n}(x|q).\end{aligned}\ ] ] to get ii ) we utilize assertion vi ) of the lemma [ wlasnosci ] . in particular we have : if @xmath337 , then @xmath338 @xmath339 @xmath340 @xmath341 . we have @xmath342 and @xmath343 so @xmath344 and @xmath345 which reduces to @xmath346 in particular kurtosis of @xmath347 distributions is equal to @xmath348 . hence it is negative and less , for @xmath349 than that of @xmath1normal which is also negative ( equal to @xmath350 ) . assertion i ) of the proposition [ mn ] leads to the generalization of the multidimensional @xmath1normal distribution that allows different one - dimensional and other marginals . a distribution in @xmath0 having density equal to @xmath351where @xmath352 @xmath353 @xmath354 @xmath355,$ ] @xmath356 will be called _ modified multidimensional _ @xmath357__normal distribution _ _ ( briefly _ _ _ _ @xmath358 ) . _ _ reasoning in the similar way as in the proof of corollary [ marginal ] and utilizing observation following from proposition [ mn ] , we have immediately the following observation . let @xmath359 . then every marginal of it is also modified multidimensional @xmath1normal . in particular @xmath360 @xmath361 @xmath362 suppose that @xmath363 and define @xmath364 then the sequence @xmath365 form a non - stationary markov chain such that @xmath366 with transitional probability @xmath367 we can define another one - dimensional distribution depending on @xmath368 parameters . we have : let @xmath329,t^{2}<1/(1-q),x , y\in s\left ( q\right ) , $ ] @xmath37 . a distribution with the density @xmath369 will be called _ modified _ @xmath370__conditional normal _ _ ( briefly @xmath370mcn)_. _ we have immediate observation that follows from assertion vii ) of lemma wlasnosci . let @xmath371mcn . then for @xmath372 hence in particular one can state the following corollary . @xmath373 @xmath374 follows expressions for first two al - salam - chihara polynomials . namely we have : @xmath375 @xmath376 we can define two formulae for densities of multidimensional distributions in @xmath0 . namely one of them would have density of the form@xmath377and the other of the form@xmath378however to find marginals of such families of distributions is a challenge and an open question . in particular are they also of modified conditional normal type ? in this section we are going to study properties of @xmath121 dimensional case of multidimensional normal distribution . to simplify notation we will consider vector @xmath379 having distribution @xmath380 that is having density @xmath381 . we start with the following obvious result : conditional distribution @xmath13 has density @xmath382where we denoted @xmath383 it is in fact rewritten version of the proof of assertion ii ) of corollary [ warunkowanie ] . notice that @xmath384 is the re - scaled askey - wilson density . namely @xmath385 where @xmath386 and @xmath387 is a normalized ( that is multiplied by a constant so that its integral is @xmath388 weight function of askey - wilson polynomials . compare e.g. @xcite and @xcite . hence our results would concern properties of askey - wilson density and askey - wilson polynomials . let us denote @xmath389 our main result is the following [ main]i ) @xmath390@xmath391 _ { q}!}% g_{n}\left ( y , z,\rho _ { 1},\rho _ { 2},q\right ) , $ ] where @xmath392 is a polynomial of order @xmath288 in @xmath393 \ii ) more over polynomial @xmath392 has the following structure @xmath394 @xmath395 @xmath396 { n}{i}_{q}\rho _ { 1}^{i}\rho _ { 2}^{n - i}\theta _ { i , n - i}\left ( y , z,\rho _ { 1}\rho _ { 2}|q\right ) , $ ] where @xmath397 is a polynomial in @xmath198 of order @xmath398 and in @xmath399 of order @xmath400 . moreover @xmath401 and @xmath402 assertion i ) of the theorem [ main ] is in fact a generalization of poisson - mehler formula ( that is assertion viii ) of lemma [ wlasnosci ] ) for askey - wilson density . notice also that for @xmath403 @xmath404 is a density function of normal distribution @xmath405 and it is obvious that expectation of any polynomial is a polynomial in @xmath406 . hence it turns out that this is true for all @xmath1normal distributions for @xmath140.$ ] as a corollary we have the following result . [ postac]let @xmath407 . let us select indices @xmath408 . then@xmath409for @xmath410 constants ( depending only on @xmath411 @xmath412 and numbers @xmath413 @xmath414 @xmath415 @xmath416 . @xmath417 @xmath418 corollary bellow gives detailed form of coefficients @xmath419 for @xmath420 [ _ 3i4]let @xmath407 . let @xmath421 . then : @xmath422 where @xmath423 { n}{l}_{q}\frac{\rho _ { i-1}^{n - l}\left ( \rho _ { i}^{2}\right ) _ { n - l}\rho _ { i}^{l}\left ( \rho _ { i-1}^{2}\right ) _ { l}}{\left ( \rho _ { i-1}^{2}\rho _ { i}^{2}\right ) _ { n}},$ ] @xmath424 @xmath425 . if @xmath426 then @xmath427 _ { q}\rho _ { i-1}\rho _ { i}a_{0,-\left\lfloor n/2\right\rfloor + l}^{\left ( n\right ) } , $ ] @xmath428 if @xmath429 then @xmath430 _ { q}\rho _ { i-1}\rho _ { i}a_{0,j}^{(4)},$ ] @xmath431 and @xmath432 _ { q}^{2}\rho _ { i-1}\rho _ { i}a_{0,0}^{(4)},$ ] @xmath433 in particular : @xmath434 notice that in general conditional variance @xmath435 is not nonrandom indicating that @xmath1normal distribution does not behave as normal in this case , however if we set @xmath2 in ( [ warvar1 ] ) then we get @xmath436 as it should be in the normal case . notice that examining the form of coefficients @xmath437 for @xmath438 we can formulate the following hypothesis concerning general form of them : for @xmath67 , we have : @xmath439 { n}{l}_{q}\frac{\rho _ { i-1}^{n - l}\left ( \rho _ { i}^{2}\right ) _ { n - l}\rho _ { i}^{l}\left ( \rho _ { i-1}^{2}\right ) _ { l}}{\left ( \rho _ { i-1}^{2}\rho _ { i}^{2}\right ) _ { n}},$ ] @xmath440 . moreover for @xmath441 and @xmath442 @xmath443 where @xmath444 is a polynomial in @xmath47 with coefficients depending only on @xmath164 and @xmath445 proof of the theorem [ main ] is based on the properties of the following function @xmath446 _ { q}!}h_{m+l}\left ( y|q\right ) h_{m+k}\left ( z|q\right ) $ ] . we will need some of its properties . namely we will prove the following proposition which is in fact a generalization and reformulation ( in terms of polynomials @xmath447 of an old result of carlitz . original result of carlitz concerned polynomials @xmath448 { n}{i}_{q}x^{i}$ ] and expressions of the form @xmath449 compare @xcite , exercise 12.3(d ) or @xcite . \ii ) for @xmath451 : @xmath452 { k}{i}_{q}q^{\binom{i}{2}% } t^{i}h_{k - i}(y|q)g_{0,i+l}(y , z , t|q ) \label{klna0l } \\ & & + ( -1)^{j}q^{\binom{j}{2}}\sum_{i = j}^{k}\qatopd [ ] { k}{i}_{q}\qatopd [ ] { i-1% } { j-1}_{q}t^{i}g_{k - i , i+l}(y , z , t|q ) . \notag\end{aligned}\ ] ] \ii ) to prove ( [ klna0l ] ) we will use formula @xmath458 { n}{j}_{q}\qatopd [ ] { m}{j}_{q}% \left [ j\right ] _ { q}!h_{n+m-2j}(x|q)$ ] . we have @xmath459 _ { q}!}h_{m+k}\left ( y|q\right ) h_{m+l}\left ( z|q\right ) \allowbreak \\ = \sum_{m\geq 0}\frac{t^{m}}{\left [ m\right ] _ { q}!}(h_{k}\left ( y|q\right ) h_{m}(y|q)\allowbreak \\ -\sum_{j=1}^{\min ( k , m)}\qatopd [ ] { k}{j}_{q}\qatopd [ ] { m}{j}_{q}\left [ j% \right ] _ { q}!h_{k+m-2j}(y|q))h_{m+l}\left ( z|q\right ) \allowbreak \\ = h_{k}\left ( y|q\right ) g_{0,l}(y , z , t)\allowbreak -\allowbreak \sum_{j=1}^{k}t^{j}\qatopd [ ] { k}{j}_{q}\sum_{m = j}^{\infty } \frac{t^{m - j}}{% \left [ m - j\right ] _ { q}!}h_{k+(m - j)-j}(y|q)h_{l+j+(m - j)}(z|q ) \\ = h_{k}\left ( y|q\right ) g_{0,l}(y , z , t)\allowbreak -\allowbreak \sum_{j=1}^{k}t^{j}\qatopd [ ] { k}{j}_{q}g_{k - j , l+j}(y , z , t).\end{gathered}\]]@xmath187 hence let us assume that ( [ klna0l ] ) is true for @xmath460 . we have after applying just obtained formula ( for @xmath461 applied for @xmath462 @xmath463 and @xmath464 @xmath465 { k}{i}_{q}q^{\binom{i}{2}}t^{i}h_{k - i}(y|q)g_{0,i+l}(y , z , t ) \\ & & + \allowbreak ( -1)^{m}q^{\binom{m}{2}}\sum_{i = m}^{k}\qatopd [ ] { k}{i}_{q}% \qatopd [ ] { i-1}{m-1}_{q}t^{i}g_{k - i , i+l}(y , z , t)\allowbreak \\ & = & \sum_{i=0}^{m-1}(-1)^{i}\qatopd [ ] { k}{i}_{q}q^{\binom{i}{2}% } t^{i}h_{k - i}(y|q)g_{0,i+l}(y , z , t)\allowbreak \\ & & + ( -1)^{m}q^{\binom{m}{2}}\qatopd [ ] { k}{m}% _ { q}t^{m}(h_{k - m}(y|q)g_{0,m+l}(y , z , t ) \\ & & -\sum_{i=1}^{k - m}\qatopd [ ] { k - m}{i}_{q}t^{i}g_{k - m - i , l+m+i}(y , z , t))% \allowbreak \\ & & + ( -1)^{m}q^{\binom{m}{2}}\sum_{i = m+1}^{k}\qatopd [ ] { k}{i}_{q}\qatopd [ ] { % i-1}{m-1}_{q}t^{i}g_{k - i , i+l}(y , z , t)\allowbreak\end{aligned}\]]@xmath187now since @xmath59 { k}{m}_{q}\qatopd [ ] { k - m}{i - m}_{q}-% \qatopd [ ] { k}{i}_{q}\qatopd [ ] { i-1}{m-1}_{q}\allowbreak = \allowbreak \qatopd [ ] { k}{i}_{q}(\qatopd [ ] { i}{m}_{q}\allowbreak -\allowbreak \qatopd [ ] { i-1}{m-1}_{q})\allowbreak = \allowbreak q^{m}\qatopd [ ] { k}{i}_{q}\qatopd [ ] { i-1}{m}_{q}$ ] and @xmath466 we have @xmath467 { k% } { i}_{q}q^{\binom{i}{2}}t^{i}h_{k - i}(y|q)g_{0,i+l}(y , z , t)\allowbreak \allowbreak \\ & & -(-1)^{m}q^{\binom{m}{2}}\sum_{i = m+1}^{k}(\qatopd [ ] { k}{m}_{q}\qatopd [ ] { % k - m}{i - m}_{q}-\qatopd [ ] { k}{i}_{q}\qatopd [ ] { i-1}{m-1}% _ { q})t^{i}g_{k - i , l+i}(y , z , t ) \\ & = & \sum_{i=0}^{m}(-1)^{i}\qatopd [ ] { k}{i}_{q}q^{\binom{i}{2}% } t^{i}h_{k - i}(y|q)g_{0,i+l}(y , z , t)\allowbreak \\ & & + ( -1)^{m+1}q^{\binom{m+1}{2}}\sum_{i = m+1}^{k}\qatopd [ ] { k}{i}_{q}\qatopd [ ] { i-1}{m}_{q}t^{i}g_{k - i , i+l}(y , z , t)\end{aligned}\]]iv ) for @xmath468 this is obviously true . now let us iterate ( [ expansion ] ) once , applied however , for @xmath469 . we will get then @xmath470 { k}{i}_{q}q^{\binom{i}{2}% } t^{i}h_{k - i}(y|q)g_{0,i}(y , z , t|q)\allowbreak + \allowbreak \left ( -1\right ) ^{k}q^{\binom{k}{2}}t^{k}g_{0,k}\left ( y , z , t|q\right ) \allowbreak = $ ] @xmath471 { k}{i}_{q}q^{\binom{i}{2}% } t^{i}h_{k - i}(y|q)g_{0,i}(y , z , t|q)\allowbreak + \allowbreak $ ] @xmath472 { k}{i}_{q}q^{\binom{i}{2}}t^{i}h_{k - i}(z|q)g_{i,0}(y , z , t|q))\allowbreak + \allowbreak q^{k(k-1)}t^{2k}g_{k.0}\left ( y , z , t|q\right ) $ ] . thus we see that since for all @xmath473 @xmath474 and @xmath475 and are of the claimed form then from ( [ wyraz ] ) it follows that @xmath476 has the claimed form . to prove i ) we will use formula viii ) of lemma [ wlasnosci ] , that is poisson - mehler expansion formula . following ( [ _ x|yz ] ) we see that @xmath477 _ { q}!}h_{i}\left ( x|q\right ) h_{i}\left ( y|q\right ) \allowbreak \times \allowbreak \sum_{i=0}^{\infty } \frac{\rho _ { 2}^{i}}{\left [ i\right ] _ { q}!}h_{i}\left ( x|q\right ) h_{i}\left ( z|q\right ) \allowbreak \\ & & /\allowbreak \sum_{i=0}^{\infty } \frac{\rho _ { 1}^{i}\rho _ { 2}^{i}}{\left [ i% \right ] _ { q}!}h_{i}\left ( y|q\right ) h_{i}\left ( z|q\right ) .\end{aligned}\]]first , let us concentrate on the quantity : @xmath478 _ { q}!}h_{n}\left ( x|q\right ) h_{n}\left ( y|q\right ) \allowbreak \times \allowbreak \sum_{m\geq 0}\frac{\rho _ { 2}^{n}}{\left [ n\right ] _ { q}!}h_{n}\left ( x|q\right ) h_{n}\left ( z|q\right ) .\]]we will apply identity ( [ identity ] ) , distinguish two cases @xmath479 is even and @xmath479 is odd@xmath480 denote @xmath481 @xmath482 or @xmath483 depending om the case and sum over the set of @xmath484@xmath485 . we have @xmath486 _ { q}!\left [ m\right ] _ { q}!}h_{n}\left ( y|q\right ) h_{m}\left ( z|q\right ) \allowbreak \sum_{j=0}^{\min \left ( n , m\right ) } \qatopd [ ] { n}{j}_{q}\qatopd [ ] { m}{j}% _ { q}\left [ j\right ] _ { q}!h_{n+m-2j}\left ( x|q\right ) \\ = \sum_{k=0}^{\infty } \frac{h_{2k}(x|q)}{\left [ 2k\right ] _ { q}!}\allowbreak \sum_{j=0}^{\infty } \allowbreak \sum_{m = j}^{2k+j}\frac{\left [ 2k\right ] _ { q}!\rho _ { 1}^{m}\rho _ { 2}^{2k+2j - m}}{\left [ m - j\right ] _ { q}!\left [ 2k-(m - j)% \right ] _ { q}!\left [ j\right ] _ { q}}h_{m}(y|q)h_{2k+2j - m}(z|q)\allowbreak \\ + \sum_{k=0}^{\infty } \frac{h_{2k+1}(x|q)}{\left [ 2k+1\right ] _ { q}!}% \allowbreak \sum_{j=0}^{\infty } \allowbreak \sum_{m = j}^{2k+1+j}\frac{\left [ 2k+1\right ] _ { q}!\rho _ { 1}^{i+j}\rho _ { 2}^{2k+1-i+j}}{\left [ j\right ] _ { q}% \left [ m - j\right ] _ { q}!\left [ 2k+1-(m - j)\right ] _ { q}!}h_{m}\left ( y|q\right ) h_{2k+1 + 2j - m}\left ( z|q\right ) \\ = \sum_{k=0}^{\infty } \frac{h_{2k}(x|q)}{\left [ 2k\right ] _ { q}!}\allowbreak \sum_{j=0}^{\infty } \allowbreak \sum_{i=0}^{2k}\frac{\left [ 2k\right ] _ { q}!\rho _ { 1}^{i+j}\rho _ { 2}^{2k - i+j}}{\left [ j\right ] _ { q}!\left [ i\right ] _ { q}![2k - i]_{q}!}h_{i+j}\left ( y|q\right ) h_{2k - i+j}\left ( z|q\right ) \allowbreak \\ + \sum_{k=0}^{\infty } \frac{h_{2k+1}(x|q)}{\left [ 2k+1\right ] _ { q}!}% \allowbreak \sum_{j=0}^{\infty } \allowbreak \sum_{i=0}^{2k+1}\frac{\left [ 2k+1\right ] _ { q}!\rho _ { 1}^{i+j}\rho _ { 2}^{2k+1-i+j}}{\left [ j\right ] _ { q}!% \left [ i\right ] _ { q}!\left [ 2k+1-i\right ] _ { q}!}h_{i+j}\left ( y|q\right ) h_{2k+1+j - i}\left ( z|q\right)\end{gathered}\]]@xmath187 we get then@xmath487 _ { q}!}\allowbreak \sum_{i=0}^{2k}\qatopd [ ] { 2k% } { i}_{q}\rho _ { 1}^{i}\rho _ { 2}^{2k - i}\sum_{j=0}^{\infty } \frac{(\rho _ { 1}\rho _ { 2})^{j}}{\left [ j\right ] _ { q}!}h_{i+j}(y|q)h_{2k - i+j}(z|q)% \allowbreak \\ + \sum_{k=0}^{\infty } \frac{h_{2k+1}(x|q)}{\left [ 2k+1\right ] _ { q}!}% \sum_{i=0}^{2k+1}\qatopd [ ] { 2k+1}{i}_{q}\rho _ { 1}^{i}\rho _ { 2}^{2k+1-i}\sum_{j=0}^{\infty } \frac{(\rho _ { 1}\rho _ { 2})^{j}}{\left [ j% \right ] _ { q}!}h_{i+j}(y|q)h_{2k+1-i+j}(z|q ) \\ = \sum_{n=0}^{\infty } \frac{h_{n}\left ( x|q\right ) } { \left [ n\right ] _ { q}!}% \allowbreak \sum_{i=0}^{n}\qatopd [ ] { n}{i}_{q}\rho _ { 1}^{i}\rho _ { 2}^{n - i}\allowbreak \sum_{j=0}^{\infty } \frac{(\rho _ { 1}\rho _ { 2})^{j}}{% \left [ j\right ] _ { q}!}h_{i+j}(y|q)h_{n - i+j}(z|q)\end{gathered}\]]@xmath187using introduced in proposition [ iter ] function @xmath488 _ { q}!}h_{m+l}\left ( y|q\right ) h_{m+k}\left ( z|q\right ) $ ] we can express both @xmath489 _ { q}!}% h_{i}\left ( y|q\right ) h_{i}\left ( z|q\right ) \allowbreak = \allowbreak g_{0,0}(y , z,\rho _ { 1}\rho _ { 2}|q)$ ] and @xmath490 _ { q}!}% \sum_{i=0}^{n}\qatopd [ ] { n}{i}_{q}\rho _ { 1}^{i}\rho _ { 2}^{n - i}g_{i , n - i}\left ( y , z,\rho _ { 1}\rho _ { 2}|q\right ) $ ] . our theorem will be proved if we will be able to show that @xmath491 @xmath492 where @xmath493 is a polynomial of order a @xmath400 in @xmath198 @xmath398 in @xmath399 . this fact follows by induction from formula ( [ wyraz ] ) of assertion iv ) of the proposition iter since it expresses @xmath476 in terms of @xmath398 functions @xmath474 and @xmath494 for @xmath495 and the fact that all @xmath496 can be expressed by @xmath474 and @xmath475 ; @xmath497 by theorem [ main ] we know that regression @xmath498 is a polynomial in @xmath235 and @xmath499 of order at most @xmath288 . to analyze the structure of this polynomial let us present it in the form @xmath500 where coefficients @xmath501 are some polynomials of @xmath502 . now let us take conditional expectation with respect to @xmath503 of both sides . on one hand we get@xmath504 on the other we get@xmath505 since @xmath501 are polynomials in @xmath235 of order at most @xmath411 we can present them in the form @xmath506 thus we have equality:@xmath507 now we use the identity ( [ identity ] ) and get @xmath508 { t}{m}_{q}\qatopd [ ] { s% } { m}_{q}\left [ m\right ] _ { q}!h_{t+s-2m}\left ( x_{j_{k}}|q\right ) .\end{gathered}\ ] ] hence we deduce that @xmath509 for @xmath510 , @xmath511 . to count the number of coefficients @xmath512 observe that we have @xmath513 coefficients @xmath514 since @xmath398 ranges from @xmath515 to @xmath516 , @xmath517 coefficients @xmath518 where @xmath398 ranges from @xmath519 to @xmath520 and so on . the proof is based on the idea of writing down system of @xmath521 ( @xmath438 ) linear equations satisfied by coefficients @xmath437 . these equations are obtained according to the similar pattern . namely we multiply both sides of identity ( [ wzor ] ) by @xmath522 and @xmath523 and calculate conditional expectation of both sides with respect to @xmath524 or with respect to @xmath525 remembering that @xmath526 and @xmath527 and similar formulae for @xmath525 . we expand both sides with respect to @xmath528 @xmath529 @xmath530 . on the way we utilize ( identity ) and compare coefficients standing by @xmath531 on both sides . thus each obtained equation involving coefficients @xmath532 can be indexed by @xmath533 and @xmath164 if we calculate conditional expectation with respect to @xmath524 of @xmath400 if we conditional expectation is calculated with respect to @xmath525 . of course if @xmath534 then @xmath164 and @xmath400 lead to the same result . formulae for @xmath535 for @xmath536 are obtained by taking @xmath537 @xmath538 @xmath539 and applying @xmath164 and @xmath400 . for @xmath540 first we consider @xmath538 @xmath539 and @xmath541 and applying @xmath542 and @xmath400 and then @xmath538 @xmath543 and @xmath544 . in this way we get @xmath121 equations . the forth one is obtained by taking @xmath538 @xmath545 @xmath546 and @xmath164 . denote @xmath547 then @xmath548 satisfies system of linear equation with matrix @xmath549 _ { q}\text{$\rho $ } \text{$_{i-1}\rho _ { i}$ } & 1+\left [ 2% \right ] _ { q}\text{$\rho $ } _ { i-1}^{2}\text{$\rho $ } _ { i}^{2 } & \left [ 2\right ] _ { q}\text{$\rho $ } \text{$_{i-1}\rho _ { i}$ } & \text{$\rho $ } \text{$_{i-1}\rho _ { i}$}% \end{array}% \right ) $ ] and with right side vector equal to : @xmath550 _ { q}\text{$\rho $ } \text{$_{i-1}\rho _ { i}$}% \end{array}% \right ) $ ] . besides formulae for coefficients @xmath551 for @xmath552 can be obtained from formulae scattered in the literature like e.g. @xcite , @xcite or @xcite . to get equations satisfied by coefficients @xmath553 for @xmath554 first @xmath555 equations are obtained by taking @xmath556 , @xmath468 and @xmath557@xmath558 if @xmath559 and @xmath557@xmath560 if @xmath429 and then applying operations @xmath164 and @xmath400 . then , in order to get remaining @xmath222 ( in case of @xmath561 or @xmath368 ( in case of @xmath562 equations one has to be more careful since it often turns out that many equations obtained for some @xmath563 and @xmath398 are linearly dependent on the previously obtained equations . in the case of @xmath564 to get remaining two linearly independent equations we took @xmath565 @xmath566 @xmath567 and applied operations @xmath164 and @xmath400 . in this way we obtained system of @xmath568 linear equations with matrix @xmath569 _ { q}\text{$\rho _ { i-1}$}\text{$\rho _ { i}$ } & 1+\left [ 2% \right ] _ { q}^{2}\text{$\rho $ } _ { i-1}^{2}\text{$\rho $ } _ { i}^{2 } & \left [ 2% \right ] _ { q}\text{$\rho _ { i-1}$}\text{$\rho _ { i}$}+\left [ 3\right ] _ { q}\text{% $ \rho $ } _ { i-1}^{3}\text{$\rho $ } _ { i}^{3 } & \left [ 3\right ] _ { q}\text{$\rho $ } % _ { i-1}^{2}\text{$\rho $ } _ { i}^{2 } & \text{$\rho _ { i-1}$}\text{$\rho _ { i}$ } & \text{$\rho $ } _ { i-1}^{2}\text{$\rho $ } _ { i}^{2 } \\ \left [ 3\right ] _ { q}\text{$\rho $ } _ { i-1}^{2}\text{$\rho $ } _ { i}^{2 } & \left [ 2% \right ] _ { q}\text{$\rho _ { i-1}$}\text{$\rho _ { i}$}+\left [ 3\right ] _ { q}\text{% $ \rho $ } _ { i-1}^{3}\text{$\rho $ } _ { i}^{3 } & 1+\left [ 2\right ] _ { q}^{2}\text{$% \rho $ } _ { i-1}^{2}\text{$\rho $ } _ { i}^{2 } & \left [ 3\right ] _ { q}\text{$\rho _ { i-1}$}\text{$\rho _ { i}$ } & \text{$\rho $ } _ { i-1}^{2}\text{$\rho $ } _ { i}^{2 } & \text{$\rho _ { i-1}$}\text{$\rho _ { i}$}% \end{array}% \right ) $ ] right hand side vector @xmath570 _ { q}\text{$\rho _ { i-1}^{2}$}\text{$\rho _ { i}$ } \\ \left [ 3\right ] _ { q}\text{$\rho $ } _ { i-1}\text{$\rho $ } _ { i}^{2}% \end{array}% \right ) $ ] if the vector of unknowns is the following @xmath571 . for @xmath572 remaining @xmath368 equations we obtained by taking : ( @xmath573 , @xmath574 @xmath575 @xmath576 @xmath577 ( @xmath578 @xmath579 @xmath575@xmath580 @xmath577 ( @xmath581 @xmath582 @xmath583 @xmath584 and @xmath585 @xmath574 @xmath586 @xmath584 . recall that in this case we have @xmath587 equations . matrix of this system has @xmath588 entries . that is why we will skip writing down the whole system of equations . to get the scent of how complicated these equations are we will present one equation . @xmath429 one of the equations ( referring to the case @xmath578 @xmath589 @xmath590 @xmath584 is @xmath591 _ { q}^{2}\left ( 1+q^{2}\right ) \left [ 3\right ] _ { q}(\left [ 4\right ] _ { q}+\left [ 5\right ] _ { q}))$]@xmath591 _ { q}^{2}\left ( 1+q^{2}\right ) \left [ 3\right ] _ { q}\allowbreak \times $ ] @xmath592 _ { q}^{2}\left ( 1+q^{2}\right ) \left [ 3\right ] _ { q}\rho _ { i-1}\rho _ { i}(\left [ 2\right ] _ { q}\allowbreak + \allowbreak \rho _ { i-1}^{2}\rho _ { i}^{2}\allowbreak + \allowbreak 3q\rho _ { i-1}^{2}\rho _ { i}^{2}\allowbreak + \allowbreak 3q^{2}\rho _ { i-1}^{2}\rho _ { i}^{2}\allowbreak + \allowbreak q^{3}\rho _ { i-1}^{2}\rho _ { i}^{2}\allowbreak + \allowbreak q^{4}\rho _ { i-1}^{2}\rho _ { i}^{2})a_{0,0}^{\left ( 4\right ) } \allowbreak + \allowbreak \left [ 2\right ] _ { q}^{2}\left ( 1+q^{2}\right ) \left [ 3\right ] _ { q}\rho 1^{2}\rho 2^{2}(\left [ 3\right ] _ { q}+(\left [ 4\right ] _ { q}+\left [ 5\right ] _ { q})\rho _ { i-1}^{2}\rho _ { i}^{2})a_{0,1}^{\left ( 4\right ) } \allowbreak + \allowbreak \left [ 2\right ] _ { q}^{2}\left ( 1+q^{2}\right ) \left [ 3\right ] _ { q}\rho 1^{3}\rho 2^{3}(1+q+q^{2}+q^{3}+\rho _ { i-1}^{2}\rho _ { i}^{2}+q\rho _ { i-1}^{2}\rho _ { i}^{2}+q^{2}\rho _ { i-1}^{2}\rho _ { i}^{2}+q^{3}\rho _ { i-1}^{2}\rho _ { i}^{2}+q^{4}\rho _ { i-1}^{2}\rho _ { i}^{2})a_{0,2}^{(4)}\allowbreak + \allowbreak $ ] @xmath591 _ { q}^{2}\left ( 1+q^{2}\right ) \left [ 3\right ] _ { q}\rho _ { i-1}\rho _ { i}a_{1,-1}^{\left ( 4\right ) } \allowbreak + \allowbreak \left [ 2% \right ] _ { q}^{2}\left ( 1+q^{2}\right ) \left [ 3\right ] _ { q}\rho _ { i-1}^{2}\rho _ { i}^{2}a_{1,0}^{\left ( 4\right ) } \allowbreak + \allowbreak % \left [ 2\right ] _ { q}^{2}\left ( 1+q^{2}\right ) \left [ 3\right ] _ { q}\rho _ { i-1}^{3}\rho _ { i}^{3}a_{1,1}^{\left ( 4\right ) } \allowbreak = \allowbreak $ ] @xmath593 _ { q}^{2}\left ( 1+q^{2}\right ) \left [ 3\right ] _ { q}\rho _ { i-1}^{3}(\left [ 4\right ] _ { q}+\left [ 5\right ] _ { q}\rho _ { i-1}^{2})\rho _ { i}.$ ] szabowski , p.j . ( 2008 ) _ probabilistic implications of symmetries of _ @xmath47__-hermite and al - salam -chihara polynomials , _ _ infinite dimensional analysis , quantum probability and related topic , 11(4 ) , 513 - 522

we present a comprehensive theoretical study of the static spin response in hgte quantum wells , revealing distinctive behavior for the topologically nontrivial inverted structure . most strikingly , the ( long - wave - length ) spin susceptibility of the undoped topological - insulator system is constant and equal to the value found for the gapless dirac - like structure , whereas the same quantity shows the typical decrease with increasing band gap in the normal - insulator regime . we discuss ramifications for the ordering of localized magnetic moments present in the quantum well , both in the insulating and electron - doped situations . the spin response of edge states is also considered , and we extract effective land-factors for the bulk and edge electrons . the variety of counter - intuitive spin - response properties revealed in our study arises from the system s versatility in accessing situations where the charge - carrier dynamics can be governed by ordinary schrdinger - type physics , mimics the behavior of chiral dirac fermions , or reflects the material s symmetry - protected topological order .

1506.08913

the two - dimensional ( 2d ) topological insulator ( ti ) realized in inverted hgte quantum wells exhibits unusual electric - transport properties @xcite that have attracted great interest @xcite . ti behavior is also found in other 2d @xcite and bulk @xcite materials . the potential for interesting interplays between a ti s electronic and magnetic degrees of freedom , e.g. , through hyperfine interaction with the material s nuclei @xcite , or local exchange interaction with magnetic dopants @xcite , has been pointed out recently @xcite . these efforts have opened up new perspectives and extended previous work devoted to understanding spin effects @xcite and magnetism @xcite in ordinary semiconductors . our present theoretical study of the spin response in hgte quantum wells reveals unconventional spin - related properties that distinguish this paradigmatic ti material from all other currently known 2d electronic systems . we thus provide alternative means for the experimental identification of the topological regime and extend current knowledge about the fundamentals of spin - response behavior in solids . the spin susceptibility contains comprehensive information about the magnetic properties of a material . in the simplest case of a spin - rotationally invariant noninteracting electron gas , the spin susceptibility is proportional to the charge - response ( lindhard ) function @xcite . noticable deviations from that situation occur , e.g. , in systems with strong spin - orbit coupling such as 2d hole gases @xcite . for metals or degenerately doped semiconductors , the spin response of only the partially filled band is typically considered . this approach misses intrinsic contributions to many - particle response functions arising from virtual interband transitions that become important in narrow - gap and , especially gapless , electron systems . examples for the latter are the 2d dirac - like electron states on the surface of a bulk ti whose magnetic properties have been discussed in refs . . the hgte quantum wells considered here present an ideal testing ground for exploring the importance and properties of intrinsic , or virtual - carrier , effects , as it is possible to tune the band gap in such systems with a single structural parameter ( i.e. , the quantum - well width @xmath2 ) . furthermore , deviations from the conical 2d - dirac dispersion in the gapless case are well characterized within a continuum - model ( bhz ) description @xcite that also gives controlled access to the full range of , and interesting transitions between , schrdinger - physics - dominated , chiral - dirac - fermion - like , and topologically nontrivial phases . in particular , definite ( i.e. , cut - off - independent ) results for the intrinsic response at long wavelengths are obtained even in the limit of vanishing band gap ; unlike in the case of the previously considered 2d - dirac models used to describe the surface states of bulk tis @xcite . before going into greater detail in the remainder of this article , we briefly highlight four major advances and central new insights gained from our work . ( i ) we find an analytical expression for the uniform static spin susceptibility of the intrinsic ( undoped ) system , @xmath3 where @xmath4 is proportional to the band gap and positive ( negative ) for the topologically trivial normal ( topologically nontrivial inverted ) regime , @xmath5 denotes the heaviside step function , @xmath6 is a band - structure parameter from the bhz model introduced further below , and the constants @xmath7 depend both on the valence - band mixing of quantum - well basis states and on the parameter @xmath8 that characterises the relative coupling strength of the conduction - band and valence - band spin degree of freedom to the physical quantity of interest . while the intrinsic spin response is suppressed with increasing band gap in the normal regime ( @xmath9 ) , it becomes independent of gap size in the topological regime where @xmath10 . this unexpected behavior is a special feature of the symmetry - protected topological phase of the bulk system associated with the existence of gapless edge states . ( a ) mean - field critical temperature @xmath11 for virtual - carrier - mediated magnetism in a hgte quantum well plotted as a function of the band - gap parameter @xmath4 . note the striking asymmetry between the topological ( @xmath12 ) and normal ( @xmath9 ) regimes . an explicit expression for @xmath13 is given in the text ; it contains the local exchange coupling between magnetic - impurity orbitals and hgte quantum - well basis states , which is roughly constant across the transition @xcite . ( b ) critical temperature @xmath14 for the electron - doped hgte quantum well plotted as a function of @xmath4 and for various charge - carrier densities ( corresponding to indicated values for the fermi wave vector @xmath15 ; where @xmath16 corresponds to a charge density @xmath17@xmath18 for a typical structure @xcite ) . we used @xmath19 for the bhz - model electron - hole - asymmetry parameter in both plots , and @xmath20 as relevant for the ( hg , cd , mn)te system @xcite.,title="fig : " ] ( a ) mean - field critical temperature @xmath11 for virtual - carrier - mediated magnetism in a hgte quantum well plotted as a function of the band - gap parameter @xmath4 . note the striking asymmetry between the topological ( @xmath12 ) and normal ( @xmath9 ) regimes . an explicit expression for @xmath13 is given in the text ; it contains the local exchange coupling between magnetic - impurity orbitals and hgte quantum - well basis states , which is roughly constant across the transition @xcite . ( b ) critical temperature @xmath14 for the electron - doped hgte quantum well plotted as a function of @xmath4 and for various charge - carrier densities ( corresponding to indicated values for the fermi wave vector @xmath15 ; where @xmath16 corresponds to a charge density @xmath17@xmath18 for a typical structure @xcite ) . we used @xmath19 for the bhz - model electron - hole - asymmetry parameter in both plots , and @xmath20 as relevant for the ( hg , cd , mn)te system @xcite.,title="fig : " ] ( ii ) a physical consequence of the unusual spin response in the intrinsic system is the asymmetric variation of the critical ( curie ) temperature for virtual - carrier - mediated magnetic order in a hgte quantum well that has been doped magnetically but not electronically . see fig . [ fig : ct1](a ) . a similarly striking asymmetry arises when charge carriers are present in the 2d conduction band ; which is illustrated in fig . [ fig : ct1](b ) . furthermore , in this situation , a rather strong , and counter - intuitive , suppression of the curie temperature with the density of itinerant charge carriers is revealed . this tendency arises from the unconventional character of conduction - band states in the topological regime . see section [ sec : dopedsr ] for details . ( iii ) we extract the effective _ g _ factors for 2d conduction electrons in both the topological and normal regimes , which are directly accessible experimentally @xcite . their strong density and gap - parameter dependences reflect the band mixing and transition between regimes dominated by schrdinger - type and dirac - like dynamics of charge carriers . full details can be found in section [ sec:2dgfac ] . these results are essential , e.g. , to enable quantitative characterization of the transition between quantum spin - hall and quantum anomalous hall phases in magnetic 2d topological insulators @xcite . ( iv ) we predict the effective _ g _ factors of the helical edge states that are present in the topological regime of the hgte quantum well . for the out - of - plane magnetic - field direction , factor assumes a constant value that is determined by band mixing in the quantum - well eigenstates . in contrast , the in - plane _ g _ factor has a strong density dependence . see section [ sec : gfactedge ] . in the following , we discuss relevant details of our theoretical analysis and present complete results for the spin response in both intrinsic and electron - doped systems . the electronic properties of hgte quantum wells are adequately captured by an effective four - band ( bhz ) hamiltonian @xcite that acts in the low - energy subspace spanned by basis states @xmath21 and explicitly reads @xmath22 0 & { \mathcal h}^*(-{{\bm{\mathrm{k } } } } ) \end{pmatrix}~ , \label{eq : bhz}\end{aligned}\ ] ] with @xmath23 , @xmath24 and @xmath25 . the parameters @xmath26 are functions of the well width @xmath2 , and their numerical values are typically obtained by a fitting procedure @xcite . the parameter @xmath27 opens a band gap , where ( in the convention @xmath28 ) the system is in the topological ( normal ) regime when @xmath29 ( @xmath30 ) . the basis functions @xmath31 are a superposition of conduction - electron and light - hole ( lh ) basis functions with a given spin projection . the associated band @xmath32 has a much lower band - edge energy and can therefore be omitted . on the other hand , the heavy - hole ( hh ) states @xmath33 also belong to the set of low - energy excitations . in the following we set @xmath34 , and employ a dimensionless description of eq . ( [ eq : bhz ] ) that is obtained by defining an energy scale @xmath35 and a scale for the wave vector @xmath36@xcite . ( in a typical hgte quantum - well structure @xcite , @xmath37 ev and @xmath38 nm@xmath39 . ) by making use of the axial symmetry of the bhz hamiltonian ( [ eq : bhz ] ) we rotate to a real basis to obtain @xmath40 , with @xmath41 and @xmath42 , where @xmath43 is the polar angle of the 2d wave vector @xmath44 . we have defined the dimensionless parameters @xmath45 and @xmath46 , which have typical values @xcite @xmath47 . the eigenvectors in the complex and real basis are related via @xmath48 , where @xmath49 distinguishes the conduction and valence bands , which are doubly degenerate in the quantum number @xmath50 for spin projection along the growth direction and have the dispersions @xmath51 \,\ , .\ ] ] the spin susceptibility is used to characterize a system s response to spin - related external stimuli within the framework of linear - response theory @xcite . in the static limit , and for quantum - well - confined electrons , it can be written as @xcite @xmath52 \rangle \ , \bigg\ } \ , , \quad \label{eq : spinsus}\end{aligned}\ ] ] with @xmath53 and @xmath54 being coordinates in the 2d plane and perpendicular to it , respectively , and @xmath55 denoting the electron spin density measured in units of @xmath56 . for a homogeneous 2d electron system , the spin susceptibility is most straightforwardly obtained in terms of the spatially fourier - transformed quantity @xmath57 via @xmath58 the dependence of @xmath57 on the coordinates @xmath54 and @xmath59 encodes the spatial profile of the quantum - well bound states . within the bhz framework , the @xmath54-dependent part of electron wave functions is contained in the four basis functions @xmath60 . the latter are spinors whose explicit form has been derived @xcite within the six - band kane - model description @xcite for the charge - carrier dynamics that includes the bands with @xmath61 and @xmath62 symmetry closest to the bulk - material s fundamental gap . as is generally the case in multi - band systems , the spin response of electrons in a hgte quantum well is strongly influenced by both the in - plane dynamics described by the bhz hamiltonian and the nontrivial spinor structure of the bhz - model basis states . we therefore need to express the spin susceptibility within the underlying six - band kane model . the coupling between some physical stimulus represented by a field @xmath63 and the @xmath61 and @xmath62-band intrinsic angular - momentum degrees of freedom @xmath64 and @xmath65 is most generally described by a term @xmath66 in the kane - model hamiltonian . see , e.g. , table c.5 in ref . . within this approach , the coefficients @xmath67 are intra - band coupling constants with appropriately renormalized values to take account of all field - induced band - coupling effects in the bulk material . to be able to discuss a wide range of spin - related phenomena , we define an effective ( pseudo-)spin operator @xmath68 such that @xmath69 . the actual value of the parameter @xmath8 depends on the physical quantity or situation of interest with @xmath70 is relevant , where @xmath71 and @xmath72 are the respective @xmath1 factors for the @xmath61 and @xmath62 bands in the kane - model band - structure description @xcite . in contrast , the intrinsic real - spin polarisation is associated with @xmath73 . see , e.g. , eq . ( 6.65 ) in ref . . to discuss carrier - mediated magnetism , the susceptibility with @xmath74 in the notation of ref . needs to be considered . finally , @xmath75 is the proper angular - momentum operator that generates spatial rotations . ] . the system s response is then fully captured by the effective spin - susceptibility tensor @xmath76 where @xmath77 denotes the fermi function , and @xmath78^\dagger \cdot \big[\hat{s}_i ( \gamma ) \ , \psi^{(s')}_{{{\bm{\mathrm{k}}}}+{{\bm{\mathrm{q}}}}\beta}(z)\big ] \nonumber \\ & & \hspace{2 cm } \times \big[\psi^{(s')}_{{{\bm{\mathrm{k}}}}+{{\bm{\mathrm{q}}}}\beta}(z')\big]^\dagger \cdot \big [ \hat{s}_j(\gamma ) \ , \psi^{(s)}_{{{\bm{\mathrm{k}}}}\alpha}(z')\big]~,\end{aligned}\ ] ] are matrix elements of the effective spin operator given in eq . ( [ eq : effkanespin ] ) . in the spirit of subband @xmath79 theory , the six - dimensional spinor wave functions @xmath80 can be expressed in terms of the bhz - model basis state spinors @xmath81 as @xmath82 where the coefficients @xmath83 are the components of the corresponding eigenvectors of the bhz hamiltonian ( see sec . [ hgteqw ] ) . the explicit form of the basis states was derived in ref . by solving a confined - particle problem in the hgte / cdte hybrid system . for instance , @xmath84 , with the detailed form of the envelope function components @xmath85 provided in the supplemental information of ref . . in the following , we consider the growth - direction - averaged spin susceptibility of charge carriers in the hgte quantum well , given by @xmath86 @xcite , with @xmath87 calculated using the kane - model - based bhz approach as described above . note that , by replacing the spin matrices in eq . ( [ eq : spinoverlap ] ) by the unit matrix and averaging over @xmath88 , we obtain the charge - response function studied in refs . . from the axial symmetry of our kane - model description , and the fact that the eigenstates have definite spin projection in the growth direction , it follows that the in - plane spin susceptibilities are the same , i.e. , @xmath89 , and only depend on the magnitude @xmath90 . this is an important difference to other semiconducting systems where hh - lh mixing occurs @xcite . in the present situation , hh - lh mixing arises only from terms linear in @xmath44 ( giving rise to dirac - like excitations ) which is a consequence of the envelope function components @xcite behaving differently ( even or odd ) under the parity transformation @xmath91 . prefactors @xmath92 and @xmath93 determining the magnitude of intrinsic in - plane and out - of - plane spin response , respectively [ see eq . ( [ eq : intrinsicq0 ] ) ] , plotted as a function of the parameter @xmath8 from the definition of the effective kane - model spin operator ( [ eq : effkanespin ] ) . ] in the insulating limit , the conduction band is empty and the fermi level lies in the band gap , i.e. , @xmath94 . in this situation , the spin susceptibility originates from virtual interband transitions across the band gap . in the following we will focus on the zero temperature limit . the analytical result for the intrinsic spin susceptibility obtained in the limit of zero momentum transfer @xmath95 is given in eq . ( [ eq : intrinsicq0 ] ) , where [ eq : coeffci ] @xmath96 and @xmath97 ( @xmath98 ) is the amount of lh admixture in the basis states @xmath31 . ( details about the calculation of the intrinsic spin susceptibility are given in appendix [ sec : intrinsicnum ] . results for @xmath99 can be calculated numerically . for completeness , some of these are also shown in appendix [ sec : intrinsicnum ] . ) equation ( [ eq : intrinsicq0 ] ) exhibits anomalous behavior in the inverted region ( @xmath100 ) in that the uniform spin response is independent of the gap size and pinned to the value for the gapless case , even though it arises from virtual inter - band transitions . in the normal region ( @xmath101 ) , the expected decrease of the response functions with increasing band gap is found . the spin susceptibility ( [ eq : intrinsicq0 ] ) is strongly anisotropic , generally exhibiting a dominant out - of - plane response except for a very small range of the parameter @xmath8 . see fig . [ fig : ci ] . a non - vanishing @xmath102 seems to imply the counterintuitive phenomenon that an applied magnetic field could generate a magnetisation of the hgte quantum - well system in the intrinsic limit where no charge carriers are present . our more detailed analysis shows , however , that this is not the case . direct calculation of the derivative of the free energy with respect to magnetic - field strength in a model based on the bhz hamiltonian ( [ eq : bhz ] ) augmented by a zeeman term reveals that the total magnetisation is strictly zero . , the grand canonical potential is given by @xmath103}\right)$ ] , where @xmath104 and @xmath105 are the dispersions of the two spin - split bands ( @xmath106 ) obtained from the effective hamiltonian . in the limit @xmath107 ( and @xmath108 ) , we obtain @xmath109 . the net - magnetisation of the bands is given by @xmath110 . numerically we find that @xmath111 , since the numerical coefficients of the two spin - split bands are related as @xmath112 . ] this is due to the fact that intrinsic - spin and orbital contributions to the magnetization cancel , as expected in a spin - orbit - coupled system . this conclusion is further underpinned by the observation that @xmath102 vanishes in the limit of zero hh - lh mixing in eq . ( [ eq : bhz ] ) and letting @xmath113 after integration ] . as it is possible to engineer and study effective exchange interactions between impurity atoms @xcite , it is tempting to consider such interactions between two localized spins in a hgte quantum well . quite generally , the rkky hamiltonian is given by @xcite @xmath114 here @xmath115 , the fourier transform of ( [ eq : wavevecspinsus ] ) , is the spin susceptibility in real space , @xmath116 denotes the @xmath117th cartesian component of an impurity spin located at position @xmath118 , and @xmath119 is the local exchange - coupling constant between the spin degree of freedom carried by band electrons and localized ( e.g. , impurity ) spins . ( the difference in exchange - coupling strengths for conduction - band and valence - band states is accounted for by the appropriate value of @xmath8 . ) in fig . [ fig : intchizzr](a ) , we plot the growth - direction - averaged effective out - of - plane real - space spin susceptibility @xmath120 as a function of the distance @xmath121 for various values of @xmath4 ( and @xmath122 corresponding to realistic situations @xcite ) . figure [ fig : intchizzr ] shows that the effective exchange interaction mediated by the intrinsic spin response of the hgte quantum well for localized spins is of ferromagnetic ( fm ) type if @xmath123 , and that there is a cross - over to antiferromagnetic ( afm ) coupling for @xmath124 that sensitively depends on @xmath4 . in particular , we find that this cross - over is shifted significantly to smaller impurity - spin separations in the case of the inverted regime as compared to the normal regime . also , the magnitude of exchange interaction in the ferromagnetic regime for fixed distance @xmath121 markedly increases when enlarging the gap in the inverted system . in contrast , the exchange interactions in the normal regime have a very small magnitude . to further illustrate the parametric dependencies of the carrier - mediated exchange interaction , fig . [ fig : intchizzr](b ) shows the various regions of preferential spin alignments as a function of @xmath125 . interestingly , inbetween the two out - of - plane localized - spin alignments ( fm@xmath126 and afm@xmath126 ) , we also find sizeable regions for the distance where it is energetically favorable for impurity spins to align ferromagnetically in - plane ( fm@xmath127 ) . moreover , for larger distances , we find the usual bloembergen - rowland @xcite behavior for the intrinsic spin susceptibilities which is characterized by an exponential decay factor that depends on the band gap , approximately given by @xmath128 where @xmath129 is the compton wave length of the band electrons . more details are given in appendix [ sec : bloembergenrowland ] . in the limit of zero gap ( @xmath130 ) , the @xmath131 decay found previously in various dirac systems @xcite is reproduced . ( a ) intrinsic real - space spin susceptibility @xmath132 plotted as a function of @xmath133 for several values of the gap parameter @xmath4 , with fixed @xmath19 and @xmath20 . ( b ) parameter domains of preferential alignments between localized mn spins having a distance @xmath121 , mediated by the intrinsic spin response of an hgte quantum - well system . the blue ( white ) domain favours a ferromagnetic ( anti - ferromagnetic ) out - of - plane spin alignment . in the yellow region the two localized spins tend to align ferromagnetically in - plane.,title="fig:",height=151 ] ( a ) intrinsic real - space spin susceptibility @xmath132 plotted as a function of @xmath133 for several values of the gap parameter @xmath4 , with fixed @xmath19 and @xmath20 . ( b ) parameter domains of preferential alignments between localized mn spins having a distance @xmath121 , mediated by the intrinsic spin response of an hgte quantum - well system . the blue ( white ) domain favours a ferromagnetic ( anti - ferromagnetic ) out - of - plane spin alignment . in the yellow region the two localized spins tend to align ferromagnetically in - plane.,title="fig:",height=151 ] having considered the case of two localized impurity spins with exchange interactions mediated by virtual excitations , we now focus on a hgte quantum well that is doped with a large number of homogeneously distributed magnetic impurities ( eg , mn ions ) , forming effectively a hg@xmath134mn@xmath135te alloy @xcite . the average distance between the spins is @xmath136 @xcite , with @xmath137 the hgte lattice constant and @xmath138 the concentration of magnetic ions . we calculate the curie temperature in the mean field limit , assuming @xmath139 which is justified when @xmath140 nm@xmath141 . since @xmath142 @xcite , this condition is practically always fulfilled . the curie temperature for ising - type ferromagnetic order with magnetization perpendicular to the quantum well growth direction is given by @xmath143 where @xmath144 , @xmath145 denotes the impurity - spin magnitude , @xmath146 is the 3d density of magnetic impurities , and @xmath147 nm is the critical well width . for obtaining eq . ( [ eq : tcmf ] ) we have used the approximation @xmath148 . in fig . [ fig : ct1](a ) , we show the curie temperature as a function of @xmath4 , where we set @xmath149 ev nm and @xmath150 ev nm@xmath151 @xcite as their dependence on @xmath2 is much weaker compared to @xmath27 . we see that the behavior of the curie temperature in its dependence on @xmath4 , or equivalently @xmath2 , provides a clear means to distinguish between topological and normal regions . knowing the spin susceptibility as a function of the wave vector allows us to go beyond the mean field limit @xcite and discuss the stability of the mean - field ground state with respect to thermally excited spin waves ( magnons ) . in our present case of interest , the magnon dispersion for @xmath152 is given by @xmath153 , with a coefficient @xmath154 . using figure [ fig : spinsusq ] in appendix [ sec : intrinsicnum ] , which shows results for the relevant set of parameters , we find @xmath155 . thus the criterion @xcite @xmath156 needed to guarantee stability of the mean - field ground state is generally violated . it may be possible that electron - electron interactions and/or spin - orbit coupling help to stabilize ferromagnetic order , as has been shown to be the case for an ordinary 2d electron gas @xcite , but this question is beyond the scope of our present work . we consider now situations where the conduction band is occupied ( @xmath157 ) . in the @xmath95 limit , an analytical expression for the extrinsic contribution to the spin susceptibility ( arising from filled conduction - band states ) is found as [ eq : chijjq0 ] @xmath158~,\\[2 mm ] \overline{\chi}_{zz}^{(\text{dop})}(\gamma;0)&=&\frac{\mathcal{c}^2_z(\gamma)}{16\pi|b|}\biggl[\frac{2\pi \tilde n(0)}{f^2}\left(\frac{6 \gamma \mathcal{d}_z(\gamma)}{\mathcal{c}^2_z(\gamma)}\ , \tilde k_{\text{f}}^2 - \frac{4 f^2 \{\mathcal{d}^2_z(\gamma)(\tilde k_{\text{f}}^2 + \xi_{\text{m}}+f)^4 + 9\gamma^2\tilde k_{\text{f}}^4\}}{\mathcal{c}^2_z(\gamma)[\tilde k_{\text{f}}^2+(\tilde k_{\text{f}}^2 + \xi_{\text{m}}+f)^2]^2}\right ) + \frac{\tilde k_{\text{f}}^2 ( 1 + 2 \xi_{\text{m } } ) + 2 \xi_{\text{m}}^2 - 2 f |\xi_{\text{m}}|)}{f(1 + 4 \xi_{\text{m}})}\biggr]~ , \nonumber \\\end{aligned}\ ] ] where the first ( second ) term between the square brackets in the expression for @xmath159 is an intraband ( interband ) contribution . in eqs . ( [ eq : chijjq0 ] ) , we used the abbreviations @xmath160 and @xmath161 . @xmath162 is related to the density of states at the fermi energy as @xmath163 ( where @xmath164 is the static charge - response function ) , and reads @xmath165 . for the general case with @xmath166 , @xmath167 can be calculated numerically and its line shape is presented in appendix [ sec : dopednum ] for various examples . due to the sharpness of the fermi surface , we find for the corresponding real - space spin susceptibilities the expected @xmath168 oscillatory decay expected from a 2d fermi liquid @xcite . next we calculate the curie temperature using the mean field approach under the premise @xmath169 , which leads to the condition @xmath170 . this condition is generally fulfilled since @xmath171 nm@xmath141 is the reliable range of the effective model @xcite . the curie temperature is given by @xmath172 where @xmath173 . note that also in the doped case ising type magnetism prevails . in fig . [ fig : ct1](b ) , we show the curie temperature as a function of @xmath4 for various values of the fermi wave vector . for low doping , we see a strong dependence of the curie temperature on @xmath4 in the topological regime , which becomes very weak in the normal regime . for large doping , on the other hand , the curie temperature is not very sensitive to @xmath4 in its whole range . furthermore , the curie temperature is generally suppressed with increased doping , but this trend is much stronger in the topological regime . for the doped case , we find the magnon dispersion @xmath174 where @xmath175 always , in contrast to the intrinsic case . see fig . [ fig : spinsusdop ] in appendix [ sec : dopednum ] . thus a necessary condition for the stability of the ferromagnetic groundstate is fulfilled @xcite . the paramagnetic response of charge carriers in quantum - confined structures is usually interpreted in terms of the pauli spin susceptibility and quantified by an effective single - particle @xmath1 factor @xcite . however , any actually measured spin - related quantities correspond almost always to averaged collective responses of the , e.g. , quasi-2d , electron system that can be crucially affected by nontrivial spin - related phenomena @xcite . here we consider the paramagnetic response of conduction - band electrons in hgte quantum wells and show how their paramagnetic response is changed as a function of the band - gap parameter that drives the transition between the topological and normal regimes . to define an effective @xmath1-factor for our system of a hgte quantum well , we introduce the bulk - material zeeman term @xmath177 , where @xmath178 is the bohr magneton and @xmath71 ( @xmath72 ) the @xmath61-band ( @xmath62-band ) @xmath1 factor @xcite . linear - response theory enables us to determine the paramagnetic response to the magnetic field , which is given by @xmath179 , where @xmath180 are the spin susceptibilities of the electron doped system for the in - plane and out - of - plane response involving @xmath181 , see eq . ( [ eq : chijjq0 ] ) for @xmath182 . we compare this with the pauli susceptibility given by @xmath183 , with @xmath184 being ( up to a minus sign ) the density of states which is the zero-@xmath185 limit of the lindhard function and @xmath186 is the land @xmath1-factor for the two directions . thus , we can extract collective @xmath1 factors for the charge carriers as @xmath187 our approach to determine @xmath1 factors via the spin susceptibility complements previous work @xcite where an effective zeeman term was derived for the bhz hamiltonian . effective @xmath1-factors associated with ( a ) the in - plane and ( b ) the out - of - plane response of an electron - doped hgte quantum well , plotted as a function of the gap parameter @xmath4 for various levels of doping and with @xmath19 , @xmath188 . ( note that @xmath189 corresponds to a charge density @xmath190@xmath18 for a typical structure @xcite.),title="fig : " ] effective @xmath1-factors associated with ( a ) the in - plane and ( b ) the out - of - plane response of an electron - doped hgte quantum well , plotted as a function of the gap parameter @xmath4 for various levels of doping and with @xmath19 , @xmath188 . ( note that @xmath189 corresponds to a charge density @xmath190@xmath18 for a typical structure @xcite.),title="fig : " ] for a narrow - gap semiconductor , the values for @xmath71 and @xmath72 are dominated by band - coupling contributions . using generic expressions resulting from kane - model descriptions @xcite , we find @xmath191 with @xmath192 in terms of the spin - orbit - splitting gap @xmath193 between the @xmath62 and @xmath194 band edges in the @xmath195 kane model . as @xmath196 for our situations of interest , we can set @xmath197 in the following . hence , in fig . [ fig : g - factor ] , we show the effective @xmath1-factors for in - plane and out - of - plane responses as a function of @xmath4 for various levels of electron doping obtained for @xmath188 . for very low densities , we see a vanishing in - plane response and a maximal @xmath198 hh - like out - of - plane response in the topological region @xmath100 . this behavior arises because the conduction - band character is dominated by the hh basis states , which experience a `` frozen '' spin orientation perpendicular to the quantum well due to the confinement - induced hh - lh energy splitting @xcite . this can be easily verified from ( [ eq : chijjq0 ] ) by taking the limit @xmath199 . departures from these results occur for larger doping levels ( larger @xmath200 ) due to increased hh - lh mixing . in the normal region ( @xmath101 ) , we encounter the situation that the conduction band is dominantly composed of the electron basis states , rendering the _ g _ factor to be sizable at any doping . the fact that the in - plane spin response is notably larger than the out - of - plane response for @xmath101 and low doping is due to the lh admixture in the conduction - band states . as the carrier density increases , the concomitantly increased hh - lh mixing results in significant modifications . more detailed exploration of parametric dependences exhibited by the collective _ g _ factors obtained here will be useful to augment previous perturbative estimates @xcite and aid the interpretation of recent measurements @xcite . we have also investigated the edge - state contributions to the spin susceptibility in the topological region , finding them to be negligible compared to the bulk contributions in almost all situations . the only exception occurs for the in - plane response function @xmath201 in the purely intrinsic situation where the fermi level lies in the minigap of the edge - state dispersions that opens up in a finite - size sample @xcite . more details and full results pertaining to edge states are given in appendices [ sec : compbulkedge ] and [ sec : spinsusedge ] . effective @xmath1-factor characterising the response of helical edge states to an in - plane magnetic field . ] upto very small finite - size corrections , the paramagnetic response of helical edge states is captured by using the effective @xmath1-factors @xmath202 given by [ eq : gfactoredge ] @xmath203 the results shown in eqs . ( [ eq : gfactoredge ] ) were obtained using eq . ( [ eq : gfactor ] ) together with eqs . ( [ eq : susedgezz])-([eq : susedgexx ] ) . the prediction of the edge - state _ factor is a major result of our work . in fig . [ fig : g - factoredge ] , we plot the in - plane @xmath1-factor as a function of @xmath200 where we used the natural cut - off scale @xmath204 . interestingly , the in - plane @xmath1-factor of edge states decreases monotonically with increasing doping level . such a behavior is in stark contrast to the bulk case shown in fig . [ fig : g - factor](a ) and to other 2d systems @xcite where the in - plane response increases with increased doping . this behavior reflects the fact that the spin - quantisation axis of the edge states is perpendicular to the 2d plane , i.e. , parallel to the @xmath54 direction , and their helical nature . we present a detailed theoretical study of the spin response in hgte quantum wells where virtual - carrier - related processes are particularly relevant and exactly tractable within the effective bhz - model description . anomalous properties of the spin susceptibility and carrier - mediated magnetism are found in the inverted regime , extending our current understanding of spin - related properties in topological materials . most strikingly , the uniform static spin susceptibility in the intrinsic limit is constant , independent of the band gap , in the topological regime [ see eq . ( [ eq : intrinsicq0 ] ) ] . this results in a distinct asymmetry between normal and inverted systems illustrated , e.g. , by the dependence of the critical temperature for virtual - carrier - mediated magnetic order in a system that is doped with magnetic ions . important differences between the two regimes are also exhibited in the situation where the conduction band becomes filled ( see fig . [ fig : ct1 ] ) . the stability of mean - field ferromagnetic ground states with respect to thermal excitations of magnons has been analysed . we find that the magnetic order is stable in the situation with finite doping . in the topological regime , quasi-1d helical edge states exist . we have investigated their spin - response properties , finding that their contribution to the spin susceptibility is thermodynamically suppressed compared to that arising from 2d quantum - well states , _ except _ in the very special and probably physically hard - to - realise situation when the chemical potential is in the mini - gap opened by the hybridisation of states from opposite edges in a finite sample . we have used our results obtained for the spin susceptibility to define effective collective @xmath1 factors for states from the quasi-2d quantum - well subbands and also for the quasi-1d helical edge states . the different character of quasi-2d - subband states in the normal and topological ( inverted ) regimes is reflected in the values for the effective @xmath1 factor . their strong dependence on charge - carrier density reveals the importance of interband mixing . the behavior of the @xmath1 factors found for the edge states reflects their helical nature and spin - quantization property . our results are directly relevant for current and potential future experimental investigations of the spin - related properties of 2d topological insulators , in particular those realised in hgte / hgcdte and inas / gasb quantum wells . for example , recent observation of josephson - junction interference patterns in an s - hgte / hgcdte - s hybrid system has enabled extraction of _ g _ factors for the quantum - well charge carriers @xcite . it would be interesting to use similar techniques @xcite to measure the edge - state _ g _ factors and compare with our predictions . furthermore , our results for the spin response in both the intrinsic and doped regimes are informative for the design of , and interpretation of measured quantities for , dilute magnetic phases in these systems @xcite . it would be interesting to extend our formalism to study the spin response in 3d topological - insulator materials @xcite . in particular , as was the case in the 2d quantum - well - based tis considered in this work , the interplay of charge - carrier dynamics and the spinor character of extended bulk states could be a source of rich variety in spin - related properties also in 3d , and the contributions of the conducting surfaces are currently not understood . e. m. h. acknowledges financial support from german science foundation ( dfg ) grant no . ha 5893/4 - 1 within spp 1666 and from the enb graduate school `` topological insulators '' . this work was started while e. m. h. and u. z. visited the kavli institute for theoretical physics at the university of santa barbara where their work was supported in part by the national science foundation under grant no . nsf phy11 - 25915 . the authors thank r. winkler for illuminating discussions about possible forms of relevant spin operators and useful comments that have improved the manuscript . intrinsic spin susceptibilities ( a ) @xmath205 and ( b ) @xmath206 as a function of @xmath207 for various values of @xmath208 ( from top to bottom ) with @xmath19 , @xmath209 and @xmath210 . the inset in ( a ) displays the normalized spin susceptibility @xmath211 in the small @xmath207 region for @xmath212 , which shows more clearly that @xmath213 in the small-@xmath185 limit also in the normal regime.,title="fig : " ] + intrinsic spin susceptibilities ( a ) @xmath205 and ( b ) @xmath206 as a function of @xmath207 for various values of @xmath208 ( from top to bottom ) with @xmath19 , @xmath209 and @xmath210 . the inset in ( a ) displays the normalized spin susceptibility @xmath211 in the small @xmath207 region for @xmath212 , which shows more clearly that @xmath213 in the small-@xmath185 limit also in the normal regime.,title="fig : " ] + the intrinsic contributions to the diagonal entries of the spin susceptibility ( the only ones that are non - zero for the bhz model ) are calculated by @xmath214 where in the zero - temperature limit the valence band is fully occupied , i.e. , @xmath215 . here and in the following we consider the growth - direction - averaged case : @xmath216 . in fig . [ fig : spinsusq ] , we plot @xmath205 [ @xmath217 in panel ( a ) [ ( b ) ] as a function of @xmath185 for various values of @xmath4 . the inset in fig . [ fig : spinsusq](a ) illustrates that @xmath218 for all of the values of @xmath4 considered , which implies that the mean - field ferromagnetic order for out - of - plane aligned magnetic impurity spins will generally be destroyed by spin - wave ( magnon ) excitations @xcite . local intrinsic spin susceptibilities @xmath219 ( solid orange curve ) as a function of @xmath220 for ( a ) @xmath221 and ( b ) @xmath222 with @xmath19 , @xmath209 and @xmath20 . the dashed green curve shows the modelled behavior in eq . ( [ eq : rkkybrmodel ] ) with coefficients @xmath223 for ( a ) and @xmath224 for ( b ) . for comparison we also show the @xmath131 decay for a gapless dirac system represented by the dot - dashed blue curve . the sharp drop of the orange curve in ( a ) is due to a sign change of @xmath219 at about @xmath225.,title="fig : " ] + local intrinsic spin susceptibilities @xmath219 ( solid orange curve ) as a function of @xmath220 for ( a ) @xmath221 and ( b ) @xmath222 with @xmath19 , @xmath209 and @xmath20 . the dashed green curve shows the modelled behavior in eq . ( [ eq : rkkybrmodel ] ) with coefficients @xmath223 for ( a ) and @xmath224 for ( b ) . for comparison we also show the @xmath131 decay for a gapless dirac system represented by the dot - dashed blue curve . the sharp drop of the orange curve in ( a ) is due to a sign change of @xmath219 at about @xmath225.,title="fig : " ] bloembergen and rowland @xcite found that the local rkky interaction of gapped systems becomes short - ranged , i.e. , is exponentially suppressed by the band - gap . the functional dependence of the local spin susceptibility on the distance for the gapped dirac system at hand can thus be modelled by @xmath226 where @xmath227 is the inverse of the compton wavelength of the system and @xmath228 is a numerical coefficient that can depend on the distance itself . to illustrate this behavior for the hgte quantum well system , we plot @xmath229 in fig . [ fig : brchir ] as a function of @xmath230 for @xmath221 [ @xmath222 ] in panel ( a ) [ ( b ) ] together with both the line shape expected for 2d massless - dirac particles and the bloembergen - rowland result . -0.5 for the case where the fermi energy is above the conduction energy band edge , i.e. , @xmath232 , the spin susceptibility receives contributions due to electron doping given by @xmath233 & & \hspace{-1.3 cm } \times\left [ \frac{\mathscr{w}_{jj({{\bm{\mathrm{k}}}},{{\bm{\mathrm{k}}}}+{{\bm{\mathrm{q}}}},+,+)}^{(s , s')}(\gamma)}{e^{(s)}_{{{\bm{\mathrm{k}}}}+}-e^{(s')}_{{{\bm{\mathrm{k}}}}+{{\bm{\mathrm{q}}}}+ } + i\hbar\eta\delta}+\frac{\mathscr{w}_{jj({{\bm{\mathrm{k}}}},{{\bm{\mathrm{k}}}}+{{\bm{\mathrm{q}}}},+,-)}^{(s , s')}(\gamma)}{e^{(s)}_{{{\bm{\mathrm{k}}}}+ } -e^{(s')}_{{{\bm{\mathrm{k}}}}+{{\bm{\mathrm{q}}}}-}+i\hbar\eta\delta}\right ] , \nonumber\\[2mm]\end{aligned}\ ] ] where for zero temperature @xmath234 , with @xmath200 being the fermi wave vector associated with the conduction band . the complete spin susceptibility in the doped case is therefore given by @xmath235 we show the line shape of @xmath236 and @xmath201 in fig . [ fig : spinsusdop ] for @xmath237 and various choices of doping . the insets of fig . [ fig : spinsusdop](a , c ) show the small @xmath185 dependence of @xmath236 , indicating the stability of the out - of - plane ising - type ferromagnetism with respect to magnon excitations . in order to compare bulk and edge contributions to the spin susceptibility we begin by considering the real - space spin - response function @xmath238\rangle\,\ , , \ ] ] where @xmath239 are spin - density operators . in our system of interest , electrons shall be confined to move freely in @xmath240 dimensions , with system size @xmath241 in all of these free directions . the position vector shall be split up into a part @xmath242 comprising the coordinate directions in which the motion is free and a part @xmath243 in whose coordinates motion is confined ; @xmath244 . the second - quantised electron operator in real - space representation can be written as @xmath245 with normalized spinor bound - state wave functions @xmath246 . a straightforward calculation yields @xmath247 with the @xmath248-dependent spin susceptibility of the @xmath2-dimensional ( @xmath2d ) system given by @xmath249 we are now interested in the homogeneous part of the spin response defined as @xmath250 where @xmath251 in the situation where both 2d bulk and 1d edge states are present , we therefore find @xmath252 \quad .\ ] ] thus @xmath253 is the well - defined quantity in the thermodynamic limit , and only those edge - related terms that scale with @xmath241 will contribute to it . following @xcite ( see also @xcite ) , the dispersions for the edge states , using open boundary conditions for a confinement along the @xmath138 direction and assuming @xmath254 , is given by @xmath255 where @xmath256 . the associated spinor wave functions are @xmath257 where @xmath258 , with @xmath259 and @xmath260 , since @xmath29 . to simplify matters , we have assumed particle - hole symmetry ( @xmath254 ) and the system length @xmath261 , i.e. , the gap in the dispersions of the edge states is negligible . note that the spinors in eq . ( [ eq : edgestatespinors ] ) are independent of the wave vector component along the @xmath262 direction . the spin susceptibility of the edge states is calculated by @xmath263 { } & & \times \frac{n_{\text{f}}[e^{(s)}_{k}]-n_{\text{f}}[e^{(s')}_{k+q}]}{e^{(s)}_{k}-e^{(s')}_{k+q}+i\hbar \eta}~,\end{aligned}\ ] ] where @xmath264^{\dagger } \cdot\hat s_j(\gamma ) \cdot \eta^{(s')}({{\bm{\mathrm{\varrho } } } } ) \nonumber\\[1 mm ] { } & & \times [ \eta^{(s')}({{\bm{\mathrm{\varrho^\prime}}}})]^\dagger\cdot\hat s_j(\gamma ) \cdot\eta^{(s)}({{\bm{\mathrm{\varrho^\prime}}}}).\end{aligned}\ ] ] with an averaging over the coordinates along the confined directions we have @xmath265 \int d{{\bm{\mathrm{\varrho}}}}~d{{\bm{\mathrm{\varrho^\prime}}}}~\mathscr{w}_{xx}^{(s , s')}(\gamma;{{\bm{\mathrm{\varrho } } } } , { { \bm{\mathrm{\varrho^\prime}}}})&= & \frac{\xi^2_x(\gamma)}{16}\begin{pmatrix } 0 & 1\\ 1 & 0 \end{pmatrix}~,\end{aligned}\ ] ] with @xmath266 and @xmath267 . thus @xmath268 , where @xmath269 is the lindhard function associated with the edge states . explicit calculation of ( [ eq : spinsusedge ] ) yields [ eq : susedgeelec ] @xmath270 where @xmath271 is a large - wave - vector cut - off . due to the special energy dispersion eq . ( [ eq : edgedispersion ] ) , @xmath272 is a constant ( independent of @xmath200 and @xmath185 ) . clearly , for the hole doped case ( @xmath273 ) , the same result as in eq . ( [ eq : susedgeelec ] ) is obtained due to the assumed particle - hole symmetry . following @xcite , we now take into account the finite size of the system . as a result , edge states at the two sides that have the same spin can couple which results in a gapped spectrum of their energy dispersions.we assume the system size @xmath241 to be large and use an approximation of the wave functions in @xcite . taking into account the various degrees of freedom of both system sides , we modify the approach of @xcite , where the solutions for the wave functions read @xmath274 where @xmath275 is a normalization constant and @xmath276 denotes the right- and left - mover . for non - zero wave vector @xmath277 , we consider the matrix @xmath278 \delta & -a k \end{pmatrix}~\end{aligned}\]]in the basis of right- and left - mover , where @xmath279 is the induced gap which is a function of the system parameters @xcite . the energy dispersions of ( [ eq : rlmover ] ) are @xmath280 where @xmath281 labels the ( spin - degenerate ) conduction and valence bands , respectively . the associated eigenstates are @xmath282 \tau \sqrt{\frac{e^{(s)}_{k\tau}-ak}{2e^{(s)}_{k\tau } } } \end{pmatrix}~.\ ] ] with this information , the spinors in eq . ( [ eq : edgestatespinors ] ) are modified to @xmath283 thus , for the present case , the spin susceptibility of the edge states is @xmath284 { } & & \times\frac{n_{\text{f}}[e^{(s)}_{k\tau}]-n_{\text{f}}[e^{(s')}_{k+q \tau'}]}{e^{(s)}_{k\tau}-e^{(s')}_{k+q\tau'}+i\hbar\eta}~,\end{aligned}\ ] ] where @xmath285^\dagger\cdot\hat s_j(\gamma)\cdot \eta^{(s')}_{k+q\tau'}({{\bm{\mathrm{\varrho}}}})\nonumber \\[1 mm ] { } & & \times [ \eta^{(s')}_{k+q\tau'}({{\bm{\mathrm{\varrho^\prime}}}})]^\dagger\cdot\hat s_j(\gamma)\cdot \eta^{(s)}_{k\tau}({{\bm{\mathrm{\varrho^\prime}}}})~.\nonumber\\[2mm]\end{aligned}\ ] ] averaging over the coordinates along the confined directions , we obtain for the overlap factor of the lindhard function @xmath286 ( a_k b_{k+q}-b_k a_{k+q})^2 & \quad ( \tau\neq\tau ' ) \end{cases}\end{aligned}\ ] ] where @xmath287 , @xmath288 . the result for the overlap factors of the spin susceptibilities is @xmath289 & & { } \times \begin{cases } [ ( a_k a_{k+q}+b_k b_{k+q})~\xi_z(\gamma ) & \quad ( \tau=\tau')\\[1 mm ] + \tau ( a_k b_{k+q}+b_k a_{k+q})~\mathcal{c}_z(\gamma)~n l \delta]^2 \\[3 mm ] [ ( b_k b_{k+q}-a_k a_{k+q})~\mathcal{c}_z(\gamma)~n l \delta&\quad ( \tau\neq\tau')\\[1 mm ] + \tau(a_k b_{k+q}-b_k a_{k+q})~\xi_z(\gamma)]^2 \end{cases}\end{aligned}\ ] ] and @xmath290 & & { } \times \begin{cases } [ a_k b_{k+q}+ b_k a_{k+q}&\quad ( \tau=\tau')\\[1 mm ] + \tau(a_k a_{k+q}+ b_k b_{k+q } ) n l \delta ] ^2\\[3 mm ] [ a_k a_{k+q}- b_k b_{k+q}&\quad ( \tau\neq \tau')\\[1 mm ] -\tau(a_k b_{k+q}- b_k a_{k+q } ) n l \delta ] ^2\\[3 mm ] \end{cases}\end{aligned}\ ] ] in obtaining ( [ eq : overap0edge])-([eq : overap0edgexx ] ) we have used @xmath291 the functional @xmath241-dependence of @xmath292 @xcite , where @xmath293 , and we have defined @xmath294 . calculating the intrinsic contribution to the spin susceptibilities of the edge states in the long - wavelength limit ( @xmath95 ) , we obtain [ eq : intrinsicq0edge1 ] @xmath295~,\nonumber\\\end{aligned}\ ] ] @xmath296~,\end{aligned}\ ] ] where @xmath297 . we note that the intrinsic contribution to the lindhard function vanishes in the limit @xmath95 , which can be inferred from ( [ eq : overap0edge ] ) . to compare this result with the one of the intrinsic bulk contribution , we divide ( [ eq : intrinsicq0edge1zz ] ) and ( [ eq : intrinsicq0edge1xx ] ) by the length @xmath241 and let @xmath241 go to infinity ( see sec . [ sec : compbulkedge ] ) . thus , we obtain next we include the contributions due to doping to the spin susceptibilities . the interband contributions read ( for @xmath300 ) [ eq : dopedq0edgeinter ] @xmath301,\nonumber\\\end{aligned}\ ] ] @xmath302,\end{aligned}\ ] ] while the intraband contributions are [ eq : dopedq0edgeintra ] @xmath303 @xmath304 which is consistent with the finding that @xmath305 [ @xmath306 are important [ unimportant ] for @xmath307 , whereas it is the other way around for the interband contributions . considering @xmath308}{l}=0~,\end{aligned}\ ] ] @xmath309}{l}=\nonumber\\[2 mm ] & & { } \hskip1.4cm\frac{g_{\text{s}}\xi^2_x(\gamma)}{32\pi|b| } \left(1-\sqrt{1 - 4|\xi_{\text{m}}|}\right),\end{aligned}\ ] ] we find that this is the same contribution as the intrinsic contribution , eq . ( [ eq : edgeq0intrinsicxx ] ) , which has however the opposite sign . thus , the sum of doped and intrinsic contributions of the edge states vanishes in the large @xmath241 limit . as a consistency check , we verify for @xmath310 that the sum of eqs . ( [ eq : intrinsicq0edge1 ] ) , ( [ eq : dopedq0edgeinter ] ) and ( [ eq : dopedq0edgeintra ] ) yields eqs . ( [ eq : susedgezz ] ) and ( [ eq : susedgexx ] ) ( multiplied by @xmath311 ) in the limit @xmath312 . here we demonstrate that the basic features of the intrinsic spin susceptibilities given in eq . ( [ eq : intrinsicq0 ] ) are robust even when effects due to structural inversion asymmetry ( sia ) are included . by taking into account the influence of a perpendicular electric field @xmath313 , it has been shown in ref . @xcite that the bhz hamiltonian is supplemented by entries that mix the spin - up and spin - down components of the bhz basis states . the leading contribution due to sia is linear in the wave vector and given by @xmath314 0 & 0 & 0 & 0 \\[2 mm ] i r_0k_+ & 0 & 0 & 0 \\[2 mm ] 0 & 0 & 0 & 0 \end{pmatrix}~,\ ] ] where @xmath315 . the necessity to avoid dielectric breakdown provides an upper limit for the electric - field magnitude through the condition @xmath316 . defining the sia - related dimensionless parameter @xmath317 , this condition translates into @xmath318)\approx 0.12 $ ] for a typical heterostructure @xcite . thus @xmath319 is generally a small parameter , and a perturbative treatment for sia effects is appropriate . to lowest order in @xmath320 , the intrinsic spin susceptibility in the limit @xmath95 is found as [ eq : intrinsicjjq0r ] @xmath321 ^ 2}\right)~ , \\ \label{eq : intrinsiczzq0r } \overline{\chi}^{(\text{int})}_{zz}(\gamma ; { { \bm{\mathrm{q}}}}= 0 ) & = & -\frac{\mathcal{c}^2_{z}(\gamma)}{16\pi |b| } \frac{1}{1 + 4\xi_{\text{m}}\theta(\xi_{\text{m}})}\ , \left(1 + \xi_{\text{r}}^2~\frac{2\{9\gamma^2 + 2 \xi_{\text{m } } \theta(\xi_{\text{m}})[\mathcal{c}^2_{z}(\gamma)+18\gamma^2]\}}{3\mathcal{c}^2_{z } ( \gamma)[1 + 4\xi_{\text{m}}\theta(\xi_{\text{m}})]^2}\right)~.\end{aligned}\ ] ] thus the lowest - order sia corrections to @xmath322 are quadratic in the small parameter @xmath320 . this means that the result given in eq . ( [ eq : intrinsicjjq0r ] ) represents already an excellent approximation . interestingly , @xmath323 turns out to not be modified by sia contributions in the inverted regime . we find that this remains true even when higher - order corrections in @xmath320 are considered . in contrast , @xmath324 in eq . ( [ eq : intrinsiczzq0r ] ) has finite sia corrections in the inverted regime given by @xmath325 . in our case where @xmath20 this amounts to a relative change that is about 1% . also in the normal regime , sia contributions to @xmath322 are at most of relative magnitude 1% . thus we conclude that the spin susceptibilities in eq . ( [ eq : intrinsicq0 ] ) generally receive only very small corrections when sia terms are included in the bhz hamiltonian . 47ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop `` , '' @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) in @noop _ _ , ( , , ) pp . @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , , ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( )

we consider the estimation of integrated covariance ( icv ) matrices of high dimensional diffusion processes based on high frequency observations . we start by studying the most commonly used estimator , the _ realized covariance _ ( rcv ) matrix . we show that in the high dimensional case when the dimension and the observation frequency grow in the same rate , the limiting spectral distribution ( lsd ) of rcv depends on the covolatility process _ not only through the targeting icv _ , _ but also on how the covolatility process varies in time_. we establish a marenko pastur type theorem for weighted sample covariance matrices , based on which we obtain a marenko pastur type theorem for rcv for a class of diffusion processes . the results explicitly demonstrate how the time variability of the covolatility process affects the lsd of rcv . we further propose an alternative estimator , the _ time - variation adjusted realized covariance _ ( tvarcv ) matrix . we show that for processes in class , the tvarcv possesses the desirable property that its lsd depends solely on that of the targeting icv through the marenko pastur equation , and hence , in particular , the tvarcv can be used to recover the empirical spectral distribution of the icv by using existing algorithms . . .

1005.1862

diffusion processes are widely used to model financial asset price processes . for example , suppose that we have multiple stocks , say , @xmath0 stocks whose price processes are denoted by @xmath3 for @xmath4 , and @xmath5 are the log price processes . let @xmath6 . then a widely used model for @xmath7 is [ see , e.g. , definition 1 in @xcite ] @xmath8 where , @xmath9 is a @xmath0-dimensional drift process ; @xmath10 is a matrix for any @xmath11 , and is called the ( instantaneous ) _ covolatility process _ ; and @xmath12 is a @xmath0-dimensional standard brownian motion . the _ integrated covariance _ ( icv ) matrix @xmath13 is of great interest in financial applications , which in the one dimensional case is known as the _ integrated volatility_. a widely used estimator of the icv matrix is the so - called _ realized covariance _ ( rcv ) matrix , which is defined as follows . assume that we can observe the processes @xmath14 s at high frequency synchronously , say , at time points @xmath15 : @xmath16 then the rcv matrix is defined as @xmath17\\[-8pt ] & & \eqntext{\mbox{where } \delta\mathbf{x}_{\ell } = \pmatrix { \delta x^{(1)}_\ell\cr \vdots\cr \delta x^{(p)}_\ell } : = \pmatrix { x^{(1)}_{\tau_{n,\ell}}-x^{(1)}_{\tau_{n,\ell-1}}\cr \vdots\cr x^{(p)}_{\tau_{n,\ell}}-x^{(p)}_{\tau_{n,\ell-1}}}.}\end{aligned}\ ] ] in the one dimensional case , the rcv matrix reduces to the _ realized volatility_. thanks to its nice convergence to the icv matrix as the observation frequency @xmath1 goes to infinity [ see @xcite ] , the rcv matrix is highly appreciated in both academic research and practical applications . the tick - by - tick data are usually not observed synchronously , and moreover are contaminated by market microstructure noise . on sparsely sampled data ( e.g. , 5-minute data for some highly liquid assets , or subsample from data synchronized by refresh times [ @xcite ] ) , the theory in this paper should be readily applicable , just as one can use the realized volatility based on sparsely sampled data to estimate the integrated volatility ; see , for example , @xcite . having a good estimate of the icv matrix @xmath18 , in particular , its spectrum ( i.e. , its set of eigenvalues @xmath19 ) , is crucial in many applications such as principal component analysis and portfolio optimization ( see , e.g. , the pioneer work of markowitz ( @xcite ) and a more recent work [ @xcite ] ) . when the dimension @xmath0 is high , it is more convenient to study , instead of the @xmath0 eigenvalues @xmath19 , the associated _ empirical spectral distribution _ ( esd ) @xmath20 a naive estimator of the spectrum of the icv matrix @xmath18 is the spectrum of the rcv matrix @xmath21 . in particular , one wishes that the esd @xmath22 of @xmath21 would approximate @xmath23 well when the frequency @xmath1 is sufficiently high . from the large dimensional random matrix theory ( ldrmt ) , we now understand quite well that in the high dimensional setting this good wish wo nt come true . for example , in the simplest case when the drift process is 0 , covolatility process is constant , and observation times @xmath15 are equally spaced , namely , @xmath24 , we are in the setting of estimating the usual _ covariance matrix _ using the _ sample covariance matrix _ , given @xmath1 i.i.d . @xmath0-dimensional observations @xmath25 . from ldrmt , we know that if @xmath26 converges to a non - zero number and the esd @xmath23 of the true covariance matrix converges , then the esd @xmath27 of the sample covariance matrix also converges ; see , for example , @xcite , @xcite , @xcite and @xcite . the relationship between the _ limiting spectral distribution _ ( lsd ) of @xmath28 in this case and the lsd of @xmath18 can be described by a marenko pastur equation through stieltjes transforms , as follows . [ propmpgen ] assume on a common probability space : for @xmath29 and for @xmath30 , @xmath31 with @xmath32 i.i.d . with mean 0 and variance 1 ; [ asmyn ] @xmath33 with @xmath34 as @xmath35 ; [ asmconvsigma ] @xmath18 is a ( possibly random ) nonnegative definite @xmath36 matrix such that its esd @xmath37 converges almost surely in distribution to a probability distribution @xmath38 on @xmath39 as @xmath35 ; @xmath18 and @xmath40 s are independent . let @xmath41 be the ( nonnegative ) square root matrix of @xmath18 and @xmath42 . then , almost surely , the esd of @xmath43 converges in distribution to a probability distribution @xmath44 , which is determined by @xmath38 in that its stieltjes transform @xmath45 solves the equation @xmath46 in the special case when @xmath47 , where @xmath48 is the @xmath36 identity matrix , the lsd @xmath44 can be explicitly expressed as follows . [ propmp ] suppose that @xmath40 s are as in the previous proposition , and @xmath47 for some @xmath49 . then the lsd @xmath44 has density @xmath50 and a point mass @xmath51 at the origin if @xmath52 , where @xmath53 the lsd @xmath44 in this proposition is called the marenko pastur law with ratio index @xmath54 and scale index @xmath55 , and will be denoted by mp@xmath56 in this article . in practice , the covolatility process is typically not constant . for example , it is commonly observed that the stock intraday volatility tends to be u - shaped [ see , e.g. , @xcite , @xcite ] or exhibits some other patterns [ see , e.g. , @xcite ] . in this article , we shall allow them to be not only varying in time but also stochastic . furthermore , we shall allow the observation times @xmath15 to be random . these generalizations make our study to be different in nature from the ldrmt : in ldrmt the observations are i.i.d . ; in our setting , the observations @xmath57 may , first , be dependant with each other , and second , have different distributions because ( i ) the covolatility process may vary over time , and ( ii ) the observation durations @xmath58 may be different . in general , for any time - varying covolatility process @xmath10 , we associate it with a constant covolatility process given by the square root of the icv matrix @xmath59.\ ] ] let @xmath60 be defined by replacing @xmath10 with the constant covolatility process @xmath61 ( and replacing @xmath62 with 0 , and @xmath12 with another independent brownian motion , if necessary ) in ( [ eqx ] ) . observe that @xmath7 and @xmath60 share the same icv matrix at time 1 . based on @xmath60 , we have an associated rcv matrix @xmath63 which is estimating the same icv matrix as @xmath64 . since @xmath64 and @xmath65 are based on the same estimation method and share the same targeting icv matrix , it is desirable that their esds have similar properties . in particular , based on the results in ldrmt and the discussion about constant covolatility case in section [ ssecldrmt ] , we have the following property for @xmath65 : if the esd @xmath23 converges , then so does @xmath66 ; moreover , their limits are related to each other via the marenko pastur equation ( [ eqstfh0 ] ) . does this property also hold for @xmath21 ? our first result ( proposition [ propnotconv ] ) shows that even in the most ideal case when the covolatility process has the form @xmath67 for some deterministic ( scalar ) function @xmath68 , such convergence results may _ not _ hold for @xmath21 . in particular , the limit of @xmath27 ( when it exists ) changes according to how the covolatility process evolves over time . this leads to the following natural and interesting question : how does the lsd of rcv matrix depend on the time - variability of the covolatility process ? answering this question in a general context without putting any structural assumption on the covolatility process seems to be rather challenging , if not impossible . for a class @xmath69 ( see section [ sectheory ] ) of processes , we do establish a result for rcv matrices that s analogous to the marenko pastur theorem ( see proposition [ propmprcv ] ) , which demonstrates clearly how the time - variability of the covolatility process affects the lsd of rcv matrix . proposition [ propmprcv ] is proved based on theorem [ thmmpweightedcov ] , which is a marenko pastur type theorem for _ weighted _ sample covariance matrices . these results , in principle , allow one to recover the lsd of icv matrix based on that of rcv matrix . estimating high dimensional icv matrices based on high frequency data has only recently started to gain attention . see , for example , @xcite ; @xcite who made use of data over long time horizons by proposing a method incorporating low - frequency dynamics ; and @xcite who studied the estimation of icv matrices for portfolio allocation under gross exposure constraint . in @xcite , under sparsity assumptions on the icv matrix , banding / thresholding was innovatively used to construct consistent estimators of the icv matrix in the spectral norm sense . in particular , when the sparsity assumptions are satisfied , their estimators share the same lsd as the icv matrix . it remains an open question that when the sparsity assumptions are not satisfied , whether one can still make good inference about the spectrum of icv matrix . for processes in class @xmath69 ( see section [ sectheory ] ) , whose icv matrices do not need to be sparse , we propose a new estimator , the _ time - variation adjusted realized covariance _ ( tvarcv ) matrix . we show that the tvarcv matrix has the desirable property that its lsd exists provided that the lsd of icv matrix exists , and furthermore , the two lsds are related to each other via the marenko pastur equation ( [ eqstfh0 ] ) ( see theorem [ thmconvedf ] ) . therefore , the tvarcv matrix can be used , for example , to recover the lsd of icv matrix by inverting the marenko pastur equation using existing algorithms . the rest of the paper is organized as the following : theoretical results are presented in section [ sectheory ] , proofs are given in section [ secproofs ] , simulation studies in section [ secsimulation ] , and conclusion and discussions in section [ secconclusion ] . _ notation . _ for any matrix @xmath70 , @xmath71 denotes its spectral norm . for any hermitian matrix @xmath70 , @xmath72 stands for its esd . for two matrices @xmath70 and @xmath73 , we write @xmath74 ( @xmath75 , resp . ) if @xmath76 ( @xmath77 , resp . ) is a nonnegative definite matrix . for any interval @xmath78 , and any metric space @xmath79 , @xmath80 stands for the space of cdlg functions from @xmath81 to @xmath79 . additionally , @xmath82 stands for the imaginary unit , and for any @xmath83 , we write @xmath84 as its real part and imaginary part , respectively , and @xmath85 as its complex conjugate . we also denote @xmath86 , @xmath87 and @xmath88 . we follow the custom of writing @xmath89 to mean that the ratio @xmath90 converges to 1 . finally , throughout the paper , @xmath91 etc . denote generic constants whose values may change from line to line . proposition [ propmpgen ] asserts that the esd of sample covariance matrix converges to a limiting distribution which is uniquely determined by the lsd of the underlying covariance matrix . unfortunately , proposition [ propmpgen ] does not apply to our case , since the observations @xmath92 under our general diffusion process setting are not i.i.d . proposition [ propnotconv ] below shows that even in the following most ideal case , the rcv matrix does not have the desired convergence property . [ propnotconv ] suppose that for all @xmath0 , @xmath93 is a @xmath0-dimensional process satisfying @xmath94,\ ] ] where @xmath95 is a nonrandom ( scalar ) cdlg process . let @xmath96 , and so that the icv matrix @xmath18 is @xmath97 . assume further that the observation times @xmath15 are equally spaced , that is , @xmath24 , and that the rcv matrix @xmath21 is defined by ( [ eqrcv ] ) . then so long as @xmath98 is not constant on @xmath99 , for any @xmath100 , there exists @xmath101 such that if @xmath102 , @xmath103 in particular , @xmath27 does not converge to the marenko pastur law mp@xmath56 . observe that mp@xmath56 is the lsd of rcv matrix when @xmath104 . the main message of proposition [ propnotconv ] is that , the lsd of rcv matrix depends on the whole covolatility process _ not only through @xmath18 _ , _ but also on how the covolatility process varies in time_. it will also be clear from the proof of proposition [ propnotconv ] ( section [ seccountereg ] ) that , the more `` volatile '' the covolatility process is , the further away the lsd is from the marenko pastur law mp@xmath56 . this is also illustrated in the simulation study in section [ secsimulation ] . to understand the behavior of the esd of rcv matrix more clearly , we next focus on a special class of diffusion processes for which we ( i ) establish a marenko pastur type theorem for rcv matrices ; and ( ii ) propose an alternative estimator of icv matrix . suppose that @xmath7 is a @xmath0-dimensional process satisfying ( [ eqx ] ) , and @xmath105 is cdlg . we say that @xmath7 belongs to class @xmath69 if , almost surely , there exist @xmath106;{\mathbb{r}})$ ] and @xmath107 a @xmath36 matrix satisfying @xmath108 such that @xmath109 observe that if ( [ eqclassc ] ) holds , then the icv matrix @xmath110 . we note that @xmath107 does not need to be sparse , hence neither does @xmath18 . a special case is when @xmath111 . this type of process is studied in proposition [ propnotconv ] and in the simulation studies in section [ secsimulation ] . a more interesting case is the following . [ propsubclassc ] suppose that @xmath112 satisfy @xmath113 where @xmath114;{\mathbb{r}})$ ] are the drift and volatility processes for stock @xmath115 , and @xmath116 s are ( one - dimensional ) standard brownian motions . if the following conditions hold : the _ correlation matrix _ process of @xmath117 @xmath118 is constant in @xmath119 $ ] ; [ eqrpos ] @xmath120 for all @xmath121 ; and the correlation matrix process of @xmath122 @xmath123 is constant in @xmath119 $ ] ; then @xmath122 belongs to class @xmath69 . the proof is given in the supplementary article [ @xcite ] . equation ( [ eqx2 ] ) is another common way of representing multi - dimensional log - price processes . we note that if @xmath112 are log price processes , then over short time period , say , one day , it is reasonable to assume that the correlation structure of @xmath122 does not change , hence by this proposition , @xmath122 belongs to class @xmath69 . observe that if a diffusion process @xmath7 belongs to class @xmath69 , the drift process , and @xmath15 s and @xmath68 are independent of @xmath12 , then @xmath124 where `` @xmath125 '' stands for `` equal in distribution , '' @xmath126 is the nonnegative square root matrix of @xmath127 , and @xmath128 consists of independent standard normals . therefore the rcv matrix @xmath129 where @xmath130 . this is similar to the @xmath43 in proposition [ propmpgen ] , except that here the `` weights '' @xmath131 may vary in @xmath132 , while in proposition [ propmpgen ] the `` weights '' are constantly @xmath133 . motivated by this observation we develop the following marenko pastur type theorems for weighted sample covariance matrices and rcv matrices . [ thmmpweightedcov ] suppose that assumptions and in proposition [ propmpgen ] hold . assume further that : for @xmath29 and @xmath30 , @xmath31 with @xmath32 i.i.d . with mean 0 , variance 1 and finite moments of all orders ; @xmath18 is a ( possibly random ) nonnegative definite @xmath36 matrix such that its esd @xmath23 converges almost surely in distribution to a probability distribution @xmath38 on @xmath39 as @xmath35 ; moreover , @xmath38 has a finite second moment ; [ asmconvw ] the weights @xmath134 are all positive , and there exists @xmath135 such that the rescaled weights @xmath136 satisfy @xmath137 moreover , almost surely , there exists a process @xmath138;{\mathbb{r}}_+)$ ] such that @xmath139 there exists a sequence @xmath140 and a sequence of index sets @xmath141 satisfying @xmath142 and @xmath143 such that for all @xmath1 and all @xmath132 , @xmath131 may depend on @xmath144 but only on @xmath145 ; [ asmsigmanormrcv ] there exist @xmath146 and @xmath147 such that for all @xmath0 , @xmath148 almost surely . define @xmath149 . then , almost surely , the esd of @xmath43 converges in distribution to a probability distribution @xmath150 , which is determined by @xmath38 and @xmath151 in that its stieltjes transform @xmath152 is given by @xmath153 where @xmath154 , together with another function @xmath155 , uniquely solve the following equation in @xmath156 : @xmath157 assumption ( a.i@xmath158 ) can undoubtedly be weakened , for example , by using the truncation and centralization technique as in @xcite and @xcite ; or , a closer look at the proof of theorem [ thmmpweightedcov ] indicates that as long as @xmath32 has finite moments up to order @xmath159 , the theorem is true and can be proved by exactly the same argument . [ rmkthm2 ] if @xmath160 , then @xmath161 , and theorem [ thmmpweightedcov ] reduces to proposition [ propmpgen ] . moreover , if @xmath162 is not constant , that is , @xmath163 on @xmath99 , then except in the trivial case when @xmath38 is a delta measure at @xmath164 , the lsd @xmath165 , where @xmath44 is the lsd in proposition [ propmpgen ] determined by @xmath166 . see the supplementary article [ @xcite ] for more details . theorem [ thmmpweightedcov ] is proved in section [ secpfmpw ] . a direct consequence of this theorem and lemma [ lemmadriftnegligible ] below is the following marenko pastur type result for rcv matrices for diffusion processes in class @xmath2 . we note that , thanks to lemma [ lemmadriftnegligible ] below ( see the remark after the proof of lemma [ lemmadriftnegligible ] for more explanations ) , regarding the drift process , except requiring them to be uniformly bounded , we put no additional assumption on them : they can be , for example , stochastic , cdlg and dependant with each other . furthermore , we allow for dependence between the covolatility process and the underlying brownian motion in other words , we allow for the leverage effect . in the special case when @xmath167 does not change in @xmath0 , is nonrandom and bounded , and the observation times are equally spaced , the ( rather technical ) assumptions ( b.iii ) and ( b.iv ) below are trivially satisfied.=1 [ propmprcv ] suppose that for all @xmath0 , @xmath168 is a @xmath0-dimensional process in class @xmath69 for some drift process @xmath169 , covolatility process @xmath170 and @xmath0-dimensional brownian motion @xmath171 . suppose further that : there exists @xmath172 such that for all @xmath0 and all @xmath173 , @xmath174 for all @xmath175 almost surely ; @xmath176 satisfies assumption and in theorem [ thmmpweightedcov ] ; there exists a sequence @xmath140 and a sequence of index sets @xmath141 satisfying @xmath142 and @xmath143 such that @xmath177 may depend on @xmath178 but only on @xmath179 ; moreover , there exists @xmath180 such that for all @xmath0 , @xmath181 for all @xmath175 almost surely ; additionally , almost surely , there exists @xmath182;{\mathbb{r}})$ ] such that @xmath183 the observation times @xmath15 are independent of @xmath7 ; moreover , there exists @xmath135 such that the observation durations @xmath184 satisfy @xmath185 additionally , almost surely , there exists a process @xmath186 such that @xmath187 } \rightarrow \upsilon_s:=\int_0^s \upsilon_r\ , dr \qquad\mbox{as } n\rightarrow\infty\mbox { for all } 0\leq s\leq1,\ ] ] where for any @xmath188 , @xmath189 $ ] stands for its integer part . then , as @xmath35 , @xmath190 converges almost surely to a probability distribution @xmath150 as specified in theorem [ thmmpweightedcov ] for @xmath191 . proposition [ propmprcv ] demonstrates explicitly how the lsd of rcv matrix depends on the time - variability of the covolatility process . hence , the rcv matrix by itself can not be used to make robust inference for the esd @xmath23 of the icv matrix . if @xmath192 [ and hence @xmath193 is known , then in principle , the equations ( [ eqstfhweighted ] ) and ( [ eqmwtm ] ) can be used to recover @xmath23 . however , in general , @xmath192 is unknown and estimating the process @xmath192 can be challenging and will bring in more complication in the inference . moreover , the equations ( [ eqstfhweighted ] ) and ( [ eqmwtm ] ) are different from and more complicated than the classical marenko pastur equation ( [ eqstfh0 ] ) , and in order to recover @xmath23 based on these equations , one has to extend existing algorithms [ @xcite , @xcite and @xcite etc . ] which are designed for ( [ eqstfh0 ] ) . developing such an algorithm is of course of great interest , but we shall not pursue this in the present article . we shall instead propose an alternative estimator which overcomes these difficulties . @xmath194suppose that a diffusion process @xmath7 belongs to class @xmath69 . we define the _ time - variation adjusted realized covariance _ ( tvarcv ) matrix as follows : @xmath195 where for any vector @xmath196 , @xmath197 stands for its euclidean norm , and @xmath198 let us first explain @xmath199 . consider the simplest case when @xmath200 , @xmath68 deterministic , @xmath201 , and @xmath202 . in this case , @xmath203 where @xmath204 and @xmath205 s are i.i.d . standard normal . hence , @xmath206 . however , as @xmath35 , @xmath207 , hence @xmath208 , the latter being the usual sample covariance matrix . we will show that , first , @xmath209 ; and second , if @xmath7 belongs to class @xmath69 and satisfies certain additional assumptions , then the lsd of @xmath199 is related to that of @xmath210 via the marenko pastur equation ( [ eqstfh0 ] ) , where @xmath211 hence , the lsd of @xmath212 is also related to that of @xmath18 via the same marenko pastur equation . we now state our assumptions . observe that about the drift process , again , except requiring them to be uniformly bounded , we put no additional assumption . furthermore , we allow for the dependence between the covolatility process and the underlying brownian motion , namely , the leverage effect . _ assumptions _ : [ asmdrift ] there exists @xmath172 such that for all @xmath0 and all @xmath173 , @xmath174 for all @xmath175 almost surely ; [ asmgammadep ] there exist constants @xmath213 , a sequence @xmath214 and a sequence of index sets @xmath141 satisfying @xmath142 and @xmath143 such that @xmath167 may depend on @xmath178 but only on @xmath179 ; moreover , there exists @xmath215 such that for all @xmath0 , @xmath216 for all @xmath175 almost surely ; [ asmsigmabd ] there exists @xmath217 such that for all @xmath0 and for all @xmath115 , the individual volatilities @xmath218 for all @xmath219 $ ] almost surely ; [ asmtrsigmagrow ] @xmath220 @xmath221 almost surely ; [ asmsigmalsd ] almost surely , as @xmath35 , the esd @xmath222 converges to a probability distribution @xmath38 on @xmath39 ; [ asmsigmanorm ] there exist @xmath223 and @xmath224 such that for all @xmath0 , @xmath225 almost surely ; [ asm2delta ] the @xmath226 in [ asmgammadep ] and @xmath227 in [ asmsigmanorm ] satisfy that @xmath228 [ asmpn ] @xmath229 as @xmath35 ; and [ asmobstime ] there exists @xmath230 such that for all @xmath1 , @xmath231 moreover , @xmath15 s are independent of @xmath7 . we have the following convergence theorem regarding the esd of our proposed estimator tvarcv matrix @xmath232 . [ thmconvedf ] suppose that for all @xmath0 , @xmath233 is a @xmath0-dimensional process in class @xmath69 for some drift process @xmath234 , covolatility process @xmath235 and @xmath0-dimensional brownian motion @xmath178 , which satisfy assumptions [ asmdrift]@xmath236[asm2delta ] above . suppose also that @xmath0 and @xmath1 satisfy [ asmpn ] , and the observation times satisfy [ asmobstime ] . let @xmath232 be as in ( [ eqtvarcv ] ) . then , as @xmath35 , @xmath237 converges almost surely to a probability distribution @xmath44 , which is determined by @xmath38 through stieltjes transforms via the same marenko pastur equation ( [ eqstfh0 ] ) as in proposition [ propmpgen ] . the proof of theorem [ thmconvedf ] is given in section [ secpfthm ] . the lsd @xmath38 of the targeting icv matrix is in general not the same as the lsd @xmath44 , but can be recovered from @xmath44 based on equation ( [ eqstfh0 ] ) . in practice , when one has only finite number of samples , the articles [ @xcite , @xcite and @xcite etc . ] studied the estimation of the population spectral distribution based on the sample covariance matrices . in particular , applying theorem 2 of @xcite to our case yields . [ coresdrec ] let @xmath238 , and define @xmath239 as in theorem 2 of el karoui ( @xcite ) . if @xmath240 are bounded in @xmath0 , then , as @xmath35 , @xmath241 almost surely . therefore , when the dimension @xmath0 is large , based on the esd of tvarcv matrix @xmath232 , we can estimate the spectrum of underlying icv matrix @xmath18 well . we collect some either elementary or well - known facts in the following . the proofs are given in the supplemental article [ @xcite ] . [ lemmadriftnegligible ] suppose that for each @xmath0 , @xmath242 and @xmath243 , are all @xmath0-dimensional vectors . define @xmath244 if the following conditions are satisfied : @xmath33 with @xmath245 ; there exists a sequence @xmath246 such that for all @xmath0 and all @xmath132 , all the entries of @xmath247 are bounded by @xmath248 in absolute value ; @xmath249 almost surely . then @xmath250 almost surely , where for any two probability distribution functions @xmath44 and @xmath251 , @xmath252 denotes the levy distance between them . [ lemmatrdiff ] let @xmath83 with @xmath253 , @xmath70 and @xmath73 be @xmath36 with @xmath73 hermitian , and @xmath254 . then @xmath255 the following two lemmas are similar to lemma 2.3 in @xcite . [ lemmanormsum ] let @xmath256 with @xmath257 , and @xmath70 be an hermitian nonnegative definite matrix . then @xmath258 . [ lemmadiffest ] let @xmath259 with @xmath260 and @xmath261 , @xmath70 be a @xmath36 hermitian nonnegative definite matrix , @xmath73 any @xmath36 matrix , and @xmath254 . then : @xmath262 @xmath263 [ lemmanormdiff ] for any hermitian matrix @xmath70 and @xmath83 with @xmath264 , @xmath265 . both lemmas [ lemmanormsum ] and [ lemmadiffest ] require the real part of @xmath266 ( or @xmath267 , @xmath268 ) to be nonnegative . in our proof of theorem [ thmmpweightedcov ] , the requirements will be fulfilled thanks to the following lemma . [ lemmarealpos ] let @xmath269 with @xmath270 , @xmath70 be a @xmath36 hermitian nonnegative definite matrix , @xmath254 , @xmath271 . then @xmath272 [ lemmapostrace ] let @xmath269 with @xmath270 , @xmath70 be any @xmath36 matrix , and @xmath73 be a @xmath36 hermitian nonnegative definite matrix . then . [ lemmaunitm ] suppose that @xmath273 . then for any @xmath274 , the equation @xmath275 admits at most one solution in @xmath276 . the following result is an immediate consequence of lemma 2.7 of @xcite . [ lemmaconcstdnormal ] for @xmath277 where @xmath278 s are i.i.d . random variables such that @xmath279 , and @xmath280 for some @xmath281 , there exists @xmath282 , depending only on @xmath283 , @xmath284 and @xmath285 , such that for any @xmath36 nonrandom matrix @xmath70 , @xmath286 [ propstieltjesconv ] supposethat @xmath287 are real probability measures with stieltjes transforms @xmath288 . let @xmath289 be an infinite set with a limit point in @xmath290 . if @xmath291 exists for all @xmath292 , then there exists a probability measure @xmath293 with stieljes transform @xmath294 if and only if @xmath295 in which case @xmath296 in distribution . by assumption , @xmath68 is positive and non - constant on @xmath99 , and is cdlg , in particular , right - continuous ; moreover , . hence , there exists @xmath297 and @xmath298\subseteq[0,1]$ ] such that @xmath299.\ ] ] therefore , if @xmath300\subseteq[c , d]$ ] , @xmath301 where @xmath128 consists of independent standard normals . hence , if we let @xmath302\subseteq[c , d]\}$ ] and @xmath303 then for any @xmath304 , by weyl s monotonicity theorem [ see , e.g. , corollary 4.3.3 in @xcite ] , @xmath305\bigr).\ ] ] now note that @xmath306 , hence if @xmath307 , by proposition [ propmp ] , @xmath308 will converge almost surely to the marenko pastur law with ratio index @xmath309 and scale index @xmath55 , which has density on @xmath310 $ ] with functions @xmath311 and @xmath312 defined by ( [ eqmpsupp ] ) . by the formula of @xmath312 , @xmath313 hence , for any @xmath100 , there exists @xmath314 such that for all @xmath315 , @xmath316 that is , @xmath317 by ( [ eqspecmono ] ) , when the above inequality holds , @xmath318 to prove theorem [ thmmpweightedcov ] , following the strategies in @xcite , @xcite , @xcite , we will work with stieltjes transforms . proof of theorem [ thmmpweightedcov ] for notational ease , we shall sometimes omit the sub / superscripts @xmath0 and @xmath1 in the arguments below : thus , we write @xmath319 instead of @xmath40 , @xmath320 instead of @xmath131 , @xmath321 instead of @xmath18 , @xmath79 instead of @xmath43 , etc . also recall that @xmath322 , which converges to @xmath323 . by assumption ( a.vi ) we may , without loss of generality , assume that the weights @xmath320 are independent of @xmath319 s . this is because , if we let @xmath324 be the result of replacing @xmath325 with independent random variables with the same distribution that are also independent of @xmath320 , and @xmath326 , then @xmath327 , and so by the rank inequality @xmath328 [ see , e.g. , lemma 2.2 in @xcite ] , @xmath329 and @xmath79 must have the same lsd . we proceed according to whether @xmath38 is a delta measure at @xmath164 or not . if @xmath38 is a delta measure at @xmath164 , we claim that @xmath150 is also a delta measure at @xmath164 , and the conclusion of the theorem holds . the reason is as follows . by assumption [ asmconvw ] , @xmath330 hence by weyl s monotonicity theorem again , for any @xmath304 @xmath331 however , it follows easily from proposition [ propmpgen ] that @xmath332 converges to the delta measure at @xmath164 , hence so does @xmath333 . below we assume that @xmath38 is _ not _ a delta measure at @xmath164 . let @xmath334 be the @xmath36 identity matrix , and @xmath335 be the stieltjes transform of @xmath333 . by proposition [ propstieltjesconv ] , in order to show that @xmath333 converges , it suffices to prove that for all @xmath336 with @xmath270 sufficiently large , @xmath337 exists , and that @xmath294 satisfies condition ( [ eqconditionst ] ) . we first show the convergence of @xmath288 for @xmath336 with @xmath270 sufficiently large . since for all @xmath1 , @xmath338 , it suffices to show that @xmath339 has at most one limit . for notational ease , we denote by @xmath340 . we first show that @xmath341 where @xmath342 . in fact , by lemma [ lemmaconcstdnormal ] and assumptions ( a.i@xmath158)and [ asmsigmanormrcv ] , for any @xmath343 , @xmath344 using markov s inequality we get that for any @xmath100 , @xmath345 hence , choosing @xmath346 , using borel cantelli and that @xmath347 yield @xmath348 the convergence ( [ eqrinorm ] ) follows . next , let @xmath349 where @xmath350 note that by lemma [ lemmarealpos ] , for any @xmath132 , @xmath351 we shall show that @xmath352 observe the following identity : for any @xmath36 matrix @xmath73 , @xmath353 and @xmath354 for which @xmath73 and @xmath355 are both invertible , @xmath356 see equation ( 2.2 ) in @xcite . writing @xmath357 taking the inverse , using ( [ eqmultidecom ] ) and the definition ( [ eqmn ] ) of @xmath358 yield @xmath359 taking trace and dividing by @xmath0 we get @xmath360 where @xmath361 by ( 5.2 ) in the proof of lemma [ lemmarealpos ] in the supplementary article [ @xcite ] , @xmath362 . hence , @xmath363 therefore in order to show ( [ eq2sttrsame ] ) , by assumption [ asmconvw ] , it suffices to prove @xmath364 define @xmath365 where @xmath366 . observe that for every @xmath132 , @xmath367 is independent of @xmath319 . for any @xmath336 with @xmath270 and any @xmath368 , @xmath369 define @xmath370 which belongs to @xmath276 by lemma [ lemmapostrace ] , and @xmath371 then by a similar argument for ( [ eqbddcoef ] ) and using assumption [ asmconvw ] , @xmath372 hence , it suffices to show that @xmath373\\[-8pt ] \max_{\ell=1,\ldots , n } p^{\varepsilon}\bigl|m_{(\ell)}(z ) - { \widetilde}{m}_{(\ell ) } ( z)\bigr| & \rightarrow & 0\qquad \mbox{almost surely.}\nonumber\end{aligned}\ ] ] we shall only prove the second convergence . in fact , @xmath374 where @xmath375 since for all @xmath115 , @xmath376 it suffices to show that @xmath377 to prove this , recall that @xmath378 , by lemma [ lemmaconcstdnormal ] and the independence between @xmath319 and @xmath379 , for any @xmath343 , @xmath380 where in the last line we used lemma [ lemmanormdiff ] and assumption [ asmsigmanormrcv ] . hence , for any @xmath381 , choosing @xmath382 and using borel cantelli again , we get @xmath383\\[-8pt ] & & \hspace*{44pt}\qquad { } - \frac{{\operatorname{tr } } ( { \sigma}^{1/2 } ( s_{(j,\ell ) } - zi)^{-1}{\sigma}^{1/2})}{p}\biggr| \rightarrow0.\nonumber\end{aligned}\ ] ] furthermore , by lemma [ lemmatrdiff ] and assumption [ asmsigmanormrcv ] , recall that @xmath384 , @xmath385\\[-8pt ] & & \qquad\leq 2 \frac { \|{\sigma}\| } { p v } \leq\frac{c p^\delta}{pv}. \nonumber\end{aligned}\ ] ] the convergence ( [ eqzeta ] ) follows . we now continue the proof of the theorem . recall that @xmath386 , and @xmath319 consists of i.i.d . random variables with finite moments of all orders . by lemma [ lemmadiffest](ii ) and ( [ eqwipos ] ) , @xmath387 where in the last line we used lemma [ lemmanormdiff ] , assumption [ asmsigmanormrcv ] , the assumption that @xmath147 ( and hence @xmath388 ) and ( [ eqdiffm ] ) , and ( [ eqrinorm ] ) . furthermore , similar to ( [ eqdifftracenorm0 ] ) , by lemma [ lemmaconcstdnormal ] and the independence between @xmath319 and @xmath389 , for any @xmath343 , @xmath390 where in the last line we use lemmas [ lemmanormdiff ] , [ lemmanormsum ] and ( [ eqwipos ] ) , and assumption [ asmsigmanormrcv ] . hence , choosing @xmath346 and using borel cantelli again , we get @xmath391\\[-8pt ] & & \hspace*{23.2pt}\quad { } - \frac{{\operatorname{tr } } ( { \sigma}^{1/2 } ( s_{(\ell ) } - zi)^{-1}(m_{(\ell ) } { \sigma } + i)^{-1}{\sigma}^{1/2})}{p}\biggr| \rightarrow 0 . \nonumber\end{aligned}\ ] ] furthermore , by lemmas [ lemmadiffest](i ) , [ lemmanormsum ] and ( [ eqwipos ] ) , the assumption that @xmath147 ( and hence @xmath392 ) and ( [ eqdiffm ] ) , and assumption [ asmsigmanormrcv ] , @xmath393\\[-8pt ] & & \qquad\leq \max_{\ell=1,\ldots , n } \bigl|m_{(\ell ) } - m_n\bigr|\cdot\bigl\|\bigl(s_{(\ell ) } - zi\bigr)^{-1 } \bigr\|\cdot\|{\sigma}\|^2\nonumber\\ & & \qquad\leq \max_{\ell=1,\ldots , n } \bigl|m_{(\ell ) } - m_n\bigr|\cdot\frac{c p^{2\delta}}{v } \rightarrow0 . \nonumber\end{aligned}\ ] ] finally , similar to ( [ eqdifftrace0 ] ) , by lemmas [ lemmatrdiff ] and [ lemmanormsum ] , and assumption [ asmsigmanormrcv ] , @xmath394 combining ( [ eqdiffnorm ] ) , ( [ eqconcnum ] ) , ( [ eqdifftrace ] ) and ( [ eqdifftrace2 ] ) , we see that ( [ eqdito0 ] ) , and hence ( [ eq2sttrsame ] ) holds . now we are ready to show that @xmath339 admits at most one limit . [ claimconvm ] suppose that @xmath395 converges to @xmath294 , then @xmath396 where @xmath155 is the unique solution in @xmath397 to the following equation : @xmath398 writing @xmath399 right - multiplying both sides by @xmath400 and using ( [ eqmultidecom ] ) we get @xmath401 taking trace and dividing by @xmath1 we get @xmath402 where , recall that , @xmath403 is the stieltjes transform of @xmath333 . hence , if @xmath404 , then @xmath405 however , by the same arguments for ( [ eqconcnum0 ] ) and ( [ eqdifftrace0 ] ) we have @xmath406 and @xmath407 where , recall that @xmath408 which belongs to @xmath276 by lemma [ lemmapostrace ] . then by ( [ eqconvtm ] ) , assumption [ asmconvw ] and lemma [ lemmaunitm ] , @xmath409 must also converge , and the limit , denoted by @xmath410 , must be the unique solution in @xmath276 to the equation ( [ eqntm ] ) . now by ( [ dfnmtilde ] ) , ( [ eqmmtilde ] ) and assumption [ asmconvw ] , we get the convergence for @xmath411 in the claim . that @xmath412 follows from the expression and that @xmath413 . we now continue the proof of the theorem . by the convergence of @xmath414 to @xmath415 and the previous claim , @xmath416 but ( [ eq2sttrsame ] ) implies that @xmath417 observing that @xmath412 , @xmath418 , and @xmath38 is not a delta measure at @xmath164 , we obtain that @xmath419 . hence @xmath420 , and by ( [ eqntm ] ) , @xmath421 . based on this , we can get another expression for @xmath154 , as follows . by ( [ eqnm ] ) , we have @xmath422 & = & -\frac{1}{z}\cdot\frac{1}{y { \widetilde}{m}(z)}\cdot\biggl(1- \int_0 ^ 1 \frac{1}{1+y { \widetilde}{m}(z ) w_s } \,ds\biggr)\nonumber\\[-9.5pt]\\[-9.5pt ] & = & -\frac{1}{z}\cdot\frac{1}{y { \widetilde}{m}(z)}\cdot\bigl(1- \bigl(1-y\bigl ( 1 + z m(z ) \bigr)\bigr)\bigr)\nonumber\\[-2pt ] & = & -\frac{1}{z}\cdot\frac{1 + zm(z)}{{\widetilde}{m}(z ) } , \nonumber\vspace*{-1pt}\end{aligned}\ ] ] where in the third line we used the definition ( [ eqntm ] ) of @xmath155 . we can then derive another formula for @xmath155 . by ( [ eqnm ] ) , @xmath423 by using that @xmath38 is a probability distribution . dividing both sides by@xmath424 and using ( [ eqnmf2 ] ) yield @xmath425 and hence since @xmath412 , @xmath426 observe that by lemma [ lemmapostrace ] and ( [ eqwipos ] ) , for any @xmath1 , both @xmath288 and @xmath427 belong to @xmath276 , hence so do @xmath294 and @xmath154 . we proceed to show that for those @xmath336 with @xmath428 sufficiently large , there is at most one triple @xmath429 that solves the equations ( [ eqnm ] ) , ( [ eqnm ] ) and ( [ eqntmf2 ] ) . in fact , if there are two different triples @xmath430 , both satisfying ( [ eqnm ] ) , ( [ eqnm ] ) and ( [ eqntmf2 ] ) . then necessarily , @xmath431 and @xmath432 . now by ( [ eqnm ] ) , @xmath433 by ( [ eqntmf2 ] ) , @xmath434 therefore , @xmath435\\[-9.5pt ] & & { } \times \int_{\tau\in{\mathbb{r}}}\frac{\tau^2}{(\tau m_1(z ) + 1)(\tau m_2(z ) + 1 ) } \,d{h}(\tau ) . \nonumber\vspace*{-1pt}\vadjust{\goodbreak}\end{aligned}\ ] ] however , since @xmath436 , @xmath437 , @xmath438 and @xmath439 hence , for @xmath336 with @xmath428 sufficiently large , ( [ eq1 ] ) can not be true . it remains to verify ( [ eqconditionst ] ) , that is , @xmath440 in fact , using ( [ eqnm ] ) we get that @xmath441 since @xmath442 , @xmath443 for all @xmath444 . moreover , by ( [ eqnm ] ) and that @xmath445 , @xmath446 , hence by the dominated convergence theorem , the right - hand side of ( [ eqmcond ] ) converges to @xmath447 as @xmath448 . the tvarcv matrix has the form of weighted sample covariance matrices as studied in theorem [ thmmpweightedcov ] ; however , assumption [ asmwdep ] therein is not satisfied , and we need another proof . theorem [ thmconvedf ] is a direct consequence of the following two convergence results . [ proprvldp ] under assumption [ asmtrsigmagrow ] , namely , suppose that @xmath449 then , almost surely , @xmath450 the proof is given in the supplemental article [ @xcite ] . next , recall that @xmath210 and @xmath199 are defined by ( [ eqsigmabr ] ) and ( [ eqsigmawt ] ) , respectively . [ propconvedf ] @xmath194under the assumptions of theorem [ thmconvedf ] , both @xmath451 and @xmath452 converge almost surely . @xmath451 converges to @xmath453 defined by @xmath454 the lsd @xmath455 of @xmath199 is determined by @xmath453 in that its stieltjes transform @xmath456 satisfies the equation @xmath457 this can be proved in very much the same way as theorem [ thmmpweightedcov ] , by working with stieltjes transforms . however , a much simpler and transparent proof is as follows . proof of proposition [ propconvedf ] the convergence of @xmath451 is obvious since @xmath458 we now show the convergence of @xmath452 . as in the proof of theorem [ thmmpweightedcov ] , for notational ease , we shall sometimes omit the superscript @xmath0 in the arguments below : thus , we write @xmath459 instead of @xmath460 , @xmath68 instead of @xmath177 , @xmath107 instead of @xmath461 , etc . first , note that @xmath462 where @xmath463 by performing an orthogonal transformation if necessary , without loss of generality , we may assume that the index set @xmath464 . then by assumptions [ asmgammadep ] and [ asmobstime ] , for @xmath465 , @xmath205 are i.i.d . @xmath466 . write @xmath467 and @xmath468 . with the above notation , @xmath199 can be rewritten as @xmath469 by assumptions [ asmdrift ] , [ asmgammadep ] and [ asmobstime ] , there exists @xmath470 such that @xmath471 for all @xmath115 and @xmath132 , hence @xmath472 s are uniformly bounded . we will show that @xmath473\\[-8pt ] & & \qquad= \max_{\ell=1,\ldots , n } |\mathbf{z}_\ell^t\breve{\sigma}_p \mathbf{z}_\ell / p - 1 |\rightarrow0 \qquad\mbox{almost surely},\nonumber\end{aligned}\ ] ] which clearly implies that @xmath474 to prove ( [ eqldpendo ] ) , write @xmath475 where @xmath476 and @xmath477 are @xmath478 and @xmath479 matrices , respectively . then @xmath480 by a well - known fact about the spectral norm , @xmath481 in particular , by assumptions [ asmgammadep ] , [ asmsigmanorm ] and [ asm2delta ] , @xmath482 hence @xmath483 . now using the fact that @xmath484 consists of i.i.d . standard normals and by the same proof as that for ( [ eqrinorm0 ] ) we get @xmath485 to complete the proof of ( [ eqldpendo ] ) , it then suffices to show that @xmath486 we shall only prove the first convergence ; the second one can be proved similarly . we have @xmath487 observe that for all @xmath488 , by assumption [ asmgammadep ] , @xmath489 by the burkholder davis gundy inequality , we then get that for any @xmath490 , there exists @xmath491 such that @xmath492 now we are ready to show that @xmath493 . in fact , for any @xmath100 , for any @xmath490 , by markov s inequality , ( [ equlnorm ] ) , hlder s inequality and ( [ eqrtnmoments ] ) , @xmath494}{p^{k } { \varepsilon}^{k}}\\ & \leq & c p^{1+k\delta_2 + k\delta_1 - k}.\end{aligned}\ ] ] by assumption [ asm2delta ] , @xmath495 , hence by choosing @xmath283 to be large enough , the right hand side will be summable in @xmath0 , hence by borel cantelli , almost surely , @xmath496 . we now get back to @xmath199 as in ( [ eqwtsigmaneat ] ) . by ( [ eqldpnorm ] ) , for any @xmath100 , almost surely , for all @xmath497 sufficiently large , for all @xmath498 , @xmath499 hence , almost surely , for all @xmath1 sufficiently large , @xmath500 where @xmath501 . hence , by weyl s monotonicity theorem , for any @xmath304 , @xmath502 next , by lemma [ lemmadriftnegligible ] , @xmath503 has the same lsd as @xmath504 . moreover , by using the same trick as in the beginning of the proof of theorem [ thmmpweightedcov ] , @xmath505 has the same limit as @xmath506 , where @xmath507 , and @xmath324 consists of i.i.d . standard normals . for @xmath506 , it follows easily from proposition [ propmpgen ] that it converges to @xmath455 . moreover , by theorems 1.1 and 2.1 in @xcite , @xmath455 is differentiable and in particular continuous at all @xmath508 . it follows from ( [ eqfsandwich ] ) that @xmath452 must also converge to @xmath455 . in this section , we present some simulation studies to illustrate the behavior of esds of rcv and tvarcv matrices . in particular , we show that the esds of rcv matrices that have the same targeting icv matrix @xmath18 can be quite different from each other , depending on the time variability of the covolatility process . our proposed estimator , the tvarcv matrix @xmath232 , in contrast , has a very stable esd . we use in particular a reference curve which is the marcenko pastur law . the reason we compare the esds of rcv and tvarcv matrices with the marcenko pastur law is that the marcenko pastur law is the lsd of @xmath509 defined in ( [ eqrcv0 ] ) , which is the rcv matrix estimated from sample paths of constant volatility that has the same targeting icv matrix as @xmath510 . as we will see soon in the following two subsections , when the covolatility process is time varying , the esd of rcv matrix can be very different from the marcenko pastur law , while the esd of tvarcv matrix always matches the marcenko pastur law very well . in the simulation below , we assume that @xmath511 , or in other words , @xmath7 satisfies ( [ eqxsimplest ] ) with @xmath68 a deterministic ( scalar ) process , and @xmath12 a @xmath0-dimensional standard brownian motion . the observation times are taken to be equidistant : @xmath512 . we present simulation results of two different designs : one when @xmath68 is piecewise constant , the other when @xmath68 is continuous ( and non - constant ) . in both cases , we compare the esds of the rcv and tvarcv matrices . results for different dimension @xmath0 and observation frequency @xmath1 are reported . in all the figures below , we use red solid lines to represent the lsds of @xmath513 given by the marcenko pastur law , black dashed line to represent the esds of rcv matrices , blue bold longdashed line to represent the esds of tvarcv matrices . we first consider the case when the volatility path follows piecewise constants . more specifically , we take @xmath514 to be @xmath515$,\cr \sqrt{0.0001 } , & \quad $ t\in[1/4,3/4)$.}\ ] ] in figure [ figp100n1000dzn3 ] , we compare the esds of rcv and tvarcv matrices for different pairs of @xmath0 and @xmath1 , with the lsd of @xmath513 given by the marcenko pastur law as reference . , @xmath516 ; right panel : @xmath517 , @xmath516 . ] ) as shown in the legend , which are such that the targeting icv matrix is the same ; the red solid curve is the marcenko pastur law ( the lsd of @xmath518 ) . @xmath0 and @xmath1 are both taken to be @xmath519 . left panel : rcv ; right panel : tvarcv . ] we see from figure [ figp100n1000dzn3 ] that : * the esds of rcv matrices are very different from the lsd given by the marcenko pastur law ( the lsd of @xmath518 ) ; * the esds of tvarcv matrices follow the lsd given by the marcenko pastur law very well , for both pairs of @xmath0 and @xmath1 , even when @xmath0 is small compared with @xmath1 . in fact , the dependence of the esd of rcv matrix on the time variability of covolatility process can be seen more clearly from figure [ figcomprv ] , where we consider the same design but different values for @xmath68 : @xmath520$,\cr b^{1/2}\times10^{-2 } , & \quad $ t\in[1/4,3/4)$,}\qquad \mbox{where } a + b = 8.\ ] ] we plot the esds of rcv and tvarcv matrices for the case when @xmath521 , in the left and right panel , respectively . the curves corresponding parameters @xmath522 are reported in the legend . note that since all pairs of @xmath522 have the same summation , in all cases the targeting icv matrices are the same . we see clearly from figure [ figcomprv ] that , the esds of rcv matrices can be very different from each other even though the rcv matrices are estimating the same icv matrix ; while for tvarcv matrices , the esds are almost identical . we illustrate in this subsection the case when the volatility processes have continuous sample paths . in particular , we assume that @xmath7 satisfies ( [ eqxsimplest ] ) with @xmath523.\ ] ] we see from figure [ figp100n1000cos ] similar phenomena as in design i about the esds of rcv and tvarcv matrices for different pairs of @xmath0 and @xmath1 . , @xmath516 ; right panel : @xmath517 , @xmath516 . ] we have shown theoretically and via simulation studies that : * the _ limiting spectral distribution _ ( lsd ) of rcv matrix depends not only on that of the icv matrix , but also on the time - variability of covolatility process ; * in particular , even with the same targeting icv matrix , the _ empirical spectral distribution _ ( esd ) of rcv matrix can vary a lot , depending on how the underlying covolatility process evolves over time ; * for a class @xmath69 of processes , our proposed estimator , the _ time - variation adjusted realized covariance _ ( tvarcv ) matrix , possesses the following desirable properties as an estimator of the icv matrix : as long as the targeting icv matrix is the same , the esds of tvarcv matrices estimated from processes with different covolatility paths will be close to each other , sharing a unique limit ; moreover , the lsd of tvarcv matrix is related to that of the targeting icv matrix through the same marcenko pastur equation as in the sample covariance matrix case . furthermore , we establish a marcenko pastur type theorem for weighted sample covariance matrices . for a class @xmath69 of processes , we also establish a marcenko pastur type theorem for rcv matrices , which explicitly demonstrates how the time - variability of the covolatility process affects the lsd of rcv matrix . in practice , for given @xmath0 and @xmath1 , based on the ( observable ) esd of tvarcv matrix , one can use existing algorithms to obtain an estimate of the esd of icv matrix , which can then be applied to further applications such as portfolio allocation , risk management , etc . we are very grateful to the editor , the associate editor and anonymous referees for their very valuable comments and .

we address the thermodynamics ( equilibrium density profiles , phase diagram , instability analysis ... ) and the collapse of a self - gravitating gas of brownian particles in dimensions , in both canonical and microcanonical ensembles . in the canonical ensemble , we derive the analytic form of the density scaling profile which decays as , with . in the microcanonical ensemble , we show that decays as , where is a non - trivial exponent . we derive exact expansions for and in the limit of large . finally , we solve the problem in , which displays rather rich and peculiar features . laboratoire de physique quantique ( umr 5626 du cnrs ) , universit paul sabatier , + 118 , route de narbonne , 31062 toulouse , france + e - mail : _clement.sire@irsamc.ups-tlse.fr & chavanis@irsamc.ups-tlse.fr_

cond-mat0204303

in an earlier paper @xcite , we studied a model of self - gravitating brownian particles confined within a three - dimensional spherical box . we considered a high friction limit in which the equations of the problem reduce to a smoluchowski - poisson system with appropriate constraints ensuring the conservation of energy ( in the microcanonical ensemble ) or temperature ( in the canonical ensemble ) @xcite . the equilibrium states ( maximum entropy states ) correspond to isothermal configurations which are known to exist only above a critical energy or a critical temperature ( see , e.g. , @xcite ) . when no hydrostatic equilibrium exists , we found that the system generates a finite time singularity ( i.e. , the central density becomes infinite in a finite time ) and we derived self - similar solutions describing the collapse . this study was performed both in the microcanonical and canonical ensembles , with emphasize on the inequivalence of ensembles for such a nonextensive system . in the canonical ensemble , we showed that the scaling exponent for the density is @xmath2 and we determined the invariant profile @xmath8 , satisfying @xmath9 for @xmath10 , analytically . in the microcanonical ensemble , the scaling exponent @xmath11 and the corresponding invariant profile @xmath8 were determined numerically . in this paper , we propose to extend our analysis to a space of arbitrary dimension @xmath0 . the interest of this extension is twofold . first , we shall consider an infinite dimension limit @xmath12 in which the problem can be solved analytically . in particular , it is possible to determine the scaling exponent @xmath13 and the profile @xmath14 in the microcanonical ensemble by a systematic expansion procedure in powers of @xmath15 ( sec . [ sec_micro ] ) , while the canonical value is always @xmath2 and the profile can be calculated exactly for any dimension ( sec . [ sec_cane ] ) . we show that , already up to order @xmath16 , the results of the large @xmath0 expansion agree remarkably well with those found numerically for @xmath17 . moreover , we show that the nature of the problem changes at two particular dimensions @xmath7 and @xmath18 . in sec . [ sec_statics ] , we compute the equilibrium phase diagram as a function of the dimension . for @xmath19 , the @xmath20 curve has a spiral shape like in @xmath21 . for @xmath22 and @xmath23 , the @xmath20 curve is monotonous . the dimension @xmath7 is _ critical _ and requires a particular attention that is given in sec . [ sec_twod ] . we show that for @xmath7 the system generates a dirac peak ( containing a finite fraction of mass ) while for @xmath24 , the central singularity contains no mass ( at the collapse time ) . the case @xmath7 has interest in theoretical physics regarding 2d gravity @xcite and string theory @xcite ( in connection with the liouville field theory ) . it has also applications in the physics of random surfaces @xcite and random potentials @xcite , 2d turbulence @xcite and chemotaxis @xcite ( for bacterial populations ) . finally , the dynamical equations considered in this paper and in ref . @xcite are receiving a growing interest from mathematicians who established rigorous results concerning the existence and unicity of solutions for arbitrary domain shape without specific symmetry . we refer to the papers of rosier @xcite and biler _ et al . _ @xcite , and references therein , for the connection of our study with mathematical results . consider a system of particles with mass @xmath25 interacting via newtonian gravity in a space of dimension @xmath0 . the particles are enclosed within a box of radius @xmath26 so as to prevent evaporation and make a statistical approach rigorous . let @xmath27 denote the distribution function of the system , i.e. @xmath28 gives the mass of particles whose position and velocity are in the cell @xmath29 at time @xmath30 . the integral of @xmath3 over the velocity determines the spatial density @xmath31 the total mass of the configuration is @xmath32 in the meanfield approximation , the total energy of the system can be expressed as @xmath33 where @xmath34 is the kinetic energy and @xmath35 the potential energy . the gravitational potential @xmath36 is related to the density by the newton - poisson equation @xmath37 where @xmath38 is the surface of a unit sphere in a @xmath0-dimensional space and @xmath39 is the constant of gravity . finally , we introduce the boltzmann entropy @xmath40 and the free energy ( more precisely the massieu function ) @xmath41 where @xmath42 is the inverse temperature . if the system is isolated , the equilibrium state maximizes the entropy @xmath43 at fixed energy @xmath44 and mass @xmath45 ( microcanonical description ) . alternatively , if the system is in contact with a heat bath which maintains its temperature fixed , the equilibrium state maximizes the free energy @xmath46 at fixed mass @xmath45 and temperature @xmath47 ( canonical description ) . it can be shown that for systems interacting via a long - range potential like gravity , this meanfield description is _ exact _ so that our `` thermodynamical approach '' is rigorous . to solve this variational problem , we shall proceed in two steps . we first maximize @xmath43 ( resp . @xmath46 ) at fixed @xmath45 , @xmath44 ( resp . @xmath47 ) _ and _ @xmath48 . this yields the maxwell distribution @xmath49 it is now possible to express the energy and the entropy in terms of @xmath48 as @xmath50 @xmath51 where we have omitted unimportant constant terms in the entropy ( [ maxent9 ] ) . the entropy and the free energy are now functionals of @xmath48 and we consider their maximization at fixed energy or temperature . introducing lagrange multipliers to satisfy the constraints , the critical points of @xmath43 or @xmath46 are given by the boltzmann distribution @xmath52 then , the equilibrium state is obtained by solving the boltzmann - poisson equation @xmath53 and relating the lagrange multipliers to the appropriate constraints . note that a similar variational problem occurs in the context of two - dimensional turbulence ( @xmath7 ) to characterize large - scale vortices considered as maximum entropy structures @xcite . it is easy to show that there is no global maximum of entropy at fixed mass and energy for @xmath24 ( see , e.g. , ref . @xcite for @xmath17 ) . we can make the entropy diverge to @xmath54 by approaching an arbitrarily small fraction of particles in the core ( @xmath55 ) so that the potential energy goes to @xmath56 . since the total energy is conserved , the temperature must rise to @xmath54 and this leads to a divergence of the entropy to @xmath54 . note that if we collapse _ all _ particles in the core , the entropy would diverge to @xmath56 . therefore , the formation of a dirac peak is not thermodynamically favorable in the microcanonical ensemble . for @xmath7 , there exists a global entropy maximum for all energies . on the other hand , there is no global maximum of free energy at fixed mass and temperature for @xmath24 ( see , e.g. , ref . @xcite for @xmath17 ) and if @xmath57 for @xmath7 ( see appendix [ sec_absence ] ) . we can make the free energy @xmath46 diverge to @xmath54 by collapsing all particles at @xmath58 . therefore , a canonical system is expected to form a dirac peak . for @xmath7 and @xmath59 , there exists a global maximum of free energy . for @xmath23 , there exists a global maximum of entropy and free energy for all accessible values of energy and temperature . we refer to refs . @xcite for a rigorous proof of these results . when no global maxima of entropy or free energy exist , we can nevertheless look for local maxima since they correspond to metastable states which can be relevant for the considered time scales . of course , the critical points of entropy at fixed @xmath44 and @xmath45 are the same as the critical point of free energy at fixed @xmath47 and @xmath45 . only the onset of instability ( regarding the second order variations of @xmath43 or @xmath46 with appropriate constraints ) will differ from one ensemble to the other . for @xmath17 , this stability problem was considered by antonov @xcite and padmanabhan @xcite in the microcanonical ensemble and by chavanis @xcite in the canonical ensemble , by solving an eigenvalue equation . it was also studied by lynden - bell & wood @xcite and katz @xcite by using an extension of poincar theory of linear series of equilibria . we shall give the generalization of these results in sec . [ sec_stab ] to the case of a system of arbitrary dimension @xmath0 . to determine the structure of isothermal spheres , we introduce the function @xmath60 , where @xmath61 is the gravitational potential at @xmath58 . then , the density field can be written @xmath62 where @xmath63 is the central density . introducing the notation @xmath64 and restricting ourselves to spherically symmetrical configurations ( which maximize the entropy for a non - rotating system ) , the boltzmann - poisson eq . ( [ maxent11 ] ) reduces to the form @xmath65 which is the @xmath0-dimensional generalization of the emden equation @xcite . for @xmath24 , eq . ( [ emden2 ] ) has a simple explicit solution , the singular sphere @xmath66 the regular solution of eq . ( [ emden2 ] ) satisfying the boundary conditions @xmath67 must be computed numerically . for @xmath68 , we can expand the solution in taylor series and we find that @xmath69 to obtain the asymptotic behavior of the solutions for @xmath70 , we note that the transformation @xmath71 , @xmath72 changes eq . ( [ emden2 ] ) in @xmath73 for @xmath24 , this corresponds to the damped oscillations of a fictitious particle in a potential @xmath74 , where @xmath75 plays the role of the position and @xmath30 the role of time . for @xmath76 , the particle will come at rest at the bottom of the well at position @xmath77 . returning to original variables , we find that @xmath78 therefore , the regular solution of the emden equation ( [ emden2 ] ) behaves like the singular solution for @xmath79 . to determine the next order correction , we set @xmath80 with @xmath81 . keeping only terms that are linear in @xmath82 , eq . ( [ emden6 ] ) becomes @xmath83 the discriminant associated with this equation is @xmath84 . it exhibits two critical dimensions @xmath7 and @xmath18 . for @xmath19 , we have @xmath85 the density profile ( [ emden9 ] ) intersects the singular solution ( [ emden3 ] ) infinitely often at points that asymptotically increase geometrically in the ratio @xmath86 ( see , e.g. , fig . 1 of ref . @xcite for @xmath17 ) . for @xmath87 , we have @xmath88 for @xmath7 , eq . ( [ emden6 ] ) can be solved explicitly and we get @xmath89 this result has been found by various authors in different fields ( see , e.g. , @xcite ) . note that @xmath90 at large distances instead of the usual @xmath91 behavior obtained for @xmath24 . this implies that the mass of an unbounded isothermal sphere is finite in @xmath7 , although it is infinite for @xmath24 . for @xmath23 , we can neglect the r.h.s . in eq . ( [ emden7 ] ) and we get @xmath92 where @xmath93 is a constant depending on the dimension @xmath0 . for @xmath94 , eq . ( [ emden2 ] ) can be solved exactly , yielding the result @xmath95 establishing @xmath96 . as is well - known @xcite , isothermal spheres satisfy a homology theorem : if @xmath97 is a solution of the emden equation , then @xmath98 is also a solution , with @xmath99 an arbitrary constant . this means that the profile of isothermal configurations is always the same ( characterized intrinsically by the function @xmath100 ) , provided that the central density and the typical radius are rescaled appropriately . because of this homology theorem , the second order differential equation ( [ emden2 ] ) can be reduced to a _ first order _ differential equation for the milne variables @xmath101 taking the logarithmic derivative of @xmath102 and @xmath103 with respect to @xmath104 and using eq . ( [ emden2 ] ) , we get @xmath105 @xmath106 taking the ratio of the foregoing equations , we obtain @xmath107 the solution curve in the @xmath108 plane is plotted in fig . [ uv3 ] for different values of @xmath0 . the curve is parameterized by @xmath104 . it starts from the point @xmath109 with a slope @xmath110 corresponding to @xmath111 . the points of horizontal tangent are determined by @xmath112 and the points of vertical tangent by @xmath113 . these two lines intersect at @xmath114 , which corresponds to the singular solution ( [ emden3 ] ) . for @xmath19 , the solution curve spirals indefinitely around the point @xmath115 . for @xmath116 , the curve reaches the point @xmath115 without spiraling . for @xmath7 , we have the explicit solution @xmath117 so that @xmath118 for @xmath79 . for @xmath23 , @xmath119 for @xmath79 ( see fig . [ uvd1et2 ] ) . more precisely , @xmath120 where we have defined @xmath121 . for @xmath94 , @xmath122 . for bounded isothermal systems , the solution of eq . ( [ emden2 ] ) is terminated by the box at a normalized radius given by @xmath123 . we shall now relate the parameter @xmath124 to the temperature and energy . according to the poisson equation ( [ maxent4 ] ) , we have @xmath125 introducing the dimensionless variables defined previously ( using @xmath126 ) , we get @xmath127 we note that , for @xmath7 , the parameter @xmath128 is independent on @xmath26 . this is a consequence of the logarithmic form of the newtonian potential in two dimensions . the computation of the energy is a little more intricate . first , using an extension of the potential tensor theory developed by chandrasekhar ( see , e.g. , ref . @xcite ) , we can show that the potential energy in @xmath0-dimensions can be written @xmath129 for @xmath130 . now , the boltzmann - poisson equation ( [ maxent11 ] ) is equivalent to the condition of hydrostatic equilibrium @xmath131 with an equation of state @xmath132 . substituting this relation in eq . ( [ thermo3 ] ) and integrating by parts , we obtain @xmath133 where @xmath134 is the volume of a hypersphere with unit radius . ( [ thermo5 ] ) is the form of the virial theorem in @xmath0-dimension . the total energy @xmath135 can be written @xmath136 expressing the pressure in terms of the emden function , using @xmath137 and eq . ( [ emden1 ] ) , and using eq . ( [ thermo2 ] ) to eliminate the temperature , we finally obtain @xmath138 it turns out that the normalized temperature and the normalized energy can be expressed very simply in terms of the values of the milne variables at the normalized box radius . indeed , writing @xmath139 and @xmath140 and using eqs . ( [ thermo2])([thermo7 ] ) , we get @xmath141 @xmath142 the curves @xmath143 , @xmath144 are plotted in figs . [ alphaetaall]-[alphalambdaall ] . for @xmath19 , they exhibit damped oscillations toward the values @xmath145 and @xmath146 , corresponding to the singular solution ( [ emden3 ] ) . for @xmath116 the curves are monotonous . for @xmath7 , we have explicitly @xmath147 the expression of the energy has been obtained directly from eq . ( [ maxent8 ] ) with the boundary condition @xmath148 . the inverse temperature increases monotonically with @xmath124 up to the value @xmath149 . using eq . ( [ emden11 ] ) and returning to the original variables , we can write the density profile in the form @xmath150 this density profile is represented in fig . [ profil2d ] for different temperatures . at the critical inverse temperature @xmath149 , all the particles are concentrated at the center of the domain . the density profile approaches the dirac distribution @xmath151 which has an infinite ( negative ) energy . for @xmath23 , the curves @xmath143 and @xmath144 tend to @xmath54 and @xmath152 as @xmath153 . for @xmath94 , we have explicitly @xmath154 note that according to eq . ( [ thermo3 ] ) , the energy is necessarily positive for @xmath23 , so the region @xmath155 is forbidden . returning to original variables , the density profile is given by @xmath156 where we recall that @xmath157 . for @xmath153 , the profile tends to a dirac peak @xmath158 . in figs . [ lambdaeta3]-[lambdaeta2 ] , we have plotted the equilibrium phase diagram @xmath159 , giving the temperature as a function of the energy , for different dimensions @xmath0 . for @xmath19 , the curve spirals around the limit point @xmath160 corresponding to the singular solution . for @xmath116 , the curve is monotonous until the limit point . for @xmath7 , the curve is explicitly given by @xmath161 and is represented in fig . [ lambdaeta2 ] , together with the case @xmath94 . we stress that the preceding results , obtained in the meanfield approximation , are exact in the thermodynamical limit @xmath162 such that @xmath128 and @xmath163 are kept fixed . if the box radius is given , this implies that @xmath164 and @xmath165 . alternatively , if the temperature and the energy per particle are given , the thermodynamical limit is such that @xmath162 with @xmath166 constant ( for @xmath24 ) . for @xmath19 , the curve @xmath143 presents an extremum at points @xmath167 such that @xmath168 . using eqs . ( [ thermo8 ] ) and ( [ milne3 ] ) , we find that this condition is equivalent to @xmath169 since the curve @xmath170 passes through the center of the spiral in the @xmath108 plane , this determines an infinity of solutions ( see fig . [ uvparabole3 ] ) , one at each extremum of @xmath103 ( since @xmath171 ) . asymptotically , the @xmath167 follow a geometric progression ( see ref . @xcite for more details ) : @xmath172 in fig . [ alphaetaall ] , we see that an equilibrium state exists only for @xmath173 this determines a maximum mass ( for given @xmath47 and @xmath26 ) or a minimum temperature ( for given @xmath45 and @xmath26 ) beyond which no equilibrium state is possible . in that case , the system is expected to undergo an _ isothermal collapse _ ( see sect . [ sec_cane ] ) . for @xmath7 and for @xmath116 , the @xmath143 curve is monotonous . an equilibrium state exists only provided that @xmath174 @xmath175 we get comparable results for the energy . for @xmath19 , the curve @xmath144 presents an extremum at points @xmath176 such that @xmath177 . using eqs . ( [ thermo9 ] ) and ( [ milne2])([milne3 ] ) , we find that this condition is equivalent to @xmath178 we can check that the limit point @xmath115 is solution of this equation . therefore , the intersection of the parabola @xmath179 defined by eq . ( [ min5 ] ) with the spiral in the @xmath108 plane determines an infinity of points @xmath176 at which the energy is extremum ( see fig . [ uvparabole3 ] ) . on fig . [ alphalambdaall ] , we see that an equilibrium state exists only for @xmath180 this determines a minimum energy ( for given @xmath45 and @xmath26 ) or a maximum radius ( for given @xmath45 and @xmath44 ) beyond which no equilibrium state exists . in that case , the system is expected to collapse and overheat ; this is called _ gravothermal catastrophe _ ( see sect . [ sec_micro ] ) . for @xmath116 , the curve @xmath144 is monotonous . an equilibrium state exist only for @xmath181 for @xmath7 , there exists an equilibrium state for each value of energy ( see fig . [ lambdaeta2 ] ) : there is no gravothermal catastrophe in the microcanonical ensemble in two - dimensions @xcite . for @xmath23 , there exists an equilibrium state for all accessible values of energy and temperature . we now study the thermodynamical stability of self - gravitating systems in various dimensions . we start by the canonical ensemble which is simpler in a first approach . a critical point of free energy at fixed mass and temperature is a local maximum if , and only if , the second order variations @xmath182 are negative for any variation @xmath183 that conserves mass to first order , i.e. @xmath184 this is the condition of thermodynamical stability in the canonical ensemble . introducing the function @xmath185 by the relation @xmath186 and following a procedure similar to the one adopted in ref . @xcite , we can put the second order variations of free energy in the quadratic form @xmath187 the second order variations of free energy can be positive ( implying instability ) only if the differential operator which occurs in the integral has positive eigenvalues . we need therefore to consider the eigenvalue problem @xmath188 with @xmath189 . if all the eigenvalues @xmath190 are negative , then the critical point is a maximum of free energy . if at least one eigenvalue is positive , the critical point is an unstable saddle point . the point of marginal stability , i.e. the value of @xmath124 in the series of equilibria @xmath143 at which the solutions pass from local maxima of free energy to unstable saddle points , is determined by the condition that the largest eigenvalue is equal to zero ( @xmath191 ) . we thus have to solve the differential equation @xmath192 with @xmath193 . introducing the dimensionless variables defined previously , we can rewrite this equation in the form @xmath194 with @xmath195 . if @xmath196 denotes the differential operator which occurs in eq . ( [ stab7 ] ) , we can check by using the emden eq . ( [ emden2 ] ) that @xmath197 therefore , the general solution of eq . ( [ stab7 ] ) satisfying the boundary conditions at @xmath111 is @xmath198 using eq . ( [ stab8 ] ) and introducing the milne variables ( [ milne1 ] ) , the condition @xmath199 can be written @xmath200 this relation determines the points at which a new eigenvalue becomes positive ( crossing the line @xmath191 ) . comparing with eq . ( [ min1 ] ) , we see that a mode of stability is lost each time that @xmath128 is extremum in the series of equilibria , in agreement with the turning point criterion of katz @xcite in the canonical ensemble . in particular , the series becomes unstable at the point of minimum temperature ( or maximum mass ) @xmath201 . secondary modes of instability appear at values @xmath202 , @xmath203 , ... we obtain the same results by considering the dynamical stability of isothermal gaseous spheres with respect to the navier - stokes equations ( see ref . @xcite for @xmath17 ) . therefore , dynamical and thermodynamical stability criteria coincide for isothermal gaseous spheres . according to eq . ( [ stab3 ] ) , the perturbation profile that triggers a mode of instability at the critical point @xmath191 is given by @xmath204 where @xmath205 is given by eq . ( [ stab8 ] ) . introducing the milne variables ( [ milne1 ] ) , we get @xmath206 the density perturbation @xmath183 becomes zero at point(s ) @xmath207 such that @xmath208 . the number of zeros is therefore given by the number of intersections between the spiral in the @xmath108 plane and the line @xmath209 ( see fig . [ uvparabole3 ] ) . for the @xmath210-th mode of instability we need to follow the spiral to the @xmath210-th extremum of @xmath103 ( since @xmath167 corresponds to an extremum of @xmath128 , hence @xmath103 ) . therefore , the density perturbation @xmath183 corresponding to the @xmath210-th mode of instability has @xmath210 zeros @xmath211 . asymptotically , the zeros follow a geometric progression with ratio @xmath212 @xcite . in the microcanonical ensemble , the condition of thermodynamical stability requires that the equilibrium state is an entropy maximum at fixed mass and energy . this condition can be written @xmath213 for any variation @xmath183 that conserves mass to first order ( the conservation of energy has already been taken into account in obtaining eq . ( [ stab10 ] ) ) . now , following a procedure similar to that of ref . @xcite in @xmath17 , the second variations of entropy can be put in a quadratic form @xmath214 with @xmath215 the problem of stability can therefore be reduced to the study of the eigenvalue equation @xmath216 with @xmath217 . the point of marginal stability ( @xmath191 ) will be determined by solving the differential equation @xmath218 with @xmath219 introducing the dimensionless variables defined previously , this is equivalent to @xmath220 with @xmath221 and @xmath195 . using the identities ( [ stab7b ] ) , we can check that the general solution of eq . ( [ stab16 ] ) satisfying the boundary conditions for @xmath111 and @xmath222 is @xmath223 the point of marginal stability is then obtained by substituting the solution ( [ stab18 ] ) in eq . ( [ stab17 ] ) . using the identities @xmath224 @xmath225 which result from simple integrations by parts and from the properties of the emden equation ( [ emden2 ] ) , it is found that the point of marginal stability is determined by the condition ( [ min5 ] ) . therefore , the series of equilibria becomes unstable at the point of minimum energy in agreement with the turning point criterion of katz @xcite in the microcanonical ensemble . the structure of the perturbation that triggers the instability can be determined with the graphical construction described in ref . it is found that the first mode of instability has a core - halo structure ( i.e. , two nodes ) in the microcanonical ensemble , unlike the first mode of instability in the canonical ensemble @xcite . the thermodynamical stability analysis presented in this section also shows that the equilibrium states for @xmath226 and @xmath116 are always stable since the series of equilibria do not present turning points of energy or temperature . we now consider the dynamics of a system of self - gravitating brownian particles in a space of dimension @xmath0 . like in ref . @xcite , we consider a high friction limit in order to simplify the problem . we thus study the smoluchowski equation @xmath227 coupled to the newton - poisson equation ( [ maxent4 ] ) . in the microcanonical ensemble , the temperature @xmath228 evolves in time so as to satisfy the energy constraint ( [ maxent8 ] ) . in the canonical ensemble , the temperature is constant . the smoluchowski equation can be obtained from a variational principle called the maximum entropy production principle @xcite . this variational approach is interesting as it makes a direct link between the dynamics and the thermodynamics . in the microcanonical description , the rate of entropy production can be put in the form @xcite @xmath229 for a stationary solution , @xmath230 and we obtain the boltzmann distribution ( [ maxent10 ] ) which is a critical point of entropy . considering a small perturbation around equilibrium , we can establish the identity @xcite : @xmath231 where @xmath190 is the growth rate of the perturbation defined such that @xmath232 . this relation shows that a stationary solution of the smoluchowski - poisson system is dynamically stable if and only if it is a local entropy maximum . we get similar results in the canonical ensemble with @xmath46 in place of @xmath43 . the relation ( [ brown5 ] ) has been found for other kinetic equations satisfying a @xmath233-theorem ( chavanis , in preparation ) . in @xmath7 , an equation morphologically similar to the smoluchowski equation ( [ brown3 ] ) can been derived from the @xmath234-body liouville equation for a gas of point vortices @xcite . from now on , we set @xmath235 . the equations of the problem become @xmath236 @xmath237 @xmath238 with boundary conditions @xmath239 for @xmath24 . for @xmath7 , we take @xmath240 on the boundary . integrating eq . ( [ dim2 ] ) once , we can rewrite the smoluchowski - poisson system in the form of a single integrodifferential equation @xmath241 the smoluchowski - poisson system is also equivalent to a single differential equation @xmath242 for the quantity @xmath243 which represents the mass contained within the sphere of radius @xmath244 . the appropriate boundary conditions are @xmath245 it is also convenient to introduce the function @xmath246 satisfying @xmath247 we look for self - similar solutions of the form @xmath248 in terms of the mass profile , we have @xmath249 and @xmath250 in terms of the function @xmath251 , we have @xmath252 substituting the _ ansatz _ ( [ dim6s ] ) into eq . ( [ seq ] ) , we find that @xmath253 where we have set @xmath254 . the variables of position and time separate provided that there exists @xmath124 such that @xmath255 where @xmath256 is a constant . in that case , eq . ( [ dim8 ] ) reduces to @xmath257 assuming that such a scaling exists implies that @xmath258 is a constant that we arbitrarily set equal to @xmath124 ( note that this convention is different from the one adopted in ref . this leads to @xmath259 so that the central density becomes infinite in a finite time @xmath260 . the scaling equation now reads @xmath261 for @xmath10 , we have asymptotically @xmath262 in the canonical ensemble in which the temperature @xmath47 is a constant , we have is treated in the appendix [ sec_cold ] ] @xmath263 in that case , the scaling equation ( [ scalingd ] ) can be solved analytically . following a procedure similar to the one developed in ref . @xcite , we find that @xmath264 then , eqs . ( [ dim6s ] ) and ( [ dim7 ] ) yield @xmath265 according to eqs . ( [ dim5 ] ) and ( [ dim11 ] ) , the central density evolves with time like @xmath266 according to eqs . ( [ dim5 ] ) and ( [ dim6 ] ) , the core radius and the core mass evolve like @xmath267 note that for @xmath24 , the core mass goes to zero at the collapse time . at @xmath268 , we get the singular profile @xmath269 in the microcanonical ensemble , the exponent @xmath124 is not determined by simple dimensional analysis . in ref . @xcite , we found numerically that the scaling equation ( [ scalingd ] ) has physical solutions only for @xmath270 , with @xmath271 for @xmath17 . we also argued that the system will select the exponent @xmath5 , since it leads to a maximum increase of entropy . in this section , we show that in the limit of large dimension @xmath0 , we can explicitly understand the occurrence of such a @xmath6 . in addition , we will present the derivation of perturbative expansions for @xmath6 and the scaling function @xmath43 , in powers of @xmath15 . ( [ scalingd ] ) can be formally integrated as a first order differential equation ( writing @xmath272 $ ] ) , leading to an expression of @xmath273 as a function of @xmath274 , @xmath273 itself , and @xmath275 : @xmath276 . \label{simp}\ ] ] we now define @xmath277 , such that @xmath278 . since @xmath43 should be analytic , the foregoing relation implies for @xmath279 , @xmath280 where @xmath281 is the function which occurs in the denominator of the integral in eq . ( [ simp ] ) . from eq . ( [ micro1 ] ) , we must have @xmath282 for @xmath283 , which implies @xmath284 and @xmath285 . these conditions can be rewritten @xmath286 @xmath287_{x_0}. \label{condd2}\ ] ] this preparatory work now allows the introduction of a systematic expansion in large dimension @xmath0 for the scaling function @xmath43 , the scaling exponent @xmath124 , and @xmath277 . in this limit , let us neglect the contribution of the terms which are not of order @xmath0 in the right - hand side of eq . ( [ scalingd ] ) . this actually amounts to take @xmath288 in eq . ( [ simp ] ) . within this approximation , we find @xmath289 which is an analytic function only if @xmath2 . this leads to @xmath290 , and to the more explicit form for @xmath43 , @xmath291 @xmath292 remains undetermined , and will be fixed by the next order approximation . indeed , we can iteratively solve the full scaling equation eq . ( [ scalingd ] ) , by reinserting the zeroth order solution into eq . ( [ simp ] ) , and eventually continue this process with the new improved scaling function , and so forth ... thus , expressing the conditions of eq . ( [ condd1 ] ) and eq . ( [ condd2 ] ) , and defining @xmath293 ( which will be of order @xmath294 ) , we obtain @xmath295 and @xmath296+o(d^{-2 } ) . \label{alphad1}\ ] ] eq . ( [ alphad1 ] ) provides a relation between the possible values for @xmath124 and the associated value of @xmath297 . note that the function of @xmath75 in the right - hand side of eq . ( [ alphad1 ] ) has a well defined maximum . hence , up to order @xmath298 , we find that @xmath299 , with @xmath300 which is associated to the value @xmath301 or @xmath302 . as @xmath124 is necessarily greater than 2 ( as the temperature can not vanish ) , a solution exists for any @xmath303 $ ] . as already mentioned , @xmath304 is dynamically selected as it leads to the maximum divergence of the entropy and the temperature ( see eq . ( [ atheta ] ) below ) . inserting eq . ( [ sinf ] ) into eq . ( [ simp ] ) , we find the next order approximation for @xmath43 @xmath305^{\frac{\alpha\phi}{2 } } , \label{snext}\ ] ] where @xmath277 is given by eq . ( [ x0d ] ) , and @xmath306 and @xmath307 are defined by @xmath308+o(d^{-2 } ) . \label{phix1}\ ] ] again , the analyticity condition imposes that @xmath309 , which exactly leads to eq . ( [ alphad1 ] ) , and to the following explicit form for @xmath43 : @xmath310^{-1}. \label{snextexp}\ ] ] this improved scaling function can be inserted again into the conditions expressed by eq . ( [ condd1 ] ) and eq . ( [ condd2 ] ) , leading to the next order term in the expansion of @xmath124 . after elementary but cumbersome calculations , we end up with @xmath296 + \frac{8}{d^2}\left[\frac{5}{z}-\frac{26}{z^2}+ \frac{31}{z^3}-\frac{6}{z^4 } -\left(\frac{1}{z}-\frac{7}{z^2}+ \frac{14}{z^3}-\frac{8}{z^4}\right)\ln z \right]+ o(d^{-3 } ) . \label{alphad2}\ ] ] this function has again a well defined maximum for @xmath311 associated to the value @xmath312 this expansion gives @xmath313 in @xmath17 , in fair agreement with the exact value @xmath314 obtained numerically in @xcite . in addition , the exponent @xmath2 is associated to @xmath315 . in principle , these expansions can be systematically pursued to the prize of increasingly complicated calculations . finally note that eqs . ( [ dim5 ] ) and ( [ dim11 ] ) lead to the following exact asymptotic for the central density @xmath316 : @xmath317 where we have used @xmath318 and the definition of @xmath75 . the function @xmath319 is determined implicitly by eq . ( [ alphad2 ] ) , up to order @xmath16 . for the special cases @xmath2 and @xmath320 , we respectively find @xmath321 which shows that @xmath322 is substantially greater than @xmath323 ( twice bigger in the infinite @xmath0 limit , the ratio being even bigger for finite @xmath0 , as @xmath324 ) . this substantial difference was noted in ref . @xcite , in the case @xmath17 . finally , as expected in the microcanonical ensemble , the temperature diverges during the collapse as @xmath325 with @xmath326 ; see eqs . ( [ dim5])([dim9])([dim11 ] ) . the strongest divergence is obtained for @xmath320 . according to eq . ( [ alphaexp2 ] ) , we have @xmath327 if we plug @xmath17 in eq . ( [ atheta ] ) , we find the estimate @xmath328 in fair agreement with the exponent measured numerically in @xcite , @xmath329 . in two dimensions , the dynamical equation ( [ sca ] ) for the mass profile reads @xmath330 after the change of variable @xmath331 has been effected . looking for a stationary solution , and using @xmath332 , eq . ( [ dyn2d ] ) is readily integrated leading to @xmath333 note that @xmath334 , ensuring that the whole mass is included in this solution . using @xmath335 , we find that the density profile is given by @xmath336 with @xmath337 this solution exists provided that @xmath338 , which defines the collapse temperature . we have thus recovered the result ( [ thermo11 ] ) by a slightly different method . note that the value of @xmath339 and the dependance of @xmath340 and @xmath63 with the temperature coincide with the exact results obtained within conformal field theory @xcite . in the following , @xmath47 is set constant as we have already seen that the gravothermal does not exist in the microcanonical ensemble in two dimensions . we now address the dynamics at the critical temperature @xmath342 . we note that contrary to what happens in other dimensions , the central density diverges at @xmath343 . thus , in analogy with critical phenomena , we anticipate a scaling solution for @xmath344 , of the form @xmath345 which preserves the scaling form obtained above @xmath343 , and which satisfies the boundary condition @xmath346 . the corresponding density profile is @xmath347 the central density @xmath348 is expected to diverge for @xmath76 , so that @xmath349 is also expected to diverge . inserting the _ ansatz _ eq . ( [ tc2d ] ) into eq . ( [ dyn2d ] ) shows that the left - hand term is indeed negligible compared to both terms of the right - hand side , to leading order in @xmath350 . so far , this prevents us from determining a dynamical equation for @xmath350 . in order to achieve that , we must solve eq . ( [ dyn2d ] ) to the next order in @xmath351 . we thus look for a solution of the form @xmath352 where @xmath353 is expected to be of order @xmath294 , and should satisfy @xmath354 and @xmath355 ( the total integrated mass should be 0 and 1 , respectively , for @xmath356 and @xmath357 ) , and @xmath358 , which ensures that eq . ( [ rhoa ] ) is exactly obeyed , defining @xmath349 without any ambiguity . the contribution of @xmath359 in the left - hand side of eq . ( [ dyn2d ] ) is dominated by the time derivative of eq . ( [ tc2d ] ) : @xmath360= \frac{u(1-u)}{(1+au)^2}\frac{da}{dt},\ ] ] which will be checked self - consistently hereafter . in addition , non linear terms in @xmath361 in the right - hand side are also negligible . therefore , @xmath361 satisfies @xmath362 actually , for a given time , this equation becomes an ordinary differential equation involving only one variable @xmath102 , as @xmath350 and @xmath363 appear as parameters . ( [ eqh ] ) can be integrated leading to a first order equation in @xmath361 , which can be easily solved . defining @xmath364 , we finally get @xmath365 , \nonumber\end{aligned}\ ] ] which depends on time only through the variable @xmath350 and @xmath363 . now , @xmath363 is determined by imposing the boundary condition @xmath355 , which leads to @xmath366 . \label{dadt}\ ] ] one can solve iteratively eq . ( [ dyn2d ] ) , by adding the time derivative of the above solution to the left - hand side , in order to compute an improved @xmath361 . to leading order , the solution of eq . ( [ h ] ) is preserved . however , new terms are generated which are important for @xmath367 ( @xmath368 ) , and which generate terms of order @xmath369 is the expansion for @xmath363 . this explains the form of the error term in eq . ( [ dadt ] ) . integrating eq . ( [ dadt ] ) for large time , we get the exact asymptotic expansion for large time @xmath370 . \label{at}\ ] ] for @xmath76 , the central density diverges like @xmath349 and the core radius goes to zero like @xmath371 . in addition , the scaling solution ( [ scarho2d ] ) at @xmath372 goes to a dirac peak containing the whole mass ( see eq . ( [ thermo12 ] ) ) , as the decay exponent of the scaling function is 4 , which is strictly greater than 2 . for @xmath7 , the scaling equation associated to eq . ( [ sca ] ) does not display any physical solution when numerically solved . in this section , we thus present a special treatment adapted to this case . the principal difference with other dimensions is the divergence of the central density at @xmath343 , and the occurrence of a scaling solution at this temperature . strictly below @xmath343 , we expect a finite time collapse . close to the center , the solution takes the form @xmath374 where again the left - hand side of eq . ( [ dyn2d ] ) is negligible compared to each term on the right - hand side . we thus look for a solution of the type @xmath375 where @xmath361 is now of order @xmath294 as it contains a finite fraction of the total mass , since the first term contains a mass of order @xmath376 . inserting this _ ansatz _ in the dynamical equation eq . ( [ dyn2d ] ) , we obtain @xmath377 one can look for a scaling solution of the type @xmath378 so that the mass included in this scaling profile @xmath379 . with this definition , the density profile decays for large distance as @xmath380 , with @xmath381 . inserting this _ ansatz _ in eq . ( [ ew ] ) , we obtain @xmath382\frac{da}{dt}a^{-2}=vh''+2 \left(\frac{v}{1+v}h\right)'+ 2a^{\gamma-1}h h ' , \label{scacol2d}\ ] ] where derivatives are with respect to the variable @xmath103 . we are free to choose @xmath383 , so that @xmath384 . for small @xmath103 , eq . ( [ scacol2d ] ) leads to @xmath385 eq . ( [ scacol2d ] ) has a global scaling solution only for @xmath386 . however , we know that in this case the scaling equation obtained by setting @xmath386 does not display any physical solution . thus , we conclude that there is no scaling solution obtained by imposing that all terms in eq . ( [ scacol2d ] ) scale the same way . however , as we will see in the section devoted to numerical simulations , the direct simulation of eq . ( [ dyn2d ] ) seems to display a scaling solution with @xmath387 for numerically accessible densities . strictly speaking , this is totally excluded by the above equation , except if one allows @xmath388 to very slowly depend on the density or @xmath350 . for a given @xmath350 , we thus want to solve eq . ( [ scacol2d ] ) , where the boundary conditions will ultimately select the effective value of @xmath388 , and that of @xmath363 . more precisely , once we impose @xmath384 , and the condition of eq . ( [ dadtg ] ) , we end up with a shooting problem for @xmath389 and @xmath388 . for large @xmath350 , and @xmath390 , it is clear that the non linear term of the right - hand side of eq . ( [ scacol2d ] ) becomes irrelevant , and we drop it from now on . in order to understand the origin of this shooting problem , and to obtain an accurate estimate of @xmath388 , let us solve eq . ( [ scacol2d ] ) in the limit of very large @xmath350 , in the range @xmath391 . in this regime , eq . ( [ scacol2d ] ) simplifies to the following equation @xmath392\omega = vh''+2h ' , \label{sca2dsimp}\ ] ] where @xmath393 let us now multiply this equation by @xmath394 and integrate the resulting equation . after elementary manipulations , we obtain @xmath395h=-\frac{\omega c}{4 t } v^{1-\gamma}- \frac{\omega a^{1-\gamma}}{2-\gamma}v^{-1}+(2-\gamma)(3-\gamma)v^{1-\gamma } \int_v^{+\infty}w^{\gamma-3}h(w)\,dw,\label{eqf}\ ] ] where @xmath396 , which has been defined in eq . ( [ asymf ] ) , appears here as an integration constant . then , one can integrate this differential equation which leads to the following self - consistent relation for @xmath233 : @xmath397 where @xmath398 is defined as the opposite of the right - hand side of eq . ( [ eqf ] ) : @xmath399 eq . ( [ sca2dsimp ] ) implies that @xmath400 , when @xmath401 ( of course , this apparent divergence does not occur in the full dynamical equation eq . ( [ scacol2d ] ) ) . considering the prefactor @xmath402 in eq . ( [ eqff ] ) , this behavior can be obtained if and only if @xmath403 as @xmath404 is expected to go to zero for large @xmath350 as @xmath405 ( see eq . ( [ omega ] ) ) , the dominant contribution of the integral of the third term in the definition of @xmath398 comes from the large @xmath406 region , for which @xmath233 can be replaced by its asymptotic form . hence , defining @xmath407 and @xmath408 , and using eq . ( [ omega ] ) , the condition expressed in eq . ( [ cond1 ] ) can be rewritten @xmath409 as @xmath410 is of order @xmath294 , we find that @xmath411 as @xmath412 . more precisely , in this limit , @xmath413 is the solution of the following implicit equation @xmath414 where @xmath415 . finally , we obtain @xmath416^{-1})\right ) . \label{expe}\ ] ] in conclusion , although the solution is not strictly speaking a true scaling solution , the explicit dependence of @xmath388 on @xmath350 is so weak that an apparent scaling should be seen with an effective @xmath388 almost constant for a wide range of values of @xmath350 . hence , the total density profile is the sum of the scaling profile obtained at @xmath343 with a @xmath417 weight ( behaving as a dirac peak of weight @xmath417 , at @xmath268 ) and of a pseudo - scaling solution associated to an effective scaling exponent slowly converging to @xmath2 . let us illustrate quantitatively the time dependence of @xmath381 . for example , taking arbitrarily @xmath418 ( the dependence on @xmath34 is weak and vanishes for large @xmath350 ) , eq . ( [ impe ] ) and eq . ( [ expe ] ) respectively lead to @xmath419 and @xmath420 , and to @xmath421 and @xmath422 ( note that the error between the asymptotic expansion of eq . ( [ expe ] ) and the implicit expression first grows before slowly decaying for @xmath423 ! ) . finally , for the maximum value of @xmath350 accessible numerically of order @xmath424 , we expect to observe an apparent scaling solution with @xmath425 , or @xmath426 . in this subsection , we present numerical simulations concerning the two - dimensional case . indeed , the three - dimensional case has been extensively studied in @xcite . it has been shown that the scaling function as well as the corrections to scaling ( which have been calculated for the canonical ensemble in @xcite ) are perfectly described by the theory . in addition , as the system behaves qualitatively the same for any dimension @xmath24 in both thermodynamical ensembles , we have decided to focus on the numerical study of the @xmath7 case , which displays some very rich behaviors , as exemplified in the present section . we consider the system in the canonical ensemble , as the gravitational collapse does not occur in the microcanonical ensemble . in fig . [ colld_fig1 ] , we show the scaling function at @xmath343 , as given by eq . ( [ scarho2d ] ) , finding a perfect agreement . in fig . [ colld_fig2 ] , we also display @xmath427 as a function of @xmath428 , and find an asymptotic behavior fully compatible with that given by eq . ( [ dadt ] ) . below @xmath343 , and in the accessible range of @xmath350 ( up to @xmath429 ) , we find an apparent scaling regime with @xmath430 , as predicted in sec . [ sec_tinftc ] . this is illustrated in fig . [ colld_fig3 ] , for @xmath431 . note that the effective @xmath388 or @xmath124 can also be extracted from the time evolution of @xmath349 or the central density ( see eq . ( [ dadtg ] ) ) . in fig . [ colld_fig4 ] , we show that this way of measuring @xmath388 is fully compatible with the value of the effective exponents @xmath430 when an equilibrium state exists , there is little hope to be able to solve the full smoluchowski - poisson system analytically in order to study the relaxation towards equilibrium . we shall consider a simpler problem in which a test particle evolves in a medium of field particles at statistical equilibrium ( thermal bath approximation ) . the particles are assumed to create a stationary potential @xmath432 which induces a drift of the test particle along the gradient of @xmath433 . in addition , the test particle is assumed to experience a diffusion process . if @xmath434 denotes the density probability of finding the test particle in @xmath435 at time @xmath30 , we expect the evolution of @xmath434 to be determined by a smoluchowski equation of the form @xmath436 where @xmath432 is solution of the boltzmann - poisson equation ( [ maxent11 ] ) . this means that we replace the true potential by its equilibrium value but still allow the density @xmath434 to vary with time . as we shall see , it is possible to solve the smoluchowski equation ( [ schro1 ] ) analytically in @xmath94 by using an analogy with a problem of quantum mechanics . an equation of the form ( [ schro1 ] ) has been proposed in refs . @xcite to model the motion of a test vortex in a bath of field vortices at statistical equilibrium . in that context , ( [ schro1 ] ) can be formally derived from the @xmath234-body liouville equation of the system by using projection operator technics . it is well - known that a fokker - planck equation like eq . ( [ schro1 ] ) can be formally transformed into a schrdinger equation with imaginary time . indeed , performing the change of variable @xmath437 we find that the evolution of @xmath100 is determined by an equation of the form @xmath438 this can be written as a schrdinger - type equation @xmath439 with a potential @xmath440 . so far , this transformation is general . if we now consider the one - dimensional case , the boltzmann - poisson equation ( [ schro1 ] ) can be solved analytically and the potential @xmath441 determined explicitly . introducing the notations @xmath442 and @xmath443 and using eq . ( [ emden13 ] ) we can rewrite eq . ( [ schro3 ] ) in the form @xmath444 a separation of the variables can be effected by the substitution @xmath445 where @xmath307 is solution of the ordinary differential equation @xmath446 where we have set @xmath447 . the solutions of this schrdinger equation are described in detail in ref . the spectrum of positive energies is continuous . the spectrum of negative energies is discrete and reduces to @xmath448 ( fundamental state ) . the first excited state in the continuum is @xmath449 . we can check that the corresponding eigenfunctions are @xmath450 and @xmath451 . in order to obtain the qualitative behavior of the time dependent solution of eq . ( [ schro1 ] ) , we neglect the contribution from the continuum states with @xmath452 , only keeping the @xmath453 and @xmath454 eigenstates . within this approximation and for sufficiently large times , we obtain @xmath455 where @xmath99 and @xmath456 are constant . returning to original variables , we get @xmath457 where @xmath458 is given by eq . ( [ thermo14 ] ) and @xmath459 is a constant . we find that the relaxation time is given by @xmath460 . in this paper , we have studied the boltzmann - poisson equation and the smoluchowski - poisson system in various dimensions of space . our study shows in particular how the nature of the problem changes as we pass from @xmath17 to @xmath7 . we showed that the dimension @xmath7 is critical in the sense that the results obtained for @xmath24 diverge if they are naively extrapolated to @xmath7 . on a physical point of view , the two - dimensional problem differs from the @xmath24 case in two respects : in the 2d case , the central density of the equilibrium state is infinite at the critical temperature @xmath339 while it is finite at @xmath339 in higher dimensions . on the other hand , in @xmath7 , the self - similar collapse results in a dirac peak which contains a finite fraction of mass , while for @xmath24 , the mass contained in the core tends to zero at the collapse time ( but a dirac peak is always formed in the canonical ensemble after @xmath260 as discussed in appendix [ sec_cold ] ) . we have also evidenced another characteristic dimension @xmath18 at which the nature of the problem changes . for @xmath116 the classical spiral characterizing isothermal spheres in the physical three - dimensional space disappears . however , since the points in the spiral correspond to unstable states , that are therefore unphysical , this transition at @xmath18 is not so critical and indeed the nature of the self - similar collapse does not show any transition at that dimension . it is interesting to note that the dependance of the phase diagram in the @xmath461 and @xmath108 planes with the dimension of space @xmath0 shows some resemblance with the dependance of the phase diagram of confined polytropic spheres with the index @xmath210 of the polytrope @xcite . an extension of our study would be to relax the high friction limit and consider solutions of the kramers - poisson system and other relaxation equations described in ref . these equations are expected to display qualitatively similar behaviors than those described here ( i.e. gravitational collapse , finite time singularity , self - similar solutions , ... ) but their study appears to be of considerable difficulty since we now need to consider the evolution of the system in phase space . we hope to come to that problem in future publications . we give a proof similar to the one given in ref . @xcite for @xmath17 . in two dimensions , we consider a homogeneous disk of mass @xmath45 and radius @xmath350 at temperature @xmath47 . it is easy to show that the total energy ( [ maxent8 ] ) of this disk is @xmath462 with the convention @xmath463 at large distances . according to eqs . ( [ maxent6 ] ) and ( [ maxent9 ] ) , its free energy reads @xmath464 for @xmath465 , the free energy behaves like @xmath466 therefore , if @xmath57 the free energy goes to @xmath54 when we contract the system to a point . this is sufficient to prove the absence of a global maximum of free energy below @xmath339 . this also shows the natural tendency ( in a thermodynamical sense ) of a canonical self - gravitating system to collapse to a dirac peak for @xmath467 . for @xmath468 , eq . ( [ seq ] ) reduces to @xmath469 looking for a self - similar solution of the form ( [ dim6s ] ) and imposing the conditions ( [ dim9 ] ) and ( [ dim11 ] ) , we find that the scaling profile satisfies @xmath470 of course , for @xmath468 , the exponent @xmath124 can not be determined on dimensional grounds , as the definition @xmath471 is not relevant anymore . as we will see , @xmath124 will be determined by imposing that the scaling solution is analytic . ( [ scalingt0 ] ) can be readily solved leading to the following implicit equation for @xmath43 : @xmath472 where @xmath34 is an integration constant . now , from the definition of @xmath43 , we expect a small @xmath274 expansion of the form @xmath473 , which first implies that @xmath474 and that @xmath475 , which finally leads to @xmath476 in terms of the scaling function @xmath477 associated to the mass profile , eq . ( [ st0 ] ) can be rewritten @xmath478^{\frac{d+2}{d } } , \label{st0p}\ ] ] where @xmath479 is arbitrary . this leads to the exact large @xmath274 asymptotic behavior @xmath480 moreover , using eq . ( [ dim11 ] ) and ( [ s00 ] ) , we get the exact universal asymptotic behavior of the central density @xmath481 finally , we note that the implicit equation ( [ st0p ] ) can be written as a parametric set of equations @xmath482^{\frac{1}{d}}.\ ] ] these results can be obtained by a different , more physical , method . we have indicated in ref . @xcite that , for @xmath468 , the particles have a deterministic motion with equation @xmath483 for a spherically symmetrical system , this can be rewritten @xmath484 where @xmath485 is the mass within @xmath244 . if @xmath350 denotes the initial position of the particle located at @xmath244 at time @xmath30 , we have @xmath486 so eq . ( [ penston2 ] ) can be integrated explicitly in @xmath487 if @xmath488 behaves like @xmath489 close to the origin ( which is a regular expansion ) , then @xmath490 introducing the collapse time @xmath491 and considering the limit @xmath492 , we obtain @xmath493 introducing the scaling variables @xmath494 we can put the solution in a parametric form @xmath495 where @xmath496 is a constant . at the collapse time @xmath268 , @xmath497 these results are of course equivalent to those obtained previously . we can now use this method to study the evolution of the system for @xmath498 ( post - collapse solution ) . for @xmath499 , according to eqs . ( [ penston3 ] ) and ( [ penston9 ] ) , the mass contained inside the sphere of radius @xmath500 at @xmath268 has collapsed at @xmath58 , resulting in a dirac peak of weight @xmath501 note that in a bounded domain the final collapse to a central dirac peak containing the whole mass occurs in a finite time after @xmath260 . for @xmath502 ( associated to @xmath503 ) , one has @xmath504 @xmath505 introducing the scaling variables @xmath506 we obtain @xmath507 subtracting the dirac peak at @xmath58 , and considering @xmath508 , for which @xmath509 , we find that the leading contribution to the mass profile for small @xmath244 is @xmath510 hence the density profile does not diverge at @xmath511 for @xmath498 . instead , the density approaches the constant value @xmath512 which decreases with time . the density profile is thus depleted on a scale @xmath513 , which increases with time . for @xmath514 , the density profile remains essentially unaffected . in principle , the same phenomenon arises for any @xmath515 : the density profile obtained at @xmath260 ultimately collapses into a central dirac peak at a time @xmath516 . this solves the apparent paradox that the solution at @xmath268 has a vanishing central mass and a finite free energy @xcite . in fact , if we allow singular profiles to develop , the evolution continues for @xmath498 and the dirac peak with infinite free energy ( predicted by statistical mechanics @xcite ) is formed during the post collapse regime of our brownian model . , the results should be different in the microcanonical ensemble . we shall reserve for a future communication the full description of the post - collapse regime . ] in practice , degeneracy effects ( of quantum or dynamical origin ) lead to a finite small core of finite density , controlled by the maximum allowed degeneracy @xcite .

singlet fission , the molecular process through which photons are effectively converted into pairs of lower energy triplet excitons , holds promise as a means of boosting photovoltaic device efficiencies . in the preceding article of this series , we formulated a vibronic theory of singlet fission , inspired by previous experimental and theoretical studies suggesting that vibronic coupling plays an important role in fission dynamics . here , we extend our model in order to simulate two - dimensional electronic spectra , through which the theory is further validated based on a comparison to recent measurements on pentacene crystals . moreover , by means of such spectral simulations , we provide new insights into the nature of the correlated triplet pair state , the first product intermediate in the fission process . in particular , we address a disagreement in the literature regarding the identification , energies , and transition dipole moments of its optical transitions towards higher - lying triplet states .

1703.01174

the shockley - queisser efficiency limit for single - junction solar cells@xcite can be circumvented@xcite by a molecular process called singlet fission in which an optically excited singlet state is converted into a pair of triplet excitons , which are transiently correlated in a spin - zero state . the prospect of enhancing photovoltaic efficiencies through such an exciton multiplication mechanism has stimulated intense interest in singlet fission.@xcite recent time - resolved spectroscopic studies have provided indications that vibronic coupling plays an integral role in the excited state dynamics of singlet fission materials.@xcite however , a detailed understanding of the mechanisms through which singlet fission occurs remains lacking , and calls for a theoretical study in which a microscopic set of electronic degrees of freedom interacting with vibrational modes are treated non - perturbatively . in the preceding article of this series , hereafter referred to as i , we formulated a vibronic exciton model tailored to meet this demand . using crystalline pentacene as a prototypical fission material we characterized the excited states contributing to linear absorption , paying special attention to the energy and spatial configuration of the correlated triplet pair state ( tt@xmath0 ) , which acts as the first conversion intermediate in singlet fission . the experimental detection of the fission product states is hindered by the fact that they are essentially optically dark with respect to the ground state , a feature that results from their multi - exciton character and the triplet nature of their constituents . on the picosecond time scale , singlet fission can be observed through recombination into singlet excitons by means of delayed fluorescence.@xcite transient absorption@xcite and two - photon photoemission spectroscopy,@xcite on the other hand , allow for a sufficient time resolution to resolve the sub - ps fission dynamics in materials such as crystalline pentacene . accordingly , it is assumed that the product states may be detected through allowed transitions towards other ( auxiliary ) excited states , typically higher - lying triplets or ionized states . however , the disadvantage of such approaches is that the optical signal simultaneously bears the imprints of both the fission product and the auxiliary states . for example , optical transitions between tt@xmath0 and higher - lying triplet states inform on the relative energy difference of the involved states , but do not reveal their energetic separation from the ground state or the singlet excitation through which the fission process is initiated . this magnifies the issue that such triplet signals are commonly mapped onto the bright response from the singlet state , rendering the overall spectra congested with features . perhaps for this reason , some of the identified triplet transitions in bulk pentacene@xcite seem inconsistent with measurements@xcite and calculations on non - interacting pentacene molecules,@xcite discrepancies that remain to be clarified.@xcite most of the aforementioned complications can be circumvented through two - dimensional electronic spectroscopy ( 2des),@xcite which by projecting the optical response to an extra dimension allows one to resolve otherwise overlapping spectral signals.@xcite 2des has served as an important experimental technique for the study of sub - ps quantum dynamics in photosynthesis@xcite as well as synthetic materials@xcite for over a decade , but only recently found its first application in the context of singlet fission in a study on polycrystalline pentacene by bakulin _ _ et al.__@xcite one of the key results of that study was the direct detection of the optical transition between tt@xmath0 and the ground state . although this transition is extremely weak , this observation was accomplished by creating a coherent vibrational wavepacket in the pair state potential , yielding spectral beats that were enhanced through intensity borrowing . this experimental result helped clarify the energetic separation of tt@xmath0 from the ground and singlet excited states , and was used in i to parametrize properties of the triplet excitons in the proposed model for pentacene crystals . by further analyzing the 2des measurements through our model , additional crucial information about the excited states involved in singlet fission is readily obtainable . in the present article , we extend our theoretical model in order to simulate 2des of fission materials . this allows us to subject the model to additional tests by comparing simulations to the 2des measurements on polycrystalline pentacene . perhaps more importantly , however , it allows us to further investigate the microscopic nature of tt@xmath0 through its contribution to time - resolved spectroscopy . in particular , by reproducing 2des features associated with this state , we are able to shed light on its optical transitions towards higher - lying triplet states . in doing so , we address questions regarding the identification of these transitions , while at the same time describing important molecular properties that impact estimates of the associated signal strengths . this paper is organized as follows . after a few remarks on the observation of singlet fission through time - resolved spectroscopy , we review in sec . [ sec_theory ] the main aspects of the theory introduced in i. subsequently , the necessary extensions of the vibronic basis set and hamiltonian are discussed , followed by the theoretical framework employed to simulate 2des . the section closes with a summary of the associated parameters for pentacene crystals . in sec . [ sec_results ] , we present results obtained for crystalline pentacene through the model . first , static 2d spectra are compared to experimental measurements , as well as oscillatory spectra which serve to detect tt@xmath0 . what follows is an evaluation of the optical transitions between triplet states , through which tt@xmath0 becomes visible in time - resolved spectra , and a discussion of several material properties which affect the strengths of these transitions . we conclude in sec . [ sec_conclusions ] . in order to streamline the discussion , we begin by showing in fig . [ fig_fissionscheme ] a schematic representation of how sub - ps fission dynamics is typically observed within time - resolved spectroscopy . this illustration is minimal in the sense that it represents the many excited states found in fission materials by a maximally reduced set : a singlet ground state ( s@xmath1 ) , a singlet excited state ( s@xmath0 ) , a correlated triplet pair ( tt@xmath0 ) , and an analogues state that is higher in energy ( tt@xmath2 ) . these are assumed to be adiabatic states , and as such belong to the quantum basis that is spectroscopically detectable . in a similarly simplified fashion , time - resolved spectroscopic experiments can be regarded as consisting of two optical excitation events separated in time by some varying interval . the first event ( commonly referred to as `` pump '' ) excites the molecular system from its initial state s@xmath1 to the excited state s@xmath0 . subsequently , some non - adiabatic mechanism induces a transition from s@xmath0 to tt@xmath0 , after which a second excitation event ( the `` probe '' ) brings the system to the higher - lying pair state , tt@xmath2 . the observed signal associated with this second transition is used as a time - dependent probe of the fission process.@xcite we note that experiments using two - photon photoemission spectroscopy involved ionized states instead of tt@xmath2.@xcite we will exclude such states from our discussion , and as such limit ourselves to ( third - order ) purely optical techniques such as 2des and transient absorption . ) , singlet excited state ( s@xmath0 ) , and the low- and high - energy triplet pair states ( tt@xmath0 and tt@xmath2 ) . two optical excitation events are indicated with wiggling lines . the first event ( `` pump '' ) bring the molecular system from the ground state to s@xmath0 . this triggers the fission process ( arrow ) . the second event ( `` probe '' ) couples the product tt@xmath0 state to a triplet pair state that is higher in energy . together , the states span three different manifolds , between which transitions are only significant when induced by an optical pulse . the labels `` singly - excited '' and `` doubly - excited '' refer to the number of optical excitation events required to populate the manifolds starting from the ground state , and are not to be confused with the number of involved electron - hole pairs ( excitons ) . for example , tt@xmath0 is sometimes referred to as a double excitation , being composed of two excitons , but nevertheless resides in the singly - excited manifold as defined here . numbers indicate energies ( in ev ) obtained for pentacene crystals in our preceding article . also depicted is the vibrationally dressed triplet pair state ( tt@xmath3 ) , found to be quasi - resonant with s@xmath0 in bulk pentacene ( dash ) in ref . and in the preceding article . ] the states depicted in fig . [ fig_fissionscheme ] span three manifolds , referred to as ground state , singly - excited , and doubly - excited ( the last two names refer to the number of optical excitation events required to populate the manifolds starting from the ground state , not to the number of electron - hole pairs ) . electronic couplings between states from different manifolds are usually negligible , and hence transitions between them are only significant when induced by an electromagnetic field . nevertheless , rather than the minimal set of states depicted in fig . [ fig_fissionscheme ] , each manifold in reality consists of a large number of coupled states . in i , we characterized the excited states constituting the singly - excited manifold of crystalline pentacene , while making a connection to the simplified representation of singlet fission as shown in fig . [ fig_fissionscheme ] . this yielded the adiabatic energies reported in this figure , where it should be noted that tt@xmath0 was found to consist of a handful of nearly - degenerate states , in contrast to the single bright state that represents s@xmath0 . also depicted is the vibrationally dressed analogue of tt@xmath0 ( tt@xmath3 ) , which appears prominently in our results , and which was emphasized earlier by bakulin _ _ et al.__@xcite the underlying nearly - degenerate states are found to be quasi - resonant with s@xmath0 , and have been predicted to be of great importance to the fission dynamics in crystalline pentacene.@xcite finally , we note that the non - adiabatic dynamics referred to in fig . [ fig_fissionscheme ] will be a topic of a future study , and is not included in the modeling presented here . although the contribution of dynamical fluctuations are not present in the calculated spectra , our approach allows one to reproduce the detection of the correlated triplet pair state through a coherently prepared vibrational wavepacket,@xcite which is the primary focus of the present study . in i , we presented in detail a diabatic vibronic basis set tailored to provide a description of the singly - excited manifold . the electronic subspace of this basis comprises a microscopic set of singlet states ( s@xmath0 ) , triplet states ( t@xmath0 ) , cationic states ( c ) , and anionic states ( a ) , where triplets always come in pairs and cations and anions always appear together , such that all basis states have zero spin and are charge neutral . diabatic states are consistently denoted with lower case labels in order to differentiate them from adiabatic states such as s@xmath0 and tt@xmath0 , which in turn consist of mixtures of diabatic states of different character . beside the electronic degrees of freedom , the basis includes intramolecular vibrational states accompanying the electronic excitations . these are denoted @xmath4 , @xmath5 , @xmath6 , and @xmath7 , for the singlet , triplet , cation and anion , respectively , and are always described in the vibrational potential pertaining to the electronic state of the molecule under consideration . in addition , the basis contains purely vibrational excitations , where @xmath8 vibrational quanta reside at a molecule in its singlet ground state ( s@xmath1 ) . the initial steps of singlet fission take place entirely within the singly - excited manifold . when considering linear absorption , or when characterizing the involved excited states , restricting the basis to this manifold alone suffices , as was done in i. however , when considering time - resolved spectroscopic experiments on fission materials , the ground state manifold and doubly - excited manifolds must be included as well . similarly to i , we adapt the `` two - particle '' approximation,@xcite such that the ground state manifold is spanned by the states @xmath9 and @xmath10 here the first state represents a purely vibrational excitation involving @xmath11 quanta in the s@xmath1 potential of molecule @xmath12 . in the second state , such a vibration is accompanied by a vibrational excitation involving @xmath13 quanta located at site @xmath14 . ( note that @xmath15 is imposed in order to avoid double counting . ) additionally , there is the unique state denoted @xmath16 , in which all molecules are both electronically and vibrationally unexcited . the doubly - excited manifold , on the other hand , consists of triplet pairs in which one member resides in a higher - lying state . higher - lying states contributing to time - resolved spectroscopic measurements of singlet fission have been associated with the molecular excitations t@xmath17 and t@xmath18.@xcite we will nevertheless remain non - specific for the moment and consider for each set of calculations a single higher - lying state referred to as t@xmath2 , which suffices to account for the 2des measurements , while leaving a discussion on the nature of this state to sec . [ sec_triplettransitions ] . the doubly - excited triplet pairs are thus expressed as @xmath19 representing a t@xmath0 excitation at molecule @xmath12 , accompanied by @xmath5 vibrational quanta , and a t@xmath2 excitation at molecule @xmath14 involving @xmath20 quanta in the associated potential . similarly to pair excitations in the singly - excited manifold , we assume this multi - exciton state to be entangled in a zero spin configuration , following the symmetry - adapted linear combination analogous to that defined for the ( singly - excited ) correlated triplet pair state in ref . . the rationale behind this is that optical transitions are spin - conserving , and considering spin dephasing to be insignificant at the ultra - fast timescales considered in this work,@xcite only such overall singlet states can be populated . in principle , the doubly - excited manifold also includes pairs of singly - excited basis states , which can become populated via two optical excitations . however , for crystalline pentacene , the associated spectroscopic signals do not benefit from the large dipole moment associated with the transition between triplets ( as discussed in sec . [ sec_triplettdm ] ) , and are therefore much weaker than the signals associated with the doubly - excited triplet pairs . considering this , and the fact that such basis states would render the calculations prohibitively expensive , we have excluded such states from the doubly - excited manifold . from the diabatic basis , adiabatic states are obtained upon solving the time - independent schrdinger equation , @xmath21 , where @xmath22 denotes the hamiltonian of the material under consideration , and @xmath23 and @xmath24 represent an ( adiabatic ) eigenstate and the associated eigenenergy , respectively . ( similarly to i , we apply @xmath25 throughout this work . ) since we are concerned with three non - interacting excitation manifolds , the hamiltonian can be subdivided according to @xmath26 where the contributions with subscript 0 , 1 , and 2 act in the subspace corresponding to the ground state , singly - excited , and doubly - excited manifolds , respectively . for the ease of discussion , we will also explicitly differentiate between the eigenfunctions pertaining to these manifolds , denoting them as @xmath27 , @xmath28 , and @xmath29 , respectively . we refer to i for a formulation and discussion of the singly - excited hamiltonian @xmath30 , which includes the couplings and energies associated with the vibronic s@xmath0 , t@xmath0 , c and a states , and continue to formulate @xmath31 and @xmath32 . taking the energy of s@xmath1 as a reference , the ground state hamiltonian contains only the vibrational energy , @xmath33 , given by @xmath34 here , @xmath35 and @xmath36 represent the vibrational ladder operators pertaining to the s@xmath1 potential at molecule @xmath12 , and @xmath37 is the vibrational quantum . the doubly - excited hamiltonian , on the other hand , contains , in addition to vibrational energy , a contribution from electronic energy and a vibronic coupling term , @xmath38 here , @xmath39 is the diagonal energy of the diabatic pair state involving the higher - lying triplet ( assumed to be independent of the @xmath12 and @xmath14 combination ) . in the vibronic coupling term , @xmath40 runs over the molecular excitations t@xmath0 and t@xmath2 , and @xmath41 is the associated vibronic coupling strength , the square of which corresponds to the huang - rhys ( hr ) factor . it should be noted that both @xmath31 and @xmath32 are diagonal operators in the diabatic basis , and consequently , the adiabatic states in the associated manifolds are identical to the diabatic ones . however , whereas intermolecular couplings in the ground state manifold are typically very weak , substantial couplings between doubly - excited states are plausible . the effect of such couplings is touched upon in sec . [ sec_triplettdm ] . of central importance to spectroscopy , be it linear absorption or time - resolved optical techniques , is the transition dipole moment operator @xmath42 of the material under consideration ( bold notation is used throughout to refer to three - dimensional vector properties ) . similarly to the hamiltonian , we subdivide this operator into contributions associated with different ( pairs of ) excited state manifolds as @xmath43 the first term couples the ground state and the singly - excited manifolds through the optical transition from s@xmath1 to s@xmath0 , @xmath44 where @xmath45 is the associated transition dipole moment of molecule @xmath12 . in contrast , the singly - excited manifold is coupled to the doubly - excited manifold through optical transitions between t@xmath0 and t@xmath2 , @xmath46 where @xmath47 is the corresponding transition dipole moment . note that since @xmath48 exclusively contributes to linear absorption , only this part of the transition dipole moment operator was considered in i. however , as depicted in fig . [ fig_fissionscheme ] , an evaluation of 2des as presented here requires both contributions to @xmath42 to be included . since we are concerned with one - component crystals ( as opposed to cocrystals ) , all molecular transition dipole moments can be considered identical , aside from the description of the orientation . we can therefore set @xmath49 and @xmath50 , where the unit vector @xmath51 represents the direction of the i@xmath52f transition dipole moment at molecule @xmath12 , and @xmath53 is the associated ( @xmath12-independent ) magnitude . furthermore , for the sake of clarity , it is helpful to additionally formulate transition dipole moment operators for which the magnitudes are divided out , @xmath54 and @xmath55 such that the original transition dipole moment operators are retained through @xmath56 and @xmath57 . , @xmath58 , and @xmath59 , respectively ( with or without prime ) . on the left , @xmath60 , @xmath61 , and @xmath62 indicate the `` pump '' time , waiting time , and `` probe '' time . ] whereas linear absorption is readily calculated via fermi s golden rule , 2des is more naturally evaluated in a time - resolved framework , through the calculation of the associated optical response functions.@xcite these response functions derive from four light - matter interactions , separated in time by the intervals @xmath60 , @xmath61 , and @xmath62 . the first two interactions constitute the aforementioned `` pump '' event , whereas the second interaction pair makes up the `` probe '' event , while the time interval @xmath61 , the waiting time , is varied to dynamically study the molecular sample . what distinguishes 2des @xcite ( as well as its infrared analogue @xcite ) from transient absorption is the ability to simultaneously resolve the pump and probe frequencies . nevertheless , the underlying response functions are formally equivalent . the response functions of interest to singlet fission can be divided into four essentially different forms , based on fundamentally different ways in which the density matrix is affected by the four optical interactions . these are conveniently represented by means of double - sided feynman diagrams , describing the time - evolution of the molecular density matrix , as shown in fig . [ fig_diagrams ] . the first difference lies in the relative phase accumulations during the pump and probe events , which could have equal signs ( non - rephasing , nr ) or opposite signs ( rephasing , r ) . the second difference derives from the two physical processes that independently contribute to the spectral signal . one process , referred to as excited state absorption ( ea ) , is light absorption during the probe event by molecules that have been promoted to the singly - excited manifold by the pump event . the other relates to the bleaching of ground state absorption during the probe event , due to molecules that have been excited by the pump event , which is referred to as ground state bleach ( gb ) . under experimental conditions , it is fundamentally impossible to disentangle gb and ea signals , and the detected spectrum can at most be dissected into nr and r contributions . this is in marked contrast to numerical calculations , which proceed by separate evaluation of gb and ea , providing an easy venue for comparing the associated signals side - by - side . singlet fission studies which used transient absorption @xcite or 2des @xcite have monitored triplet transitions that occur via ea as a dynamic probe of the fission process . from this perspective , gb is an undesirable contributor which complicates the interpretation of the measurements due to spectral congestion . we further note that , in addition to ea and gb , a third process commonly contributes to time - resolved spectra , in which the probe event induces emission from singly - excited molecules . however , fission in bulk pentacene is known to depopulate the singlet state within @xmath6380 fs,@xcite and radiative transitions from triplet pair states are generally very weak . as a result , such stimulated emission is not observed for this material,@xcite or at most at very early times,@xcite and is therefore neglected in our model . ( for the same reasons , ea pathways involving transitions from the singlet state to higher - lying singlet excitations are neglected . ) the practical implementation of time - resolved spectroscopy relies on the realization of ultra - short laser pulses , particularly so for 2des , which implies a broad laser spectrum in the frequency domain . nevertheless , the spectral widths of fission materials typically exceed 1 ev , which is unreachable even by most 2des setups . a simple method to account for a finite laser bandwidth is to multiply the transition dipole moment operator with an adiabatic window function , @xmath64.@xcite accordingly , the transition dipole moment operator is replaced by a laser - corrected analogue according to @xmath65 where @xmath66 and @xmath67 run over all manifolds . we have adopted this method , while following the example of ref . by using a discrete window function @xmath68 , where @xmath69 is the heaviside step function , and @xmath70 and @xmath71 characterize the lower and upper bounds of the laser spectrum , respectively . although noting that more sophisticated approaches for incorporating the laser spectrum are available,@xcite we find the current approach sufficient for our purpose . the response functions depicted in fig . [ fig_diagrams ] are formulated as @xmath72 @xmath73 @xmath74 @xmath75 here , @xmath76 , @xmath77 , @xmath78 , and @xmath79 label the four pulse polarizations . the minus sign appearing in the ea response functions , resulting from an uneven number of optical interactions at the bra of the molecular density matrix , is representative of the absorptive character of the associated spectral signals . all response functions are formulated in the frame of reference of the molecular crystal , and @xmath80 refers to the @xmath76-polarized component of @xmath81 ( while the other transition dipole moment operators are labeled analogously ) . however , by having all pulse polarizations in the summation run independently over the unit vectors in the molecular frame ( @xmath40 , @xmath82 , and @xmath83 ) , and by weighing the contributions by the appropriate tensor values @xmath84,@xcite we effectively consider an isotropic collective of molecular clusters . this is used as a proxy for a polycrystalline material , as discussed in sec . [ sec_2des ] . lastly , @xmath85 is the molecular propagator , accounting for the time - evolution of the sample in between the ( impulsive ) light - matter interactions . note that @xmath86 ( identity matrix ) is omitted . fourier transforming the response functions over @xmath60 and @xmath62 yields the 2d spectra associated with waiting time @xmath61 as a function of the pump frequency @xmath87 and the probe frequency @xmath88 . most of the parameters for pentacene crystals used in the present article have been presented and discussed in i. we proceed to mention only those parameters that specifically pertain to 2des , and which have therefore not been presented in i. we note , however , that all of these parameters are concerned with t@xmath2 , the higher - lying triplet level included in our basis set . such high - energy states are challenging to evaluate through computations based on first principles , and equally difficult to characterize spectroscopically . as a result , relatively little _ a priori _ knowledge about them is available . we have therefore motivated the values adapted for these parameters based on the agreement of our calculations with 2des measurements in secs . [ sec_2des ] and [ sec_ft ] . accordingly , the high - energy triplet pair state energy is taken to be @xmath89 ev ( 28825 @xmath90 ) . for the hr factors associated with t@xmath2 , an _ ad hoc _ value of @xmath91 is applied , while varying values of this quantity are explored in sec . [ sec_triplettdm ] . the direction and magnitude of the t@xmath0 to t@xmath2 transition dipole moments are determined based on the crystal structure of pentacene,@xcite similar to the procedure for the s@xmath1 to s@xmath0 transition in i. whereas @xmath92 is oriented along the short axis of molecules @xmath12 , @xmath93 is found to be oriented along the long axis instead@xcite and adapted accordingly . regarding their magnitudes @xmath94 and @xmath95 , we first note that these quantities solely act as constants contributing to the response functions given by eq . [ eq_responses ] . specifically , the constant contributing to the gb pathways is given by @xmath96 , whereas the equivalent for ea is @xmath97 . hence , upon setting @xmath94 to an arbitrary value , the magnitude of the triplet transition effectively acts as a weighing of the ea signal relative to the gb analogue through its scaling to @xmath98 . for this reason , and in the light of a current inconclusiveness concerning the strength of the triplet transitions,@xcite we use @xmath95 as a free parameter , expressed in terms of @xmath94 , while leaving a thorough discussion of this quantity to sec . [ sec_triplettdm ] . in i , we evaluated the convergence of the photophysics of pentacene with respect to the applied crystal size , and found that the linear optical properties are well accounted for when limiting the crystal to 3@xmath993 unit cells extending in the crystallographic @xmath100-plane ( while imposing periodic boundary conditions ) . for pentacene , having 2 inequivalent molecules per unit cell , this amounts to 18 molecular sites . still , the vibronic basis set employed in our model scales rather unfavorably with the number of electronic degrees of freedom , yielding a total of 2430 states , despite an optimal truncation of the basis set . simulations of 2des involving this number of states are just within the computational capacities at our disposal , and we have adapted the crystal size accordingly . in doing so , the diabatic bases spanning the ground state and doubly - excited state manifolds are truncated analogously to the singly - excited basis discussed in i. specifically , the maximum t@xmath0 to t@xmath2 separation is set to 6.2 , the separation between the two - particle states in the ground state manifold is limited to 20 , and the total number of vibrational quanta per basis state is limited to 2 . linear absorption has proven indispensable to validate microscopic models of singlet fission.@xcite nevertheless , with the recently reported 2des measurements on polycrystalline pentacene , the opportunity to subject models to an even more stringent test arises . the ability of 2des to project the optical response of a molecular material to an additional spectral dimension allows one to directly resolve the correlations between excitations over time . on the one hand , non - oscillatory signals are representative of energy transfer pathways and couplings between excited states.@xcite on the other hand , oscillatory signals represent coherent dynamics.@xcite in 2des experiments on pentacene , the optical transition between the ground state and the correlated triplet pair state has been directly detected with the help of such quantum beats.@xcite before turning our attention to this spectroscopic result , we first focus on the non - oscillatory signals by comparing our simulations to the measurements of static 2d spectra . in calculating the static 2d spectra , the underlying response functions ( eq . [ eq_responses ] ) are evaluated while scanning the pump ( @xmath60 ) and probe ( @xmath62 ) times from 0 to 200 fs using a 1 fs time resolution . as a convenient way to realize spectral broadening , a multiplication by the exponential @xmath101 is carried out before performing the 2d fourier transforms , resulting in lorentzian peak profiles along @xmath87 and @xmath88 . in doing so , the lifetime parameter @xmath102 is adjusted such that the lorentzian width corresponds to 27 mev ( 214 @xmath90 ) . while evaluating the crystal , a weighted average is taken over all possible polarization sequences , as embodied by eq . [ eq_responses ] , effectively regarding the crystal as a member of an isotropic ensemble . in i , this approach was justified for linear absorption of polycrystalline pentacene , since the molecular s@xmath103s@xmath0 transitions are polarized in the @xmath100-plane , for which an effective isotropy occurs in the polycrystalline state . however , the t@xmath104t@xmath2 transition is oriented parallel to the crystallographic @xmath105-axis , a direction along which polycrystalline pentacene has been shown to be reasonably ordered.@xcite nevertheless , the 2des measurements were recorded by orienting the laser beams at a 50@xmath106 angle relative to this axis,@xcite in order to compromise between maximizing the t@xmath104t@xmath2 signal and minimizing the strong absorption cross - section in the @xmath100-plane ( due to extended in - plane crystal sizes).@xcite we argue that the isotropic calculations presented here , by its weighing of both weak in - plane and strong out - of - plane t@xmath104t@xmath2 signals , forms a reasonable proxy for such an experimental setup . furthermore , we point out that the 2des spectral features are , by construction , qualitatively insensitive to such polarization details , with quantitative differences anticipated only in the relative intensities of gb and ea signals . another factor that has only a quantitative impact on the features observed in 2des is the magnitude of the triplet transition dipole moment , @xmath95 , relative to that of the singlet transition . as argued in sec . [ sec_parms ] , this quantity effectively modulates the ea : gb signal ratio via its @xmath107 scaling . we accordingly utilize the magnitude of this quantity as a parameter to reach agreement with experiment regarding the weighing of oscillatory spectral features from gb and ea in sec . [ sec_ft ] , yielding @xmath108 . this value is significantly larger than the magnitude of @xmath109 used in the 2des simulations ref . , a difference that originates primarily from the much weaker tt@xmath0 oscillator strength obtained in i , which was found to be more realistic with respect to linear absorption features . in sec . [ sec_triplettdm ] , we highlight several physical factors that additionally impact the ea signal strength , and which allow to account for the 2des measurements using smaller values of @xmath110 . finally , we note that the quantitative accuracy of our modeling of the gb and ea signals is inherently limited , since aspects such as dynamical population of tt@xmath0 and realistic pulse shapes are not accounted for . before focusing on the spectral region near the energy of tt@xmath0 , we first compare static 2des calculated through our model with measurements in the higher - energy region where multiple bright exciton states render congested spectra , making for a challenging test case . shown in fig . [ fig_highenergy ] are experimental real - valued 2d spectra from ref . . the spectra were recorded in the nr and r phase sequences , at a waiting time @xmath111 fs , and using a laser bandwidth ranging from @xmath112 ev ( 16500 @xmath90 ) to @xmath113 ev ( 19000 @xmath90 ) . by setting the window function ( eq . [ eq_laserdipole ] ) accordingly , we obtain the calculated results also shown in fig . [ fig_highenergy ] . since the t@xmath104t@xmath2 transition lies outside this window , we instead included for these higher - energy region calculations a higher - energy t@xmath114 ( @xmath115 ) state using @xmath116 ev ( 35800 @xmath90 ) , based on a comparison with the experimental 2d spectra . despite not being otherwise parametrized to reproduce the 2d spectra , our model reaches a very high level of agreement , reproducing the dominating diagonal peak at 2.15 ev present in both the nr and r spectra , as well as the associated cross - peak at probe energy @xmath117 ev and the series of cross - peaks extending over the range of pump energies from @xmath118 ev up to 2.35 ev . the main discrepancies concern the spectral fine structure and the intensity of negative signal . based on our comparison between linear absorption bands calculated for a 3@xmath993 and a 6@xmath996 pentacene crystal in i , the former can be attributed to the finite crystal dimension used in the 2des calculations , for which the fine structure in the high - energy spectral region has not fully converged . the lack of negative signal , on the other hand , is mostly attributable to the absence of fission dynamics in our model , which , when included , would considerably enhance static ea contributions through a buildup of triplet product states . furthermore , the negative features appearing in the experimental spectra at a detection energy close to @xmath119 are likely artifacts due to the laser pulse that go beyond the windowing effect applied through eq . [ eq_laserdipole ] , being consistent with theoretical predictions from ref . ( see fig . 2 in that reference ) . we note , however , that most of the features are reproduced at least qualitatively . to s@xmath0 and tt@xmath0 . shown are 2d spectra at a waiting time of 80 fs . the presentation is analogous to fig . [ fig_highenergy ] . ] a similar comparison is shown in fig . [ fig_lowenergy ] for the 2d spectra recorded at 80 fs in the region from @xmath120 ev ( 12950 @xmath90 ) to @xmath121 ev ( 15800 @xmath90 ) , which covers the optical transition between the s@xmath1 and tt@xmath0 . in marked contrast to the higher - energy equivalents , both the static nr and r spectra show very little structure apart from a pronounced diagonal peak located at 1.85 ev and associated with s@xmath0 . when compared to our calculated results , this peak is somewhat red - shifted in the measurements , which is a known distortion owing to the finite pulse bandwidth.@xcite regardless , both the experimental and numerical spectra show no discernible features around @xmath122 ev , which corresponds to the energy of tt@xmath0 , implying that the s@xmath103tt@xmath0 transition between these states is indiscernible through static 2d spectra . however , as was first demonstrated in ref . , this transition can nonetheless be unveiled by fourier transforming the oscillatory 2d signal relative to the waiting time , yielding fourier transform amplitude maps . numerical fourier transform ( ft ) amplitude maps are derived from 2d spectra calculated at 5 fs intervals of the waiting time @xmath61 between 0 and 1500 fs . in order to keep the computational costs manageable , we have truncated @xmath60 and @xmath62 to 100 fs , which restricts the resolution of the conjugate energies @xmath87 and @xmath88 , but which is still sufficient to produce satisfactory spectral results . for each ( @xmath123 ) pair , complex - valued time traces are composed by taking the associated @xmath61-dependent signal from the 2d spectra . these traces are fitted to an exponential , whereupon a fast fourier transform is applied to the residue , the result of which forms a complex - valued frequency spectrum . the absolute value of this spectrum at a certain frequency @xmath124 , added to its analogue at @xmath125 , yields the ft amplitude map associated with @xmath124 as a function of @xmath87 and @xmath88 . mev , derived from the non - rephasing ( top row ) and rephasing ( bottom - row ) 2d spectra in the region covering the optical transitions from s@xmath1 to s@xmath0 and tt@xmath0 . measured results are shown on the left , whereas the calculated equivalents are shown on the right . all maps are normalized individually . ] the experimental fourier transform amplitude maps from ref . in the spectral region of tt@xmath0 are shown in fig . [ fig_ftmaps ] . the associated fourier transform frequency corresponds to @xmath126 mev ( 1360 @xmath90 ) , which is approximately the symmetric stretching vibration @xmath37 . interestingly , the signals resolved upon fourier transforming the optical response with respect to @xmath61 are obviously much richer in features than the static spectra shown in fig . [ fig_lowenergy ] would suggest . in the nr map , the features are roughly concentrated in two peaks in the diagonal region , labeled @xmath127 and @xmath128 in the numerical maps in order of descending pump energies . whereas @xmath128 appears as a single peak , @xmath127 seems to be composed of a strong peak at the diagonal overlapping a weaker feature slightly blue - shifted along @xmath87 . in the r map , on the other hand , two peaks are observable at off - diagonal locations . in both phasing sequences , the peak intensity at higher @xmath87 amounts to approximately three times the intensity of the lower @xmath87 peak . before turning our attention to the total nr and r maps resulting from our model , we first consider in fig . [ fig_ftmapsdissect ] the separate contributions from gb and ea to the nr oscillatory signal , which are inseparable under experimental conditions . these data substantiate that @xmath127 has two contributors , originating from gb and ea , whereas @xmath128 consists of a single feature owing to ea . moreover , since the ea signal is modulated by the triplet transition dipole moment through its scaling to @xmath98 , agreement in the relative intensities of @xmath127 and @xmath128 can be reached with the measurements by adjusting this parameter in our model . the matching of these intensities motivates @xmath108 . using this value , good overall agreement is reached for both the total r and nr maps , as can be seen in fig . [ fig_ftmaps ] . nevertheless , we highlight in sec . [ sec_triplettdm ] various molecular properties that additionally impact the @xmath127 and @xmath128 peak intensities , while leaving the oscillatory features qualitatively unaffected . by means of these properties , agreement for the ft amplitude maps can be reached using different values of @xmath95 . mev , derived from non - rephasing ground state bleach ( left ) and excited state absorption ( right ) signals . both are normalized to the calculated non - rephasing map shown in fig . [ fig_ftmaps ] . ] the spectral features observed in the gb and ea maps shown in fig . [ fig_ftmapsdissect ] are readily characterized based on the associated diagrams from fig . [ fig_diagrams ] combined with the excited state energies of crystalline pentacene as determined in i. these feynman pathways are shown in fig . [ fig_diagramsspec ] . we note that in addition to the vibrationless ground state s@xmath1 , the relevant diagrams include a single - quantum purely vibrational excitation . denoting this excitation as s@xmath129 , we point out that it represents a manifold of ( degenerate ) states which were referred to as @xmath130 in sec . [ sec_theory ] . the features observed in the r maps are similarly accounted for based on the diagrams , upon inverting for each diagram the first two light - matter interactions , and these are therefore not explicitly considered here . the characterization shown in fig . [ fig_diagramsspec ] bears strong similarity with the analysis presented in ref . based on phenomenological modeling , although we note that the ea contribution to peak @xmath127 was not identified in this earlier work . as in ref . , we identify @xmath128 as a direct probe of the optical transition between s@xmath1 and tt@xmath0 as it maps the associated energy difference onto the pump axis , and is free of overlapping gb contributions . and @xmath128 in the fourier transform amplitude maps associated with @xmath126 mev . the involved states are labeled in accordance with fig . [ fig_fissionscheme ] , including the vibrationless ground state s@xmath1 , whereas s@xmath129 refers to electronic ground state dressed with a single vibration . numbers indicate the energy differences ( in ev ) between the ket and bra states , which dictate the @xmath87 ( bottom ) and @xmath88 ( top ) energies at which the associated peaks appear in the maps . ] concordant with ref . , we find that the transition between s@xmath1 and tt@xmath0 exposes weak oscillations that are observed in the ft amplitude maps , in marked contrast to the non - oscillatory ( static ) signals which show no trace of this transition . in addition , note that in i we found that the contribution of this direct transition to linear absorption is negligible . in order to better understand these differences , we proceed to examine the strength of the oscillatory signal at @xmath128 , through which the s@xmath103tt@xmath0 transition is detected . in accordance with fig . [ fig_diagramsspec ] , this strength is proportional to @xmath131 disregarding the polarization details for the moment . the first term in this product involves the transition dipole moment between s@xmath1 and tt@xmath0 , which was found to be small in i. however , the remainder of the involved transition intensities may be sizable . the last term couples s@xmath1 and the vibrationally dressed correlated triplet pair state tt@xmath3 , which , in i , was found to have a transition dipole moment whose magnitude is approximately one third of that of the bright s@xmath103s@xmath0 transition . the remaining terms involve bright transitions between triplets . this transition dipole moment sequence yields much stronger signals than the imprint of tt@xmath0 on linear absorption , given by @xmath132 , which involves two weighings with the weak s@xmath103tt@xmath0 transition dipole moment , and which does not benefit from the involvement of additional bright transitions . finally , since the triplet transitions play an essential role in the prominence of @xmath128 , and the spectroscopic detection of the correlated triplet pair state in general , we dedicate the following two sections to a more complete discussion of their nature . among the open questions related to singlet fission , the attribution of optical transitions between triplet excitations is perhaps the most controversial.@xcite as mentioned in sec . [ sec_theory ] , studies have suggested two different higher - lying triplet excitations to be involved in time - resolved spectroscopic observations of singlet fission . in transient absorption measurements on polycrystalline pentacene , an ea feature centered at a probe energy of 2.34 ev has been attributed to the transition from t@xmath0 to t@xmath18,@xcite whereas a broad feature ranging from 1.24 to 1.77 ev has been associated with the t@xmath0 to t@xmath17 transition.@xcite calculations based on an approximate coupled - cluster singles and doubles ( ccsd ) , on the other hand , have predicted the t@xmath104t@xmath17 and t@xmath104t@xmath18 transitions to be located at 0.69 ev and 1.95 ev , respectively.@xcite interestingly , whereas the latter was calculated to have a rather strong oscillator strength , the former turned out to be essentially dark , which is consistent with measurements on pentacene in the gas phase@xcite and in solution.@xcite whether crystallization of pentacene can alter the triplet transitions to the extent that such could explain these apparent differences has remained unclear.@xcite in i , we found that the adiabatic energy of tt@xmath0 lies close to the diabatic energy @xmath133 as a result of weak effective couplings among local t@xmath0 excitations . couplings between higher - lying triplets can be larger ( as discussed in sec . [ sec_triplettdm ] ) , although spectral shifts between @xmath39 and the energy of tt@xmath2 larger than a tenth of an ev are not expected . another possible explanation for this disagreement is that overlapping gb features in transient absorption obscures the interpretation of the triplet transitions . this complication is largely overcome by 2des via its potential to simultaneously resolve pump and probe energies . in the 2des measurements on polycrystalline pentacene , two static ea features were identified at pump energies of 1.70 and 1.88 ev , growing in intensity to become resolvable at @xmath134 fs.@xcite these were associated with the triplet transitions involving t@xmath17 and t@xmath18 , respectively , despite energetic differences relative to both transient absorption measurements and ccsd calculations . although further experimental research in combination with first principles modeling is needed to resolve this controversy beyond a shadow of doubt , our theoretical results provide useful information to facilitate this , especially in its reproduction of the experimental ft amplitude maps . in order to reach agreement for these maps , the doubly - excited manifold in our model is accounted for by considering a single kind of t@xmath2 state , taking part in the pairs t@xmath0t@xmath2 that are parametrized by the associated energy @xmath89 ev . as a consequence , tt@xmath2 lies 1.85 ev above tt@xmath0 , which is close to the t@xmath104t@xmath18 transition energy calculated through ccsd . consequentially , t@xmath2 is naturally identified with t@xmath18 ( and tt@xmath2 with tt@xmath18 ) . moreover , tt@xmath2 happens to lie 1.68 ev above the tt@xmath3 energy as determined in i , hence , triplet transitions occur at 1.85 ev as well as 1.68 ev . we thus find a possible explanation for both static ea features observed in the 2des measurements ( within spectroscopic accuracy ) based on t@xmath18 , but without the involvement of t@xmath17 . we furthermore point out that the triplet transition at 1.85 ev overlaps with the absorption signal of s@xmath0 , which renders it difficult to resolve using transient absorption . instead , transitions from tt@xmath0 towards vibrationally dressed tt@xmath135 states are anticipated at somewhat higher energies ( and outside the spectral window centered around the transitions from s@xmath1 to s@xmath0 and tt@xmath0 ) , which could account for the presumed t@xmath104t@xmath18 features observed through that technique . the features at lower energies may likewise originate from vibronic transitions involving multiple - phonon levels in the tt@xmath0 state potential , which seems consistent with the broadness of these features . in conclusion , within the spectral window under consideration , the observed transition energies are most consistent with the scenario in which t@xmath18 contributes exclusively . regardless the nature of t@xmath2 , its substantial transition dipole moment relative to t@xmath0 is essential to the optical detection of the s@xmath103tt@xmath0 transition , as elaborated in sec . [ sec_ft ] . more specifically , the intensity of peak @xmath128 in the ft amplitude maps , through which this transition is detected , scales as @xmath136 owing to its ea origin . peak @xmath127 , on the other hand , results from ea and gb and has therefore a different ( and weaker ) dependence on the triplet transition dipole moment . a comparison of both peak intensities in sec . [ sec_ft ] yielded @xmath108 , when applied in the framework of our model . this value compares reasonably well to ccsd calculations of @xmath137 ( 10.24 debye , ref . ) and td - dft calculations of @xmath94 ( 1.16 debye , ref . ) , while it is significantly larger than @xmath138 found by the phenomenological modeling of 2des presented in ref . ( see discussion in sec . [ sec_2des ] ) . this is illustrative of the uncertainty regarding the relative values of @xmath94 and @xmath95 in the literature.@xcite for this reason , we bring forward several material properties which , alongside @xmath95 , impact the @xmath127 and @xmath128 peak intensities , and through which the experimental ft amplitude maps can be numerically accounted for using different values of @xmath95 . this exploration is by no means intended as an accurate determination of @xmath95 , for which the unknowns in the material properties under consideration and the approximations made in our modeling are too restrictive . rather , it serves to guide ongoing research by highlighting various factors that have to be taken into account when interpreting time - resolved spectroscopic measurements of singlet fission . instead of performing expensive simulations of ft amplitude maps based on 2d spectra , we proceed to use an approximate measure for the @xmath127 and @xmath128 peak intensities . first , the gb and ea signal strengths without inclusion of the transition dipole moment magnitudes are assessed by means of the coefficients @xmath139 and @xmath140 which are based on the feynman diagram analysis presented in fig . [ fig_diagramsspec ] . note that the strengths of both ea features are approximated by the same coefficient , eq . [ eq_p_ea ] , and that the summations explicitly include all adiabatic states underlying tt@xmath0 , tt@xmath3 , and tt@xmath2 , as well as the degenerate purely vibrational excitations represented by s@xmath129 . the peak intensities are expressed in terms of these coefficients and the triplet transition dipole moment according to @xmath141 where the slight location mismatch between ea and gb features underlying @xmath128 is neglected . since the coefficients can assume positive as well as negative values , taking the absolute value is necessary to yield the amplitude . based on the measurements , the amplitude of @xmath128 amounts to roughly one third of the @xmath127 amplitude , and in our numerical evaluation , @xmath95 is consistently adjusted such that this ratio is maintained throughout . [ fig_ftmapanalysis ] shows a bar graph containing the calculated ratios between @xmath142 and @xmath143 , and the values of @xmath95 obtained through this procedure . it is important to note that the triplet transition dipole moment obtained upon keeping the model unaltered is @xmath144 , close to the value applied in the foregoing calculations . this indicates that the approximate method employed here yields results consistent with explicit calculations of the ft amplitude maps . , green bars ) , for which the transition dipole moments are disregarded , and the corresponding value of @xmath95 ( in terms of @xmath94 ) required to reproduce the fourier transform amplitude maps shown in fig . [ fig_ftmaps ] through eq . [ eq_peaks ] ( blue bars ) . results calculated in the original model ( outlined in sec . [ sec_theory ] ) are compared with modified models in which the t@xmath0 and t@xmath2 huang - rhys factors are optimized to maximize the ea : gb peak intensity , in which dipole - dipole couplings between triplet transitions are included , and where all of the above have been applied . negative values of @xmath145 are denoted with a minus sign . ] the first two material properties that expectedly affect the peak intensities through their dependence on ea signals are the hr factors of t@xmath0 and t@xmath2 , since the triplet transition is weighed by the vibrational overlaps between initial and final states . the t@xmath0 hr factor was taken to be equal to the s@xmath0 hr factor , @xmath146 . as discussed in i , the rationale behind this is that calculations in tetracene have yielded a t@xmath0 hr factor comparable to that of s@xmath0,@xcite and that tetracene and pentacene are expected to behave in a similar fashion in this respect . nevertheless , given these considerations , the value @xmath147 is somewhat arbitrary and it is therefore worthwhile to explore alternative hr factors of t@xmath0 , under the constraint that values do not strongly deviate from @xmath148 . upon doing so , we confirm the dependence of the @xmath127 and @xmath128 peak intensities on this parameter , and find a significant enhancement in @xmath145 for @xmath149 . for the hr factor of t@xmath2 we have used an _ ad hoc _ value of @xmath91 in the absence of exemplary values reported in the literature . interestingly , upon varying @xmath150 ( while leaving @xmath147 unaltered ) we find a region of parameter space where the vibrational overlaps between t@xmath0 and t@xmath2 invert the sign of @xmath142 . consequently , the ea and gb coefficients contributing to peak @xmath127 interfere destructively , see eq . [ eq_peaks ] . as a result , the @xmath127 to @xmath128 peak intensity ratio can be reproduced using significantly smaller values of @xmath95 . this effect is nicely illustrated in fig . [ fig_ftmapanalysis ] , where the case is shown for which @xmath150 is set to 2.9 , the value that minimizes the triplet transition dipole moment ( referred to as `` optimal '' ) . although the enhancement in @xmath145 is modest , its sign change induces a sizable decrease in @xmath95 . the third material property that may contribute to the @xmath127 and @xmath128 peak intensities is dipole - dipole interaction between the two higher - lying triplet pair states that couple optically to the same pair state in the singly - excited manifolds . more specifically , when considered in the purely electronic representation , the two states @xmath151 and @xmath152 both couple to @xmath153 through @xmath95 , as a result of which excitation energy can be radiatively transferred between both doubly - excited states . this is very similar to the way `` conventional '' dipole - dipole interactions transfer singlet excitations through their optical coupling to the ground state . within the commonly practiced point - dipole approximation,@xcite the associated interaction strength scales as the product of the involved transition dipole moments , @xmath154 . hence , provided that @xmath95 is substantially larger than @xmath94 , dipole - dipole interactions among triplet excitations are expectedly much stronger than those among singlet excitations . the result is a delocalization of the doubly - excited state over two diabatic pair states @xmath151 and @xmath152 , which , in combination with vibronic coupling , leads to a redistribution of oscillator strengths of the triplet transitions , akin to vibronically coupled h- and j - aggregates.@xcite in order to explore the effect of triplet dipole - dipole couplings , we add the doubly - excited hamiltonian with the term @xmath155 where @xmath156 denotes the triplet dipole - dipole interaction between molecular sites @xmath12 and @xmath14 . in order to motivate the coupling values , we first recall from i that the dipole - dipole interaction strength among nearest - neighbor singlets in pentacene crystals is @xmath157 mev ( including the effect of dielectric screening ) . considering the square dependence of such couplings on the involved transition dipole moments , and the ( relative ) values quoted for @xmath95 in the literature,@xcite we apply a conservative value of -125 mev for nearest - neighbor couplings in the @xmath158 and @xmath159 directions ( other couplings need not be defined due to the spatial truncation of the triplet pairs ) . the minus sign is a consequence of the parallel polarization of the triplet transition dipole moments , for which the point - dipole approximation predicts couplings to be negative.@xcite as shown in fig . [ fig_ftmapanalysis ] , incorporating these triplet dipole - dipole interactions enhances @xmath145 by 50 % , with a concomitant decrease of the predicted value of @xmath95 . finally , we have calculated the ea and gb coefficients while implementing all three modifications discussed above , which yields a triplet transition dipole moment close to 8 . hence , by varying material properties that have remained largely undetermined , we find about a 50 % decrease in @xmath95 upon retaining an agreement for the ft amplitude maps . conversely , for a given @xmath95 , these material properties act alongside the triplet transition dipole moment in determining the strength of ea contributions relative to those from gb . to further substantiate the significance of these effects , further research is needed in order to pinpoint the triplet hr factors and magnitude of triplet dipole - dipole couplings . we note , however , that the material properties apart from @xmath160 pertain to the doubly - excited manifold , and as such are not involved in the actual fission dynamics . instead , their main relevance is centered around the spectroscopic observation of singlet fission . the first article in this series formulated a vibronic exciton theory of singlet fission that encompasses a microscopic set of interacting electronic and vibrational degrees of freedom , treated at a non - perturbative level . in the present article , we outlined an extension of this theory which enables us to simulate 2des of fission materials . as such , this work follows up on the recently reported first application of this optical technique to study singlet fission , and in particular on the direct detection of the optical transition between the ground state s@xmath1 and the correlated triplet pair state tt@xmath0 in polycrystalline pentacene . firstly , our simulation scheme allowed us to validate our model through a comparison of simulated 2d spectra to these measurements . the overall agreement was found to be quite good , providing a firm base for the applicability of our theory . secondly , through a detailed analysis of our spectral data , we obtained information about tt@xmath0 and the associated photophysics that is not directly accessible through experiments . special attention was paid to the optical transitions between tt@xmath0 and higher - lying triplet states tt@xmath2 , through which the correlated triplet pair state is commonly observed in time - resolved spectroscopy . we addressed a controversy in the literature regarding the nature of the involved higher - lying triplet states , for which different identifications , transition energies , and transition dipole moments have been reported . within the spectral range of the ground state transitions towards tt@xmath0 and s@xmath0 , we found that all experimental features are accounted for by including in our model only a single tt@xmath2 state , using a transition energy consistent with high - level calculations for @xmath161.@xcite based on agreement with 2des , we find that the associated transition dipole moment , @xmath95 , is 8 to 15 times larger than the singlet transition dipole moment @xmath94 . we nevertheless emphasize that our determination of @xmath95 is qualitative rather than quantitative . in addition , we have brought forward several material properties that impact the optical transition strength between tt@xmath0 and tt@xmath2 in addition to @xmath95 , and which consequently interfere with a direct determination of this parameter based on spectroscopy . among the contributing factors , we have found the hr factors of t@xmath0 and t@xmath2 to be of importance , as is the dipole - dipole interaction between higher - lying triplets . most of the aspects discussed here pertain to the optical transition between tt@xmath0 and tt@xmath2 , which is of critical importance to the detection of singlet fission through time - resolved spectroscopy . nevertheless , these aspects do not impact the process of fission itself . having passed the bar set by successfully comparing simulated 2d spectra to experimental results , we will return to this process in a forthcoming article , where dynamical simulations of fission materials will be presented . in doing so , we will focus in particular on the role of high - frequency vibrational modes in facilitating ultra - fast exciton multiplication in crystalline pentacene , with the aim of unraveling principles that are common to fission materials in general . the authors wish to thank artem bakulin for sharing data of the 2des measurements on polycrystalline pentacene . r.t . acknowledges the netherlands organisation for scientific research nwo for support through a rubicon grant . d.r.r . acknowledges funding from nsf grant no . che1464802 . 59ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty http://scitation.aip.org/content/aip/journal/jap/32/3/10.1063/1.1736034 [ * * , ( ) ] http://scitation.aip.org/content/aip/journal/jap/100/7/10.1063/1.2356795 [ * * , ( ) ] http://dx.doi.org/10.1021/cr1002613 [ * * , ( ) ] http://dx.doi.org/10.1146/annurev-physchem-040412-110130 [ * * , ( ) ] http://dx.doi.org/10.1038/nphys3241 [ * * , ( ) ] http://dx.doi.org/10.1038/nchem.2371 [ * * , ( ) ] http://dx.doi.org/10.1038/nchem.2665 [ * * , ( ) ] http://dx.doi.org/10.1063/1.3495764 [ * * , ( ) ] http://dx.doi.org/10.1063/1.3664630 [ * * , ( ) ] http://dx.doi.org/10.1021/ja301683w [ * * , ( ) ] http://dx.doi.org/10.1021/ja300504t [ * * , ( ) ] http://dx.doi.org/10.1021/ja1042462 [ * * , ( ) ] http://pubs.acs.org/doi/abs/10.1021/ja201688h [ * * , ( ) ] http://link.aps.org/doi/10.1103/physrevb.84.195411 [ * * , ( ) ] http://dx.doi.org/10.1021/ja2080482 [ * * , ( ) ] http://dx.doi.org/10.1021/acs.jpclett.6b00947 [ * * , ( ) ] http://science.sciencemag.org/content/334/6062/1541 [ * * , ( ) ] http://dx.doi.org/10.1038/nchem.1436 [ * * , ( ) ] http://link.aps.org/doi/10.1103/physrevlett.99.176402 [ * * , ( ) ] http://link.aps.org/doi/10.1103/physrevb.79.235318 [ * * , ( ) ] http://link.aps.org/doi/10.1103/physrevlett.102.017401 [ * * , ( ) ] link://www.sciencedirect.com / science / article / pii / s0921452609005614 [ * * , ( ) ] @noop * * , ( ) http://dx.doi.org/10.1039/f29726801928 [ * * , ( ) ] http://www.sciencedirect.com/science/article/pii/002228527490085x [ * * , ( ) ] http://scitation.aip.org/content/aip/journal/jcp/129/21/10.1063/1.3023118 [ * * , ( ) ] http://www.sciencedirect.com/science/article/pii/s0009261498011403 [ * * , ( ) ] @noop _ _ ( , ) http://www.annualreviews.org/doi/abs/10.1146/annurev.physchem.54.011002.103907 [ * * , ( ) ] http://pubs.acs.org/doi/abs/10.1021/cr078377b [ * * , ( ) ] http://dx.doi.org/10.1038/nature03429 [ * * , ( ) ] http://www.pnas.org/content/103/34/12672.abstract [ * * , ( ) ] http://dx.doi.org/10.1038/nature05678 [ * * , ( ) ] http://dx.doi.org/10.1038/nature08811 [ * * , ( ) ] http://dx.doi.org/10.1038/nchem.2005 [ * * , ( ) ] http://dx.doi.org/10.1038/nphys3017 [ * * , ( ) ] http://dx.doi.org/10.1038/nchem.2525 [ * * , ( ) ] http://science.sciencemag.org/content/323/5912/369 [ * * , ( ) ] http://dx.doi.org/10.1038/nchem.1834 [ * * , ( ) ] @noop * * , ( ) http://scitation.aip.org/content/aip/journal/jcp/116/13/10.1063/1.1446034 [ * * , ( ) ] http://scitation.aip.org/content/aip/journal/jcp/138/11/10.1063/1.4794425 [ * * , ( ) ] http://www.sciencedirect.com/science/article/pii/s0009261497000560 [ * * , ( ) ] http://dx.doi.org/10.1021/jp9813286 [ * * , ( ) ] http://link.aps.org/doi/10.1103/physrevb.74.024303 [ * * , ( ) ] http://dx.doi.org/10.1021/acs.jpca.5b10312 [ * * , ( ) ] http://www.sciencedirect.com/science/article/pii/s0301010401002324 [ * * , ( ) ] http://dx.doi.org/10.1002/(sici)1521-3765(19991105)5:11<3399::aid-chem3399>3.0.co;2-v [ * * , ( ) ] http://scitation.aip.org/content/aip/journal/jcp/136/5/10.1063/1.3677839 [ * * , ( ) ] http://dx.doi.org/10.1021/ja208431r [ * * , ( ) ] http://scitation.aip.org/content/aip/journal/jcp/134/20/10.1063/1.3590871 [ * * , ( ) ] http://link.aps.org/doi/10.1103/physrevlett.110.226402 [ * * , ( ) ] http://scitation.aip.org/content/aip/journal/jcp/141/7/10.1063/1.4892793 [ * * , ( ) ] http://dx.doi.org/10.1021/acs.jpcc.5b07163 [ * * , ( ) ] http://dx.doi.org/10.1021/cm049563q [ * * , ( ) ] http://link.aps.org/doi/10.1103/physrevb.70.125401 [ * * , ( ) ] http://dx.doi.org/10.1021/acs.jpclett.5b02249 [ * * , ( ) ] http://dx.doi.org/10.1063/1.461317 [ * * , ( ) ] http://dx.doi.org/10.1021/ar900233v [ * * , ( ) ]

markov chain monte carlo ( mcmc ) methods allow to sample a distribution known up to a multiplicative constant . classical mcmc samplers are known to have very poor mixing properties when sampling multimodal distributions . the equi - energy sampler is an interacting mcmc sampler proposed by kou , zhou and wong in 2006 to sample difficult multimodal distributions . this algorithm runs several chains at different temperatures in parallel , and allow lower - tempered chains to jump to a state from a higher - tempered chain having an energy ` close ' to that of the current state . a major drawback of this algorithm is that it depends on many design parameters and thus , requires a significant effort to tune these parameters . in this paper , we introduce an adaptive equi - energy ( aee ) sampler which automates the choice of the selection mecanism when jumping onto a state of the higher - temperature chain . we prove the ergodicity and a strong law of large numbers for aee , and for the original equi - energy sampler as well . finally , we apply our algorithm to motif sampling in dna sequences . keywords : interacting markov chain monte carlo , adaptive sampler , equi - energy sampler , ergodicity , law of large numbers , motif sampling . author s email addresses : first-name.last-name@telecom-paristech.fr this work is partially supported by the french national research agency , under the program anr-08-blan-0218 bigmc .

1207.0662

markov chain monte carlo ( mcmc ) methods are well - known tools for sampling a target distribution @xmath0 known up to a multiplicative constant . mcmc algorithms sample @xmath0 by constructing a markov chain admitting @xmath0 as unique invariant distribution . a canonical example is the the metropolis - hastings algorithm @xcite : given the current value @xmath1 of the chain @xmath2 , it consists in proposing a move @xmath3 under a proposal distribution @xmath4 . this move is then accepted with probability @xmath5 \eqsp,\end{aligned}\ ] ] where @xmath6 stands for @xmath7 ; otherwise , @xmath8 . it is known that the efficiency of mcmc methods depends upon the choice of the proposal distribution @xcite . for example , when sampling multi - modal distributions , a metropolis - hastings algorithm with @xmath4 equal to a gaussian distribution centered in @xmath1 tends to be stuck in one of the modes . so the convergence of such an algorithm will be slow , and the target distribution will not be correctly approximated unless a huge number of points is sampled . efficient implementations of mcmc rely on a strong expertise of the user in order to choose a proposal kernel and , more generally , design parameters adapted to the target @xmath0 . this is the reason why _ adaptive _ and _ interacting _ mcmc methods have been introduced . adaptive mcmc methods consist in choosing , at each iteration , a transition kernel @xmath9 among a family @xmath10 of kernels with invariant distribution @xmath0 : the conditional distribution of @xmath11 given the past is @xmath12 where the parameter @xmath13 is chosen according to the past values of the chain @xmath14 . from the pioneering adaptive metropolis algorithm of @xcite , many adaptive mcmc have been proposed and successfully applied ( see the survey papers by @xcite , @xcite , @xcite for example ) . interacting mcmc methods rely on the ( parallel ) construction of a family of processes with distinct stationary distributions ; the key behind these techniques is to allow interactions when sampling these different processes . at least one of these processes has @xmath0 as stationary distribution . the stationary distributions of the auxiliary processes are chosen in such a way that they have nice convergence properties , hoping that the process under study will inherit them . for example , in order to sample multi - modal distributions , a solution is to draw auxiliary processes with target distributions equal - up to the normalizing constant - to tempered versions @xmath15 , this solution is the basis of the parallel _ tempering algorithm _ @xcite , where the states of two parallel chains are allowed to swap . following this tempering idea , different interacting mcmc algorithms have been proposed and studied so far @xcite . the _ equi - energy sampler _ of kou , zhou and wong @xcite is an example of such interacting mcmc algorithms . @xmath17 processes are sampled in parallel , with target distributions ( proportional to ) @xmath18 , @xmath19 . the first chain @xmath20 is usually a markov chain ; then @xmath21 is built from @xmath22 as follows : with a fixed probability @xmath23 , the current state @xmath24 is allowed to jump onto a past state of the auxiliary chain @xmath25 , and with probability @xmath26 , @xmath24 is obtained using a local " mcmc move ( such as a random walk metropolis step or a metropolis - adjusted langevin step ) . this mechanism includes the computation of an acceptance ratio so that the chain @xmath21 will have @xmath27 as target density . as the acceptance probability of such a jump could be very low , only jumps toward selected past values of @xmath22 , namely those with an _ energy _ close to that of the current state @xmath24 , are allowed . this selection step allows higher acceptance rates of the jump , and a faster convergence of the algorithm is expected . the equi - energy sampler has many design parameters : the interacting probability @xmath23 , the number @xmath17 of parallel chains , the temperatures @xmath28 , @xmath29 and the selection function . it is known that all of these design parameters play a role on the efficiency of the algorithm . @xcite suggest some values for all these parameters , designed for practical implementation and based on empirical results on some simple models . @xcite discuss the choice of the interacting probability @xmath23 in similar contexts ; @xcite discuss the choice of the temperatures @xmath30 of the chains for the parallel tempering algorithm . recently , an algorithm combining parallel tempering with equi - energy moves have been proposed by @xcite . in this paper , we discuss the choice of the energy rings and the selection function , when the jump probability @xmath23 , the number @xmath17 of auxiliary processes and the temperatures are fixed . we introduce a new algorithm , called _ adaptive equi - energy sampler _ in which the selection function is defined adaptively based on the past history of the sampler . we also address the convergence properties of this new sampler . different kinds of convergence of adaptive mcmc methods have been addressed in the literature : convergence of the marginals , the law of large numbers ( lln ) and central limit theorems ( clt ) for additive functionals ( see e.g. @xcite for convergence of the marginals and weak lln of general adaptive mcmc , @xcite or @xcite for lln and clt for adaptive metropolis algorithms , @xcite and @xcite for convergence of the marginals , lln and clt for general adaptive mcmc algorithms - see also the survey paper by @xcite ) . there are quite few analysis of the convergence of interacting mcmc samplers . the original proof of the convergence of the equi - energy sampler in @xcite ( resp . @xcite ) contains a serious gap , mentioned in @xcite ( resp . @xcite ) . @xcite established a strong lln of a simplified version of the equi - energy sampler , in which the number of levels is set to @xmath31 and the proposal during the interaction step are drawn uniformly at random in the past of the auxiliary process . finally , fort , moulines and priouret @xcite established the convergence of the marginals and a strong lln for the same simplified version of the equi - energy sampler ( with no selection ) but have removed the limitations on the number of parallel chains . the paper addresses the convergence of an interacting mcmc sampler in which the proposal are selected from energy rings which are constructed adaptively at each levels . in this paper , we obtain the convergence of the marginals and a strong lln of a smooth version of the equi - energy sampler and its adaptive variant . we illustrate our results in several difficult scenarios such as sampling mixture models with well - separated " modes and motif sampling in biological sequences . the paper is organized as follows : in section [ presentation ] , we derive our algorithm and set the notations that are used throughout the paper . the convergence results are presented in section [ main : results ] . finally , section [ illustration ] is devoted to the application to motif sampling in biological sequences . the proofs of the results are postponed to the appendix . let @xmath32 be a measurable polish state space and @xmath33 be a markov transition kernel on @xmath32 . @xmath33 operates on bounded functions @xmath34 on @xmath35 and on finite positive measures @xmath36 on @xmath37 : @xmath38 the @xmath39-iterated transition kernel @xmath40 , @xmath41 is defined by : @xmath42 by convention , @xmath43 is the identity kernel . for a function @xmath44 , we denote by @xmath45 the v - norm of a function @xmath46 : @xmath47 if @xmath48 , this norm is the usual uniform norm . let @xmath49 . we also define the v - distance between two probability measures @xmath50 and @xmath51 by : @xmath52 when @xmath48 , the v - distance is the total - variation distance and will be denoted by @xmath53 . let @xmath54 be a measurable space , and @xmath55 be a family of markov transition kernels ; @xmath56 can be finite or infinite dimensional . it is assumed that for all @xmath57 , @xmath58 is @xmath59))$]-measurable , where @xmath60)$ ] denotes the borel @xmath61-field on @xmath62 $ ] . let @xmath0 be the probability density of the target distribution with respect to a dominating measure @xmath63 on @xmath32 . in many applications , @xmath0 is known up to a multiplicative constant ; therefore , we will denote by @xmath64 the ( unnormalized ) density . we denote by @xmath33 the metropolis - hastings kernel with proposal density kernel @xmath65 and invariant distribution @xmath0 defined by : @xmath66 where @xmath67 is the acceptance ratio given by @xmath68 the equi - energy ( ee ) sampler proposed by @xcite exploits the fact that it is often easier to sample a tempered version @xmath69 , @xmath70 , of the target distribution than @xmath0 itself . this is why the algorithm relies on an auxiliary process @xmath71 , run independently from @xmath72 and admitting @xmath69 as stationary distribution ( up to a normalizing constant ) . this mechanism can be repeated yielding to a multi - stages equi - energy sampler . we denote by @xmath17 the number of processes run in parallel . let @xmath73 . choose @xmath17 temperatures @xmath74 and set @xmath75 ; and @xmath17 mcmc kernels @xmath76 such that @xmath77 . @xmath17 processes @xmath78 , @xmath79 , are defined by induction on the probability space @xmath80 . the first auxiliary process @xmath81 is a markov chain , with @xmath82 as transition kernel . given the auxiliary process @xmath22 up to time @xmath39 , @xmath83 , and the current state @xmath84 of the process of level @xmath85 , the equi - energy sampler draws @xmath86 as follows : * ( metropolis - hastings step ) with probability @xmath87 , @xmath88 . * ( equi - energy step ) with probability @xmath23 , the algorithm selects a state @xmath89 from the auxiliary process having an energy close to that of the current state . an acceptance - rejection ratio is then computed and if accepted , @xmath90 ; otherwise , @xmath91 . in practice , @xcite only apply the equi - energy step when there is at least one point in each ring . in @xcite , the distance between the energy of two states is defined as follows . consider an increasing sequence of positive real numbers @xmath92 if the energies of two states @xmath93 and @xmath94 belong to the same energy ring , i.e. if there exists @xmath95 such that @xmath96 , then the two states are said to have `` close energy '' . the choice of the energy rings is most often a difficult task . as shown in figure [ comparaison_10d][right ] , the equi - energy sampler is inefficient when the energy rings are not appropriately defined . the efficiency of the sampler is increased when the variation of @xmath64 in each ring is small enough so that the equi - energy move is accepted with high probability . we propose to modify the equi - energy sampler by adapting the energy rings on the fly " , based on the history of the algorithm . our new algorithm , so called _ adaptive equi - energy _ sampler ( aee ) is similar to the equi - energy sampler of @xcite except for the equi - energy step , which relies on adaptive boundaries of the rings . for the definition of the process @xmath21 , @xmath97 , adaptive boundaries computed from the process @xmath22 are used . for a distribution @xmath98 in @xmath56 , denote by @xmath99 , @xmath100 the bounds of the rings , computed from r.v . with distribution @xmath98 ; by convention , @xmath101 . define the associated energy rings @xmath102 for @xmath103 . we consider selection functions @xmath104 of the form @xmath105 where @xmath106 measures the distance between @xmath107 and the ring @xmath108 . by convention @xmath109 if @xmath110 . we finally introduce a set of selection kernels @xmath111 for all @xmath112 defined by @xmath113 where @xmath114 @xmath115 is associated to the equi - energy step when defining @xmath21 : a draw under the selection kernel proportional to @xmath116 is combined with an acceptance - rejection step . the acceptance - rejection step is defined so that when @xmath117 , @xmath18 is invariant for @xmath115 @xcite . this equi - energy step is only allowed when each ring contains at least one point ( of the auxiliary process @xmath22 up to time @xmath39 ) . we therefore introduce , for all positive integer @xmath118 , the set @xmath119 : @xmath120 with these notations , aee satisfies for any @xmath41 and @xmath121 , @xmath122 = { \mathbb{e}}[f(y_{n+1}^{(k)})|y_n^{(k ) } , y_m^{(k-1 ) } , 1 \leq m \leq n ] = p_{{\theta}_n^{(k-1)}}^{(k)}f(y_n^{(k ) } ) \eqsp,\ ] ] where @xmath123 is the filtration defined by @xmath124 ; the transition kernel is given by @xmath125 and for @xmath97 , @xmath126 and @xmath127 is the empirical distribution @xmath128 different functions @xmath129 can be chosen . for example , the function given by @xmath130 yields to a selection function @xmath131 such that @xmath132 iff @xmath133 are in the same energy ring and @xmath134 otherwise . in this case , the acceptance - rejection ratio @xmath135 is equal to @xmath136 upon noting that by definition of the proposal kernel , the points @xmath93 and @xmath94 are in the same energy ring . by using this hard " distance during the equi - energy jump , all the states of the auxiliary process having their energy in the same ring as the energy of the current state are chosen with the same probability , while the other auxiliary states have no chance to be selected . other functions @xmath129 could be chosen , such as soft " selections of the form @xmath137 where @xmath138 is fixed . with this `` soft '' distance , given a current state @xmath24 , the probability for each auxiliary state @xmath139 , @xmath140 , to be chosen is proportional to @xmath141 . then , the soft " selection function allows auxiliary states having an energy in a @xmath142-neighborhood of the energy ring of @xmath143 to be chosen , as well as states having their energy in this ring . nevertheless , this selection function yields an acceptance - rejection ratio @xmath144 which may reveal to be quite costly to evaluate . the asymptotic behavior of aee will be addressed in section [ main : results ] . the intuition is that when the empirical distribution @xmath145 of the auxiliary process of order @xmath146 converges ( in some sense ) to @xmath147 , the process @xmath148 will behave ( in some sense ) as a markov chain with transition kernel @xmath149 . to highlight the interest of our algorithm , we consider toy examples : the target density @xmath0 is a mixture of @xmath150-valued gaussianschreck / index.html ] . this model is known to be difficult , as illustrated ( for example ) in @xcite for a random walk metropolis - hastings sampler ( srwm ) , an ee - sampler and a parallel tempering algorithm . indeed , if the modes are well separated , a metropolis - hastings algorithm using only local moves " is likely to remain trapped in one of the modes for a long - period of time . in the following , aee is implemented with ring boundaries computed as described in section [ ring : construction ] . figure [ comparaison].(a ) displays the target density @xmath0 and the simulated one for three different algorithms ( srwm , ee and aee ) in one dimension . the histograms are obtained with @xmath151 samples ; for ee and aee , the probability of interaction is @xmath152 , the number of parallel chains is equal to @xmath153 and the number of rings is @xmath154 . for the adaptive definition of the rings in aee , we choose the hard " selection ( [ eq : hard - distance ] ) and the construction of the rings defined in section [ ring : construction ] . in the same vein , figure [ comparaison_2d ] displays the points obtained by the three algorithms when sampling a mixture of two gaussian distributions in two dimensions . as expected , in both figures , srwm never explores one of the modes , while ee and aee are far more efficient . to compare ee and aee in a more challenging situation , we consider the case of a mixture with two components in ten dimensions . we run ee and aee with @xmath155 parallel chains with respective temperatures @xmath156 , the probability of jump @xmath23 is equal to @xmath157 , and the number of rings is @xmath158 . both algorithms are initialized in one of the two modes of the distribution . for the metropolis - hastings step , we use a symmetric random walk with gaussian proposal ; the covariance matrix of the proposal is of the form @xmath159 where @xmath160 is calibrated so that the mean acceptance rate is approximatively @xmath161 . figure [ comparaison_10d ] displays , for each algorithm , the @xmath162-norm of the empirical mean , averaged over @xmath163 independent trajectories , as a function of the length of the chains . in order to show that the efficiency of ee depends crucially upon the choice of the rings , we choose a set of boundaries so that in practice , along one run of the algorithm , some of the rings are never reached . figure [ comparaison_10d](a ) compares ee and aee in this extreme case : even after @xmath164 iterations , all of the equi - energy jumps are rejected for the ( non - adaptive ) ee , and the algorithm is trapped in one of the modes . this does not occur for aee , and the @xmath162-error tends to zero as the number of iterations increases . this illustrates that our adaptive algorithm avoids the poor behaviors that ee can have when the choice of its design parameters is inappropriate . we now run ee in a less extreme situation : we choose ( fixed ) energy rings so that the sampler can jump more easily than in the previous experiment between the modes . figure [ comparaison_10d](b ) illustrates that the adaptive choice of the energy rings speeds up the convergence , as it makes the equi - energy jumps be more often accepted . to have a numerical comparison , the equi - energy jumps were accepted about ten times more often for aee than for ee . [ cols="^,^ " , ] in this section , the convergence of the @xmath17-stages adaptive equi - energy sampler is established . in order to make the proof easier , we consider the case when the distance function @xmath129 in the definition of the selection function ( [ eq : fonctionselectionee ] ) is given by ( [ eq : soft : selection : function ] ) . @xcite provide sufficient conditions for the convergence of the marginals and the strong lln ( s - lln ) of interacting mcmc samplers . we use their results and show the convergence of the marginals i.e. @xmath165 = \pi(f ) \eqsp,\ ] ] for any continuous bounded functions @xmath34 . note that this implies that this limit holds for any indicator function @xmath166 such that @xmath167 where @xmath168 denotes the boundary of @xmath169 ( * ? ? ? * theorem 2.1 ) . we then establish the s - lln : for a wide class of continuous ( un)bounded functions @xmath34 , @xmath170 our results are established for target distributions @xmath0 satisfying [ ee1 ] a. [ ee1:a ] @xmath0 is the density of a probability distribution on the measurable polish space @xmath171 and @xmath172 and for any @xmath173 $ ] , @xmath174 . b. [ ee1:b ] @xmath0 is continuous and positive on @xmath35 . usually , the user knows @xmath0 up to a normalizing constant : hereafter , @xmath64 will denote this available ( unnormalized ) density . as in @xcite , we first introduce a set of conditions that will imply the geometric ergodicity of the kernels @xmath175 , and the existence of an invariant probability measure for @xmath175 ( see conditions e[ee3 ] ) . we finally introduce conditions on the boundaries of the adaptive energy rings ( see conditions e[ee : ring ] ) . examples of boundaries satisfying e[ee : ring ] and computed from quantile estimators are given in section [ ring : construction ] ( see also @xcite for stochastic approximation - based adapted boundaries ) . convergence of adaptive and interacting mcmc samplers is addressed in the literature by assuming containment conditions and diminishing adaptations ( so called after @xcite ) . assumptions e[ee3 ] is the main tool to establish a ( generalized ) containment condition . in our algorithm , the adaptation mechanism is due to ( a ) the interaction with an auxiliary process and ( b ) the adaption of the rings . therefore , assumptions e[ee3 ] and e[ee : ring ] are related to the diminishing adaptation condition ( see e.g. lemma [ lemme : cv : dv : proba ] in section [ secproof : ergo_ees ] ) . [ ee3 ] for each @xmath176 : a. [ ee3:a ] @xmath177 is a @xmath178-irreducible transition kernel which is feller on @xmath179 and such that @xmath180 . b. [ ee3:b ] there exist @xmath181 , @xmath182 and @xmath183 such that @xmath184 with @xmath185 by convention , @xmath186 . c. [ ee3:c ] for all @xmath187 , the sets @xmath188 are 1-small for @xmath177 . note that by definition of @xmath189 and e[ee1][ee1:a ] , @xmath190 and @xmath191 . e[ee3 ] is satisfied for example if for each @xmath85 , @xmath177 is a symmetric random walk metropolis hastings kernel ; and @xmath0 is a sub - exponential target density @xcite . in our algorithm , @xmath81 is a markov chain with transition kernel @xmath82 . as discussed in @xcite[chapters 13 and 17 ] , e[ee3 ] is sufficient to prove ergodicity and a s - lln for @xmath81 . e[ee3 ] also implies uniform @xmath192-moments for @xmath81 . these results , which initializes our proof by recurrence of the convergence for the process number @xmath17 , is given in proposition [ prop : lgn : first ] . define the probability distributions @xmath193 [ prop : lgn : first ] assume e[ee1][ee1:a ] , e[ee3 ] and @xmath194 < \infty$ ] . then , a. [ ee2:c ] for all bounded measurable functions @xmath34 , @xmath195 = { \theta}_\star^{(1)}(f)$ ] . b. [ ee2:a ] @xmath196 , and for any measurable function @xmath34 in @xmath197 , @xmath198 a.s . c. [ ee2:b ] @xmath199 < \infty$ ] . [ ee : ring ] a. [ ring : c ] for any @xmath200 , @xmath201 . b. [ ring : a ] for any @xmath202 and @xmath203 , @xmath204 w.p.1 c. [ ring : b ] there exists @xmath205 such that for any @xmath206 , any @xmath207 , and any @xmath208 , @xmath209 w.p.1 . note that by definition of @xmath210 ( see ( [ eq : fonctionselectionee ] ) ) @xmath211 condition e[ee : ring][ring : a ] states that the rings @xmath212 converge to @xmath213 w.p.1 ; therefore , e[ee : ring][ring : c ] is satisfied as soon as the limiting rings are of positive probability under the distribution of @xmath214 when @xmath215 . when the energy bounds are fixed , the conditions e[ee : ring][ring : a]-[ring : b ] are clearly satisfied and e[ee : ring][ring : c ] holds under convenient choice of the rings . we will discuss in section [ ring : construction ] how to check the condition e[ee : ring ] with adaptive energy bounds . proposition [ drift : small : aees ] shows that the kernels @xmath216 satisfy a geometric drift inequality and a minorization condition , with constants in the drift independent of @xmath217 for @xmath218 ( @xmath219 being defined in ( [ eq : thetam ] ) ) . the proof is in appendix [ secproof : drift : small : aees ] . [ drift : small : aees ] assume e[ee1][ee1:a ] and e[ee3 ] . for all @xmath176 : a. there exist @xmath220 and @xmath221 such that for all @xmath222 and any @xmath223 , @xmath224 for all @xmath225 and all @xmath226 , the sets @xmath227 are @xmath228-small for @xmath175 and the minorization constants depend neither upon @xmath217 nor on @xmath118 . b. [ drift : small : aees : item2 ] for all @xmath229 , there exists a probability measure @xmath230 invariant for @xmath231 . in addition , @xmath232 for @xmath233 . theorem [ theo : tout ] is proved in section [ proof ] . theorem [ theo : tout ] shows that there exists @xmath234 such that w.p.1 , for all @xmath39 large enough @xmath235 belongs to some @xmath236 . note that in @xcite , a s - lln for the equi - energy sampler is established by assuming that there exists a deterministic positive integer @xmath118 such that w.p.1 , @xmath237 for any @xmath39 . such a condition is quite strong since roughly speaking , it means that after @xmath39 steps ( even for small @xmath39 ) , all the rings contain a number of point which is proportional to @xmath39 , w.p.1 . this is all the more difficult to guarantee in practice , that the rings have to be chosen prior to any exploration of @xmath0 . our approach allows to relax this strong condition . the convergence of the marginals and the law of large numbers both require the convergence in @xmath39 ( @xmath85 fixed ) of @xmath238 for some functions @xmath34 . such a convergence is addressed in theorem [ theo : tout ] . we will then have the main ingredients to establish the convergence results for the processes @xmath21 , @xmath239 . [ theo : tout ] assume e[ee1 ] , e[ee3 ] , e[ee : ring ] and @xmath240 < \infty$ ] for all @xmath241 . a. [ prop : subsettheta ] there exists @xmath234 such that for all @xmath200 @xmath242 b. [ prop : cvg : pi ] for any @xmath243 , any @xmath244 and any continuous function @xmath245 , @xmath246 c. for any @xmath241 and for all bounded continuous function @xmath247 , @xmath248 = \theta_{\star}^{(k)}(f)$ ] . d. let @xmath249 . for any @xmath250 and for all continuous function @xmath34 in @xmath251 @xmath252 observe that , for the process @xmath253 , the family of functions for which the law of large numbers holds depends _ ( i ) _ upon @xmath254 given by ee[ee : ring]([ring : b ] ) i.e. in some sense , depends upon the adaptation rate ; and _ ( ii ) _ the temperature ladder . in the case @xmath189 can be chosen arbitrarily close to @xmath255 for any @xmath85 ( see comments after ( * ? ? ? * theorem 4.1 and 4.3 ) ) , this family of functions only depends upon @xmath254 and the lowest inverse temperature : it is all the more restrictive than @xmath256 is small . to our best knowledge , we are the first to prove such convergence results for aee ( and ee ) : previous works @xcite consider the simpler case when there is no selection i.e. @xmath257 . we propose to choose the adaptive boundaries @xmath258 as the @xmath259-quantile of the distribution of @xmath260 when @xmath261 is sampled under the distribution @xmath98 . this section proves that empirical quantiles of regularly spaced orders are examples of adaptive boundaries @xmath262 satisfying e[ee : ring ] . let @xmath263 be the cumulative distribution function ( cdf ) of the r.v . @xmath260 when @xmath264 : @xmath265 we denote the quantile function associated to @xmath260 by : @xmath266 with this definition , for @xmath267 , we set @xmath268 . with this choice of the boundaries , the condition e[ee : ring][ring : c ] holds : by ( [ eq : cs : ee : ringa ] ) , e[ee : ring][ring : c ] is satisfied because @xmath0 is continuous . the conditions e[ee : ring][ring : a]-[ring : b ] require the convergence of the quantile estimators and a rate of convergence of the variation of two successive boundaries . to prove such conditions , we use an hoeffding - type inequality . [ prop : vitesse : quantiles ] assume a. [ hyp : fd ] the cumulative distribution function @xmath269 where @xmath270 is given by ( [ eq : thetastar ] ) , is differentiable with positive derivative on @xmath271 . b. there exists @xmath272 such that @xmath81 is a @xmath272-uniformly ergodic markov chain with initial distribution satisfying @xmath273 < \infty$ ] . then e[ee : ring][ring : a]-[ring : b ] hold with @xmath274 and @xmath31 . the proof is in section [ secproof : prop : vitesse : quantiles ] . extensions of proposition [ prop : vitesse : quantiles ] to the case when @xmath81 is not a uniformly ergodic markov chain is , to our best knowledge , an open question . therefore , our convergence result of aee when the boundaries are the quantiles defined by inversion of the cdf of the auxiliary process applies to the @xmath275-stage level and seems difficult to extend to the @xmath17-stage , @xmath276 . we proved recently in @xcite that when the quantiles are defined by a stochastic approximation procedure , the conditions e[ee : ring][ring : a]-[ring : b ] hold even under very weak conditions on the auxiliary @xmath21 , @xmath97 . in this case , the convergence of the @xmath17-level aee with @xmath276 is established . one of the challenges in biology is to understand how gene expression is regulated . biologists have found that proteins called transcription factors play a role in this regulation . indeed , transcription factors bind on special motifs of dna and then attract or repulse the enzymes that are responsible of transcription of dna sequences into proteins . this is the reason why finding these binding motifs is crucial . but binding motifs do not contain deterministic start and stop codons : they are only random sequences that occurs more frequently than expected under the background model . several methods have been proposed so far to retrieve binding motifs @xcite , which yields to a complete bayesian model @xcite . among the bayesian approach , one effective method is based on the gibbs sampler @xcite - it has been popularized by software programs @xcite . nevertheless , as discussed in @xcite , it may happen that classical mcmc algorithms are inefficient for this bayesian approach . therefore , @xcite show the interest of the equi - energy sampler when applied to this bayesian inverse problem ; more recently , @xcite proposed a gibbs - based algorithm for a similar model ( their model differs from the following one through the assumptions on the background sequence ) . we start with a description of our model for motif sampling in biological sequences - this section is close to the description in @xcite but is provided to make this paper self - contained . we then apply aee and compare it to the interacting mcmc of ( * ? ? ? * section 3 ) ( hereafter called i - mcmc ) , and to a metropolis - hastings algorithm ( mh ) . comparison with gibbs - based algorithms ( namely bioprospector and alignace ) can be found in the paper of @xcite . the available data is a dna sequence , which is modeled by a background sequence in which some motifs are inserted . the background sequence is represented by a vector @xmath277 of length @xmath278 . each element @xmath279 is a nucleotide in @xmath280 ; in this paper , we will choose the convention @xmath281 . the length @xmath282 of a motif is assumed to be known . the motif positions are collected in a vector @xmath283 , with the convention that @xmath284 iff the nucleotide @xmath279 is located at position number @xmath285 of a motif ; and @xmath286 iff @xmath279 is not in the motif . the goal of the statistical analysis of the data @xmath287 is to explore the distribution of @xmath169 given the sequence @xmath287 . we now introduce notations and assumptions on the model in order to define this conditional distribution . we denote by @xmath288 the probability that a sub - sequence of length @xmath282 of @xmath287 is a motif . it is assumed that the background sequence is a markov chain with ( deterministic ) transition matrix @xmath289 on @xmath290 ; and the nucleotide in a sequence are sampled from a multinomial distribution of parameter @xmath291 , @xmath292 being the probability for the @xmath285-th element of a motif to be equal to @xmath293 . in practice , it has been observed that approximating @xmath294 by the frequency of jumps from @xmath293 to @xmath285 in the ( whole ) sequence @xmath287 is satisfying . it is assumed that the r.v . @xmath295 are independent with prior distribution @xmath296 and @xmath297 ; @xmath298 is a dirichlet distribution with parameters @xmath299 and @xmath297 is a beta distribution with parameters @xmath300 . @xmath301 , @xmath302 and @xmath303 are assumed to be known . therefore , given @xmath295 , @xmath304 is a markov chain described as follows : * if @xmath305 then @xmath306 ; else @xmath307 . * if @xmath308 , @xmath309 ; else @xmath310 is drawn from a multinomial distribution with parameter @xmath311 . the chains are initialized with @xmath312 ; the distribution of @xmath313 given @xmath314 and @xmath315 ( resp . given @xmath316 and @xmath315 ) is uniform on @xmath317 ( resp . a multinomial distribution with parameter @xmath318 ) . this description yields to the following conditional distribution of @xmath169 given @xmath287 : ( up to a multiplicative constant ) - see @xcite for similar derivation - @xmath319 where * @xmath320 is the number of elements of @xmath169 equal to @xmath228 . * @xmath321 is the number of elements of @xmath169 equal to @xmath322 . * @xmath323 is the number of pairs @xmath324 equal to @xmath325 . to highlight the major role of the equi - energy jumps , and the importance of the construction of the rings to make the acceptance probability of the jumps large enough , we compare aee to i - mcmc , and to mh . the data are obtained with values of @xmath326 and @xmath315 similar to those of @xcite : @xmath327 , @xmath328 , @xmath329 , @xmath330 for all @xmath285 , @xmath293 , and @xmath331 we sample a sequence @xmath287 of length @xmath332 and the size of the motif is @xmath333 . we now detail how the mh and the metropolis - hastings steps of aee and i - mcmc are run . for the metropolis - hastings stage , the proposal distribution @xmath334 is of the form @xmath335 where we set @xmath336 . the proposed state @xmath337 of the metropolis - hastings step is then sampled element by element ; the distributions are designed to be close to the previous model : @xmath338 equal to @xmath339 if @xmath340 , and else , @xmath338 is sampled under a bernoulli distribution of parameter @xmath341 the replacement constant @xmath342 is fixed by the users and @xmath343 is given by @xmath344 - where @xmath160 is a value fixed by the users . @xmath345 is the bernoulli distribution with parameter ( [ eq : proposal : bio ] ) . finally , the candidate @xmath337 is accepted with probability @xmath346 figure [ comparaison_bio ] displays the results obtained by aee , i - mcmc and a mh sampler . each subplot displays two horizontal lines with length equal to the length of the observed dna sequence . the upper line represents the actual localization of the motifs , and the lower line represents in gray - scale the probability for each position to be part of a motif computed by one run of each algorithm after @xmath347 iterations . for aee and i - mcmc , we choose @xmath152 , @xmath153 , @xmath348 . the acceptance rate of the jump for aee was about five times higher than for i - mcmc , which confirms the interest of the rings . as expected , aee performs better than the other algorithms : there were @xmath349 actual motifs , and aee retrieved @xmath163 motifs , whereas the i - mcmc and the mh retrieved respectively @xmath350 and @xmath351 motifs . as illustrated by the numerical examples , the efficiency of ee depends upon the choice of the energy rings . the adaptation we proposed improves this efficiency since it makes the probability of accepting a jump more stable . it is known that adaptation can destroy the convergence of the samplers : we proved that aee converges under quite general conditions on the adapted bounds and these general conditions can be used to prove the convergence of aee when applied with other adaptation strategies @xcite . it is also the first convergence result for an interacting mcmc algorithm including a selection mechanism . our sketch of proof can be a basis for the proof of other interacting mcmc such as the simcmc algorithm of @xcite , the non - linear mcmc algorithms described in ( * ? ? ? * section 3 ) or the pteem algorithm of @xcite . define @xmath352 the case @xmath353 is a consequence of e[ee3 ] since @xmath354 for any @xmath98 so that @xmath355 . we now consider the case @xmath356 : in the proof below , for ease of notations we will write @xmath33 , @xmath357 , @xmath358 , @xmath359 and @xmath360 instead of @xmath177 , @xmath175 , @xmath361 , @xmath362 and @xmath363 . _ ( a ) _ let @xmath222 and @xmath364 . by definition of @xmath131 ( see ( [ eq : fonctionselectionee ] ) ) and of @xmath219 ( see ( [ eq : thetam ] ) ) , @xmath365 . moreover , by e[ee3][ee3:b ] @xmath366 we have by ( [ eq : definitionkthetaaee ] ) , ( [ eq : definitionw ] ) and ( [ eq : thetatilde ] ) @xmath367 by ( [ eq : definitionalphaaee ] ) , @xmath368 defining @xmath369 by @xmath370 gives the upper bound @xmath371}z(1-z^{\sigma } ) \leq \psi(\sigma ) $ ] . hence , @xmath372 . this yields @xmath373 with @xmath374 and @xmath375 . the minorization condition comes from the lower bound @xmath376 . _ ( b ) _ let @xmath222 and @xmath364 . by e[ee3][ee3:a ] , @xmath33 is @xmath377-irreducible and so is @xmath9 ; @xmath9 possesses a @xmath228-small set and is thus aperiodic . in addition , @xmath378 , with @xmath379 and @xmath380 is a @xmath228-small set for @xmath9 . by ( * ? ? ? * chapter 15 ) , @xmath381 exists and @xmath382 . [ lemme : unif : ergo ] assume e[ee1][ee1:a ] and e[ee3 ] . then for all @xmath244 , for all @xmath222 and all @xmath223 , there exist @xmath383 and @xmath384 such that for all @xmath385 and any @xmath386 and any @xmath241 , @xmath387 let @xmath388 and assume in addition that @xmath389 w.p.1 . then for any positive integer @xmath65 , on the set @xmath390 @xmath391 the proof in the case @xmath353 is a consequence of e[ee3 ] and ( * ? ? ? * chapter 15 ) since @xmath354 . consider the case @xmath97 . here again , the dependence upon @xmath85 is omitted : @xmath392 denote @xmath393 and @xmath127 . _ proof of ( [ control : normev ] ) _ let @xmath244 and set @xmath394 . by the jensen s inequality and proposition [ drift : small : aees ] , there exists @xmath395 and @xmath396 such that for any @xmath397 and any @xmath364 , @xmath398 let @xmath222 and @xmath364 . by ( * ? ? * lemma 2.3 . ) , ( [ control : normev ] ) holds and there exist constants @xmath399 such that for any @xmath364 , @xmath400 where @xmath401 is the minorizing constant of @xmath9 on the set @xmath402 . _ proof of ( [ lemme : unif : ergo : tool1 ] ) _ for all @xmath403 , @xmath404 since @xmath405 w.p.1 , @xmath406 w.p.1 . thus showing that on the set @xmath407 , @xmath408 . this implies ( [ lemme : unif : ergo : tool1 ] ) . let @xmath409 . define for any positive integer @xmath65 and any @xmath410 , @xmath411 by convention , @xmath412 for any @xmath413 . [ lemme : unif : moments ] assume e[ee1][ee1:a ] , e[ee3 ] and @xmath414 < \infty$ ] for any @xmath415 . then for any @xmath243 , @xmath416 < \infty \eqsp.\ ] ] the proof is by induction on @xmath85 . the case @xmath353 is a consequence of e[ee3 ] since @xmath417 . assume the property holds for @xmath418 . in this proof , @xmath419 , @xmath420 will be denoted by @xmath421 . by ( [ eq : interactingmcmc ] ) and proposition [ drift : small : aees ] we obtain , for @xmath422 @xmath423 & \leq { \mathbb{e}}\left [ p_{\theta_{j-1 } } w(x_{j-1 } ) \un_{a_{q , j-1}^{(k)}}\right ] \\ & \leq \tilde{\lambda } { \mathbb{e}}\left [ w(x_{j-1 } ) \un_{a_{q , j-2}^{(k ) } } \right ] + \tilde{b } \ , m_\star \ , { \mathbb{e}}\left[\theta_{j-1}(w ) \un_{a_{q , j-1}^{(k-1 ) } } \right ] \\ & \leq \tilde{\lambda } { \mathbb{e}}\left [ w(x_{j-1 } ) \un_{a_{q , j-2}^{(k ) } } \right ] + \tilde{b } \ , m_\star \ , \sup_l { \mathbb{e}}\left[w(y_l ) \un_{a_{q , l-1}^{(k-1 ) } } \right ] \eqsp.\end{aligned}\ ] ] since @xmath424 , the induction assumption implies that @xmath425 < \infty$ ] . iterating this inequality allows to write that for some constant @xmath426 @xmath427 \leq c ' \ { \mathbb{e}}\left [ w(x_q ) \right ] \eqsp.\ ] ] finally , by definition of @xmath428 , either @xmath429 if @xmath430 , or @xmath431 otherwise ; note that if @xmath432 then @xmath433 . since both @xmath33 and @xmath357 for @xmath434 satisfy a drift inequality ( see e[ee3 ] and proposition [ drift : small : aees ] ) , @xmath435 < \infty$ ] by ( [ eq : interactingmcmc ] ) . @xmath436 [ hyp : proof : subset ] there exists @xmath409 such that @xmath437 . by proposition [ prop : lgn : first ] , the conditions r[hyp : proof : ergo ] and r[hyp : proof : lln ] hold for @xmath353 ; r[hyp : proof : pi ] also holds for @xmath353 since @xmath445 for any @xmath98 . we assume that for any @xmath446 , for @xmath388 , the conditions r[hyp : proof : subset](@xmath447 ) , r[hyp : proof : pi](@xmath285 ) , r[hyp : proof : ergo](@xmath285 ) and r[hyp : proof : lln](@xmath285 ) hold . we prove that r[hyp : proof : subset](@xmath85 ) , r[hyp : proof : pi](@xmath448 ) , r[hyp : proof : ergo](@xmath448 ) and r[hyp : proof : lln](@xmath448 ) hold . to make the notations easier , the superscript @xmath85 is dropped from the notations : the auxiliary process @xmath21 will be denoted by @xmath449 , and the process @xmath450 by @xmath451 ; @xmath452 and @xmath453 are resp . denoted by @xmath454 and @xmath455 . _ ( proof of r[hyp : proof : subset](@xmath85 ) ) _ we prove there exist an integer @xmath234 and a positive r.v . @xmath465 such that @xmath466 to that goal , we prove that with probability @xmath228 , for all @xmath39 large enough , @xmath467 and use the assumption e[ee : ring][ring : c ] . for all @xmath93 and @xmath98 , there exists a ring index @xmath468 such that @xmath469 . upon noting that @xmath470 , it holds @xmath471 we write @xmath472 by definition of @xmath473 , @xmath474 is continuous and bounded . therefore , by r[hyp : proof : lln](k ) , lemma [ lemme : conv : h ] and e[ee : ring][ring : a ] , the proof of ( [ eq : tool2:prop : subsettheta ] ) is concluded by @xmath475 first of all , observe that by definition of @xmath360 ( see proposition [ drift : small : aees ] ) and the expression of @xmath357 , @xmath476 . we check the conditions of ( * ? ? ? * theorem 2.11 ) . by proposition [ prop : subsettheta ] it is sufficient to prove that for any @xmath477 , @xmath478 w.p.1 . lemma [ lemme : unif : ergo ] and r[hyp : proof : lln](@xmath85 ) show that on the set @xmath479 , @xmath480 and @xmath481 w.p.1 . equicontinuity of the class @xmath482 , where @xmath34 is a bounded continuous function on @xmath35 , will follow from lemmas [ lem : regux : g ] to [ lem : equicont ] . finally , the weak convergence of the transition kernels is proved in lemma [ lem : ascvgnoyaux ] . following the same lines as in the proof of ( * ? ? ? * theorem 3.5 ) , it can be proved that the above discussion for @xmath34 bounded and proposition [ drift : small : aees ] imply @xmath483 w.p.1 . for any continuous function @xmath34 such that @xmath484 . [ lem : regux : alpha ] assume e[ee1][ee1:a ] . for all @xmath222 , there exists a constant @xmath490 such that for all @xmath491 and @xmath492 @xmath493 \eqsp , \\ \label{majoration : alpha : y } \left| \alpha_\theta(x , y ) - \alpha_{\theta}(x , y ' ) \right| & \leq c_m \left[\left| \pi^{\beta_k-\beta_{k+1}}(y)-\pi^{\beta_k-\beta_{k+1}}(y ' ) \right| + \left| \pi(y ) - \pi(y')\right| \right ] \eqsp.\end{aligned}\ ] ] by definition of @xmath494 ( see ( [ eq : definitionalphaaee ] ) ) , @xmath495 , with @xmath496 note that @xmath497 . by symmetry , we can assume that @xmath498 and this implies @xmath499 since @xmath500 . therefore , @xmath501 \eqsp.\end{gathered}\ ] ] the proof of ( [ majoration : alpha : x ] ) is concluded by lemma [ lem : regux : g ] . the proof of ( [ majoration : alpha : y ] ) is on the same lines and omitted . let @xmath34 be a continuous function on @xmath35 , bounded by @xmath228 . let @xmath222 and @xmath364 . we have @xmath503 where @xmath504 is given by ( [ eq : thetatilde ] ) . this yields to @xmath505 we have @xmath506 where @xmath507 is given by ( [ eq : thetatilde ] ) . so lemmas [ lem : regux : g ] and [ lem : regux : alpha ] imply that for all @xmath508 , there exists a constant @xmath490 such that for all @xmath233 : @xmath509 the proof is concluded since @xmath33 is feller and @xmath0 is continuous . [ lem : ascvgnoyaux ] let @xmath222 . assume e[ee1 ] , e[ee : ring][ring : a ] and r[hyp : proof : lln](k ) . for all @xmath463 , there exists a set @xmath510 such that @xmath511 and for all @xmath512 and any bounded continuous function @xmath34 @xmath513 following the same lines as in the proof of ( * ? ? ? * proposition 3.3 . ) , it is sufficient to prove that for any @xmath514 and any bounded continuous function @xmath34 , @xmath515 w.p.1 on the set @xmath516 . let @xmath34 and @xmath93 be fixed . we write @xmath517 where @xmath504 is given by ( [ eq : thetatilde ] ) . moreover , @xmath518 this yields to @xmath519 where @xmath520 . there exists a constant @xmath490 such that on the set @xmath521 , ( see the proof of lemma [ lem : regux : alpha ] for similar upper bounds ) @xmath522 where @xmath523 is defined by ( [ eq : thetatilde ] ) . we write by definition of the function @xmath131 ( see ( [ eq : fonctionselectionee ] ) ) @xmath524 by lemma [ lemme : conv : h ] and e[ee : ring][ring : a ] , the first term converges to zero w.p.1 . since @xmath525 is continuous , r[hyp : proof : lln](k ) implies that the second term tends to zero w.p.1 . therefore , on the set @xmath526 , @xmath527 converges to zero w.p.1 , as well as @xmath528 , and @xmath529 . note that by lemma [ lem : regux : alpha ] , @xmath530 is bounded and continuous . therefore , following the same lines as above , it can be proved that under r[hyp : proof : lln](k ) and e[ee : ring][ring : a ] , on the set @xmath526 , @xmath531 = 0 w.p.1 we check the conditions of ( * ? ? ? * theorem 2.1 ) . let @xmath34 be a bounded continuous function on @xmath35 . by r[hyp : proof : pi](k+1 ) , @xmath532 w.p.1 . let @xmath533 . by proposition [ prop : subsettheta ] , there exists @xmath477 such that @xmath534 . following the same lines as in the proof of ( * ? ? ? * theorem 3.4 ) , it can be proved by using lemmas [ lemme : unif : ergo ] , [ lemme : conv : h ] and [ lemme : cv : dv : proba ] and the condition e[ee : ring][ring : b ] that @xmath535 = 0 $ ] . this concludes the proof . by definition of @xmath357 , for all function @xmath34 bounded by @xmath228 , ( [ diff : p ] ) holds . so @xmath538 the term @xmath539 is equal to @xmath540 with @xmath541 note that @xmath542 . therefore , for all @xmath543 , @xmath544 the term @xmath545 is upper bounded by @xmath546 moreover , @xmath547 \right| \leq 2s \sup_{l , x } |h_{\theta , l}(x)-h_{\theta',l}(x)| \eqsp.\ ] ] this concludes the proof . let @xmath249 and set @xmath548 . we check the conditions of ( * ? ? ? * theorem 2.7 ) . by proposition [ drift : small : aees ] , condition a3 of @xcite holds . by r[hyp : proof : pi](@xmath448 ) , @xmath549 w.p.1 for any continuous function @xmath34 in @xmath550 . condition a4 ( resp . a5 ) of @xcite is proved in lemma [ lemma : a5 ] ( resp . lemma [ lemma : a6 ] ) . by r[hyp : proof : subset](j ) for all @xmath555 , it is sufficient to prove that for any positive integer @xmath65 @xmath556 where @xmath557 is defined in appendix [ app : lemme : unif : moments ] . following the same lines as in the proof of lemma [ lemme : unif : moments ] , it can be proved that @xmath558 w.p.1 . by lemma [ lemme : unif : ergo ] and r[hyp : proof : lln](k ) , on the set @xmath559 , @xmath560 w.p.1 . therefore , we have to prove that @xmath561 w.p.1 . following the same lines as in the proof of lemma [ lemme : cv : dv : proba ] , we obtain that on the set @xmath557 , there exists a constant @xmath490 such that @xmath562 set @xmath563 , such that @xmath564 . by e[ee : ring][ring : b ] , there exists a r.v . @xmath261 finite w.p.1 such that @xmath565-a.s . @xmath566 therefore , it holds @xmath567 \\ \leq \sum_{j \geq q } j^{-1/s } \left ( j^{-{\gamma}/s } + j^{-1/s } \right ) { \mathbb{e}}\left [ \left(\|\theta_j\|_v + \|\theta_{j-1}\|_v \right)^{\frac{1}{s } } v^{\frac{1}{s}}(x_j ) \un_{a_{q , j}^{(k ) } } \right ] \eqsp.\end{gathered}\ ] ] we have , @xmath568^{1/2 } \ \sup_j { \mathbb{e}}\left[v(x_j)^{\frac{2}{s } } \un_{a_{q , j}^{(k ) } } \right]^{1/2 } \ \eqsp,\ ] ] where @xmath569 is finite since @xmath570 and @xmath571 . since @xmath572 , lemma [ lemme : unif : moments ] implies that @xmath573 < \infty$ ] . in addition , since @xmath574 we have , by jensen s inequality , @xmath575 & \leq { \mathbb{e}}\left[\left ( \frac{1}{j } \sum_{p=1}^j v(y_p ) \right)^{\frac{2}{s } } \un_{a_{q , j}^{(k-1 ) } } \right ] \leq { \mathbb{e}}\left [ \frac{1}{j } \sum_{p=1}^j v^{\frac{2}{s}}(y_p ) \un_{a_{q , j}^{(k-1 ) } } \right ] \\ & \leq \sup_p { \mathbb{e}}\left [ w(y_p ) \un_{a_{q , p-1}^{(k-1 ) } } \right]\end{aligned}\ ] ] which is finite under lemma [ lemme : unif : moments ] . similarly , we prove that @xmath576 w.p.1 , upon noting that @xmath577 . by r[hyp : proof : subset](j ) for all @xmath555 , it is sufficient to prove that for any positive integer @xmath65 @xmath579 where @xmath557 is defined in appendix [ app : lemme : unif : moments ] . let @xmath580 . by lemma [ lemme : unif : ergo ] , @xmath581 w.p.1 ; and , as in the proof of lemma [ lemme : unif : moments ] , it can be proved that @xmath582 < \infty$ ] . the proof is concluded since @xmath583 . [ hoeffding ] let @xmath584 be a markov chain on @xmath585 , with transition kernel @xmath586 and initial distribution @xmath587 . assume @xmath586 is @xmath272-uniformly ergodic , and denote by @xmath588 its unique invariant distribution . then there exists a constant @xmath17 such that for any @xmath589 and for any bounded function @xmath590 @xmath591 \eqsp.\ ] ] [ lemme : appl : hoeffding ] assume that there exists @xmath272 such that @xmath71 is a @xmath272-uniformly ergodic markov chain with initial distribution @xmath587 with @xmath592 . let @xmath459 and @xmath593 ; and set @xmath594 . for all @xmath595 and any @xmath596 , @xmath597 where @xmath598 . let @xmath599 . we write @xmath600 . since @xmath601 iff @xmath602 , @xmath603 proposition [ hoeffding ] is then applied with @xmath604 . as @xmath605 we obtain @xmath606\right ) \eqsp.\end{aligned}\ ] ] for some constant @xmath17 independent of @xmath607 . similarly , @xmath608 \right ) \eqsp,\end{aligned}\ ] ] which concludes the proof . _ proof of proposition [ prop : vitesse : quantiles ] _ let @xmath609 and @xmath610 be defined by @xmath611 where @xmath17 is given by lemma [ lemme : appl : hoeffding ] . note that under ( [ hyp : fd ] ) , @xmath612 since @xmath613 . by ( [ hyp : fd ] ) , @xmath614 is differentiable and we write when @xmath615 @xmath616 hence @xmath617 for @xmath39 large enough . similarly , @xmath618 for @xmath39 large enough . so when @xmath39 is large enough , @xmath619 with @xmath620 defined in lemma [ lemme : appl : hoeffding ] . by lemma [ lemme : appl : hoeffding ] , for @xmath39 large enough , to @xmath621 as @xmath622 , the borel - cantelli lemma yields @xmath623 w.p.1 . this concludes the proof . c. andrieu , a. jasra , a. doucet , and p. del moral . non - linear markov chain monte carlo . in _ conference oxford sur les mthodes de monte carlo squentielles _ , volume 19 of _ esaim proc . _ , pages 7984 . 2007 . timothy l. bailey and charles elkan . fitting a mixture model by expectation maximization to discover motifs in biopolymers . in _ second international conference on intelligent systems for molecular biology _ , 1994 . x. liu , douglas l. brutlag , and jun s. liu . bioprospector : discovering conserved dna motifs in upstream regulatory regions of co - expressed genes . in _ pacific symposium on biocomputing _ , pages 127138 , 2001 . g. d. stormo and g. w. hartzell . identifying protein - binding sites from unaligned dna fragments . in _ proceedings of the national academy of sciences of the united states of america _ , volume 86 , pages 11831187 , february 1989 .

we calculate the probability of the creation of a circular phonon ( c - phonon ) in he ii by a c - photon of the resonator . it is shown that this probability has sharp maxima at frequencies , where the effective group velocity of the c - phonon is equal to zero ; the density of states of c - phonons strongly grows at such frequencies . for he ii , these frequencies correspond to a roton and a maxon . from the probability of the c - roton creation , we calculate the roton line width which is found to approximately agree with the experimental one . we conclude that the roton line observed in the super - high - frequency ( shf ) absorption spectrum of helium is related to the creation of c - rotons . a possible interpretation of the stark effect observed for the roton line is also proposed . * keywords * : liquidhe , microwave resonator , circular roton , supernarrow absorption line .

1009.4435

in the recent works @xcite , an unconventional effect was discovered . in a dielectric disk - shaped resonator placed in liquid @xmath0he , the azimuth ( @xmath1-)modes were excited . these modes are of the `` whispering - gallery '' type and represent a superposition of standing and running resonance electromagnetic ( em ) waves of the shf range in a narrow frequency interval . at the roton frequency @xmath2 the supernarrow absorption line with the width @xmath3 was observed in the shf spectrum of the resonator . this width is by six orders less than that of a roton peak measured in neutron experiments and is comparable with the width of line in the mssbauer effect . it was assumed in @xcite that the line is related to the van hove singularity caused by a high density of states of plane ( p- ) rotons near the roton minimum of the dispersion curve . in this case , however , one is faced with the problem to satisfy the momentum conservation law , since the p - roton momentum is greater than the momentum of a p - photon with the same energy by six orders . therefore , it was supposed @xcite that the excess momentum of a p - roton is transferred to helium as a whole . to clarify this and other points , it is necessary to calculate the probability of the creation of a roton and the widths and the forms of lines for various possible processes , and then to choose a process explaining the experiment . it is necessary to take into account that the em field of a disk - shaped resonator has the circular ( c- ) symmetry and is concentrated only near the disk according to measurements @xcite and the theory @xcite . since namely the em field induces the transition , the latter must be characterized by the c - symmetry . in @xcite , it was shown that a phonon near a disk possesses also the c - symmetry . therefore , we assume that the narrow line corresponds to the creation of a c - roton by the em field of the resonator . below , we will find the probability of this process and the width of the corresponding absorption line . a number of results required for calculations was obtained in @xcite ; formulas ( n ) from that work will be denoted here by ( n@xmath4 ) . a part of the results of the present work was briefly published in @xcite . to calculate the line width , it is necessary to know the probability of the c - photon @xmath5 c - phonon process . no problems concerning the conservation laws appear for this process , because both c - photon and c - phonon have no momentum , but have angular momenta @xmath6 and @xmath7 , respectively @xcite . moreover , the condition @xmath8 is easily satisfied . as will be seen , the field of the resonator contains @xmath9 photons with close frequencies . at such occupation number , the photon field can be considered as an external classical perturbing harmonic field with frequency @xmath10 acting on helium . we now calculate the probability of the creation of a c - phonon by this field . the width of the azimuth mode , on which the roton line is observed , is about 2.5mhz , the line width @xmath11mhz , and the frequency @xmath12 ghz . it is seen that the mode width is very small as compared with the frequency ; therefore , the latter can be considered constant . it is worth noting that , while calculating the transition amplitude , we can take no care of the conservation laws . they are satisfied automatically ( if a transition contradicts some conservation law , this will manifest itself in the disagreement of the symmetries of the initial and final states , and the amplitude will become equal to zero ) . the probability ( per unit time ) of the creation of a c - phonon in he ii due to the action of the em field of a resonator is @xcite @xmath13 @xmath14 where @xmath15 and @xmath16 are the phase volumes of all nuclei and all electrons , @xmath17 and @xmath18 are the wave functions ( wfs ) of the initial and final states of helium , respectively , @xmath19 is the energy of a c - phonon , and @xmath20 . the explicit formula for @xmath21 follows from the perturbation operator @xmath22 if the wavelength @xmath23 of the em field is much more than the size of the system , then the problem is solved in the dipole approximation @xcite . in our case , this approximation is not suitable , since the system size exceeds @xmath23 by one order . in addition , a photon is spent on the excitation of fluid helium as a whole , i.e. on the creation of a c - phonon which is related to the motion of atoms as united objects , rather than on the excitation of electron shells of a single atom or many atoms . therefore , we will use a general approach , by considering the action of the em field directly on the charged particles in an atom , i.e. , on electrons and the nucleus . but the atoms interact with one another . as a result , the em field creates the collective excitation , a phonon which is electrically neutral as a whole . it should be noted that a sound wave is associated with a variable concentration gradient . in this case , a variable local electric field arises in the interatomic space , since helium atoms polarize one another @xcite . however , this gives only a negligible correction to the effect . for a charge particle in the em field , we have @xcite @xmath24 where @xmath25 , and @xmath26 and @xmath27 are the charge and the radius - vector of a particle , respectively . the term with @xmath28 induces two - photon transitions which will not be considered here . the field @xmath29 should be real and can be presented in the form @xmath30 the quantity @xmath31 is calculated in @xcite , formula ( 6@xmath4 ) . since we are interesting in the field @xmath29 in helium , we set @xmath32 in ( 6@xmath4 ) . as a result , we obtain @xmath33 , i.e. the divergence is determined by the field @xmath34 , i.e. the time - independent part of @xmath29 . it can be always set to zero , by adding a gradient of the corresponding function to @xmath35 in this case , the measurable quantities @xmath36 and @xmath37 are not changed . therefore , we set @xmath38 , so that @xmath39 . in what follows , we omit index 0 in @xmath40 ( [ a5 ] ) . so , we have @xmath41 consider the set of he ii atoms in the field @xmath29 . an atom of @xmath0he consists of the nucleus with the charge @xmath42 and two electrons , each possessing the charge @xmath43 . therefore , for he ii , we have @xmath44 here , @xmath45 are coordinates of the nucleus of the @xmath46-th atom , @xmath47 and @xmath48 are coordinates of the electrons of this atom , @xmath49 is the nucleus mass of @xmath0he atom , and @xmath50 is the electron mass . to calculate @xmath51 we need to know the wfs @xmath18 and @xmath17 . let the initial state characterized by the wf @xmath17 be the ground state of helium , and let the final state with @xmath18 be the ground state plus one c - phonon . then @xmath52 where @xmath53 is the wf of all electrons , and @xmath54 is the wf of all nuclei of helium atoms . the modern microscopic models describe the properties of he ii well enough . according to them , a helium atom is a very elastic object . therefore , we can consider with a good accuracy that the electron shell of each atom follows the nucleus without inertia and is not deformed by the interaction with neighboring atoms . in this case , @xmath55 coincides with @xmath56 written as a function of the coordinates of atoms , and the electron part looks as @xmath57 where @xmath58 , @xmath59 . for the wf of the ground state of @xmath0he atom , we use the well - known one - parameter approximation with @xmath60 ( @xmath61 ) . in reality , the electron shells of atoms are perturbed by neighbors and as a result are somewhat deformed @xcite . these deformations are `` directed '' to the atom , with which the interaction occurs . since the adjacent atoms surround the given atom from all sides and chaotically on the average , the polarizations induced by them cancel one another to a significant degree . as a result , the mean polarization of an atom in he ii turns out small , @xmath62 @xcite , though it is greater by one order of magnitude than the polarization induced by a single atom located at the mean interatomic distance @xmath63 . the literature presents the assertions that the presence of large - amplitude zero oscillations of he ii with the amplitude @xmath64 implies that the electron shell of a helium atom is strongly deformed and `` smeared '' over several interatomic distances . however , this should not be accepted : zero oscillations mean only that , due to the strong interatomic interaction , the helium atoms move with a large ( even at @xmath65 ) mean velocity , which leads to a small density of helium and exhausts the condensate . but the electron shells of atoms are deformed slightly in this case , as follows from the smallness of @xmath66 . we will neglect these deformations and apply formula ( [ a13 ] ) . for the wf of a state with a single c - phonon , we use the formula @xcite @xmath67 in view of it , we obtain @xmath68 let us take into account that @xmath69 @xmath70 @xmath71\psi_{0}^{el}. \label{a18}\ ] ] then @xmath72 @xmath73 @xmath74 we note that the matrix element @xmath76 in ( [ a1 ] ) is constructed on the basis of the wfs @xmath17 and @xmath18 not containing the time factor ( the latter is included into the @xmath77-function ) . in formula ( 50@xmath4 ) for @xmath78 the coordinates of atoms can be considered as the coordinates of nuclei . omitting the factor @xmath79 , we present @xmath80 in the form @xmath81 below @xmath82 . whence , with regard for ( 56@xmath4 ) and ( 58@xmath4 ) , we get @xmath83 since @xmath84 , the first integral in ( [ a23 ] ) is equal to zero . as for the second integral , the integration is carried on over the region , where the helium atoms are positioned , i.e. , over the region outside of the disk . therefore , we describe the field @xmath29 with the use of solution ( 31@xmath4)(33@xmath4 ) without the factor @xmath79 ( taken into account in ( [ a5 ] ) ) . denoting @xmath85 , we have @xmath86 g(r ) . \label{a24 } \end{aligned}\ ] ] using ( 72@xmath4 ) and the relation @xmath87 , we obtain @xmath88iq_{\rho}\times \nonumber \\ & & \times \left [ a_{1}(\rho_{1},z_{1})e^{-i(\varphi_{1}-\varphi_{q } ) } - a_{2}(\rho_{1},z_{1})e^{i(\varphi_{1}-\varphi_{q } ) } \right ] \times \nonumber \\ & & \times e^{[iq_{z}(z_{1}-z_{2 } ) + iq_{\rho}\rho_{1}\cos{(\varphi_{1}-\varphi_{q } ) } - iq_{\rho}\rho_{2}\cos{(\varphi_{2}-\varphi_{q } ) } ] } , \label{a25 } \end{aligned}\ ] ] where @xmath89 . in the cylindrical coordinate system , @xmath90 and @xmath91 . the relation @xmath92 ( @xmath93 is arbitrary ) yields @xmath94 } = \nonumber \\ & & = 2\pi i^{l\pm 1}e^{il\varphi_{q}}j_{l\pm 1}(q_{\rho}\rho_{1 } ) , \label{a26 } \end{aligned}\ ] ] @xmath95 } = \nonumber \\ & & = 2\pi i^{-l_{c}}e^{-il_{c}\varphi_{q}}j_{l_{c}}(q_{\rho}\rho_{2 } ) . \label{a27 } \end{aligned}\ ] ] with regard for ( 32@xmath4 ) , ( 33@xmath4 ) and ( [ a22 ] ) , we write the required matrix element as @xmath96e^{[iq_{z}(z_{1}-z_{2})-ik_{z}z_{2}]}\times \nonumber \\ & & \times\left [ a_{1}(\rho_{1},z_{1})j_{l-1}(q_{\rho}\rho_{1})+a_{2}(\rho_{1},z_{1})j_{l+1}(q_{\rho}\rho_{1 } ) \right ] . \label{a28 } \end{aligned}\ ] ] integrating @xmath97 over the whole volume ( including the disk ) , we obtain @xmath98 } [ s(q_{z},q_{\rho})-1]= \nonumber \\ & & = 2\pi e^{-ik_{z}z_{1}}[s(-k_{z},q_{\rho})-1 ] . \label{a29 } \end{aligned}\ ] ] we pass in ( [ a28 ] ) to discrete @xmath99 , by replacing @xmath100 and using for @xmath99 the quantization conditions @xcite @xmath101 @xmath102 where @xmath103 and @xmath104 are , respectively , the height and the radius of the disk resonator , and @xmath105 and @xmath106 are the same for the chamber with helium ; the numbers are indicated for the experiment @xcite . the quantity @xmath107 is defined in @xcite : for small @xmath108 @xmath109 ( in particular , for the smallest @xmath108 we have @xmath110 ) , and @xmath107 decreases down to @xmath111 with increase in @xmath112 . with regard for ( 64@xmath4 ) , we will take the integral required in the calculations of ( [ a28 ] ) : @xmath113\times \nonumber \\ & \times & j_{l}(q_{\rho}\rho_{2})j_{l}(k_{\rho}\rho_{2})j_{l\pm 1}(q_{\rho}\rho_{1 } ) \approx\nonumber \\ & \approx & k_{\rho}[s(k)-1]j_{l\pm 1}(k_{\rho}\rho_{1 } ) \frac{b(k_{\rho}r_{\infty},l)}{1-\tilde{r}_{d}/r_{\infty } } , \label{a31 } \end{aligned}\ ] ] @xmath114 at greater @xmath108 the function @xmath115 , whereas it is close to 1 ( @xmath116 ) at lower @xmath112 ( @xmath117 ) and can be calculated numerically . taking the disk into account gives @xmath118 . since @xmath119 , @xmath120 , and @xmath121 , this consideration changes @xmath122 only slightly , which can be neglected . from ( [ a28])([a31 ] ) , we now obtain @xmath123 @xmath124 , \label{a32b}\ ] ] @xmath125 . \label{a33 } \end{aligned}\ ] ] as is seen , we must integrate with respect to the variables @xmath126 and @xmath127 over the region outside of the disk with regard for the distribution of the em field calculated in @xcite . we denote the values of the integral @xmath128 taken over regions i and ii from ( 32@xmath4)(33@xmath4 ) by , respectively , indices 1 and 2 : @xmath129 it will be seen below that values of the functions @xmath75 and @xmath130 are maximum in two limiting cases : @xmath131 and @xmath132 . in the first case , the simple , but sufficiently bulky calculations give @xmath133 @xmath134 @xmath135}}{\pi/\tilde{h}_{d}-k_{z } } + \nonumber \\ & + & ( k_{z}\rightarrow -k_{z } ) , \label{a36 } \end{aligned}\ ] ] where @xmath136 , @xmath137 , @xmath138 , and @xmath139 is the effective angle . using the latter , we set the line of sewing @xmath140 in ( 32@xmath4 ) , ( 33@xmath4 ) at @xmath141 in the form @xmath142 . the condition @xmath143 determines this angle at the beginning of the line of sewing . in calculations , we used the properties @xmath144 and @xmath145 , as well as the asymptotics of the bessel and neumann functions . since @xmath146 , the relation @xmath147 is valid . consider the case where @xmath132 . now , @xmath112 is small . the smallest @xmath112 are determined by the equality @xmath148 ( see ( 46@xmath4 ) ) , where @xmath149 . using the formulas @xcite @xmath150 @xmath151 @xmath152 we determine three first zeros of the bessel function : @xmath153 , @xmath154 , and @xmath155 . in order to calculate @xmath156 and @xmath157 we take into account that , at small @xmath108 the component @xmath158 . then we have @xmath159 @xmath160 . \label{n2 } \end{aligned}\ ] ] for three smallest @xmath161 ( @xmath162 ) , we determined numerically the following values : @xmath163 , -0.00049 , and 0.00029 . by calculating integral ( [ a33 ] ) , we have @xmath164 j_{l-1}(q_{1}r_{d})}}{(\pi/\tilde{h}_{d}-k_{z})^{2}ctg\theta_{0 } } \nonumber \\ & & \times r_{d}\left [ j_{l-1}(k_{\rho}r_{d})+j_{l+1}(k_{\rho}r_{d } ) \right ] + ( k_{z}\rightarrow -k_{z } ) , \label{n3 } \end{aligned}\ ] ] which implies that @xmath165 is negligible : @xmath166 . the electron part , @xmath130 , of the total matrix element ( [ a19 ] ) can be calculated analogously . the nonzero value of @xmath130 is determined by the difference @xmath167 where @xmath168 , @xmath169 and we took into account that , as @xmath170 is much more than the atom size ( @xmath171 ) , @xmath172 @xmath173 representing @xmath29 in the form @xmath174 we obtain @xmath175 @xmath176/\rho , \label{e5 } \end{aligned}\ ] ] @xmath177 . \nonumber \end{aligned}\ ] ] using relations ( 72@xmath4 ) , ( [ i64])([i57 ] ) , we can verify that @xmath178\frac{b(k_{\rho}r_{\infty},l)}{1-\tilde{r}_{d}/r_{\infty } } , \label{e7}\ ] ] @xmath179 @xmath180 where the integration is carried on over the volume occupied by helium . analogously to ( [ a33 - 2 ] ) , we divide @xmath181 into the sum of integrals over regions i and ii . then we have @xmath182 @xmath183 @xmath184 , \nonumber \end{aligned}\ ] ] @xmath185 , \nonumber \end{aligned}\ ] ] where @xmath186 is defined above , @xmath187 , @xmath188 , and @xmath189 . the integrals @xmath190 and @xmath191 are calculated analogously to @xmath192 and @xmath165 . eventually , we get @xmath193\right \ } , \label{e16 } \end{aligned}\ ] ] @xmath194 at @xmath195 the integral @xmath190 is negligible : @xmath196 . formulas ( [ a34 ] ) and ( [ e16 ] ) do not involve the condition @xmath197 for the disk surface . for its consideration , we need to make replacements @xmath198 , @xmath199 in the formulas , which decreases @xmath192 and @xmath190 still further . at small @xmath108 we obtain @xmath200 @xmath201 . \nonumber \end{aligned}\ ] ] for @xmath161 with @xmath202 we get numerically : @xmath203 , and @xmath204 . in addition , @xmath205 } } { ( \pi/\tilde{h}_{d}-k_{z})^{2}ctg\theta_{0 } } \times \nonumber \\ & & \times j_{l-1}(q_{1}r_{d})j_{l}(k_{\rho}r_{d } ) \times \label{n6 } \\ & & \times\left \{(1-l)n_{l-1}(q_{1}^{h}r_{d } ) - ( 1+l)b_{l}n_{l+1}(q_{1}^{h}r_{d } ) + \right . \nonumber \\ & & \left . + q_{1}^{h}r_{d } [ j^{\prime}_{l-1}(q_{1}^{h}r_{d } ) - b_{l } j^{\prime}_{l+1}(q_{1}^{h}r_{d})]\right \}+ \nonumber \\ & + & ( k_{z}\rightarrow -k_{z } ) \sim \frac{i_{el}^{(1)}(k_{z } \rightarrow \infty)}{kr_{d } } \ll i_{el}^{(1)}(k_{z } \rightarrow \infty ) . \nonumber \end{aligned}\ ] ] with regard for ( [ a32a ] ) , ( [ a32b ] ) , ( [ e4])([e8 ] ) , the performed calculations allow us to write finally @xmath206 @xmath207b(k_{\rho}r_{\infty},l)}+ 1 \right ) . \label{e21}\ ] ] we have @xmath208 , \label{e22}\ ] ] whence the total probability of the creation of a c - phonon with `` momentum '' @xmath209 is @xmath210^{2}\delta ( e_{c}(k)-\hbar\omega ) . \nonumber \end{aligned}\ ] ] moreover , @xmath211 in all formulas . the main contribution to sum ( [ e27 ] ) is given by the regions ( @xmath212 , @xmath213 ) and ( @xmath214 , @xmath215 ) . we now determine these contributions . \a ) region of small @xmath216 ( here , @xmath213 ) . according to ( [ a33 - 2])([a36 ] ) and ( [ e12 ] ) , ( [ e16 ] ) , ( [ e17 ] ) , we have @xmath217 , @xmath218 . in ( [ e27 ] ) , we now pass from the sum @xmath219 to the integral @xmath220 and then , with the help of the relation @xmath221 to an integral over @xmath209 . due to the @xmath77-function , this integral is easily calculated . after some transformations , we obtain @xmath222 @xmath223^{2}\times \nonumber \\ & \times & \sum\limits_{n_{z } } \left \{\frac{\sin{(\frac{\pi h_{d}}{2\tilde{h}_{d}}-n_{z}\frac{\pi h_{d}}{h - h_{d } } ) } } { \frac{\pi h_{d}}{2\tilde{h}_{d}}-n_{z}\frac{\pi h_{d}}{h - h_{d } } } + ( k_{z } \rightarrow -k_{z})\right \}^{2}\approx \nonumber \\ & \approx & \frac{4.3 ( h - h_{d})}{\pi h_{d } } \left [ ( l-1)j_{l-1}(q_{1}r_{d})\sin{\alpha_{l}(k r_{d})}\right ] ^{2 } , \label{e30 } \end{aligned}\ ] ] where @xmath224 , @xmath225 , @xmath226 , and the value of @xmath209 is determined from the condition @xmath227 for @xmath228 we used the quantization laws ( [ i42a ] ) , which assumes the zero boundary conditions on the container walls . \b ) region of small @xmath112 . with the use of @xmath229 ( see ( [ i42a ] ) ) and @xmath230 , relations ( [ a33 - 2 ] ) , ( [ n1])([n3 ] ) , ( [ e12 ] ) , ( [ n4])([n6 ] ) yield @xmath231 @xmath232 @xmath233 where @xmath234 , @xmath235 . formulas ( [ e30 ] ) and ( [ e34 ] ) include @xmath236 and @xmath237 . due to the zero boundary conditions , @xmath238 , whence @xmath239 . in view of the same conditions , we set @xmath240 . using ( [ a32a ] ) , ( [ a32b ] ) , ( 75@xmath4 ) , and the formula for the volume of helium @xmath241 , we obtain finally : @xmath242^{2 } + \frac{2f\cdot a(k , l)}{1- f}\right ) . \nonumber \end{aligned}\ ] ] we note that the main contribution to @xmath243 and @xmath244 is given by the electron part of @xmath76 , and the transition probability @xmath245 is determined by the quantities @xmath246 and @xmath247 i.e. , by the creation of c - phonons with the smallest @xmath216 and large @xmath112 and with the smallest @xmath112 and large @xmath216 ( almost plane c - phonons ) . we now evaluate these contributions quantitatively for the roton line and conditions of the experiment @xcite . according to @xcite , the relation @xmath248 is valid at @xmath149 . to calculate @xmath249 in ( [ e34 ] ) , we use the roton value @xmath250 ( for @xmath251 ) . moreover , we consider that the relation @xmath252 holds as @xmath253 @xmath110 at @xmath254 @xcite , and the quantity @xmath107 decreases down to @xmath111 with increase in @xmath46 . we calculated values of @xmath255 numerically by formula ( [ i57 ] ) : @xmath256 , @xmath257 , and @xmath258 . the contribution of the following @xmath112 ( with @xmath259 ) to @xmath249 is small . whence we find @xmath260 . moreover , since @xmath261 the functions @xmath249 and @xmath262 are almost independent of @xmath209 . but if @xmath209 are such that @xmath263 , then the coefficient @xmath249 increases by several orders of magnitude , approximately as @xmath264 , and becomes @xmath265 . but , for the probability @xmath245 , such a growth is cancelled by the factor @xmath266 in ( [ e37 ] ) in front of the large parentheses . using these numbers , the relation @xmath267 , and ( [ e34 ] ) , we obtain that the probability of the creation of c - phonons with large @xmath112 is @xmath268 times more than that for the almost planes ones ( with small @xmath112 ) . thus , a resonator creates c - phonons mainly of two `` extreme '' types : almost completely circular and almost plane . the high probability of the creation of c - phonons with an almost plane structure is a somewhat unexpected result . at the same time , it is clear that the completely plane phonons can not be created due to the angular momentum conservation law : a phonon must possess a certain `` twist '' ( @xmath269 , @xmath270 ) in order to carry away the angular momentum of a c - photon . as was mentioned above , the spectrum of the shf emission of a disk resonator contains the very narrow absorption line at the roton frequency ( @xmath271 ) . let the line width be the distance between the points , which located on both sides from the line center , and for which the intensity of a signal is about 0.8 of the background one ( the signal far from the line ) . then it follows from the experiment @xcite that the line width is about @xmath272 at @xmath273 and decreases , at lower @xmath274 to the resolving power of a spectroscope ( @xmath275 ) . moreover , the width stops to decrease at @xmath276 and approaches a constant @xmath275 . the minimum experimental temperature @xmath277 , but formula ( [ e37 ] ) is applicable , if there are no c - rotons in the initial state of helium , i.e. , at @xmath65 . because the width does not depend on @xmath278 already at @xmath276 , we may assume that the width at @xmath65 is the same as that at @xmath277 . the creation of a c - phonon ( or a c - roton ) is caused by the c - photon @xmath5 c - phonon transition . in order to calculate its probability , we should divide @xmath245 ( [ e37 ] ) by the number of c - photons in a resonator , @xmath279 we can calculate the latter by dividing the total energy of the em field of the resonator , @xmath280 by the energy of a c - photon , @xmath281 . since the field outside of the disk is weak , we will consider only the field inside the disk determined by formulas ( 17@xmath4 ) and ( 18@xmath4 ) . since @xmath282 ^ 2 \approx 2\cdot 10^{-3}r_{d}^{2 } \label{a54}\ ] ] and @xmath283 , we can obtain @xmath284 , \label{a55 } \end{aligned}\ ] ] where the small terms contributing to @xmath285 are omitted . using ( [ a55 ] ) , it is easy to evaluate the total energy and the number of c - photons in the pumping band ( @xmath286 ) for the roton mode ( @xmath287 ) and the roton frequency : @xmath288 @xmath289 in a vicinity of the roton minimum , we have @xmath290 with @xmath291 , @xmath292 . then the probability @xmath293 of the c - photon @xmath5 c - roton process for the experimental width @xmath294 is @xmath295^{2 } + \frac{fa(k , l)}{1- f}\right ) \times \nonumber \\ & \times & \left ( \frac{0.95 cm}{r_{d}}\right ) ^{2 } \approx 3.36\cdot 10^{-7}\omega_{rot}. \label{a57 } \end{aligned}\ ] ] this process weakens the flow of photons propagating from the resonator . we now determine the absorption line width . to this end , we consider properties of the resonator . in the experiment @xcite , the pumping signal with the frequency band @xmath296 and the power @xmath297w was firstly switched - on . but , as a result of losses , the resonator received @xmath298w . the resonator accumulates and amplifies the pumping signal , but , in this case , the energy losses in the resonator increase also , until a stationary equilibrium state is established . in this state , the losses in the resonator on the emission are equal to the pumping . in other words , the condition of equilibrium for the em field of the resonator with frequencies in the pumping band is the equality of the pumping energy flow and the losses : @xmath299 where @xmath300 and @xmath301 are , respectively , the number of c - photons in the resonator and the mean emission time of a c - photon by the resonator ( obviously , it is also the period of renewal of em modes of the resonator ) . by the value of @xmath302 we can estimate the electrical signal formed by an antenna catching photons emitted by the resonator as @xmath303w . it follows from the distribution of the stationary em field of the resonator @xcite that this field induces in the antenna the signal @xmath304w , weaker by 13 orders . in other words , the signal generated by the antenna is due to photons emitted by the resonator , rather than to the stationary em field of the latter . when the frequency of the em field approaches the roton one , then , according to ( [ a57 ] ) , the probability of the emission of c - rotons by the resonator becomes large . we now consider the frequency interval @xmath305 , which is much less than the line width and the pumping band but contains a macroscopic number of quanta . if the emission of c - rotons occurs , then the condition of equilibrium for the band @xmath306 takes the form @xmath307 where @xmath308 is a duration of the emission of a c - roton by c - photon of the resonator , and @xmath309 is the number of c - rotons emitted for a time interval @xmath310 by c - photons from the band @xmath306 . in this case , the losses of the resonator are separated into the channels of emission of c - photons and c - rotons . respectively , the flow of emitted c - photons decreases , which is manifested in the resonator spectrum as the absorption . as was mentioned above , the signal is equal to @xmath311 of the background one on the edge of lines . hence , @xmath312 of the losses of the resonator are transferred into c - rotons , whereas @xmath311 pass into c - photons . thus , each c - photon of the resonator in the frequency band @xmath306 for the time @xmath301 emits a c - photon with a probability of 0.8 and a c - roton with a probability of 0.2 . this yields @xmath313 , which allows us to write the following condition for the line edge : @xmath314 relations ( [ a56 ] ) , ( [ eq1 ] ) yield @xmath315s , which corresponds to the experimental value @xmath316s . using the last value of @xmath301 and ( [ lin ] ) , we obtain @xmath317 for the line edge : @xmath318 , which is only by a factor of 1.5 less than the theoretical value of @xmath317 ( [ a57 ] ) . k. @xmath319 experiment @xcite , dotted line theory ( by the present work ) . the frequency is equal to that indicated in the figure plus @xmath320 ( this number is the roton energy at @xmath321k ) . , width=321 ] the theoretical and experimental lines are shown in fig . 1 . at @xmath322 , the amplitude of the former is given by the formula @xmath323 . in this case , the error of the line width is about one order ; it related to the approximate character of the solutions for the wf of a c - phonon ( not completely correct consideration of boundary conditions ) and for the em field of the resonator , as well as to the neglect of a deviation of a symmetry of the system far from the disk ( antenna , container s wall , etc . ) from the cylindrical symmetry . as is seen from fig . 1 , the theoretical line corresponds approximately to the experimental one by width , but does not by shape . the density of states @xmath324 is high in the region above the energy @xmath325 of a roton ( @xmath326 at @xmath327 ) . below @xmath328 no roton states are present , and the density of states decreases sharply to a value corresponding to the linear phonon curve : @xmath329 . therefore , for the c - photon @xmath5 c - roton transition , the line must sharply fall on the side of lower frequencies , whereas the experimental line is almost symmetric . this difference means that the process is more complicated and involves at least two particles , rather than one particle ( c - photon @xmath5 c - roton ) , as was accepted above . in particular , one more c - phonon can be created , or a part of the angular momentum of a c - photon can be absorbed by the disk . we hope to clarify the mechanism in the future investigations . it is necessary also to take into account that a spreading of the c - roton energy which is due to the interaction between quasiparticles can also contribute to the line width . however , this leads to a very large line width of the order of magnitude of the width @xmath330 in neutron experiments . this value is larger by 6 orders than the width of the narrow shf line and by 4 orders than the mode width . therefore , this spreading can not be observed for a single mode , but it is revealed as the `` pedestal '' ( base ) on many modes @xcite . the number of photons approaching the receiver is diminished not only due to the creation of c - rotons by the field of the resonator , but also due to their creation by c - photons already emitted by the resonator . the time - of - flight of these photons to the receiver is @xmath331 which is less by 5 orders than the duration of the emission of a c - roton by a photon of the resonator . in this case , the probabilities of the emission of a c - roton for a standing photon of the resonator and a running photon are of the same order , because a standing c - photon is simply a superposition of two radial c - photons propagating toward each other . therefore , the line is formed not due to the absorption of running photons , but , as we have assumed , due to the absorption of `` standing '' photons of the resonator . while calculating formula ( [ a57 ] ) , we used the solution for the em field @xcite , for which the sewing in the corner region ( @xmath332 ) is approximate . in this region , the em field is weak . but , nevertheless , it influences the value of @xmath317 . to evaluate this influence , we calculated the probability @xmath317 for two other distributions of the em field in the corner region : @xmath333 and @xmath334 $ ] , @xmath335 . the latter can be sewed with solutions in other regions , but it does not completely satisfy the equation . moreover , if we use @xmath336 instead of @xmath337 , then the equation is satisfied , but the sewing can not be executed . the value of @xmath317 for such solutions differs from ( [ a57 ] ) by several times ; therefore , @xmath317 is not very sensitive to the sewing . the analysis indicates that it is of importance to correctly set the solution in the corner region near the joint with adjacent regions , and it is not so important in bulk . solution ( 32@xmath4 ) , ( 33@xmath4 ) satisfies this condition , so that its use is justified . as was mentioned above , the probability ( [ a57 ] ) of the c - photon @xmath5 c - roton process grows strongly at @xmath338 , i.e. , near the points of an extremum of the dispersion curve @xmath339 . this peculiarity explains why the narrow line is observed namely at the roton frequency and predicts the possibility to find one more line at the frequency of the maxon maximum , @xmath340 @xcite ( up to now , the frequencies @xmath341 were studied ) . let us substitute the maxon parameters ( @xmath342 , @xmath343 , @xmath344 ) in ( [ a57 ] ) , and let us take into account that , in the denominator , the number @xmath345 , and , according to ( 19@xmath4 ) and ( 20@xmath4 ) , it should be @xmath346 for the resonance mode . we obtain @xmath347 . thus , without regard for an additional factor ( see above ) , we have that if @xmath301 is approximately identical for the maxon and roton lines , then the width of the maxon line must be larger by a factor of 1.7 ( at @xmath65 ) than the width of the roton line . the peculiarity at @xmath338 is well known in solid - state physics as the van hove singularity . at @xmath348 the states of c - rotons falling in the small given energy interval are strongly concentrated . respectively , the transition probability in this energy interval sharply increases . in crystals , the narrow lines of light absorption @xcite and neutron scattering @xcite were registered a lot of times . however , the width of the latter is larger by several orders than those of shf lines in helium and corresponds to the pedestal . thus , the narrow shf line in helium is related to the van hove singularity , like the lines of crystals , but its widening is caused by another mechanism . in neutron experiments with liquid helium , the analogous very narrow peaks must be observed on the scattering curve @xmath349 at the frequencies of the roton and maxon extrema . however , the high error of neutron measurements ( @xmath350 ) does not allow one to register these peaks . in addition to the processes considered above , one more channel is possible : c - photon @xmath5 p - phonon + the transfer of a momentum to the disk and the transfer of an angular momentum to the disk or for the creation of a vortex rotating around the disk . such processes must be less probable , since the greater the number of quasiparticles participating in a process , the less is the probability of the process . moreover , what is more important , the overlapping of the wave functions of the em field and a p - phonon is slight due to different symmetries . in order to explain the appearance of the narrow line of absorption , the authors of work @xcite proposed to consider the following process : p - photon @xmath5 p - roton + the transfer of a momentum to helium as a whole . in our opinion , the transfer of a momentum to the disk can be more probable in such an approach , since the disk is a tougher system as compared with helium . the approach with a plane photon and a plane phonon is the main alternative to the above - considered process with c - rotons . but the latter is , apparently , more probable by two reasons . + 1 ) as was mentioned above , the line is formed by the stationary em field of the resonator , rather than by running photons . the em field outside of the resonator can be represented as a superposition of plane waves ( wave packets ) . but the total em field of the resonator is a sum of c - photons localized outside and inside the disk ( see @xcite ) . at the creation of a c - roton , a c - photon disappears _ as a whole _ inside and outside of the disk . but , due to different values of @xmath351 of helium and the disk , such a c - photon can not be presented in the form of a superposition of photons which are plane in both helium and the disk ( a photon is not plane or in helium , or in the disk ) . in other words , we must be based in the input equations on the _ circular _ field of the resonator . in addition , the properties of this field are unlike those of plane waves , because the field of the resonator sharply drops with increase in the distance to the resonator . + 2 ) in addition , the approach with p - photons requires the expansion of the field in multipoles . here , the main contribution is given by the term with the dipole moment ( dm ) @xmath352 of a roton . it was assumed in work @xcite that such a stationary dm arises in a roton due to the mutual polarization of atoms , and the probability of the p - photon @xmath5 p - phonon transition is proportional to @xmath353 . it was also proposed in @xcite that a vortex ring possesses an intrinsic dm . however , both quasiparticles create the reciprocal motion of atoms : a part of atoms moves forward , but the similar part moves backward . in this case , the separated direction is set by the velocity of a quasiparticle . therefore , the appearance of a dm is related to the asymmetry of a quasiparticle relative to the forward - backward directions . the question about the presence of such an asymmetry can be clarified with the help of the following reasoning ( which belongs to yu.v . shtanov ) . by definition , the stationary dm of a quasiparticle is equal to @xmath354 where @xmath355 is the wf of a quasiparticle . let us make inversion of time @xmath356 . in this case , we have @xcite @xmath357 , and dm ( [ vr ] ) is not changed . but the dm must be directed along the velocity of a quasiparticle , i.e. , it should change the sign . this implies that the dm is zero . it was considered in @xcite that , by the cpt - theorem , the charges also change their signs at @xmath356 , which gives @xmath358 . however , the change @xmath356 in the equation can be performed formally without any connection with the time arrow . then the charges conserve their signs at @xmath359 and the dm turns out zero . the real quasiparticle is a wave pocket , but it is clearly that in this case @xmath360 also . such a consideration is not valid if the state of a quasiparticle is degenerate . in other words , at given @xmath361 and @xmath362 there are the states with dms @xmath363 and @xmath364 which can transit to each other at @xmath365 however , we have no reasons to consider that a roton or a ring has such a degeneration . for @xmath366 there appears another possibility , if the reflected state is unstable and transits in a stable one with the inverse dm . this internal irreversibility can be related to the ordering of deformations of the electron shells of atoms induced by the interaction with neighbors . for clearness , we note that this is similar to a flag on a moving car . such a flag points out always the direction opposite to one of the motion . at @xmath367 we obtain a flag indicating the direction of motion , i.e. , we obtain the unstable state . a similar structure of a ring or a roton is possible in principle . but it is improbable , especially for a roton representing reciprocal oscillations of the density . it is of importance that such a change of the symmetry , like that at @xmath368 , will happen at the reflection of a quasiparticle from the wall . at each reflection , the quasiparticle must loss energy . hence , it will be unstable . such an instability is possible for rings , and it could explain why the rings are not discovered in the spectrum of quasiparticles or by the contribution to the heat capacity till now . but such an instability seems impossible for phonons and rotons . thus , a roton has no stationary dm . this is also the case for a ring , most probably . this reasoning implies that , while explaining the narrow line , the circular symmetry of the problem must be taken into account already in the input equations . as is known from the general theorems of quantum mechanics , a many - particle system located in a finite volume possesses a discrete energy spectrum . liquid @xmath0he in a vessel is the system of this kind . the energy levels of he ii can be determined from the wf of he ii calculated for the zero boundary conditions . from here , it is obvious that the real energy spectrum of he ii is not a landau continuous curve , but it is a collection of separate disconnected points very densely lying on this curve . according to ( [ i42a ] ) , ( [ i42b ] ) , the observed roton line @xcite consists of @xmath370 individual roton lines . if the experiment will be executed with a film of helium @xmath371 in thickness , the distances between lines increases by 6 orders of magnitude , and they will become resolvable . in this case , instead of a single roton line , we will measure many lines in the wide range of the frequencies : @xmath372 . thus , it will be possible to observe for the first time the line spectrum of a fluid consisting of a huge number of discrete lines , like the spectrum of an atom . it is only necessary that the intensities of lines be sufficiently high . however , if the resonator disk is only covered by a helium film , no lines will be observed . indeed , by ( [ e37 ] ) , the intensities of lines will be of the same order of magnitude as those for a thick layer of helium @xcite . but the line registered for such a layer consists of @xmath370 individual lines which are too weak to be observed separately . to resolve them , one needs to increase their intensity by 4 - 5 orders of magnitudes . this is a task for future studies . the experiment @xcite revealed one more unusual effect : as the power @xmath373 of a heat gun increases , the amplitude @xmath374 of the roton absorption peak decreases , and this occurs stepwise . this fact testifies to the `` quantization '' of the roton line amplitude . it was noted in @xcite that this effect can be related to the quantization of the azimuth velocity @xmath375 ( around the resonator ) , but the nature of this connection is not clear yet . since the growth of @xmath374 means a decrease in the number of c - rotons created by shf - photons , we assume that each step of @xmath374 means a decrease in the number of created c - rotons by some integer . in he ii near the resonator , two competing processes occur : c - photons create c - rotons and _ vice versa_. in this case , a c - roton can transit only in a c - photon with the same energy and the same @xmath1 . from the state @xmath376 with @xmath300 c - photons and @xmath377 c - rotons , the transition in two following states is possible : _ i _ ) the state @xmath378 , if a c - roton creates a c - photon ; _ ii _ ) the state @xmath379 , if a c - photon creates a c - roton . since circular photons and rotons are bosons , we can associate the creation operators with them . within the formalism of secondary quantization for bosons , we obtain the transition probability ( _ i _ ) @xmath380 for ( _ ii _ ) , we have @xmath381 whence @xmath382 where @xmath383 at @xmath384 , @xmath385 . in other words , @xmath386 is the above - calculated probability ( [ a57 ] ) of the c - photon @xmath5 c - roton transition in the case where there is a single c - photon in the initial state of the system , and there are no c - rotons . if @xmath387 , then the roton line in the spectrum of an shf signal is a line of absorption of photons ; but if @xmath388 the roton line is a line of their emission ( `` maser '' effect ) . in the experiment , the absorption line becomes weaker , as the power of a heat gun increases and transits to the emission line at @xmath389 . we assume that these facts are related to the forced creation of c - rotons by a gun . though all details of the mechanism are unclear up to now , we indicate several points . \a ) a gun is directed along a tangent to the disk , i.e. , so that the transfer of an angular momentum should be maximum . the p - rotons have no angular momentum , but the c - rotons have ( @xmath7 ) . therefore , their creation is accompanied by the transfer of both the energy and the angular momentum . it was noticed in @xcite that a step of @xmath373 corresponds to an increase in @xmath390 on the output of a gun by a value coinciding with the quantum @xmath391 of a circular velocity of helium near the resonator , according to the formula @xmath392 . the maximum experimental number of steps was @xmath393 . such a quantization of the velocity can mean that a gun creates a vortex in the superfluid component near the disk , whose axis coincides with the disk axis , and each step corresponds to the increase of the circulation by 1 . on the other hand , the action of the operator of azimuth momentum @xmath394 on the factor @xmath395 of the wf of a c - roton gives @xmath396 . dividing it by the @xmath0he atom mass , we obtain the azimuth velocity @xmath397 . thus , a c - roton induces the quantized rotation of helium atoms around the resonator with the velocity @xmath398 , like a vortex . but this rotation is accompanied by simultaneous oscillations . in turn , an increase in the circular velocity of helium @xmath399 by an external factor must stimulate the creation of c - phonons ( including c - rotons ) . apparently , a gun creates a macroscopic vortex in helium ( or a superposition of vortices ) and many plane and circular phonons and rotons , so that the ensemble of quasiparticles near the disk is quite complicated . moreover , some processes ( e.g. , decays ) are allowed for c - phonons and forbidden for p - phonons . \b ) the results of experiments with a gun testify unambiguously that an em wave creates namely c- , rather than p - rotons . if the latter would be created , then a decrease in the amplitude of the roton absorption line would be related , according to ( [ w - w ] ) , only to that a gun increases their number . but it increases also at a simple increase in the temperature ( without switching - on a gun ) ; however , the measurements show that this does not cause a weakening of the roton line and its transition into an emission line . \c ) in a disk resonator , `` left '' ( l ) and `` right '' ( r ) c - photons differing by the sign of @xmath1 are created . a waveguide captures the total signal from the resonator , i.e. , the summary field of l- and r - waves . in this case , the amplitude of one of these waves in the resonator is @xmath400 times greater than another one @xcite . however , a gun creates c - rotons with only a single polarization , l- or r- , which depends on the position of a gun . since an l c - photon can induce only an l c - roton ( the same is true , respectively , for right ones ) , eq . ( [ w - w ] ) should be written separately for l- and r - quanta . then we obtain that the summary absorption line disappears if the relation @xmath401 holds true for the dominant photon mode . a gun is able to weaken the line if the signs of @xmath1 for a c - roton and the dominant c - photon coincide and can not weaken if the signs do not coincide . thus , let a gun weaken the line at a given configuration of the em field . but if a gun is reoriented so that it twists he ii in the opposite direction , it must stop to weaken the line . it seems to us that this assertion can be easily verified in experiments . finally , we mention the observation of the linear stark effect in helium - ii the roton absorption line splits into two lines in a constant electric field @xmath402 directed in the disk plane . the distance between them increases @xmath403 @xcite . in @xcite , the authors advanced the idea of the relation of the effect to a possible quadrupole or instantaneous dipole , respectively , moment of a p - roton . we agree that a p - roton as a wave packet can possess an instantaneous dm @xmath352 with fluctuating ( or pulsing ) value and direction . this dm is due to , in particular , the interaction of the roton with neighboring quasiparticles located nonuniformly . such a dm induces an addition @xmath404 to the roton energy . since the projection of @xmath352 on @xmath402 takes the values in the continuous band @xmath405 $ ] , the mentioned addition transforms the roton level to the band . however , experiments demonstrate the splitting of the line into two ones , rather than a single band . thus , the instantaneous dm of a roton can not explain the observed line splitting . according to quantum mechanics , the effect can be explained if the roton energy level possesses at least a twofold degeneration which is taken off by the field @xmath402 . in our opinion , the effect is determined by the existence of right ( @xmath406 ) and left ( @xmath407 ) c - rotons with the same energy , whose superposition is described by the wave function @xmath408 since a perturbing potential is obviously proportional to the applied field @xmath402 , we obtain the splitting of a twofold degenerate energy level which is proportional to @xmath409 it can be evaluated by the well - known formulas of perturbation theory for a degenerate state . the field @xmath402 induces the polarization of the dielectric resonator , which is directed along @xmath410 and takes off the degeneration , by breaking the circular symmetry . but the multiple degeneration by @xmath411 remains ( several tens of small @xmath411 significantly contribute to the line ) . so , if the field @xmath402 will be directed along the @xmath412 axis , then the line must be split into two lines due to the removal of the degeneration relative to a change in the sign of @xmath411 . at the switching - on of a heat gun , we may expect the following . in the field @xmath410 the eigenfunctions of the hamiltonian of he ii are not the @xmath170- and @xmath413-functions , @xmath414 and @xmath415 , but their superpositions @xmath416 characteristic of a twofold degenerate level @xcite . the gun oriented along a tangent to the disk creates the quantized rotation of helium in a _ single _ direction and increases the number of c - rotons with one ( @xmath170 or @xmath413 ) polarization . it is obvious that , in this case , the gun can not excite the states @xmath416 ( @xmath417-superposition of two c - rotons with the counter rotation ) . therefore , if the field @xmath402 was already switched - on ( earlier than the gun ) , the gun will not decrease the peaks of the split roton line and will not induce the maser effect . it would be of interest to verify this prediction in experiments in similar fashion . it is seen from the above - presented analysis that the experiment @xcite has revealed the existence of particular excitations in he ii circular rotons which are azimuth sound waves . in the present work , we have approximately calculated the probability of the creation of a circular roton by the em field of the resonator and , on its basis , have evaluated the width of the absorption line at the roton frequency . the theoretical line is close to the experimental one by width but differs by shape . we have also advanced the assumption that the splitting of the line into two ones in a constant electric field is caused by the presence of right and left c - rotons and , respectively , by the twofold degeneration of the energy level of a circular roton . 200 a. rybalko , s. rubets , e. rudavskii , v. tikhiy , s. tarapov , r. golovashchenko , and v. derkach , phys . b * 76 * , 140503(r ) ( 2007 ) . a.s . rybalko , s.p . rubets , e.ya . rudavskii , v.a . tikhiy , yu.m . poluektov , r.v . golovashchenko , v.n . derkach , s.i . tarapov , and o.v . usatenko fiz . nizk . temp . * 35 * , 1073 ( 2009 ) [ low temp . phys . * 35 * , 837 ( 2009 ) ] . v.m . loktev and m.d . tomchenko , ukr . * 55 * , 901 ( 2010 ) [ www.ujp.bitp.kiev.ua/files/file/papers/ 55/8/550807p.pdf ] . loktev and m.d . tomchenko , phys . b * 82 * , 172501 ( 2010 ) . l.d . landau , e.m . lifshitz , _ quantum mechanics . non - relativistic theory _ ( pergamon , new york , 1980 ) . berestetskii , e.m . lifshitz , and l.p . pitaevskii , _ relativistic quantum theory _ ( pergamon press , oxford , 1982 ) . w. byers brown and d.m . whisnant , mol . phys . * 25 * , 1385 , * 26 * , 1105 ( 1973 ) . v. m. loktev and m. d. tomchenko , j. phys . b : at . mol . . phys . * 44 * , 035006 ( 2011 ) ; dop . nats . akad . n 5 , 76 ( 2010 ) ( in russian ) . tomchenko , arxiv : cond - mat/1003.4389 ( 2010 ) . e. janke , f. emde , f. lsch , _ tafeln hherer funktionen _ ( teubner , stuttgart , 1960 ) . rybalko , private communication . gibbs , k.h . andersen , w.g . stirling , h. schober , j. phys . cond . mat . * 11 * , 603 ( 1999 ) . davydov , _ theory of molecular excitons _ ( plenum , new york , 1971 ) . d. pines , _ elementary excitations in solids _ ( benjamin , new york , 1963 ) . v.m . loktev and m.d . tomchenko , fiz . nizk . temp . * 34 * , 337 ( 2008 ) [ low temp . phys . * 34 * , 262 ( 2008 ) ] . rybalko , s.p . rubets , e.ya . rudavskii , v.a . tikhiy , r.v . golovashchenko , v.n . derkach , s.i . tarapov , fiz . nizk . temp . * 34 * , 326 ( 2008 ) [ low temp 34 * , 254 ( 2008 ) ] . rybalko , s.p . rubets , e.ya . et al . _ , arxiv : cond - mat/0807.4810 ( 2008 ) . melnikovsky , arxiv : cond - mat/0808.1188 ( 2008 ) . mineev , pisma v zh . * 90 * , 866 ( 2009 ) [ jetp lett . * 90 * , 866 ( 2009 ) ] . vakarchuk , _ quantum mechanics _ ( lviv national university , lviv , 2004 ) ( in ukrainian ) .

we demonstrate that a log - linear relation does not provide an adequate description of the correlation between the masses of super - massive black - holes ( smbh , ) and the velocity dispersions of their host spheroid ( ) . an unknown relation between and may be expanded to second order to obtain a log - quadratic relation of the form ^ 2 $ ] . we perform a bayesian analysis using the local sample described in tremaine et al . ( 2002 ) , and solve for , and , in addition to the intrinsic scatter ( ) . we find unbiased parameter estimates of , and . at the 80% level the relation does not follow a uniform power - law . indeed , over the velocity range 70km / skm / s the logarithmic slope of the best fit relation varies between 2.7 and 5.1 , which should be compared with a power - law estimate of . the addition of the 14 galaxies with reverberation smbh masses and measured velocity dispersions ( onken et al . 2004 ) to the local smbh sample leads to a log - quadratic relation with the same best fit as the local sample . furthermore , assuming no systematic offset , single epoch virial smbh masses estimated for agn ( barth et al . 2005 ) follow the same log - quadratic relation as the local sample , but extend it downward in mass by an order of magnitude . the log - quadratic term in the relation has a significant effect on estimates of the local smbh mass function at , leading to densities of smbhs with that are several orders of magnitude larger than inferred from a log - linear relation . we also estimate unbiased parameters for the smbh - bulge mass relation using the sample assembled by haering & rix ( 2004 ) . with a parameterization ^ 2 $ ] , we find and . we determined an intrinsic scatter which is% larger than the scatter in the relation . = = = = = = = = # 1 # 1 # 1 # 1 @mathgroup@group @mathgroup@normal@groupeurmn @mathgroup@bold@groupeurbn @mathgroup@group @mathgroup@normal@groupmsamn @mathgroup@bold@groupmsamn = `` 019 = ' ' 016 = `` 040 = ' ' 336 = " 33e = = = = = = = = # 1 # 1 # 1 # 1 = = = = = = = = # 1 latexl-.36em.3ex-.15em t-.1667em.7ex-.125emx [ section ] [ firstpage ] black - holes - galaxies : formation

astro-ph0503435

observations of bulges in nearby galaxies reveal the presence of massive dark objects whose dynamical influence on the surrounding stars is consistent with their being smbhs ( e.g. kormandy & richstone 1995 ) . moreover , the masses of these smbhs correlate with properties of the host galaxy , including the luminosity of the bulge ( kormendy & richstone 1995 ) , the mass of the bulge ( magorrian et al . 1998 ) , the stellar velocity dispersion of the bulge ( ferrarese & merritt 2000 ; gebhardt et al . 2000 ) and the concentration of the bulge ( graham et al . 2002 ) . the tightest relation , with intrinsic scatter of @xmath23 dex , appears to be between smbh mass and bulge velocity dispersion ( tremaine et al . 2002 , hereafter t02 ) . significant effort has been invested in determining parameters that describe this relation , which is usually parameterised using the log - linear functional form @xmath24 the value of the power - law slope @xmath5 has been a matter of some debate ( see t02 for a review ) . recent estimates have @xmath25 ( t02 ) and @xmath26 ( ferrarese & ford 2004 ) . these values differ by @xmath222-sigma , a difference which may be attributable to systematic differences in the velocity dispersions used by different groups ( t02 ) . although the connection between smbhs and their host galaxies is not yet clear , it seems very likely that their evolution is closely intertwined . it also seems likely , given the small intrinsic scatter in the relation that the value of @xmath5 will yield important clues regarding the physics of smbh evolution ( e.g. silk & rees 1998 ; wyithe & loeb 2003 ; king 2004 ; miralda - escude 2004 ; saznov et al . moreover , studies of smbh demographics ( e.g. yu & tremaine 2002 ; shankar et al . 2004 ) rely on this relation to estimate quantities of astrophysical interest like the local smbh mass density . for these reasons , and given the heroic efforts that have been made to measure smbh masses and host velocity dispersions , it is important to make statistically robust estimates of parameters that describe the @xmath12 relation . rather than assume a powerlaw relation between @xmath0 and @xmath1 we take a more general approach in this paper . suppose we have an unknown relation @xmath27 . this relation may be expanded in a taylor series in @xmath3 about @xmath28km / s , which yields to second order @xmath29 ^ 2.\end{aligned}\ ] ] the zeroth and first order coefficients in this expansion correspond to @xmath7 and @xmath5 in the usual log - linear relation . however in addition to @xmath7 and @xmath5 we also consider the possibility of a log - quadratic term . the parameter @xmath6 therefore provides a measure of the deviation from a pure powerlaw of the @xmath12 relation . we show that there is a non - linear contribution to the @xmath30 relation , and that the assumption of a log - linear relation has biased estimates of the power - law slope . for the local sample of t02 we show that the variation in log - linear slope over the range of @xmath1 in the sample is larger than both the statistical uncertainty in the slope of the log - linear relation and the 2-sigma difference between the estimates of t02 and ferrarese ( 2002 ) . note that ( with the exceptions of [ choice ] and the appendices ) we restrict our attention to the statistical bias in estimation of the parameters @xmath5 and @xmath6 from smbh masses and galaxy velocity dispersions summarised in t02 . for further discussion of possible systematic errors in the measured parameters @xmath0 and @xmath1 themselves we refer the reader to t02 and to merritt & ferrarese ( 2001 ) . an outline of the paper is as follows . we first discuss possible bias in the local sample of smbhs with respect to parameter fitting ( [ biasfit ] ) . then in [ msig ] we repeat the parameter fitting analysis of t02 and demonstrate that the value of @xmath5 in a log - linear relation depends systematically on which low velocity dispersion galaxies are included in the sample . we then show that a log - quadratic form for the @xmath31 relation provides an improved fit which is not systematically sensitive to the galaxy sample ( [ quadmsig ] ) . the dependence of this conclusion on the method of smbh mass estimation and resolution of the smbh sphere of influence , as well as the definition of the velocity dispersion is discussed in . a bayesian approach to parameter estimation in the @xmath12 relation and quantification of the systematic bias in parameter estimation are described in [ bayes ] . this approach allows an unbiased analysis for the local sample which is presented in [ results ] . in [ results_bulge ] we apply our analysis to the relation between smbh and bulge mass using the sample of haering & rix ( 2004 ) . we present some discussion in [ discussion ] , before summarising our conclusions in [ conclusion ] . finally , in two appendices we analyse alternative samples to the one described in t02 . a first question that should be considered concerns whether or not the local smbh sample , and therefore the resulting estimates of the @xmath31 relation are statistically fair with respect to parameter estimation . the best - fit log - linear @xmath12 relation ( t02 ; ferrarese 2002 ) drops below @xmath32 for velocity dispersions smaller than @xmath33km / s , still within the range of @xmath1 in the local sample . however no kinematic detections of smbh masses have been published below @xmath34 . does this lack of low mass smbhs result in biased estimates of the slope in a log - linear @xmath12 relation ? ferrarese ( 2003 ) has suggested that the reason for the dearth of smbhs with @xmath35 is partly instrumental sensitivity . smbhs with masses smaller than @xmath36 are expected to reside in dwarf galaxies , with velocity dispersions below 50km / s . the sphere of influence of these smbhs would be observable only in the most nearby cases . ferrarese ( 2003 ) has produced a scatter plot of the local smbh population in smbh mass and distance based on a combination of the cfa redshift survey ( huchra et al . 1990 ) and the relation between smbh mass and bulge luminosity . for the nearest group of galaxies at @xmath37mpc , resolving the smbh sphere of influence limits detection to smbhs with masses in excess of @xmath34 . ferrarese ( 2003 ) goes on to suggest that in the nearby dwarf galaxies , individual stars become resolved , but are too faint to allow dynamical studies , and that enlargement of the kinematically detected sample to include smbhs below @xmath38 will require campaigns using telescopes with apertures in excess of 8 m , combined with resolution of better than 0.02 . one might therefore suppose that there is a selection bias in the smbh sample for smbhs that are larger than @xmath38 . such a selection bias would lead to biased estimates of the @xmath12 relation . however inspection of the local sample shows that one of the smallest smbhs observed ( n7457 ) lies at a distance well beyond where its sphere of influence would have been resolved by hst ( ferrarese 2003 ) . moreover , in the velocity range 70km@xmath39s@xmath40380km@xmath39s@xmath41 there are no galaxies in which a kinematic search has been conducted by the nuker team , and where a smbh was not found ( s. tremaine private communication ) . since these galaxies are not selected based on their expected smbh mass , there should be no bias against selection of low mass ( or high mass ) smbhs _ at a fixed velocity dispersion_. one can therefore estimate the expected smbh mass given a velocity dispersion from the local smbh sample , without bias introduced through selection . on the other hand , the sample _ is _ biased if one wants to estimate the velocity dispersion at a fixed smbh mass . the low surface brightness of galaxies with large velocity dispersions results in a maximum velocity dispersion where smbhs can be measured ( ferrarese 2003 ) . this combined with the exponential decline in the number density of such galaxies may result in an upper cutoff in the velocity dispersions found in the observed sample . a parametric fit that treated velocity dispersion rather than smbh mass as the dependent variable would therefore lead to a biased estimate of the parameters in the @xmath12 relation . clearly the selection function for the sample of smbhs is complex . in particular , the distribution of velocity dispersions in the sample does not reflect the overall distribution of galaxies . the non - uniformity of the distribution of velocity dispersions results in statistical inferences for parameters in equation ( [ fit ] ) that vary with the statistic used ( merritt & ferrarese 2001 ) . this implies that some or all of the estimates are biased . in order to make an unbiased estimate for a parameter like the power - law slope @xmath5 one must determine the bias inherent in the particular statistic used . in this paper we overcome the selection bias introduced by the inhomogeneous distribution of velocity dispersions via an investigation of the smbh mass vs. velocity dispersion relation in mock monte - carlo samples . given that we have access to a sample which is unbiased in @xmath42 as a function of @xmath1 ( the local sample , summarised in t02 ) , and in the absence of a theory where velocity dispersion is regulated by smbh mass ( theoretical prejudice ) , we follow tradition and restrict our attention to parameter fitting where @xmath43 is the dependent variable as a function of @xmath3 . tremaine et al . ( 2002 ) have compiled a list of @xmath44 galaxies ( the local sample ) with reliable determinations of both smbh mass ( @xmath0 ) and central velocity dispersion ( @xmath1 , defined as the luminosity weighted dispersion in a slit aperture of half length @xmath45 , the effective radius of the spheroid ) . the sample ( sch " odel et al . references for smbh masses in table 1 of t02 are verolme et al . ( 2002 ) ; tremaine ( 1995 ) ; kormendy & bender ( 1999 ) ; bacon et al . ( 2001 ) ; gebhardt et al . ( 2003 ) ; bower et al . ( 2001 ) ; greenhill & gwinn ( 1997 ) ; sarzi et al . ( 2001 ) ; kormendy et al . ( 1996 ) ; barth et al . ( 2001 ) ; kormendy et al . ( 1998 ) ; gebhardt et al . ( 2000 ) ; herrnstein et al . ( 1999 ) ; ferrarese , ford , & jaffe ( 1996 ) ; cretton & van den bosch ( 1999 ) ; harms et al . ( 1994 ) ; macchetto et al . ( 1997 ) ; ferrarese & ford ( 1999 ) ; van der marel & van den bosch ( 1998 ) ; cappellari et al . ( 2002 ) . ] is shown in the left panel of figure [ fig1 ] which illustrates the correlation between @xmath0 and @xmath1 . we begin by repeating the analysis of t02 who estimate the parameters @xmath7 and @xmath5 through minimisation of a @xmath46 variable that accounts for uncertainties in both @xmath1 and @xmath0 . the @xmath46 variable used is @xmath47 where @xmath48 and @xmath49 are the logarithm of smbh mass in solar masses and the logarithm of velocity dispersion in units of 200km / s respectively . the variables @xmath50 and @xmath51 are the uncertainties in dex for these parameters . this expression ( equation [ chi2 ] ) is symmetric in @xmath43 and @xmath3 . one might expect this to be a favourable property since the fit does not include any preconceived notions of the physical origin of the relation . in t02 an estimate of the intrinsic scatter @xmath52 in the @xmath12 relation [ as defined in the @xmath53 ( or @xmath0 ) direction ] was established by adding @xmath54 to the denominator in equation ( [ chi2 ] ) . the intrinsic scatter that resulted in a reduced minimum @xmath46 of unity was @xmath55 dex . the best fit solution for a linear @xmath12 relation from t02 has @xmath56 and @xmath57 , resulting in residuals for the three smallest galaxies that are greater than zero . in addition , the largest three galaxies also have residuals that are greater than zero . this behaviour is symptomatic of a scenario where a linear relation has been fitted to a non - linear sample . the sample described in t02 is dominated by galaxies with velocity dispersions in the range @xmath58km / s . there are a handful of smbhs at the centers of galaxies with smaller velocity dispersions , including the milky - way . these smbhs all have masses in excess of @xmath59 . the four galaxies with the lowest dispersions have a large influence over the slope inferred for a log - linear relation . figure [ fig1 ] shows how the estimate of @xmath5 varies as the galaxies with the lowest velocity dispersions are removed from the sample . two trends are apparent . firstly , the uncertainty in @xmath5 is reduced by more than 50% by the presence of the smallest few galaxies . secondly , the lowest velocity dispersions reduce the estimate of the slope from @xmath60 to @xmath61 . the estimate remains near @xmath62 as the next 6 smallest velocities are removed from the sample . the systematic variation may be seen visually in the left panel of figure [ fig1 ] , where the best - fit relations are plotted over the corresponding ranges of velocity dispersion . the systematic trend of a slope that increases by 1.5-sigma when the smallest 3 of @xmath63 galaxies are removed from the sample points to a fit that is systematically biased . one possibility is that a log - linear relation is not a good description of the data set . note that a poor fit can not be identified through a reduced value of @xmath46 that is significantly greater than unity . this is because the value of intrinsic scatter ( @xmath64 ) is determined by the condition that the reduced value of @xmath46 equal unity for the best fit solution . however we can try to find a functional form that provides a better description of the data . one possible scenario involves a characteristic velocity dispersion , with different powerlaws for large and small galaxies . to test this idea we attempt to locate a value of @xmath1 around which the slope of the relation changes by fitting two power - law slopes , one either side of a characteristic velocity dispersion . we find that the fit is improved by allowing the slope at high @xmath1 to be steeper than at low @xmath1 , and that the data prefers a break velocity somewhere in the range @xmath65km / s ( 1-sigma ) . of course , while there is at least some theoretical motivation for a log - linear relation , there is no reason to suppose that the @xmath12 relation should be a double power - law . in the following we take a more general approach . as discussed in the introduction , a general relation @xmath66 can be expanded in a taylor series to second order yielding @xmath67 ^ 2.\end{aligned}\ ] ] here the coefficients @xmath5 and @xmath6 represent log - linear and log - quadratic contributions to the @xmath12 relation . we have ( arbitrarily ) expanded about @xmath28km / s , which corresponds to the median velocity dispersion in the relation . the value of @xmath6 provides a measure of whether or not a log - linear relation provides a good description of the data . we have repeated the @xmath46 minimisation using equation ( [ fit_quad ] ) . the 1 , 1.5 , 2 and 2.5-sigma ellipsoids and the minimum @xmath46 is 1 , 2.71 , 4 and 6.63 respectively . when these contours are projected onto an individual parameter axis , they correspond to the 68% , 90% , 95% and 99% confidence intervals on those parameters . in the text we refer to these as the 1 , 1.5 , 2 and 2.5-sigma error elipsoids . ] for @xmath5 and @xmath6 are shown in the upper right - hand panel of figure [ fig2 ] for the local sample . an intrinsic scatter of @xmath68 dex results in a reduced @xmath46 of unity for the best fit solution . thus a log - quadratic fit admits a slightly smaller intrinsic scatter than the log - linear relation . the solution shows evidence for a positive log - quadratic term at around the 85% level . the most likely solution has @xmath69 and @xmath70 with 1-sigma uncertainties of 0.35 and 1.1 respectively . the best fit log - quadratic relation is plotted in the upper - left panel of figure [ fig2 ] . parameter estimation using a log - linear relation corresponds to the conditional probability for @xmath5 given @xmath71 . the horizontal grey line in the upper right panel of figure [ fig2 ] shows the cut through the bi - variate probability distribution for @xmath5 and @xmath6 . the contours of the bi - variate distribution cross this line centered around @xmath72 , which is consistent with expectations from the log - linear fit . however the most probable value @xmath57 for the log - linear fit lies near the 1.5-sigma contour of the log - quadratic fit . this illustrates the point that a log - quadratic form provides a much improved description of the local sample . unlike the log - linear case , the four galaxies with the lowest dispersions do not have a large influence on the log - quadratic fit . figure [ fig2 ] shows that the estimates of @xmath5 and @xmath6 do not vary systematically as the galaxies with the lowest velocity dispersions are removed from the sample . the most likely solution is similar for samples including all galaxies and for samples including only the largest 20 galaxies , indicating that there is evidence for a log - quadratic term in the main group of massive galaxies , and that the deviation from a power - law is not dominated by inclusion of low mass galaxies in the sample . the variation of the best - fit relation as small galaxies are removed from the sample may be seen visually in the left panel of figure [ fig2 ] , where the best - fit relations are plotted over the corresponding ranges of velocity dispersion . the above discussion relates to an expansion of the general relation @xmath73 which is truncated at second order . we found that the data prefers a non - zero contribution from a quadratic term . before continuing we mention the possibility of a non - zero contribution from an additional log - cubic term . we have repeated the above analysis for the third order relation @xmath74 ^ 2\\ & + & \beta_3\left[\log(\sigma/200{\rm km } \,{\rm s}^{-1})\right]^3.\end{aligned}\ ] ] by minimising @xmath46 over the other parameters , we find a solution @xmath75 , @xmath76 and @xmath77 . not surprisingly there is a large correlation between the odd terms @xmath5 and @xmath78 . this correlation increases the uncertainties on individual parameters . despite having an additional free parameter , the log - cubic relation requires a larger intrinsic scatter to achieve a @xmath46 of unity than the log - quadratic relation . moreover there is no evidence from the @xmath46 minimization for a log - cubic term in the local sample . therefore in the remainder of the paper we restrict our attention to the log - quadratic form for the @xmath12 relation . the local smbh sample described in t02 contains galaxies with smbh mass estimates obtained from data that did not resolve the sphere of influence ( e.g. marconi & hunt 2003 ) , and the accuracy of these masses has been called into question ( e.g. merritt & ferrarese 2001 ; ferrarese & ford 2004 ; marconi & hunt 2003 ) . in addition , the sample has smbhs with masses inferred using four separate techniques , including stellar dynamics , stellar proper motions , astrophysical masers and dynamics of gaseous disks . moreover , adding to this inhomogeneity are the different definitions for the velocity dispersion that have been supported by different groups . in this section we investigate the effect that these properties of the sample may have on the conclusion of a log - quadratic @xmath12 relation . these issues are also discussed further in two appendices ( [ app1 ] and [ app2 ] ) . some of the original smbh mass estimates based on ground based stellar dynamical data ( magorrian et al . 1998 ) lie systematically above the @xmath12 relation defined by other galaxies in the local sample with _ more secure _ mass estimates by as much as 2 orders of magnitude ( ferrarese & merritt 2000 ) . the correlation between the offset and the galaxies distance has been interpreted as evidence that the lack of resolution in these early observations resulted in a systematic resolution dependent error in the magorrian et al . ( 1998 ) masses ( merritt & ferrarese 2001 ) . of course it is not clear that one can assume masses to be incorrect because they do nt agree with the log - linear @xmath12 relation as described by the remaining smbhs . this would require the argument that the _ observational _ mass estimates must be wrong because they disagree with the a _ theoretical _ preconception that the @xmath12 relation is log - linear . however concern has been expressed that that smbh masses estimated using data that does not resolve the sphere of influence are unreliable ( e.g. ferrarese & ford 2004 ) . marconi & hunt ( 2003 ) have ordered the smbh sample in terms of the degree to which the smbh sphere of influence is resolved . there are 4 galaxies for which the sphere of influence is smaller than the resolution ( the half - width - at - half - maximum ) of the data . in addition , marconi & hunt ( 2003 ) classed some of the mass estimates as unreliable for other reasons such as unknown disk inclination , resulting in the removal of a further 3 galaxies from the sample . in the upper left panel of figure [ fig3 ] we show the remaining 24 galaxies ( dark points ) . those smbh masses designated by marconi & hunt ( 2003 ) as being unreliable , and those with spheres of influence that are not resolved are plotted in grey . in the right panel of figure [ fig3 ] we show the log - quadratic fit to these 24 galaxies . the result should be compared with the fit to the full sample , the 1-sigma error ellipse for which is also plotted ( thick contour ) . we see that the removal of the 7 _ unreliable _ galaxies has slightly increased the allowed range of @xmath6 . however the solution still prefers a non - linear term , the 1-sigma error ellipse lies near the @xmath71 line . in the lower right- and lower - left panels of figure [ fig3 ] we show the log - quadratic fits to the sample of galaxies with smbh masses determined via stellar dynamics , and by other methods respectively . each of these samples also shows evidence for a non - linear contribution to the @xmath12 relation . the sample containing smbhs with resolved spheres of influence , and the sample containing only smbhs with masses determined via stellar dynamics have very similar solutions , indicating that smbhs with masses determined via stellar dynamics in cases where the sphere of influence is unresolved do not skew the analysis and may be included in the sample . the question of the dependence of resolution on the accuracy of the smbh mass determined was investigated by gebhardt et al . they re - evaluated the smbh masses for galaxies where the smbh mass had been previously determined with ground - based imaging using higher resolution space - based data . gebhardt et al . ( 2003 ) showed that _ i ) _ the magorrian et al . ( 1998 ) masses were indeed overestimated , by up to a factor of 3 , due to the assumption of 2-integral models ; and _ ii ) _ comparison of 3-integral modeling of low - resolution ground based and high resolution space based data demonstrates that the smbh mass determinations described in gebhardt et al . ( 2003 ) contain no resolution induced bias . in other words , provided that the modeling is sufficiently general , the high resolution data increases the precision but not the accuracy of smbh mass estimates ( see figure 8 of gebhardt et al . this implies that if the resolution were not sufficient , then the lower limit on bh mass would be @xmath79 ( i.e. no definite detection ) . thus the effect of spatial resolution is built into the estimate of the range of allowable mass . the masses of smbhs in galaxies where the sphere of influence is not resolved should therefore be just as reliable as those where it is resolved , in the sense that the determination is just as accurate but carries less precision ( richstone et al . 2004 ) . at this point it should be noted that recent work from valluri , merritt & emsellem ( 2004 ) has suggested that stellar dynamical estimates of smbh masses are unreliable even when the smbh sphere of influence is resolved . this unreliability is due to a degeneracy between the mass - to - light ratio and smbh mass , which was shown to become more prominent as the number of orbits used in the modeling is increased . on the other hand , richstone et al . ( 2004 ) have countered this claim . they suggest that while it is true that the allowed range of smbh mass increases with the number of orbits assumed , this range asymptotes to provide a true estimate of smbh mass . thus they conclude that provided the number of orbits is sufficient , smbh mass estimation from reconstruction of stellar orbits using kinematic data can lead to a robust smbh mass estimate . richstone et al . ( 2004 ) estimate the number of orbits that are required to achieve convergence in the mass estimate and conclude that published mass estimates have been made using a sufficiently large orbit library and are therefore reliable . finally , the definition of the independent variable may be an important factor in the evaluation of any non - linear contribution to the @xmath12 relation . a central velocity dispersion has been advocated by ferrarese & merritt ( 2001 ) , while an effective velocity dispersion was used by gebhardt et al . the pros and cons of these different variables were summarised by merritt & ferrarese ( 2001 ) . the central velocity dispersion is more easily measured than the effective velocity dispersion which requires both surface photometry and spatially resolved spectroscopy over the effective radius of the galaxy . as a result , central velocity dispersions have been measured for many galaxies , while the effective velocity dispersions require significant effort to obtain the required data . on the other hand the effective velocity dispersion better represents the velocity dispersion over the stellar spheroid , and may therefore be expected to offer a better representation of the depth of the spheroids gravitational potential well . moreover t02 found that the smbh itself could effect the value of the central velocity by up to 30% in some cases . in order to evaluate the effect of changing the independent variable from an effective to a central velocity dispersion , we repeat the above analysis on sub - samples of the local sample of smbhs ( values of central velocity dispersion are available in merritt & ferrarese 2001 & ferrarese & ford 2004 ) . the results are shown in figure [ fig4 ] , correspond to , and are qualitatively similar to to those in figure [ fig3 ] , where the effective velocity dispersion was the independent variable . in particular , if the smbhs whose spheres of influence are not resolved are removed from the sample , then the sample still prefers a non - linear contribution . this non - linearity is at higher significance where the central rather than effective velocity dispersion is used . similarly , samples of smbhs with masses determined via stellar dynamics and by other methods each prefer a non - linear @xmath12 relation . this analysis shows the previously known result that use of the central velocity dispersion results in an estimate of the @xmath31 relation that is steeper than where the effective velocity dispersion is used . in summary , samples where the smbh masses were estimated by stellar dynamics , samples where the smbh masses were determined via other methods , and samples where the smbh masses are _ secure _ ( marconi & hunt 2003 ) , each have a fit that prefers the inclusion of a log - quadratic term . the sample of smbh masses determined via other methods shows the strongest non - linear tendency , but with the largest uncertainty . this non - linear component of the @xmath12 relation is present whether the independent variable is considered to be the central velocity dispersion ( ferrarese & merritt 2000 ) or the effective velocity dispersion ( t02 ) . in light of these results , we consider only the full local sample as described in t02 , with the effective velocity dispersion as the independent variable , in the remainder of this paper . our aim in this paper is to determine the parameters @xmath7 , @xmath5 and @xmath6 , as well as the intrinsic scatter ( designated @xmath8 ) of smbhs around equation ( [ fit_quad ] ) . our general approach adds two features relative to the @xmath46 analysis described in the previous sections . first , our approach explicitly includes the intrinsic scatter ( labeled @xmath8 ) about the mean @xmath31 relation as a third free parameter , and we consistently solve for all of @xmath7 , @xmath5 , @xmath6 and @xmath8 . the second improvement is to include the asymmetric errors quoted for @xmath42 . in the absence of errors in @xmath0 and @xmath1 we may define the contribution @xmath80 of smbh @xmath81 to the likelihood @xmath82 for the set @xmath83 . however both @xmath0 and @xmath1 contain observational uncertainty . each of the @xmath80 must therefore be averaged over the uncertainty in both @xmath0 and @xmath1 , hence @xmath84 the uncertainty in @xmath1 is described by @xmath85 for all galaxies except the milky - way for which , following t02 we assume a larger uncertainty , @xmath86 . here the notation @xmath87 refers to the value of a gaussian distribution with mean @xmath88 and variance @xmath89 at @xmath90 . the observed uncertainty in @xmath91 is assumed to follow a distribution @xmath92 here we have defined @xmath93 and @xmath94 as the uncertainty ( in dex ) above and below the observed value for smbh @xmath81 . the values of @xmath7 , @xmath5 , @xmath6 and @xmath8 may then be estimated by maximising the product of likeli - hoods for the @xmath95 residuals @xmath96 note that we have not yet specified the definition of @xmath97 . for a finite sample size , the solution for @xmath7 , @xmath5 , @xmath6 and @xmath8 is sensitive to the definition of @xmath97 . as a result , an inappropriate choice for @xmath97 can lead to a biased estimate of the parameters . in this paper we are analysing a sample where smbh mass is unbiased at a fixed value of @xmath1 . we therefore treat @xmath1 as the independent variable and model the intrinsic scatter as a gaussian with variance @xmath8 , hence @xmath98 . \end{aligned}\ ] ] the joint a - posteriori probability distributions for the parameters @xmath7 , @xmath5 , @xmath6 and @xmath8 may be found from @xmath99 where the @xmath100 are prior probabilities , which we assume to be flat in this paper , i.e. @xmath101 a - posteriori distributions for combinations of these parameters may then be obtained by marginalising over the other dimensions , for example @xmath102 or @xmath103 before presenting the probability distributions for the parameters @xmath5 , @xmath6 and @xmath8 that describe the local sample , we asses the bias in the fitting procedure by fitting parameters to monte - carlo realisations of mock smbh samples . we generate samples of @xmath63 smbhs using the following procedure . for each of the @xmath63 galaxies we select a value of @xmath1 drawn from the observed estimate @xmath104 . this value is used to select a smbh mass from an input mean relation , offset randomly according to the input intrinsic scatter @xmath105 . this mass is then further offset by a value drawn randomly from the quoted uncertainty in @xmath0 for the corresponding smbh in the observed sample . we assumed input values of @xmath106 and @xmath107 . in the following section we show that these values lie near the best fit relation for the local sample . the intrinsic scatter of the relations was assumed to be @xmath108 , defined in the @xmath53 or @xmath0 direction . for each mock sample we find the most likely values for @xmath5 and @xmath6 from equation ( [ d4p ] ) . the cumulative distributions used to calculate the mock smbh mass . ] of the best fit solutions are plotted in the upper panels of figure [ fig5 ] . the estimates are not skewed , but the parameters determined are slightly biased ( by @xmath109 and @xmath110 ) as may be seen by the comparing the median ( dashed line ) and mean ( denoted by the large dot ) to the true value ( solid vertical line ) in each case . having assessed the bias in our parameter estimation , we are in a position to estimate the parameters describing the local sample . in figure [ fig6 ] we show the a - posteriori marginalised probability distributions for the parameters @xmath5 , @xmath6 and @xmath8 computed using equations ( [ marginalised]-[marginalised2 ] ) in combination with the likelihood equation ( [ lh1 ] ) . the values of @xmath5 and @xmath6 corresponding to a fixed likelihood have been corrected a - posteriori for the biases @xmath111 and @xmath112 . in the central row we show joint distributions for @xmath6 and @xmath5 ( left ) and for @xmath8 and @xmath5 ( right ) . the contours ( dark lines ) refer to 0.036 , 0.14 , 0.26 and 0.64 of the peak height corresponding to the 4 , 3 , 2 , and 1-sigma limits of a gaussian distribution . in the bottom rows ( dark lines ) we show differential ( solid lines ) and cumulative ( dashed lines ) distributions for @xmath5 ( left ) and @xmath6 ( right ) . the vertical dotted lines show the variance . the contours show little correlation between parameters . using the likelihood defined in equation ( [ lh1 ] ) we find that the marginalised distributions imply @xmath9 and @xmath10 ( figure [ fig6 ] ) , and an intrinsic scatter of @xmath113 . the normalisation was found to be @xmath114 . the best fit ( solid line ) is plotted over the data in the upper left panel of figure [ fig6 ] , together with curves showing the level of intrinsic scatter ( dotted lines ) . in the upper right panel of figure [ fig6 ] we plot the residuals in @xmath43 , together with horizontal dotted lines showing the value of the best fit intrinsic scatter . we have assumed an intrinsic scatter that is constant with @xmath1 . inspection of the residuals in figure [ fig6 ] indicates that there is no systematic trend . the cumulative probability distribution for @xmath6 shows that there is a positive contribution ( at @xmath115 confidence ) from a log - quadratic term in the @xmath12 relation that describes the local sample . this may be interpreted as indicating that at the 80% level the @xmath12 relation does not follow a single powerlaw between @xmath33km / s and @xmath116km / s . an alternative statistical question regarding the significance of this result concerns the frequency with which one might measure a log - quadratic term as large as the best fit of @xmath117 , assuming an intrinsically log - linear relation . we have performed fits to mock samples assuming an input intrinsic slope of @xmath118 and an input @xmath119 , which corresponds to the best fit log - linear relation ( t02 ) . the cumulative distributions for the best fit @xmath5 and @xmath6 are plotted in the lower panels of figure [ fig5 ] . we find that the estimate of @xmath5 is unbiased in this case , and that the best fit value of @xmath6 is greater than 1.6 around @xmath120 of the time . this is consistent with the statement that the local sample does not follow a single power - law at the @xmath121 level . the inclusion of a log - quadratic term in the fit leads to residuals that do not lie systematically above the mean relation at high or low @xmath1 . we can therefore use the behaviour of these residuals to interpret the significance of the log - quadratic term . our monte - carlo simulations ( figure [ fig5 ] ) show that ( after accounting for the bias ) @xmath122 of samples with @xmath123 and @xmath117 produce best fits with @xmath124 . as we saw in figure [ fig1 ] the smallest three galaxies all have smbhs that are more massive than the average implied by a log - linear fit . each smbh has @xmath125 chance in 2 of lying above the mean relation . this situation would therefore arise by chance for @xmath125 sample in 10 ( @xmath126 ) , which provides an intuitive understanding of the finding that there is a log - quadratic term in the @xmath12 relation described by the local sample that is positive at the 90% level . in this section we apply our analysis to the sample of smbhs and bulge masses summarised in haering & rix ( 2004 ) . this sample closely resembles that described in t02 . the smbh - bulge mass relation is shown in the upper left panel of figure [ fig7 ] , and can be described by a log - quadratic relation of the form @xmath128 ^ 2.\end{aligned}\ ] ] we have repeated our @xmath46 analysis on the @xmath129 sample ( not shown ) . we find that the slope estimated using a log - linear fit systematically increases by an amount that is greater than the statistical uncertainty when the smallest galaxies are removed from the sample . a @xmath46 minimisation of a log - quadratic relation provides a significantly better fit , with the power - law relation ruled out at the 1-sigma level . the parameters of the log - quadratic fit do not vary systematically as galaxies are removed from the sample . motivated by the success of the log - quadratic fit for the @xmath130 relation in a @xmath46 minimisation , we perform a bayesian analysis in analogy to the one described in the previous section . we define a likeli - hood function @xmath131 where @xmath132 is the likelihood of a set of @xmath133 and @xmath0 given a relation described by @xmath134 , @xmath135 , @xmath136 and @xmath137 . following haering & rix ( 2004 ) the uncertainty in @xmath133 is described by @xmath138 for all galaxies . at fixed bulge mass , we model the intrinsic scatter as a gaussian with variance @xmath137 , hence we define a likeli - hood in analogy with equation ( [ lh1 ] ) @xmath139 ^ 2\right ] . \end{aligned}\ ] ] we performed monte - carlo simulations as before and find that the estimator described in equation ( [ lh1_bulge ] ) leads to a slightly biased estimate for the true values @xmath140 and @xmath141 . we find biases of @xmath142 and @xmath143 . in figure [ fig7 ] we show the a - posteriori marginalised probability distributions for the parameters @xmath135 , @xmath144 and @xmath137 computed using the likelihood equation ( [ lh1_bulge ] ) and corrected for bias . in the central row we show joint distributions for @xmath136 and @xmath145 ( left ) and for @xmath137 and @xmath135 ( right ) . the contours ( dark lines ) refer to 0.036 , 0.14 , 0.26 and 0.64 of the peak height corresponding to the 4 , 3 , 2 , and 1-sigma limits of a gaussian distribution . in the bottom rows we show differential ( solid lines ) and cumulative ( dashed lines ) distributions for @xmath135 ( left ) and @xmath136 ( right ) . the vertical dotted lines show the variance . we find that the marginalised distributions imply @xmath146 , @xmath147 and an intrinsic scatter of @xmath148 . the normalisation is @xmath149 . the best fit relation has a positive log - quadratic term . however this term is only required by the data at the 1-sigma level . haering & rix ( 2004 ) suggested an upper limit on the intrinsic scatter of @xmath23 dex , though the definition of this scatter was not specified . we have included intrinsic scatter self consistently in the analysis and find that the intrinsic scatter in smbh mass at fixed bulge mass is @xmath150 dex . this scatter is larger than the quoted uncertainties on either smbh or bulge mass . the scatter in the @xmath127 relation is also @xmath2250% larger than the scatter in the @xmath12 relation . the best fit log - quadratic relation is plotted over the data ( solid line ) in the upper left panel of figure [ fig7 ] together with curves showing the level of intrinsic scatter ( dotted lines ) . in the upper right panel of figure [ fig7 ] we plot the resulting residuals , together with horizontal dotted lines showing the value of the best fit intrinsic scatter . inspection of the residuals does not indicate any systematic trend of residuals as a function of @xmath133 . to quantify the extent of the departure of the local sample from a log - linear relation we have plotted the logarithmic derivatives @xmath14 and @xmath151 of the @xmath12 and @xmath127 relations as a function of @xmath1 and @xmath133 in figure [ fig8 ] . the solid lines show the slopes of the best fit relations . the dashed lines show the slope of relations at the edges of the 1-sigma error ellipsoids in @xmath152 ( left panel ) or @xmath153 ( right panel ) space , while the grey stripes show the 1-sigma range for the slope of the log - linear fits from t02 and haering & rix ( 2004 ) respectively . the figure shows that in both cases the logarithmic slope of the log - quadratic relation varies substantially more over the range of @xmath1 or @xmath154 in the smbh sample than the statistical uncertainty in the slope @xmath5 of a log - linear fit . this fact has contributed to statistical bias and inconsistent results when comparing the log - linear slopes of the @xmath12 relation between different samples of smbhs . in the remainder of this section we discuss various implications of a log - quadratic @xmath12 relation . the possibility of a log - quadratic @xmath0-@xmath1 relation leads to several interesting consequences . the first concerns the velocity dispersion of galaxies in which the most massive smbhs reside . the mass of the largest smbhs in the sloan digital sky survey ( sdss ) survey volume was discussed in wyithe & loeb ( 2003 ) . we repeat part of the discussion here . consider a smbh of mass @xmath155 . we can estimate the co - moving density of black - holes of this mass from the observed quasar luminosity function and an estimate of quasar lifetime . if we assume eddington accretion , then at the peak of quasar activity , quasars powered by @xmath156 smbhs had a comoving density of @xmath157gpc@xmath158 . wyithe & loeb ( 2003 ) found that the number of quasars relative to the number of dark - matter halos at @xmath159 implies a duty cycle for quasars of @xmath160 . the comoving density of smbhs with masses in excess of @xmath161 at @xmath159 was therefore @xmath162gpc@xmath158 . as this corresponds to the peak of smbh growth , the smbh density should match the value observed today . this density implies that the nearest smbh of @xmath163 should be at a distance @xmath164mpc which is comparable to the distance of m87 , a galaxy known to possess a smbh of this mass ( ford et al . 1994 ) . what is the most massive smbh that can be detected dynamically in the sdss ? the sdss probes a volume of @xmath125gpc@xmath165 out to a distance @xmath166 times that of m87 . at the peak of quasar activity , the density of the brightest quasars implies that there should be @xmath167 smbhs with masses greater than @xmath168 per gpc@xmath165 , the nearest of which will be at a distance @xmath169mpc , or 5 times the distance to m87 . the radius of gravitational influence of the smbh scales as @xmath170 . we therefore find that for the nearest @xmath171 and @xmath172 smbhs , the angular radius of influence should be similar . thus the dynamical signature of the nearest @xmath168 smbhs on their stellar host should be detectable . in order to find these most massive smbhs one needs to identify the galaxies in which they reside . we estimate the value of @xmath1 implied for smbhs of mass @xmath173 using the mean @xmath31 relation . if one adopts the mean log - linear relation ( t02 ) , then smbhs with masses in excess of @xmath174 should reside in galaxies with @xmath175km / s ( see figure [ fig11 ] ) . however no such galaxies exist is the sdss ( e.g. sheth et al . 2003 ) , where largest values of galaxy velocity dispersion are found to be @xmath176km@xmath39s@xmath41 . the mean log - quadratic relation reaches @xmath177 at a more modest but still unobserved velocity dispersion , @xmath178km / s . however the intrinsic scatter in the @xmath31 relation is @xmath179 dex . we find that smbhs of mass @xmath173 differ by @xmath180 from the mean log - quadratic relation at @xmath176km / s . the most massive smbhs with masses of @xmath173 inferred from quasars at @xmath159 should therefore exist in the sdss and would lie at @xmath181 above the extrapolated mean log - quadratic relation for galaxies with the largest measured velocity dispersions of @xmath182km / s . this helps to reconcile the number of luminous quasars observed at @xmath159 with both the local @xmath12 relation and the lack of super massive galaxies . a positive log - quadratic term has a significant effect on the upper end of the smbh mass - function . in figure [ fig9 ] we show the smbh mass - function computed by combining the @xmath12 relation with the velocity dispersion function of sheth et al . two mass functions are shown . the grey stripe and dashed lines represent the ranges of uncertainty in the estimate of the mass - function obtained assuming log - linear and log - quadratic @xmath12 relations respectively . the ranges at fixed @xmath0 in each case correspond to the 16th and 84th percentiles , and were computed taking into account gaussian uncertainties in both the parameters of the @xmath31 relation and the velocity - dispersion function . the solid line shows the most likely estimate for the mass - function corresponding to the log - quadratic @xmath12 relation . figure [ fig9 ] shows that inclusion of a log - quadratic term in the @xmath12 relation results in an estimate for the number density of smbhs with masses of @xmath173 that is several orders of magnitude larger than inferred from a log - linear relation . we have estimated the mass density ( for a hubbles constant of @xmath183km@xmath39s@xmath184mpc@xmath41 ) of smbhs in the local universe . for a log - linear @xmath12 relation we find a density of @xmath185mpc@xmath158 , while for a log - quadratic @xmath12 relation we obtain @xmath186mpc@xmath158 . hence the addition of a log quadratic term does not effect estimates of the total mass density which is dominated by smbhs in the range @xmath187 ( aller & richstone 2002 ) . the sample of kinematically detected smbhs will not grow by a large factor in the foreseeable future ( ferrarese 2003 ) . progress in understanding the statistical properties of the smbh population may instead come via estimates of smbh masses that are based on reverberation mapping studies ( gebhardt et al . 2000b ; ferrarese et al . 2001 ; onken et al . 2004 ; peterson et al . 2004 ; nelson et al 2004 ) . onken et al . ( 2004 ) have presented a sample of 14 smbhs with virial masses ( @xmath188 ) determined via reverberation mapping studies , in galaxies with measured velocity dispersions . reverberation masses are determined up to a geometric factor @xmath189 , hence the smbh mass is @xmath190 . onken et al . ( 2004 ) have determined the average value @xmath191 by comparing the reverberation masses of these 14 smbhs , to the masses of smbhs in the local sample using a log - linear @xmath12 relation . they found @xmath192 over a range of values for the slope @xmath5 that include @xmath193 ( ferrarese 2002 ) and @xmath57 ( t02 ) . assuming this scaling factor we can combine the 14 galaxies used to obtain @xmath191 in onken et al . ( 2004 ) with the local sample of smbhs to form a sample of 45 galaxies . the left panel of figure [ fig10 ] shows the @xmath12 relation for the local sample ( solid dots ) , and its comparison to the agn smbh masses and velocity dispersions ( open circles ) presented in onken et al . we have repeated the @xmath46 minimisation using equation ( [ fit_quad ] ) for the combined sample of @xmath194 smbhs . in the right - hand panel we show the 1 , 1.5 , 2 and 2.5-sigma ellipsoids for @xmath5 and @xmath6 . the most likely solution has @xmath195 and @xmath196 with 1-sigma uncertainties of 0.3 and 1.0 respectively . the resulting best - fit @xmath12 relation is over - plotted in the left panel ( solid line ) . this solution is very close to the one obtained using only smbhs in the local sample ( @xmath197 , @xmath198 ) . the intrinsic scatter required for a reduced @xmath46 of unity in this fit was @xmath68 dex , indicating that the addition of the agn smbhs does not increase the scatter of the @xmath12 relation . the addition of agn smbhs results in a fit with a log - quadratic term which is non - zero at around the 1.5-sigma level . the sample of agn smbhs with masses determined from reverberation mapping therefore strongly supports a log - quadratic @xmath12 relation . this is despite the fact that the normalization of the reverberation smbh masses was determined via the assumption of a log - linear relation . there are no smbhs in the local sample with masses below @xmath32 . attempts to extend the @xmath12 relation to lower masses through dynamical observations have centered on two cases , m33 ( merritt , ferrarese & joseph 2001 ; gebhardt et al . 2001 ) , and ngc205 ( valluri , ferrarese , merritt & joseph 2005 ) . only upper limits have been obtained in these systems . however we can compare the log - quadratic fit in the low-@xmath1 regime with smbh masses obtained from single epoch virial estimates based on observations of agn . the upper panel of figure [ fig11 ] shows the extrapolation of the best fit log - quadratic relation for the local sample , and its comparison to the agn smbh masses and velocity dispersions presented in barth , greene & ho ( 2004 ) , as well as the point for pox52 ( barth , ho , rutledge & sargent 2004 ) . the agn smbh masses in this sample were estimated using the single epoch virial mass calibration of onken et al . the log - quadratic fit derived using the local sample appears to describe the smbh masses in these agn much better than the log - linear fit ( see right - hand panel of figure [ fig11 ] ) . the residuals are symmetrically placed around the best fit log - quadratic relation . on the other hand the dashed line in the right hand panel of figure [ fig11 ] shows the additional residual of the log - quadratic relation relative to the log - linear relation . the residuals of the agn smbhs relative to the log - linear relation are systematically positive , indicating that the log - linear relation provides a poor description . the single epoch virial smbh masses for the agn have been estimated in a regime where the technique is not directly calibrated . one must therefore be cautious comparing the agn and local samples since the agn smbh mass estimates may contain some unknown systematic uncertainty relative to the dynamical estimates of the local sample ( barth , greene & ho 2005 did not perform a fit to the combined sample for this reason ) . moreover it is possible that the agn smbh sample is biased towards large smbh masses ( barth , greene & ho 2005 ) . never - the - less it is instructive to combine the local and agn smbh samples and investigate the resulting log - quadratic @xmath31 relation . we have repeated the @xmath46 minimisation using equation ( [ fit_quad ] ) for the combined sample of @xmath199 smbhs . in the lower - left panel of figure [ fig11 ] we show the combined sample with the resulting best - fit @xmath12 relation over - plotted ( solid line ) . the intrinsic scatter required for a reduced @xmath46 of unity in this fit was @xmath200 dex . in the lower - right panel of figure [ fig11 ] we show the 1 , 1.5 , 2 and 2.5-sigma ellipsoids for @xmath5 and @xmath6 . the most likely solution has @xmath201 and @xmath202 with 1-sigma uncertainties of 0.3 and 0.7 respectively . this solution is very close to the one obtained using only smbhs in the local sample ( @xmath197 , @xmath203 and @xmath204 dex ) , indicating that the agn smbhs follow the same log - quadratic @xmath12 relation as the local sample . however the addition of low - mass agn smbhs results in a fit with a log - quadratic term which is non - zero at greater than the 2-sigma level . the sample of agn smbhs therefore support a log - quadratic @xmath31 relation . the log - quadratic relation describes the masses of smbhs in galaxies with velocity dispersions ranging from 25 - 400km@xmath39s@xmath41 . if one extrapolates this to lower @xmath1 then the log - quadratic relation offers the tantalising prediction that there is a minimum smbh mass of @xmath205 , and that this minimum smbh mass resides in bulges with @xmath206km@xmath39s@xmath41 . interestingly , this velocity dispersion corresponds to the minimum value of the virial velocity for a dark - matter halo within which gas accreted from the igm can cool via atomic transitions in hydrogen . finally we would like to make the following point regarding the relationship between the power - law slope of the @xmath12 and @xmath127 relations . an estimate of the bulge mass may be made using the virial mass , @xmath207 , where @xmath45 is the scale radius of the bulge . the projection of the fundamental plane yields a relation between radius and velocity dispersion of @xmath208 ( bernardi et al . consider the power - law slope at the mean velocity of the galaxy sample ( @xmath209km / s ) . we have @xmath210 , which implies @xmath211 , or @xmath212 . this is in excellent agreement with the best fit parameters of @xmath213 and @xmath214 derived in this paper . furthermore , a value of @xmath215 for the @xmath12 relation is ruled out at @xmath121 significance . if it is to be consistent with the @xmath12 relation , the @xmath130 relation should therefore be steeper than linear for galaxies with @xmath209km / s . given the relation between @xmath1 and the virial mass , we would also expect the @xmath130 relation to have a positive quadratic component , given a positive quadratic component in the @xmath12 relation . using the local sample of smbhs we have demonstrated that the parameter describing a log - linear fit to the @xmath12 relation is sensitive to the inclusion of individual galaxies at a level larger than the statistical uncertainty . this indicates that a log - linear @xmath12 relation does not provide a good description of the local sample of smbhs . we expand a general relation between @xmath216 and @xmath217 to second order , and fit the data using the resulting log - quadratic relation instead . we find that a log - quadratic relation provides a substantially better fit to the local sample of smbh masses and velocity dispersions . moreover unlike the log - linear relation , the parameters describing the log - quadratic relation are not systematically dependent on the inclusion of individual galaxies in the sample . after allowing for a second - order term in the @xmath12 relation we find an unbiased estimate for the slope of the local sample at @xmath28km / s to be @xmath9 . this value is slightly ( 2/3-sigma ) larger than previous estimates for this sample . however the logarithmic slope @xmath218 of the best fit log - quadratic relation varies substantially , from 2.7 - 5.1 over the velocity range of the local sample . the coefficient of the second order term @xmath10 is greater than zero at the 90% level . this indicates that with 80% confidence the local sample describes an @xmath12 relation that does not follow a single powerlaw between @xmath33km / s and @xmath116km / s . we have tested the sensitivity of this conclusion to different sub - samples of smbh masses determined via different techniques , as well as smbhs with and without resolved spheres of influence . the log - quadratic fit implies a non - linear contribution to the @xmath31 relation in a sub - sample of galaxies that contain only smbhs whose spheres of influence were resolved , in a sub - sample containing only smbhs with masses determined through stellar dynamics , and in a sub - sample containing only smbhs with masses determined only by non stellar dynamical methods . moreover the non - linearity is present in each sub - sample whether the central or effective velocity dispersion is used as the independent variable . smbh masses in active galaxies estimated via reverberation mapping offer an avenue to increase the smbh sample for study of the @xmath31 relation . we find that the combination of the 14 galaxies with reverberation smbh masses and measured velocity dispersions ( onken et al . 2004 ) with the local smbh sample leads to a log - quadratic relation with the same best fit as the local sample alone . the @xmath12 relation can be extended to lower masses through the inclusion of single epoch virial estimates of smbh masses based on observations of agn ( barth , greene & ho 2004 ) . in a @xmath46 analysis the best - fit log - quadratic @xmath12 relation is unchanged by the addition of a sample of 16 low mass smbhs . however the uncertainty in @xmath6 is reduced . we find that in the absence of a systematic error in the normalisation of smbh masses between the local and agn samples , the @xmath12 relation described by the combined sample deviates from a power - law at greater than the 2-sigma level . the best - fit log - quadratic relation predicts a minimum mass for smbhs in galaxies of @xmath205 , which should reside in bulges with @xmath206km@xmath39s@xmath41 . a log - quadratic @xmath12 relation has important implications for smbh demography . in particular , estimates of the local smbh mass - function that utilise the log - quadratic @xmath31 relation describe densities of smbhs with @xmath219 that are orders of magnitude larger than expected for a log - linear @xmath12 relation . in addition the departure from a power - law should provide important clues regarding the astrophysics responsible for the @xmath12 relation . for example one recent model describing the effects of radiative feedback on smbh growth predicts departure from a power - law relation , including steepening of the relation at large @xmath1 ( saznov et al . 2005 ) . we have also applied our analysis to the relation between smbh and bulge mass using the sample described in haering & rix ( 2004 ) . we find evidence for a log - quadratic term in the @xmath129 relation ( at the 1-sigma level ) , with @xmath220 and @xmath20 . we find an intrinsic scatter of @xmath21 dex . since we find the intrinsic scatter in the @xmath12 relation to be @xmath113 dex , there is @xmath221 more scatter in the smbh mass at fixed bulge mass than at fixed velocity dispersion . the sample of kinematically detected smbhs will not grow by a large factor in the foreseeable future ( ferrarese 2003 ) . progress in understanding the statistical properties of the smbh population will instead come via estimates of smbh masses that are based on reverberation mapping studies . the increased sample of smbhs will offer the possibility of more clearly defining the local @xmath31 relation ( gebhardt et al . 2000b ; ferrarese et al . 2001 ; onken et al . 2004 ; peterson et al . 2004 ; nelson et al 2004 ) , and of extending its study to high redshift ( shields et al . the author would like to thank scott tremaine for some very helpful comments including the suggestion of introducing a log - quadratic term , as well as christian knigge and alister graham for pointing out errors in the original version of the paper . the author acknowledges the support of the australian research council . following the discovery of the @xmath12 relation ( gebhardt et al . 2000 , ferrarese & merritt 2000 ) , there has been debate in the literature among the different groups regarding inconsistencies in the value of the measured slope . these differences have been attributed to several causes including use of biased statistics ( merritt & ferrarese 2001 ) and systematic differences in the velocity dispersions used for the host galaxies ( t02 ) . we have shown that the slope @xmath5 of a log - linear relation is substantially changed by the presence or absence of one or a couple of the smallest galaxies in the sample . given the different samples of velocity dispersions that have been employed by the different groups , the bias introduced by the assumption of a log - linear relation may therefore be more or less significant for different samples . merritt & ferrarese ( 2001 ) presented an analysis of the log - linear relation for three different samples . the first two samples were drawn from ferrarese & merritt ( 2000 , sample 1 ) , and from the additional galaxies presented by gebhardt et al . ( 2000 , sample 2 ) . the accuracy of some of the additional smbh masses included by gebhardt et al . ( 2000 ) has been called into question on the basis that the smbh sphere of influence was not resolved ( e.g. merritt & ferrarese 2001 ) . in addition , merritt & ferrarese ( 2001 ) discuss the combined sample from these two sub - samples . for sample 2 merritt & ferrarese ( 2001 ) computed aperture corrected central velocity dispersions , so that all galaxies in the combined sample could be considered when estimating correlations between smbh mass and both a central ( ferrarese & merritt 2000 ) and an effective velocity dispersion ( gebhardt et al . merritt & ferrarese ( 2001 ) present fits for the slope @xmath5 of a log - linear relation to these samples using regression with bi - variate errors and intrinsic scatter ( their label brs ) . we first consider the correlation of smbh mass with central velocity dispersion ( ferrarese & merritt 2000 ) . merritt & ferrarese ( 2001 ) find @xmath222 and @xmath223 for samples 1 and 2 respectively . for the combined sample they get @xmath224 . samples 1 and 2 differ at the 2-sigma level , while the combined sample leads to a slope that lies between the two . to mimic this analysis for a log - quadratic relation we minimise the following @xmath46 statistic @xmath225 where @xmath48 and @xmath49 are the logarithm of smbh mass in solar masses and the logarithm of velocity dispersion in units of 200km / s respectively . the variables @xmath50 and @xmath51 are the uncertainties in dex for these parameters . the variable @xmath64 is the intrinsic scatter , adjusted to yield a reduced @xmath46 of unity for the best - fit solution . in figure [ fig12 ] we show the 1 , 1.5 , 2 and 2.5-sigma error ellipsoids for @xmath5 and @xmath6 in sample-1 , sample-2 and the combined sample . the 1-sigma ellipsoid for the combined sample is shown in the other panels for comparison , and the best fit relations are plotted over the data in the upper left panel . the grey regions represent the 1-sigma range for @xmath5 determined by merritt & ferrarese ( 2001 ) . the ellipsoids show that all three samples suggest a positive quadratic term is present in the @xmath12 relation . the values of @xmath5 determined by merritt & ferrarese ( 2001 ) are conditional probabilities for @xmath5 given @xmath71 , and so should correspond to regions where the error ellipsoids cross the @xmath71 line . figures [ fig12 ] and [ fig13 ] show this to be the case . the inclusion of a quadratic term in the relation leads to similar estimates for parameters using each sample , with @xmath226 and @xmath227 lying inside the 1-sigma ellipsoids . we also investigate the correlation between smbh mass and effective velocity dispersion ( gebhardt et al . the values for effective velocity dispersion @xmath1 in gebhardt et al . ( 2000 ) did not include an estimate of the uncertainty . in the absence of quoted uncertainties , merritt & ferrarese ( 2001 ) assumed different values for the uncertainty in effective velocity dispersion and showed that increasing the uncertainty lead to steeper best fit slopes , and to consistency between fits for samples 1 and 2 . t02 has since advocated 5% for the fractional uncertainty in effective velocity dispersion . we assume 5% errors in @xmath1 and adopt @xmath228km / s for the milky - way . in this case merritt & ferrarese ( 2000 ) found @xmath229 and @xmath230 for samples 1 and 2 respectively . for the combined sample , they found @xmath231 . in figure [ fig13 ] we show the results for the log - quadratic fit in this case . the results are consistent for a log - quadratic relation , with @xmath232 and @xmath233 lying inside the 1-sigma error ellipses of all three samples . consistency between the best fit log - quadratic relations for samples 1 and 2 therefore does not require uncertainties in @xmath1 to be larger than 5% , the value advocated by the nuker team ( gebhardt et al . 2003 ) . finally we compare the estimates of @xmath5 and @xmath6 for relations between smbh mass , and the central or effective velocity dispersions using the combined sample . the use of an effective velocity dispersion ( with 5% uncertainty ) leads to smaller values of @xmath5 and @xmath6 relative to a fit using central velocity dispersions . merritt & ferrarese ( 2001 ) found that this difference could be removed by increasing the uncertainty in @xmath1 to greater than 10% . however a solution with @xmath234 and @xmath235 lies inside the 1-sigma error ellipsoids in both cases . overall we find that a solution with @xmath236 and @xmath237 ( shown by the large cross in figures [ fig12 ] and [ fig13 ] ) lies on or inside the 1-sigma error ellipsoids of all three samples using either the central or the effective velocity dispersions as the independent variable . therefore if a log - quadratic @xmath12 relation is used we find that there is no significant disparity between the different samples . in a recent review , ferrarese & ford ( 2004 ) have compiled an updated sample of local smbhs , listed in their table - ii . of the 38 mass estimates , they class 8 as unreliable . of the remaining 30 smbhs , the mass observations resolved the sphere of influence in only 25 cases . ferrarese & ford ( 2004 ) used this sample of 25 reliable mass estimates where the sphere of influence is resolved to generate a log - linear fit to the @xmath12 relation . they find a log - linear slope of @xmath238 . we have performed log - quadratic fits to this updated sample , and show the results in figure [ fig14 ] . in the upper left panel we show the sample of 25 smbhs with resolved spheres of influence , and the corresponding best - fit relation . in the upper right - hand panel we show the 1 , 1.5 , 2 and 2.5-sigma error ellipsoids of @xmath5 and @xmath6 for this sample . this sample prefers a log - quadratic relation , with the log - linear relation lying near the @xmath239 contour . the grey regions in figure [ fig14 ] represent the 1-sigma range for @xmath5 determined by ferrarese & ford ( 2004 ) . we see that the contours of @xmath46 intercept the @xmath71 line over a range of slope described by the log - linear solution . of the 5 galaxies with reliable mass - estimates , but unresolved spheres of influence in table ii of ferrarese & ford ( 2004 ) , 4 have had their stellar kinematical data modeled using both high - resolution spaced based data , and lower resolution ground based data ( gebhardt et al . gebhardt et al . ( 2003 ) found that while higher resolution imaging led to a more precise determination , there was no systematic trend of the best fit values . we therefore repeat the above analysis with the inclusion of the 5 galaxies and show the results in the lower row of figure [ fig14 ] . the additional 5 galaxies only change the slope @xmath5 of a log - linear fit by a small amount ( significantly less than the error bar ) . moreover , these 5 smbh masses are evenly spread about the best - fit relation , indicating that there is no systematic bias due to their spheres of influence not being resolved . however the inclusion of these galaxies significantly increases the significance of a log - quadratic term in the @xmath31 relation .

the feasibility of low - energy fragmentation experiments using a magnetic spectrometer is discussed . the main challenge is the multiplicity of the ionic charge states , which can hamper the identification in both z and a of the fragments . three topics are covered . first , a specific set - up for ionization chambers , based on a very large gas thickness , is presented . its satisfactory performances are discussed in light of the observations during a 500a mev pb+p experiment performed at the frs ( gsi ) . as a second topic , the possibility to use a thick layer of matter ( a degrader ) as a passive measurement device to identify the nuclear charge and the ionic charge state of fragments is discussed . this method , successfully used for z identification in experiments such as pb+p at 1a gev , fails to measure the charge states at 500a mev for the same system . it is shown that surface defects of the degrader are probably responsible for this failure . the third topic is the description of new analysis techniques developed in order to account for and subtract the contribution of polluting charge states in the spectrometer , thus making possible a clean estimation of the production cross sections of all fragments . the combination of those new experimental and analysis techniques made the 500a mev spallation experiment a success . , , , , , magnetic spectrometer , ionic charge state , production cross section , degrader , ionization chambers 29.30.aj , 29.40.cs , 34.50.bw , 25.40-h , 25.70-z

nucl-ex0412039

an in - flight fragment separation apparatus can be used in two ways : * as a recoil spectrometer , to measure production cross sections or kinematics properties of the fragments ; * as a separation device , aiming at producing high - purity secondary beams of nuclides . the fragment recoil separator ( frs ) @xcite of gsi has been very successfully used for both purposes for more than ten years . by now , several projects across the world are dedicated to the construction of more powerful devices , in terms of larger fragment acceptance ( spectrometer for fission products ) , and/or increased selectivity and high - intensity secondary - beam production . the upper energy range covered by such devices seems to be mostly a technological ( and financial ) question , as increasing the energy requires more powerful magnets and a longer flight path , not even talking about the accelerator itself . but the lower range is determined by physics : it is assumed that only a very low proportion of non - fully stripped fragments can be tolerated in order to obtain a clean separation of fragments and to measure production cross sections . therefore , a whole energy range is somehow forbidden to fragmentation experiments . this low - energy limit has been faced by a recent experiment conducted at the gsi spectrometer frs . the aim of this experiment was to use the inverse - kinematics method to measure production cross sections from residues of the spallation of lead by protons with an incident energy of 500a mev . the results of this experiment will be the subject of a separate , forthcoming publication . this paper is dedicated to the description of the experimental and analysis techniques which were developed for this experiment and to discuss whether or not they proved to be successful . part of these techniques may be used more generally than for the frs setup from gsi . first we will briefly remind the main characteristics of the frs ( section [ chap : frs ] ) . the fragment recoil separator and its set of detectors have been designed using the hypothesis that most , if not all the fragment one wishes to study are fully stripped @xcite . this is of course less and less true with decreasing energy . in section [ chap : qst ] we will discuss the variations of the charge - state probabilities with energy and their consequences on the selectivity of the spectrometer , as well as the possible choices of so - called stripping materials . with a lead beam energy as low as 500a mev , no stripper is efficient enough so that the fraction of non - fully stripped fragments would be negligible . the response of the ionization chambers ( used for the determination of the atomic number of the fragments ) in this situation will be discussed in section [ chap : z ] , as well as a new setup , developed for this experiment , based on an increased gas thickness . in section [ chap : deg ] we will describe a method to use a thick matter of layer as a passive measurement device in order to estimate the atomic number and the charge state of the fragments . we will also discuss , how unexpected problems prevented the measurement of the ionic charge state of ions ( and thus of the mass ) in the above - mentioned experiment . the final sections will be dedicated to new analysis techniques aiming at obtaining production cross sections , even in the case of an incomplete event - by - event mass identification . section [ chap : a ] will present a method to determine the most - likely mass of each fragment and to reduce the number of possible contaminating charge states . then , in section [ chap : cs ] , we will explain how to account for the large fraction of non fully - stripped ions and how to obtain isotopic production rates with low uncertainties despite the ambiguities in the mass determination . we will then conclude and try to answer whether or not the widely admitted low - energy limit has been pushed downwards . the frs , which is schematically presented in figure [ fig : frs ] , has been designed to separate and identify nuclei produced by fragmentation of relativistic nuclei , the reference beam being 1a gev u @xcite . basically it is designed as a two - stage device : fragments produced in a target located at the entrance of the spectrometer ( @xmath0 ) are selected according to their magnetic rigidity , reach the dispersive focal plane ( @xmath1 ) , are then selected again in a second magnetic section and reach the final focal plane ( @xmath2 ) . the first magnetic section is intended to measure the magnetic rigidity of the fragments right after the target as well as to remove the primary beam in order to protect the detectors located at @xmath1 and @xmath2 . and @xmath2 are usually measured using plastic scintillators , which are also used for time - of - flight measurement.__,scaledwidth=90.0% ] the behavior of a charged particle of mass @xmath3 and charge @xmath4 moving in a magnetic field @xmath5 is given by the well - known relation : @xmath6 here @xmath7 is the curvature radius of the particle trajectory in the magnetic field , @xmath8 is the speed of light and @xmath9 and @xmath10 are the usual lorentz relativistic factors . for a nucleus with @xmath11 nucleons , equation [ eqn : brho_full ] can be reduced to : @xmath12 @xmath13 being the elementary nuclear mass , @xmath14 the elementary electric charge , and @xmath15 the number of ionic charges of the considered ion . in the rest of this paper , for simplicity , we will refer to the magnetic rigidity of a fragment as a dimensionless term , @xmath16 . in experiments at the frs , the @xmath17 ratio of each nucleus is deduced from the measurement of its time of flight and magnetic rigidity in the second magnetic section , using plastic scintillators @xcite . @xmath18 is deduced from energy loss in ionization chambers set at the exit of the spectrometer . one of the main hypotheses used for the design of the frs was that most of the fragments should be fully stripped during their flight in the magnetic sections . in this case one has @xmath19 and getting the mass of each fragment is straightforward . as will be outlined in section [ chap : select ] , the insertion of a thick layer of matter at @xmath1 is a way to drastically reduce the number of fragments transmitted down to the exit of the spectrometer . detailed calculations demonstrating this effect can be found in @xcite . let us point out that the possibility to choose the exact profile of the degrader allows one to use the frs in particular magnetic optics modes : achromatic or mono - energetic . * _ achromatic : the position of the fragments at @xmath2 does not depend on their velocity at the entrance of the spectrometer . this is particulary helpful to achieve a better separation of the fragments and was the mode selected for the 500a mev experiment which is the reference for this paper . _ * _ mono - energetic : the velocity of the fragments at @xmath2 does not depend on their velocity at the entrance of the spectrometer . this mode is mainly used for implantation experiments . _ in the following sections , we will often take the @xmath20pb+p at 500a mev experiment as a reference for our discussions . in this experiment , the target consisted of 87 mg.@xmath21 liquid hydrogen enclosed in 4 ti foils , each 9 mg.@xmath21 thick ( see @xcite for a complete description ) . after this target , a stripper foil of nb with a thickness of 60 mg.@xmath21 was set . at @xmath1 the plastic scintillator had a thickness of 3 mm and was followed by a profiled degrader with a mean thickness of 1740 mg.@xmath21 of pure al . considering the energy of the fragments after the degrader ( roughly 300a mev ) , a ti foil would have been suitable for optimal stripping , but al was a good choice , nevertheless ( see chapter [ chap : strip ] for a discussion about stripper efficiency ) . from equation [ eqn : brho ] it is obvious that a magnetic separation device is sensitive to the ionic charge of the nuclei . this is also the case of energy losses ( which are the key process for many detectors ) , as the elemental energy loss in matter is governed by electromagnetic interactions and therefore depends , not on @xmath22 as one often thinks in the first place , but on @xmath23 , the square of the effective ionic charge of the nuclei . in the case of a gas medium , this effective charge state is almost equal to the charge state as the time interval between two interactions of the fragment with gas atoms is much larger than the decay time of atomic excited states of the fragment . the evolution of the ionic charge state of an ion in matter depends on two competing processes : electron stripping and electron capture . in the relativistic or quasi - relativistic regime , the stripping probability @xmath24 is nearly independent of the energy of the ion . on the other hand , the probability of electron capture leading to the presence of the electron on shell @xmath25 , @xmath26 , strongly increases with decreasing energy ( for detailed discussion of the role of the various atomic shells and related phenomena such as the de - excitation time of the capturing ion , see for example @xcite ) . one can distinguish two extreme situations : * @xmath27 : any electron carried by the ion is removed after a very short distance in the medium . similarly , any electron capture is quickly followed by the stripping of this electron . therefore the ion is mostly fully stripped . this situation corresponds to the high - energy limit ( roughly , beyond 1a gev for heavy ions such as lead ) . * @xmath28 : any electron stripped for the @xmath25 shell is quickly replaced . this is most likely to happen if the velocity of the ion is close to the one the electron would have on its shell . in this situation the ion has its @xmath25 shell filled most of the time . for example , considering heavy ions such as lead , this situation appears in the case of the k shell around 100a mev . the intermediate situation ( @xmath29 ) can be interpreted as a statistical equilibrium between capture and stripping . for a given combination ( @xmath30 , @xmath31 , @xmath32 ) one can define a probability @xmath33 for the ion to have a charge @xmath15 ( or , in other words , to travel with @xmath34 electrons ) . considering equation [ eqn : brho ] and the fact that we have no possibility to directly determine the charge state of an ion , it becomes clear that the mass identification of fragments requires , as far as possible , those fragments to be fully stripped in the spectrometer . to optimize the proportion of bare ions , it is convenient to insert layers of well - chosen materials in the device . the choice of a stripper foil results in a compromise between its stripping efficiency and its impact on the beam quality , in terms of angular and energy straggling , and in terms of induced secondary reactions in the stripper . figure [ fig : strip ] displays the stripping efficiency , evaluated from the fraction of bare ions , as function of the projectile energy . calculations were made using the code global @xcite . the percentage of nuclear reaction in stripper is also displayed ( reaction cross sections were estimated using the karol parametrization @xcite ) . in those calculations , to simplify the discussion , the stripper thickness has always been chosen as the so - called charge - equilibrium thickness . this quantity corresponds to the thickness of matter beyond which the proportion of the ionic charge states is fully independent of the initial charge - state distribution . at high energy , any high- or medium - z element will provide excellent results . nevertheless , even at 1a gev ( the highest energy available at the gsi synchrotron sis ) , results are not perfect for very heavy elements : fully stripped ions represent only 95% of pb ions , and no more than 88% for u. with decreasing energy , the efficiency of strippers is subject to large variations . nb is the most efficient stripper for beam energies higher than 600a mev , while ti gives slightly better results below this value . nevertheless , the equilibrium charge state is reached in a thicker foil in ti than in nb , which results in a higher nuclear - reaction rate . let us point out that the energy regime in which the situation can be considered as favorable varies strongly according to the considered ion beam . the 90% proportion of fully stripped ions , which can be considered as a reasonable limit , is hardly reached for a 1a gev u beam , while it allows experiments with a xe beam at energies as low as 200a mev . the thick degrader technique has been developed in order to allow the production of a high - purity secondary beam . this technique consists in the inclusion of a thick layer of matter between two magnetic selection devices . the first selection basically operates on the @xmath17 ratio of the ions . those ions loose a large part of their energy in the degrader , which basically depends on the square of their nuclear charge . the second magnetic selection can then be roughly understood as a @xmath18 selection . pb with an energy of 500a mev . a thick degrader of thickness 2000 mg.@xmath21 al is set between the two magnetic sections . fragments transmitted in the first part of the spectrometer are plotted in grey , while those also transmitted in the second part are plotted in black . only fully stripped ions are considered on the left figure , while all combinations with 0 or 1 electron in the first part and 0 , 1 , 2 or 3 electrons in the second part are considered on the right figure.__,title="fig:",scaledwidth=47.0% ] pb with an energy of 500a mev . a thick degrader of thickness 2000 mg.@xmath21 al is set between the two magnetic sections . fragments transmitted in the first part of the spectrometer are plotted in grey , while those also transmitted in the second part are plotted in black . only fully stripped ions are considered on the left figure , while all combinations with 0 or 1 electron in the first part and 0 , 1 , 2 or 3 electrons in the second part are considered on the right figure.__,title="fig:",scaledwidth=47.0% ] the selectivity of such a separation device is presented in figure [ fig : select ] . calculations have been performed in order to simulate the transmission in the frs for fragments produced in the pb+p reaction at 500a mev . we only took into account the fragmentation - evaporation products and did not consider the production cross sections of the reaction products ( these cross sections are expected to drop very rapidly for mass losses beyond 40 and completely vanish beyond 60 ) . from the hundreds of different fragments that we considered , less than 20 are transmitted up to the end of the separator . the use of slits in the middle of each magnetic sections would even reduce this number . things are less favorable when ionic charge states are taken into account . the various combinations of charge states allow for a much larger population of fragments to be transmitted , as can be seen on the right part of figure [ fig : select ] . the purity of a secondary beam is strongly reduced . consequently , production measurements require more time to achieve satisfactory statistics as a large part of the transmitted fragments may not be of interest . it is important to notice that , due to the bell shape of the isotopic distributions of evaporation or fission residues , the situation is drastically different if one considers proton - rich or neutron - rich isotopes . if the magnetic device is set to select fully stripped fragments with a given @xmath35 ratio , h - like fragments with atomic number @xmath18 and mass @xmath36 will also be transmitted , as well as he - like fragments of mass @xmath37 , and so on . the proportion of unwanted ions therefore directly depends , not only on the charge - state probability , but also on the production cross sections for the lighter fragments . let us consider fragments of a given element . in the case of neutron - rich isotopes , cross sections for lighter isotopes are considerably larger , and therefore the proportion of contaminants is expected to be very important . on the other hand , when considering proton - rich fragments , the cross sections for the production of the lightest isotopes are very small , and therefore the problem of charge states may be somehow disregarded . calculations presented in figure [ fig : music ] summarize the different possible behaviors for stripped ions entering an ionization chamber . the setup in our simulation was intended to simulate the music chambers @xcite used at the frs . in normal conditions , these chambers are filled with p10 , a gas mixture with 90% of ar and 10% of ch@xmath38 , at atmospheric pressure ( this corresponds to a gas thickness of 100 mg.@xmath21 ) . as only 2/3 of these chamber length is instrumented , the ions were propagated along all the gas thickness , but only 2/3 of the thickness was used to calculate the energy collected after deposition in the gas . the electron stripping and/or capture probability was computed from the cross sections as calculated by global @xcite for each small step of the fragment propagation through the gas , and the energy loss calculated according to the new ionic charge state of the fragment . effects such as the escape probability of the @xmath39 electrons or the drift of the electrons through the chambers were not taken into account . the contribution of effects other than the variations of the ionic charge state were roughly taken into account as a fixed 0.3% contribution to the total straggling . on the upper part of the figure , the incoming ions have an energy of 750a mev and pass through a limited gas thickness , which corresponds to a single ionization chamber filled with 1 bar gas . their mean free path related to electron capture is much larger than both the gas thickness and the mean free path related to electron stripping . the ions remain stripped along most of their way through the gas , and the achieved resolution in energy loss is excellent . the small tails associated with each peak correspond to the ions that kept an electron for some time in the gas . of course , adding a second chamber increases the resolving power of the system . if the incident energy of the fragments is lowered to 350a mev , the probability of charge exchanges during the crossing of the fragment through the gas dramatically increases . therefore , the energy - loss distribution becomes wider as a result of the fluctuations of @xmath40 , leading to an unacceptable decrease of the resolution of the ionization chamber . if one increases the total gas thickness ( up to 700 mg.@xmath21 on the figure ) the relative statistical fluctuation in @xmath40 is reduced , leading to a better definition of the total energy loss and therefore a better resolution on @xmath40 . according to this simulation , a resolution better than 10% may be expected . furthermore , complete independency regarding the incoming charge state is achieved . according to simulations , the use of ionization chambers to identify heavy , low - energy fragments seems possible if one strongly increases the amount of gas ( or the number of chambers ) with respect to high - energy setups . this technique has been used successfully during the pb+p at 500a mev experiment at gsi . the setup consisted of 4 music chambers @xcite filled with 2 bar p10 , for a total gas thickness of the order of 800 mg.@xmath21 , from which roughly 2/3 were instrumented . due to the presence of a thick degrader at the intermediate focal plane , the energy of fragments in the second part of the spectrometer was of the order of 300a mev . the resolution obtained in nuclear charge was better than 0.9% ( or 0.7/z ) for very heavy fragments and better than 0.6% ( or 0.4/z ) for fragments around nuclear charge 70 . the resolutions given here are fwhm . we will keep this convention throughout the paper , unless the contrary is specified . tl ( left ) and @xmath41yb ( right).__,title="fig:",scaledwidth=47.0% ] tl ( left ) and @xmath41yb ( right).__,title="fig:",scaledwidth=47.0% ] examples of resolutions obtained are presented on figure [ fig : music_res ] for fragments close to ( left part of the figure ) and far from ( right part ) the lead projectile . this resolution obtained by the combination of the signals of the 4 chambers can be compared with the resolution obtained with a single chamber , which can be deduced from figure [ fig : music_correlation ] . to obtain these results , a correction taking into account the velocity of fragments had to be applied . but the large gas thickness had three drawbacks . 1 . the energy loss is large enough so that a dependency regarding to the mass of the fragment was observed . this is a trivial expectation issued from the bethe formula , but the consequence is that , for a given @xmath18 , a systematic increase of the music signal was observed with decreasing masses . the experimental evidence for this effect is presented in figure [ fig : music_mass ] . one can clearly see that this effect is strong enough to lead to a misidentification of the fragments by one charge unit , if not taken into account . + + yb in the pb+p experiment at 500a mev.__,scaledwidth=70.0% ] 2 . the large amount of matter increases the parasitic nuclear - reaction rate in the gas . products from these reactions can be rejected during the off - line analysis by comparing the signals observed in the different music chambers . an example of such a correlation is presented in figure [ fig : music_correlation ] . the large quantity of energy deposited in the gas seemed to strongly alter the shape of the relation between the signal of the music and the position of the fragments in the chambers , as well as the resolution ( see section [ chap : countrate ] ) . in ordinary operating conditions , a reduction of the signal is observed with increasing distance between the point of passage of the fragment trajectory and the anodes increases . this dependency is due to the recombination of some of the drift electrons in the gas along their path toward the anodes . those recombinations may occur with p10 molecules , but are more likely to occur with impurities such as o@xmath42 or h@xmath42o . the left part of figure [ fig : rate ] presents an example of this effect . if the counting rate is increased up to some 2 khz , a drastic change of the relation between the signal and the drift distance is observed ( right part of the figure [ fig : rate ] ) . also , compared to the figure on the left , if the resolution is only slightly altered in the region close to the anodes , no identification is anymore possible for trajectories far from the anode . further illustration of the relation between count rate and resolution in the case of high event counting rates is presented on figure [ fig : rate_pos ] , in which the music signal is plotted as a function of the time duration of a beam spill of the sis synchrotron . for a section of the music close to the anodes , the signal slightly reduces with increasing beam intensity and then gets back to its nominal value at the end of the spill . this limited effect probably is the consequence of the appearance of some screening effect . as one gets further from the anode , the relation inverts : increasing intensity leads to increasing signal , up to a point that leads to a misidentification by 2 nuclear charges . as this effect depends on many parameters it could not be corrected during the analysis , resulting in a loss of some data when the identification of the ions becomes impossible . as this variation of signal depends on position , electronics can not be responsible for it . this effect is likely to come from the saturation of electron catching sites of the gas ( mainly impurities ) for high electron densities corresponding to a large number of highly ionizing nuclei . such short time variations on music signals have already been observed in the past without satisfactory explanation . they were tentatively ascribed to pressure or temperature drifts , but no clear correlation with such quantities could be established . we have shown here that this behavior is likely to be related to the electron - attachment saturation under high counting rate , or in other words to the amount of ionization in the chambers . we have described a new way to operate the ionization chambers . we have shown that such a set - up allows to identify in @xmath18 heavy ions as heavy as lead at energies as low as 300a mev . we are confident that this method would be efficient for lower energies . the limitations of this set - up have also been stressed . first , the large amount of gas induces variations of the energy loss with the mass of the fragment . this effect does not hamper the efficiency of the method but must be taken into account during the analysis process . second , as the density of ionization in the gas largely increases with decreasing energy , the response of the ionization chamber can be modified according to the counting rate . this effect can be strong enough to prevent any identification of the fragments . therefore , it may be necessary to limit the counting rate . as written in equation [ eqn : brho ] , the mass identification of a fragment requires the knowledge of its ionic charge state . to the best of our knowledge , no apparatus can provide a direct , unambiguous measurement of the ionic charge state of an ion in the energy range considered in this paper . nevertheless , it may be deduced from a fine evaluation of the energy loss in a thick layer of matter , namely the degrader set between the magnetic sections of the frs . in this section , we will consider that the fragments are correctly identified in @xmath18 , regardless of their charge state , according for example to the use of the method presented in the previous section . for the sake of simplicity , we will often call `` degrader '' the whole matter inserted in the beam line at @xmath1 , namely , the degrader plus the plastic scintillator . they were of course treated separately in the simulation . the energy loss in matter of an ion of mass @xmath11 and kinetic energy @xmath43 is given by : @xmath44 for calculations related to magnetic rigidities , it is more convenient to write equation [ eqn : de_0 ] as a function of @xmath45 instead of @xmath46 . this leads to : @xmath47 where @xmath48 is the mean value of @xmath9 in the considered layer of matter . in the energy domain we are discussing in this paper , the variations of this quantity are small and one can therefore treat it as a constant for the purpose of the demonstration . the slowing down of the fragments and the change of charge state ( if any ) induce a variation of the magnetic rigidity between magnetic sections . using the discussion that followed equation [ eqn : brho ] , this can be written as : @xmath49 schematically the first @xmath50 term ( which we will note as @xmath51 ) corresponds to the loss of energy , while the second term ( which we will note as @xmath52 ) is related to the charge - state changing . according to equation [ eqn : de ] one can write : @xmath53 according to the discussion in section [ chap : qst ] , @xmath54 can be written : @xmath55 where @xmath23 is the mean value of the square of the atomic charge of the fragment in the degrader ( as defined in chapter [ chap : qst ] ) , @xmath56 is the velocity dependence from the bethe formula , and @xmath57 describes the position dependence due to the variations of the degrader thickness . according to its definition , @xmath23 is completely independent of the ionic charge state in both parts of the spectrometer . furthermore , it is both unambiguously related to @xmath22 and very close to it . we can then replace it by @xmath22 for this calculation . we can now write @xmath58 as : @xmath59 here @xmath60 stands for the number of electrons in the second section of the spectrometer . the second part of the development of @xmath61 writes : @xmath62 to summarize , we have established that , besides the initial velocity of the fragment , the variation of magnetic rigidity depends on three physical components : * the nuclear charge of the fragment ; * the charge state of the fragment in the second part of the spectrometer ; * the change of charge state of the fragment between the two sections of the spectrometer . in the next section ( [ chap : de_app ] ) we will present the possible applications of the measurement of magnetic - rigidity variation and the associated requirements on the energy - loss resolution . in section [ chap : de_res ] we will discuss the limitations from the characteristics of the devices and from physics . in section [ chap : de_exp ] we will present and discuss the results obtained in the 500a mev experiment . among the three effects listed in the previous section , the charge - state changing is always dominant as soon as a capture or a ionization takes place in the degrader , even with a very thick degrader for which the energy loss is large . this allows to detect and identify unambiguously any change of charge state as will be shown later in figure [ fig : deg ] . the related magnetic - rigidity variation does not depend on the thickness of matter set , and therefore information on a change of charge state is accessible whatever the setup or the energy of the incoming fragment . during the data analysis , the fragments can be sorted according to their change of charge state , a quantity that we will write as @xmath63 . besides this property , and as expected from equation [ eqn : dbrho1 ] , a large thickness of matter induces a linear dependence of the variation of magnetic rigidity with @xmath18 and @xmath15 . separating peaks associated to neighboring charges with a reasonable efficiency requires the fwhm of each peak to be ( roughly ) smaller than 0.7/@xmath18 . for example in the case of lead this corresponds to a resolution of 0.9% . in high - energy experiments , the selection of the charge states of the fragment is nearly entirely obtained using the @xmath63 value . indeed , although the proportion of fully stripped ions ( that we can write ( 0,0 ) according to the number of electrons they carry in each part of the spectrometer ) is only of the order of 80% , the @xmath63 estimation allows the rejection of most of the charge - state combinations like ( 0,1 ) , ( 0,2 ) or ( 1,0 ) . the only charge - state combinations which can not be disentangled from ( 0,0 ) are ( 1,1 ) , ( 2,2 ) , and so on . but in the case of the fragmentation of a 1a gev pb beam for example , these charge states represent only 2% of the pb fragments that have a @xmath63 equal to zero . this very low proportion is due to the combination of the two strippers ( one at the entrance of each magnetic section ) which is weighted by the product of the two probabilities . these fragments might be disregarded during the identification process and receive an appropriate correction during the estimation of production rates . in this case it is therefore convenient to use the degrader as an additional measurement of the @xmath18 value , which can be combined to the one obtained from the musics and increase the overall @xmath18 resolution . this method has already been successfully used in previous spallation experiments at the frs @xcite . in the case of a low - energy experiment , the fraction of @xmath64 fragments can be as low as 40% for fragments produced from a 500a mev pb beam . among those @xmath64 fragments , roughly 40% are hydrogen - like ions . therefore , they should be explicitly identified and taken into consideration . if one relies on the musics for the measurement of the nuclear charge , according to the previous calculations , a precise estimation of the energy loss in the degrader may be used to estimate the ionic charge state of the fragments . let us first stress what exactly means the word resolution . the value we are considering is the variation of the magnetic rigidity . it is important to distinguish the two sources of the so - called uncertainty on this quantity . the first and most obvious one is the uncertainty on the measurement . the second one , by far the largest as we will see , is not an uncertainty in the usual sense of the word : it is the fluctuation in energy induced by the path through a thick layer of matter . whatever the precision of the measurement of the magnetic rigidity or time of flight , there is no way to correct the @xmath65 width induced by those fluctuations . therefore , to identify charge states , it is the sum of those two terms which must fulfill the 0.7/@xmath18 condition presented above . for convenience , one can express the straggling in energy as a fluctuation of the variation of the magnetic rigidity : @xmath66 this relation is discussed in details in appendix [ chap : brho - e ] . in the following we will express all fluctuations and uncertainties as relative uncertainties with respect to the variation of magnetic rigidity . we will now discuss separately the contributions to the total @xmath65 width of the energy straggling and of the magnetic - rigidity measurement . the resolution that can be achieved on the measurement of the change of magnetic rigidity obviously depends on the resolution in magnetic rigidity , which itself depends on the precision of the magnets and from the position - measurement devices , namely scintillators . in a given magnetic setting between the focal planes @xmath11 and @xmath5 , characterized by a magnetic rigidity @xmath67 for centered ions , the magnetic rigidity of an ion is deduced from its position at the focal planes ( respectively @xmath68 and @xmath69 ) as : @xmath70 with @xmath71 being the dispersion of the magnetic section and @xmath72 its magnification . an optimum result is obtained only if the magnetic optics exhibits no aberration and if the positions of the fragments are measured at the focal planes of each magnetic section . we will assume that the first condition is fulfilled , although there are indications that some limited magnetic aberrations may exist close to the final focal plane . for the second assumption , the situation differs according to the focal plane one considers . in the first part of the frs , the initial point ( the target ) is unambiguously defined , and the scintillator is set at the intermediate focal plane , at a fixed position . the magnification term in equation [ eqn : xbrho ] vanishes and the uncertainty is simply obtained from the uncertainty on magnetic fields ( known to be less than 10@xmath73 ) and on the scintillator response ( estimated to be at most 3 mm ) : @xmath74 the position of the final focal plane varies according to the chosen magnetic optics , and the position of the corresponding scintillator depends on the available room left by all other detectors located at this region . therefore , the two points may not coincide . if the distance between the scintillator and the focal plane is @xmath75 and the angular distribution of a fragment has a width @xmath76 , a dispersion is introduced on the measurement of the @xmath77 position : @xmath78 the contribution of this term to the uncertainty on the measured magnetic rigidity adds to the intrinsic uncertainty of the measurement : @xmath79 if the trajectory angle of the fragments is measured , the position of fragments at the final focal plane can be reconstructed . in this case , the enlargement on the position resolution due to the incoming angle can therefore be corrected . two effects are responsible for the fluctuations of the energy loss of a fragment passing through a solid @xcite . first , the energy loss is due to successive collisions , each removing a variable amount of energy from the incoming ion . the variance can be estimated with a reasonable precision by : @xmath80 where @xmath81 is the nuclear charge of the material , @xmath82 the nuclear charge of the fragment passing through , and @xmath83 the number of atoms of the material . this term will decrease the resolution as the thickness of the material increases and as the energy of the fragment increases . in the bohr approximation @xmath84 reduces to @xmath85 . the lindhard - sorensen ( ls ) theory @xcite has extended the validity of this equation to relativistic energies . in this theory @xmath84 is a much more complicated expression but the agreement with experiment in the relativistic regime is strongly improved . for our subsequent calculations of energy losses and straggling , we used this theory as implemented in the code atima @xcite . the second phenomenon inducing straggling is the fluctuation of the ionic charge state , already discussed in section [ chap : z ] . these fluctuations are of statistical nature , therefore the energy loss resolution increases with the thickness of the solid . on the other hand , fluctuations increase with decreasing energy as more and more charge states can be populated . the calculations related to this phenomenon were performed using the ionic capture and stripping cross sections from the code global @xcite . we have performed calculations of these two effects for ions of @xmath20pb ( left part of figure [ fig : sigma_de ] ) and @xmath86hf ( right part ) . for each energy we calculated the energy loss of a @xmath20pb beam in a liquid hydrogen target used in all spallation experiments performed at the frs ( see chapter [ chap : frs ] ) , but did not consider any straggling or angular broadening at this point . any energy straggling in the target is supposed to be washed out by the achromatism of the system . this is not the case of the angular distribution induced by the reaction , but we disregarded it in order to focus on the effects of the matter set at @xmath1 . the thickness of the degrader was then calculated for each energy as it is done during experiments : the total thickness of matter at the intermediate focal plane must correspond to half of the range of the beam in aluminium ( this value represents an optimum between the energy loss and the probability for parasitic nuclear reactions in these layers ) . we explicitly took into account a plastic degrader made of 3 mm of bc-400 . pb ( left plot ) and @xmath86hf ( right plot ) , as the function of the energy of the beam . the continuous line represents the quadratic sum of these two components . effect of the angular straggling is presented in the hypothesis of a distance of 1 m between the position measurement at the exit of the spectrometer and the position of the magnetic focal plane . see text for details of the simulation.__,title="fig:",scaledwidth=43.0% ] pb ( left plot ) and @xmath86hf ( right plot ) , as the function of the energy of the beam . the continuous line represents the quadratic sum of these two components . effect of the angular straggling is presented in the hypothesis of a distance of 1 m between the position measurement at the exit of the spectrometer and the position of the magnetic focal plane . see text for details of the simulation.__,title="fig:",scaledwidth=43.0% ] results of these calculations are expressed in terms of fwhm for the relative magnetic rigidity variation . at 1a gev the charge - state fluctuations are comparable to the effect of the collision straggling in the case of pb , while , due to the quadratic summation , they are nearly negligible in the case of hf . the total fwhm for pb ions is 0.49% , and 0.44% for hf . with decreasing energy , the collision straggling keeps a nearly constant value , while the change state changes intervene more and more . at 500a mev , the resolution on energy loss for lead is not better than 0.84% , the collision straggling accounting for only 20% of this value . unexpectedly , in the case of hf , the worsening of the resolution with decreasing energy is stronger than in pb and does not reach a maximum as it does for pb . this is probably related to the number of collisions in the materials , which increases faster for pb than for hf with decreasing energy , therefore reducing the statistical fluctuations on the charge state . furthermore , around 250a mev , the cross section for capture to the @xmath87 atomic shell of pb ions is already very large while the capture cross section to the @xmath88-shell remains small : this strongly favors the @xmath89 charge state and thus furthers reduces the fluctuations . in the case of hf this situation appears only at an energy lower than 200 mev . we did not extend the calculations to lower energies as the energy loss in the target would have become too large to render an experiment feasible . reducing the hydrogen thickness would not counteract this problem in a relevant way as it would also lead to an increase of the relative contribution of the target windows ( which can not be reduced ) to the production rates . we also calculated the effect of the angular straggling induced by the materials set at the intermediate focal plane . in chapter [ chap : uncer ] we have already discussed this effect : in an achromatic system like the frs , it varies linearly with the distance between the real magnetic focal plane and the point of the beam line at which the position measurement is performed ( it is therefore canceled if the position is measured at the correct point of the beam line ) . the values plotted in figure [ fig : sigma_de ] correspond to a distance of 1 m between the expected and the actual measurement points . as one can see on the figure such a misplacement has nearly no consequence on the obtained energy loss resolution . of course a larger distance could play a significant role . as we have already discussed , in an achromatic system , the position of an ion at the final focal plane does not depend of its initial velocity , nor of its initial angle . going through the degrader induces velocity fluctuations that can not be compensated by the magnetic optics . therefore , the beam is not expected to be perfectly focused at @xmath2 . the previous calculations on the energy straggling can be converted in a minimal width expected for the beam at @xmath2 : the results are 0.27% at 500a mev and 0.16% at 1a gev . the experimental values are presented in figure [ fig : beam_pos ] . beam widths at @xmath1 are also plotted for comparison ; let us recall that the dispersion observed at the intermediate focal plane is expected to be fully corrected by the achromaticity of the system . in the case of the 1a gev experiment , the width observed at @xmath1 corresponds to a magnetic - rigidity straggling of 0.06% ( fwhm relative to b@xmath90 ) . an enlargement is observed at @xmath2 , leading to a magnetic - rigidity straggling of 0.2% ( fwhm relative to b@xmath91 , thus 0.54% relative to @xmath92b@xmath7 ) . this is in good agreement with the result of the calculation presented above ( 0.49% fwhm relative to @xmath92b@xmath7 ) . and @xmath2 in the pb+p experiments at 500a mev ( left ) and 1a gev ( right ) . squares correspond to experimental points , and a gaussian fit has been performed on each spectrum ( solid curve).__,title="fig:",scaledwidth=45.0% ] and @xmath2 in the pb+p experiments at 500a mev ( left ) and 1a gev ( right ) . squares correspond to experimental points , and a gaussian fit has been performed on each spectrum ( solid curve).__,title="fig:",scaledwidth=45.0% ] in the case of the 500a mev experiment , the width observed at @xmath1 corresponds to a magnetic - rigidity straggling of 0.16% ( fwhm relative to b@xmath90 ) . a very large value is observed at @xmath2 , corresponding to a magnetic rigidity straggling of 0.4% ( fwhm relative to b@xmath91 , thus 1.24% relative to @xmath92b@xmath7 ) . this is much higher than the result of the calculation presented above ( 0.26% fwhm relative to b@xmath91 ) . let us consider the different possible reasons for this large position spread . * the absence of correction for the angle of the ions . in this discussion we only consider the beam and therefore this possibility can be ruled out . even if the scintillator is far from the magnetic focal plane ( 2 m can be considered as an upper boundary ) , the position broadening induced by the angle is not expected to be a major contributor to the overall broadening ( the angular aperture of the beam has been estimated by our calculations to be of the order of 2 mrad at this point of the beam line ) . * the energy straggling and the charge state fluctuations . this obviously plays a role : we have shown in section [ chap : e_strag ] , that precise calculations lead to a value of 0.26% ( fwhm relative to b@xmath91 ) for this effect . this induces an important broadening of the position spectra ( which is not compensated by the achromaticity of the system ) but is not sufficient to explain the observed value . * a failure in the design of the degrader leading to a chromaticity of the system . if one rules out an error in the calculation of the degrader profile , the hypothesis of surface defects or inhomogeneities remains . from the difference between the calculated widths at @xmath2 at the two considered energies , we can postulate a typical size of the inhomogeneities . we could reproduce the observed widths at both energies by considering that the degrader surface is inhomogeneous with very small defects : 27 @xmath93 m . this value is to be compared to the degrader thicknesses : roughly 6.5 mm in the 500a mev experiment and 15 mm at 1a gev . the relative straggling in magnetic rigidity difference induced solely by those inhomogeneities would be 1% at 500a mev , and only 0.3% at 1a gev . we consider the presence of such inhomogeneities plausible as their effect would be small enough not to affect experiments conducted at higher energies , which necessarily include thicker degraders . therefore they were never considered as potentially harmful for experiments . value obtained from musics in the pb+p at 500a mev experiment . the central band corresponds to fragments that did not gain or loose any electron while passing through the degrader . the upper band corresponds to fragments that lost an electron , the lower band to fragments that gained one electron.__,scaledwidth=70.0% ] the use of a thick degrader as a passive measurement device has already proved to be successful in previous experiments conducted at the frs at energies of or close to 1a gev . we have already emphasized that its use in the spallation of lead at 500a mev was of particular importance as the number of non - stripped ions was expected to be very high ( up to 60% after the degrader ) . at higher energies , the influence of degrader inhomogeneities become negligible , and the charge - state straggling is strongly reduced . our calculations indicate that a 0.5% resolution should be obtained , which allows for a satisfactory @xmath18 or @xmath15 identification . this conclusion is in good agreement with results from previous experiments . from figure [ fig : deg ] , it is obvious that during the 500a mev experiment the resolution required to separate charge states was not achieved : no separation can be seen between fully stripped and h - like fragments . they appear as a bump , clearly visible for each spot in the upper part of the most intense line , but impossible to disentangle from the fully stripped fragments . this could be expected from the discussion conducted in the previous section : the position straggling observed at @xmath2 leads to an uncertainty of 0.4% on the magnetic rigidity . as the difference of magnetic rigidity is only 31% of the magnetic rigidity in the second part of the spectrometer , the uncertainty on @xmath65 is expected to be close to 1.25% , which is very close to the observed resolution . such a value prevent identification from the various charge states of the fragments . in the discussion above we only considered the beam , which allowed us to neglect the influence of the angle on the precision of the magnetic - rigidity variation . this is not true anymore when one looks at the fragments , and a non - optimal position of the scintillator at @xmath2 may have further worsen the experimental resolution . we could not put in evidence any effect of the angle in the case of the 500a mev experiment as a failure of a device prevented the measurement of the angle of the fragments at @xmath2 . this does not affect our global conclusion on the reasons why the reduced resolution was obtained for the energy loss measurement in the degrader . based on our simulations and the experimental results , we postulate that defects on the degrader surface are the main reason for this very poor resolution , and that an improved degrader ( with a surface roughness smaller than 10 @xmath93 m ) may allow this method to be successfully used at an energy as low as 500a mev for very heavy beams such as lead . the determination of the mass of each fragment requires two steps , which will be described in this section . first , the a / q ratio is deduced from the magnetic rigidity and time of flight in the second section . then , if the use of the degrader has been successful in determining the ionic charge state of the fragments , or if the proportion of non - stripped ions is low enough to neglect them , getting the mass of each fragment is straightforward . alternatively , if the charge state is not known , it is convenient to determine a `` most - likely '' mass for each fragment . the @xmath17 ratio is deduced from the magnetic rigidity and from the time of flight in the second section of the spectrometer . various corrections of dependencies from the positions have to be applied : effect of the position of the fragment at @xmath1 and @xmath2 on the length of the trajectory , non - linear effects in the time response of the scintillators . this leads to a parametrization defined as : @xmath94 @xmath95 and @xmath96 are free parameters . they are assumed to stay constant as long as magnetic optics is not changed . their values can be adjusted considering a given setting for which masses are affected by counting , starting from the centered fragment . it is usually convenient to use a magnetic setting corresponding to fragments close to the beam , as identification is in this case quite straightforward because of the reduced velocity spread of those fragments . in contrast , in low - energy experiments the high proportion of charge states suggests a setting centered on very light isotopes . the calculated parameters can be checked by re - calculating the mass number of the ion beam . in the case of the 500a mev experiment , the mass of @xmath97 was obtained with a precision of @xmath98 . the resolution required to separate masses with a peak - over - valley ratio of 10 is 0.1% . the frs layout and detectors have been designed to fulfill this requirement . in this respect , the most difficult situation is expected to be the measurement of heavy fragments at high energy : in this case the time of flight is short and the difference in magnetic rigidity between two neighboring masses is minimum . in this respect , a low - energy experiment is expected to be a better case , as the first constraint is softened . nevertheless , the presence of polluting charge states can strongly hamper the resolution obtained on the a / q ratio . for fragments produced from a heavy , stable nucleus such as lead , the a / q ratio is close to 2.5 . therefore the a / q peaks for fragments with an odd number of electrons will fall in - between peaks associated to fragments with an even number of electrons , leading to an apparent loss of resolution . this has of course no consequence if the ionic charge state of each fragment is known , as the exact mass can be obtained . on the other hand , if the ionic charge state is not known , one can assign its most - likely mass to each fragment . also , according to the previous discussion , the fragments of mass @xmath11 are mixed with fragments @xmath99 and @xmath100 . we will show that one can use the correlation between the @xmath17 ratio and the @xmath65 values to reduce the number of actual masses that are mixed together . , right part to a setting centered on @xmath101.__,title="fig:",scaledwidth=45.0% ] , right part to a setting centered on @xmath101.__,title="fig:",scaledwidth=45.0% ] an example of such a correlation is presented in figure [ fig : aq_dbrho ] , for heavy and light hg isotopes produced in the 500a mev pb+p experiment ( only fragments with @xmath64 are represented ) . this figure also underlines the lack of resolution in @xmath65 obtained in this experiment . the large bands seen for heavy isotopes are a mix of the fully stripped ions ( low part of bands ) and h - like ions ( upper part ) . in the case of light isotopes , very few h - like ions are produced and transmitted , and therefore only the lower part of the bands remains . as no direct separation of charge states is possible , it is advantageous to get the @xmath17 value , not from a direct projection on the @xmath17 axis , but from a projection along the direction of the correlation bands . not only does it improve the apparent resolution , creating clearly separated peaks ( in the case of the 500a mev experiment , the resolution becomes close to 0.1% for all settings , while it could reach 0.4% in the most unfavorable cases ) . it also ensures that only masses @xmath100 contribute to the pollution of isotopes of mass @xmath11 . the unavoidable weakness of this method is that the group of fragments identified as being of mass @xmath11 and atomic charge @xmath102 contains a variable fraction of @xmath103 fragments , which will have to be subtracted in order to deduce the production rates . at this step unambiguous identification in mass on an event - by - event basis is of course not possible anymore . nevertheless it is convenient to assign a mass to each fragment . to calculate masses , we assume that fragments gaining zero or more electrons when passing through the degrader were fully stripped in the first part of the frs , while fragments loosing one or more electrons are fully stripped in the second part of the frs . in other terms , for a given value of @xmath63 , each fragment is assumed to be in the most probable charge state in the two parts of the frs . in the worst case ( heavy fragments for which @xmath64 ) this hypothesis is valid for only 60% of fragments . even if fragments are not unambiguously identified in mass on an event - by - event basis , we will show that it remains possible to extract isotopic production rates . this requires to evaluate and correct for the fraction of fragments that was not correctly identified because of the unresolved ionic charge states . in this chapter we will present a method to achieve those corrections . first of all it is necessary to take into account the limited acceptance of the frs . its acceptance is limited to @xmath1041.5% in magnetic rigidity ( or in longitudinal momentum ) for each given ion . this means that , if the broadening of the fragment velocity goes beyond this value , only part of the momentum distribution will be transmitted in a given setting of magnets . calculation of the production cross sections therefore requires the reconstruction of the complete kinematic distribution of each fragment , which may require data from different magnetic settings . the momentum from each fragment in the first part of the spectrometer can be calculated by applying the relation ( [ eqn : brho ] ) to the first part of the frs : @xmath105 for identified fragments , @xmath11 and @xmath15 are known as integer values . therefore , the precision obtained on the velocity depends only on the precision on the magnetic rigidity . as previously discussed , it can be estimated to be in the order of @xmath106 , which is far better than the best time - of - flight measurements for heavy ions . any error on the estimation of the charge state of fragments has a direct consequence on the calculated mass and therefore on the calculated momentum value . let us consider for example a fragment @xmath107 which was assumed to be fully stripped in both parts of the frs . if this fragment was in fact an hydrogen - like ion , its real mass is @xmath100 and its calculated momentum is related to its real one by : @xmath108 the velocity distributions of fragmentation residues are known to be of gaussian shape . as it can be seen on figure [ fig : bg ] , the misidentified fragments create a second component in the velocity spectra . according to the previous discussion about the fragmentation experiments , this component is very important for high - z , neutron - rich fragments while it slowly vanishes for low - z and/or proton - rich fragments . there are two ways to deal with this contribution : either a direct estimation through a fit , or an estimation of the whole distribution ( also by a fit ) followed by a subtraction of the part contributed by the pollution . if the first solution is well suited for fragments very close to the projectile , for which the velocity spread is small enough for the two peaks to be distinguished , its application to the case of lighter fragments , for which the two peaks partially or totally overlap , would lead to large uncertainties . therefore , we recommend the second solution , and we will now present a way to implement it in a numerically stable way . the velocity distribution of each fragment can be described as the sum of two gaussian functions , the second one accounting for the pollution related to misidentified fragments . in order to optimize its stability , a single fit is applied , for each set of data associated to a given combination of @xmath18 and @xmath63 . the coefficients describing the center ( noted @xmath109 ) and the width ( noted @xmath110 ) of each gaussian are related to one another by simple , physics - based relations derived from the morrissey systematics @xcite : @xmath111 @xmath112 defining an elementary velocity distribution as : @xmath113 the velocity function can be simply written as : @xmath114 the @xmath115 coefficient stands for the ratio between the probability of the considered charge state and the probability of the charge state responsible for contamination . it is a free parameter of the fit . the value obtained for each @xmath18 can be compared to values calculated with a code dedicated to ionic charge states like global @xcite . as can be seen on figure [ fig : qst_fit ] , in the 500a mev pb+p experiment , the overall agreement was satisfactory . the experimental production rate @xmath116 for each isotope , including the contribution from misidentified fragments , is obtained by integration of the @xmath117 function defined in equation [ eqn : velocity_fit ] . the contribution from misidentified fragments can be subtracted iteratively : @xmath118 as the production cross section of the lightest isotopes decreases exponentially , iteration can be started considering that the net production rate @xmath119 is 0 for the lightest isotopes . to calculate the charge - state probability @xmath120 we used the global code . we considered a 10% uncertainty for this calculation . this leads to a high uncertainty for the few neutron - rich isotopes , for which the correction reaches several tens of percent . at this step , production rates have been obtained for all the most likely charge - state combinations . in order to reduce the statistical uncertainty , it is appropriate to consider the production rates obtained for all combinations . this differs from high - energy experiments in which it is generally convenient to disregard the non - fully stripped fragments . the simplest way to normalize all the different production rates consists in using calculated charge - state probabilities , by example using global . but it is also possible to fit those probabilities by requiring agreement on the different normalized distributions . this process is illustrated in figure [ fig : qst_renorm ] for the hg isotopes measured in the 500a mev experiment . the agreement between the various distributions is very satisfactory . the use of the multiple charge - state combinations has two interests : a reduction of the uncertainty on the production rate and access to an experimental value for the charge - state probabilities . further check of this method can be achieved by comparing the values obtained for the charge - state probabilities to global calculations . this comparison is presented in figure [ fig : qst_compare ] for the 500a mev experiment . the agreement is better than 10% for most combinations . nevertheless a spectacular deficiency of the code is observed for li - like ions for which production is underestimated by global by a factor 2.5 . from this point , the goal of getting the cross sections can be achieved in the same way as in any experiment : one needs to divide the production rates by the number of atoms in the target and by the number of ions from the beam , and correct for the efficiency of the detectors . in the case of fragmentation reactions , multiple reactions play an important role in the production of light isotopes . this is especially true at low energy as the production cross sections fall sharply @xcite . therefore , the contribution of these multiple reactions must be carefully subtracted . a dedicated method has been developed . this is largely self - consistent and uses a limited input from the epax parametrization @xcite . this method will be presented in a forthcoming paper @xcite . it has been demonstrated that a clean @xmath18 identification can be obtained for reaction products even when they are not completely stripped by increasing the quantity of gas of the ionization chambers with respect to the quantities used for high - energy experiments . conclusions are not so clear regarding mass identification , which requires the charge - state measurement using a thick degrader . according to our calculations , this identification is possible at 500a mev , although the expected resolution is extremely close to the acceptable limit . we have shown that during the 500a mev pb+p experiment the obtained resolution was far worse than the calculated value . using additional simulations we propose that the main reason for this failure is the presence of surface inhomogeneities on the degrader . therefore , it may be possible that the technique developed for this experiment will be successfully applied in the future , when an improved degrader will be available . we have demonstrated that it is possible to obtain production rates with a high accuracy , even in the case of a failure of the charge - state measurement . this can be achieved through a systematic evaluation of the contribution of the different charge states and the suppression of the contribution of polluting ones . calculations by an atomic - physics code ( global ) are needed in order to estimate some corrections , but most of the atomic - physics data can be obtained in a self - consistent way from the data of the experiment . in the case of the 500a mev experiment , comparison between global calculations and obtained atomic - physics data confirmed the overall accuracy of this code at this energy , with the exception of the estimation of li - like ions . if one considers the low - energy limit defined in the early papers dedicated to the frs , the overall success of the 500a mev experiment is clearly a push beyond this limit . nevertheless , further decrease in energy for experiments in the system pb+p would create further problems , as further multiplication of the charge states would lead to additional complexity in the analysis . an energy loss can be written as a normalized value : in the energy range considered in this paper ( a few hundreds of mev to one gev per nucleon ) and with the degrader thickness selected as half of the range of the beam , the kinematic factor ( left part of the expression in equation [ eqn : brho - e ] ) is always close to 1 ( for example , 0.92 at 500a mev or 0.95 at 1a gev ) . this could be expected , as this term may be approximated here by @xmath129 . h. geissel et al . - nuclear instruments and methods b 70 ( 1992 ) 286 b. voss et al . - nuclear instruments and methods a 364 ( 1995 ) 150 k .- h . schmidt et al . - nuclear instruments and methods a 260 ( 1987 ) 287 p. chesny et al . - gsi annual report 97 - 1 ( 1997 ) f. rejmund et al . - nuclear physics a 683 ( 2001 ) 540 t. enqvist et al . - nuclear physics a 686 ( 2001 ) 481 t. enqvist et al . - nuclear physics a 703 ( 2002 ) 435 j. taieb et al . - nuclear physics a 724 ( 2003 ) 413 - 430 e. casajeros - phd thesis , universitad de santiago de compostela ( 2002 ) . j. lindhard and a. h. sorensen - physical review a 53 ( 1996 ) 2443 h. geissel et al - nuclear instruments and methods b 195 ( 2002 ) 3 h. weick et al - nuclear instruments and methods b 193 ( 2002 ) 1 the code atima can be downloaded at http://www.gsi.de/ weick / atima

we have identified and quantified semi - major axis drifts in near - earth asteroids ( neas ) by performing orbital fits to optical and radar astrometry of all numbered neas . we focus on a subset of 54 neas that exhibit some of the most reliable and strongest drift rates . our selection criteria include a _ yarkovsky sensitivity _ metric that quantifies the detectability of semi - major axis drift in any given data set , a signal - to - noise metric , and orbital coverage requirements . in 42 cases , the observed drifts ( au / myr ) agree well with numerical estimates of yarkovsky drifts this agreement suggests that the yarkovsky effect is the dominant non - gravitational process affecting these orbits , and allows us to derive constraints on asteroid physical properties . in 12 cases , the drifts exceed nominal yarkovsky predictions , which could be due to inaccuracies in our knowledge of physical properties , faulty astrometry , or modeling errors . if these high rates can not be ruled out by further observations or improvements in modeling , they would be indicative of the presence of an additional non - gravitational force , such as that resulting from a loss of mass of order a kilogram per second . we define the _ yarkovsky efficiency _ as the ratio of the change in orbital energy to incident solar radiation energy , and we find that typical yarkovsky efficiencies are .

1204.5990

understanding how the yarkovsky force modifies asteroid orbits has illuminated how asteroids and meteorites are transported to near - earth space from the main belt and has allowed for deeper understanding of the structure of asteroid families @xcite . the yarkovsky force is necessary for accurately predicting asteroid trajectories , including those of potentially hazardous asteroids @xcite . the yarkovsky effect ( or force ) describes the process by which an asteroid s surface thermal lag and rotation result in net thermal emission that is not aligned towards the sun @xcite . the so - called _ diurnal component _ of the yarkovsky effect operates as follows . a prograde - spinning object generally has a component of this surface thermal emission anti - aligned with the motion along the orbit , producing a net increase in the object s semi - major axis ( i.e. , @xmath4 , where @xmath5 is the semi - major axis ) . conversely , a retrograde - spinning object generally has a component aligned with its velocity , shortening its semi - major axis ( i.e. , @xmath6 ) . the maximum possible drift rate for any radiation - powered force acting on near - earth asteroids ( neas ) can be obtained by equating the incident solar radiation energy in a given time interval to the change in orbital energy during the same interval . we find @xmath7 where @xmath1 is an efficiency factor analogous to that used by @xcite , @xmath8 is the eccentricity , @xmath9 and @xmath10 are the luminosity and mass of the sun , @xmath11 is the gravitational constant , and @xmath12 and @xmath13 are the effective diameter and bulk density of the asteroid . this equation exhibits the expected dependence on the asteroid area - to - mass ratio . in convenient units , it reads @xmath14 maximum efficiency ( @xmath1=1 ) would convert all incoming solar radiation into a change in orbital energy . we will show in section [ sec : results ] that typical yarkovsky efficiencies are @xmath15 , and that typical rates are @xmath16au / myr for kilometer - sized asteroids . the low efficiency and rates are due to the fact that it is the momentum of departing thermal photons that moves the asteroid . @xcite used precise radar ranging measurements to ( 6489 ) golevka and reported the first detection of asteroidal yarkovsky drift . the drift rate for this nea of @xmath17 au / myr @xcite corresponds to an efficiency @xmath18 for @xmath12=530 m and @xmath19 kg m@xmath20 . @xcite employed the yarkovsky effect to link a 1950 observation to asteroid ( 152563 ) 1992 bf with a @xmath21 rate of @xmath22 au / myr . this corresponds to an efficiency @xmath23 for @xmath12=420 m and @xmath13=2500 kg m@xmath20 . if 1992 bf has a density closer to 1500 kg m@xmath20 , the efficiency would be @xmath24 . there have been other searches for the effects of non - gravitational forces in asteroid orbits . @xcite considered a semi - major axis drift in the orbit of ( 1566 ) icarus and found @xmath25 au / myr . our best estimate is @xmath26 au / myr . @xcite found it necessary to incorporate a non - gravitational term @xmath27 au / myr in his orbit determination of ( 4179 ) toutatis , however the availability of radar ranges in 1992 , 1996 , 2004 , and 2008 strongly suggest a drift magnitude that does not exceed @xmath28 au / myr . @xcite examined the orbits of 10 asteroids and found drifts in four asteroids , including a @xmath29 au / myr drift for ( 1862 ) apollo . @xcite used a cometary model to search for perturbations and also detected a drift associated with ( 1862 ) apollo , though a value was not reported . our best estimate is @xmath30 au / myr ( section [ sec : results ] ) . it appears that these early estimates are not aligned with modern determinations , and may have been caused by erroneous or insufficient astrometry . more recently , @xcite searched for yarkovsky signatures and reported rate estimates for 12 candidates . here we use new developments in star catalog debiasing @xcite as well as the most recent astrometric data to compute semi - major drift rates for select neas , which multiplies the number of existing measurements by a factor of @xmath24 . observations of yarkovsky rates can be used to place constraints on composition ( i.e. metal vs. rock ) , physical properties ( i.e. bulk density ) , and spin properties ( i.e. prograde vs. retrograde ) . the magnitude of the force is dependent on the object s mass , size , obliquity , spin rate , and surface thermal properties . separating how each of these quantities uniquely contributes to a measured @xmath21 is often not possible , but past yarkovsky detections have allowed for insight into the associated objects . with certain assumptions on surface thermal properties , bulk densities were determined from the measured drifts of golevka @xcite and ( 152563 ) 1992 bf @xcite . for the latter , the magnitude and direction of the drift point to an obliquity in excess of 120 degrees @xcite . the yarkovsky drift manifests itself primarily as a change in mean anomaly ( or along - track position ) , and some observational circumstances are poorly suited to detect such changes . examples include optical astrometry secured when the line - of - sight is roughly parallel to the asteroid velocity vector or when the object is at large distances from earth . in both instances the differences in astrometric positions can be much smaller than observational uncertainties , resulting in low sensitivity to the yarkovsky effect . the overall yarkovsky sensitivity depends on the orbital geometry of the nea and on the entire set of available observations . this can be quantified rigorously . for each epoch @xmath31 at which optical observations were obtained ( @xmath32 ) , we predict the position @xmath33 for the best - fit orbit ( @xmath34 ) and the position @xmath35 for the same orbit modified by a nominal non - zero @xmath21 . the value of the nominal rate is not important as long as it results in detectable ( @xmath2arcsecond ) changes in coordinates and as long as it is applied consistently to all objects ; we used @xmath21=0.1 au / myr . we then define the _ yarkovsky sensitivity _ @xmath36 as @xmath37 where @xmath38 is the positional uncertainty associated with observation @xmath39 . this root mean square quantity provides an excellent metric to assess the relative sensitivity of any given data set to a drift in semi - major axis , including drifts caused by yarkovsky influences . the metric can be applied to the entire set of available observations , or to the subset of observations that survive the outlier rejection steps described below . we computed both quantities and used the latter for our analysis . we found that data sets with scores @xmath36 below unity yield unreliable results , including artificially large rates and large error bars . out of @xmath21,250 numbered neas , only @xmath2300 have @xmath40 and @xmath2150 have @xmath41 . in this paper we focus on a subset of these neas . for this work we employed orbital fits to optical astrometry to determine semi - major axis drift rates for neas . we used the orbfit software package , which is developed and maintained by the orbfit consortium @xcite . orbfit can fit nea trajectories to astrometric data by minimizing the root mean square of the weighted residuals to the data , optionally taking into account a given non - zero rate of change in semi - major axis @xmath21 . we included perturbations from 21 asteroids whose masses were estimated by @xcite . we downloaded optical astrometry for all numbered minor planets ( numobs.txt.gz ) from the minor planet center ( mpc ) on january 31st , 2012 . we have assumed that all the astrometry has been properly converted to the j2000 system . the quality of the astrometry varies greatly , and we applied the data weighting and debiasing techniques implemented in orbfit , which appear to follow the recommendations of @xcite . data weights are based on the time the observation was performed , the method of the observation ( ccd or plate ) , the accuracy of the star catalog , and in some cases the accuracy of the observatory . correction for known star catalog biases was applied when possible . biases vary depending on the specific star catalog and region of the sky , and can reach 1.5 arcseconds in both right ascension ( r.a . ) and declination ( dec . ) . correction for these biases can substantially improve the recovery of orbital parameters from observations . however , as discussed in @xcite , not every observation can be debiased . some observations were reported to the mpc without noting the star catalog used in the data reduction . although @xcite deduced the star catalogs used by several major surveys , there remain observations from smaller observatories that do not have associated star catalogs . accordingly , a fraction of the astrometry used in this paper was not debiased . based on counts published @xcite , we estimate this fraction to be less than 7.2% of all the observations . our procedure for determining the semi - major axis drift rate included three steps : an initial fit to the debiased data , an outlier rejection step , and a search for the best - fit @xmath21 , with iteration of the last two steps when necessary . we used the orbital elements from the minor planet center s mpcorb database as initial conditions for the first fit for each object ( step 1 ) . this first fit , performed with @xmath42 and outlier rejection turned off , slightly corrected the orbital elements for our weighted , debiased observations . the orbital elements from each object s first fit became the starting orbital elements for all later fits of that object . the second fit of each object served to reject outliers and was initially performed with @xmath42 ( step 2 ) . the residual for each observation was calculated using the usual observed ( o ) minus computed ( c ) quantities : @xmath43 where @xmath44 and @xmath45 are the uncertainties for that observation in r.a . and , respectively . we rejected observations when their @xmath46 , and recovered previously rejected observations at @xmath47 , with the rejection step iterated to convergence . results are fairly robust over a large range of thresholds for rejection ( section [ sec : results ] ) . if the post - fit residuals were normally distributed , the chosen thresholds would result in @xmath48 of observations being rejected as outliers . because errors are not normally distributed , our typical rejection rates are 2 - 5% of all available astrometry . this second fit produced the set of observations which were used in the third step . the third step was a series of orbital element fits to the observations over a set of fixed @xmath21 values . during these fits , we used the set of observations defined by the second fit and did not allow further outlier rejection . the quality of a fit was determined by summing the squares of residuals @xmath49 . to locate the region with the lowest @xmath50 , we used a three - point parabolic fit or the golden - section minimization routine @xcite . a parabola was then fit to the @xmath50 curve in the vicinity of the minimum , and we used the minimum of the parabola to identify the best - fit @xmath21 value . confidence limits were estimated using @xmath50 statistics . confidence regions of 68.3% and 95.4% ( @xmath51 and @xmath52 , respectively ) were established by the range of @xmath21 values that yielded @xmath50 values within 1.0 and 4.0 of the best - fit @xmath50 value , respectively ( fig . [ fig:2100chi ] ) . value to optical astrometry of ( 2100 ) ra - shalom ( 1,281 observation epochs , 2,562 observations , 7 adjustable parameters , 2,555 degrees of freedom ) . the sums of squares of residuals corresponding to a range of @xmath21 values are shown as circles , with a parabolic fit shown as a dotted line . the @xmath21 values plotted here were determined by the golden section search algorithm @xcite as it searched for and found a minimum at @xmath53 au / myr with a reduced @xmath50 value of @xmath54 . confidence limits of @xmath55 ( @xmath51 ) are indicated by the thick dashed line , and correspond to the range @xmath56 \times 10^{-4}$ ] au / myr . the thin dashed line shows the 95.4% ( @xmath52 ) confidence region . [ fig:2100chi ] ] the initial outlier rejection step can in some cases eliminate valid observations simply because the yarkovsky influences are not captured in a dynamical model with @xmath34 . to circumvent this difficulty , we iterated the outlier rejection step with the best - fit @xmath21 value and we repeated the fitting process . in 52 out of 54 cases , the new best - fit value matched the previous best - fit value to within @xmath51 , and we accepted the new best - fit values as final . for the other objects we repeated the reject and fit processes until successive best - fit values converged within @xmath51 ( which never required more than one additional iteration ) . our results report the @xmath21 values obtained at the end of this iterative process . we restricted our study to numbered neas with the best yarkovsky sensitivity ( equation [ eq - sy ] ) , specifically @xmath57 ( fig . [ fig - xmas ] ) . we also chose to focus on objects with non - zero @xmath21 values by using a signal - to - noise ratio ( snr ) metric , defined as the ratio of the best - fit @xmath21 to its @xmath51 uncertainty . we accepted all objects with snr @xmath58 ( fig . [ fig - xmas ] ) . some asteroids have observations that precede the majority of the object s astrometry by several decades and have relatively high uncertainties . in order to test the robustness of our results , we removed these sparse observations , which were defined as ten or fewer observations over a 10-year period . fits were then repeated for these objects without the early observations . if the initial best - fit value fell within the @xmath51 error bars of the new best - fit value , the initial result was accepted , otherwise , the object was rejected . superior detections of the yarkovsky effect are likely favored with longer observational arcs , larger number of observations , and good orbital coverage . for this reason we limited the sample to those neas with an observational arc at least 15 years long , with a number of reported observations exceeding 100 , and with at least 8 observations per orbit on at least 5 separate orbits . we report on the 54 objects that met all of these criteria : sensitivity , snr , sparse test , and orbital coverage . we validated our optical - only technique whenever radar ranging observations were available on at least two apparitions . this could only be done for a fraction of the objects in our sample . in the remainder of this paper , _ optical - only _ results are clearly distinguished from _ radar+optical _ results . for the radar+optical fits , we included all available radar astrometry and disallowed rejection of potential radar outliers . the internal consistency of radar astrometry is so high that outliers are normally detected before measurements are reported . we also verified that a fitting procedure that holds successive @xmath21 values constant is equivalent to performing 7-parameter fits ( 6 orbital parameters and @xmath21 simultaneously ) . the @xmath21 values obtained with both procedures are consistent with one another . in addition to the measurements described above , we produced numerical estimates of the diurnal yarkovsky drift for each of the objects in our sample . comparing the measured and estimated rates provides a way to test yarkovsky models . in some instances , e.g. , robust observations irreconcilable with accurate yarkovsky modeling , it could also lead to the detection of other non - gravitational forces , such as cometary activity . our numerical estimates were generated as follows . at each timestep , we computed the diurnal yarkovsky acceleration according to equation ( 1 ) of @xcite , which assumes a spherical body , with the physical parameters @xcite listed in table [ tab - thercon ] and an assumption of 0@xmath59 or 180@xmath59 obliquity . we assumed that the thermal conductivity did not have a temperature dependence , but found that adding a temperature - dependent term according to the prescription of @xcite ( @xmath60 , with @xmath61 ) did not change our predictions by more than 1% . we then resolved the acceleration along orthogonal directions , and used gauss form of lagrange s planetary equations @xcite to evaluate an orbit - averaged @xmath21 . the physical parameters chosen for these predictions mimic two extremes of rocky asteroids ; one is intended to simulate a rubble pile with low bulk density , the other a regolith - free chunk of rock ( table [ tab - thercon ] ) . these parameters correspond to a thermal inertia range of @xmath62 j m@xmath63 s@xmath64 k@xmath65 , enveloping the results of @xcite , who found an average nea thermal inertia to be @xmath66 j m@xmath63 s@xmath64 k@xmath65 . in most cases , the drift rates produced by these two extreme cases encompass the drift produced by a rubble - pile object that has a regolith - free surface , or the drift produced by a solid object with regolith . lcccc composition & @xmath67 ( @xmath68 ) & @xmath69 ( @xmath70 ) & @xmath71 ( @xmath72 ) & @xmath13 ( @xmath72 ) + rubble pile & 500 & 0.01 & 1200 & 1200 + rock chunk & 500 & 0.50 & 2000 & 2000 + there is no simple relationship between these physical parameters and predicted drift rates , but for most cases the rubble pile exhibits the larger @xmath21 values due to its low bulk density ( equation [ equ : dadttyp ] ) . the smaller values of density of the surface and thermal conductivity for rubble piles produce a smaller thermal inertia , and therefore a longer thermal lag . generally , but not always , this longer thermal lag , combined with the rotation of the asteroid , allows for a larger fraction of departing thermal emission to be aligned with the asteroid s velocity , resulting in a larger drift . when available , measured values of the geometric albedo , diameter , and spin rate from the jpl small - body database @xcite were incorporated into our predictions for yarkovsky drifts . when not available , the diameter @xmath12 in km was estimated from the absolute magnitude @xmath73 using @xcite , @xmath74 where we used two values of the v - band geometric albedo @xmath75 ( 0.05 and 0.45 ) , a range that captures observed albedos for the majority of neas . when spin rate was unknown , we assumed a value of 5 revolutions / day , based on the average spin rate values for asteroids 1 to 10 km in diameter shown in fig . 1 of @xcite . emissivity was assumed to be 0.9 . bond albedo was estimated with a uniform value of the phase integral ( q=0.39 ) on the basis of the iau two - parameter magnitude system for asteroids @xcite and an assumed slope parameter g=0.15 . we have assumed @xmath76 for the purpose of quantifying the yarkovsky efficiency when the asteroid size was unknown . we measured the semi - major axis drift rate of all 1,252 numbered neas known as of march 2012 . some of the drift rates are not reliable because of poor sensitivity to yarkovsky influences ( fig . [ fig - xmas ] ) . plotted as a function of semi - major axis drift rate @xmath21 for 1,252 numbered neas . data sets with yarkovsky sensitivity below unity ( dashed line ) yield unreliable results , including large rates and large error bars . our selection criteria require @xmath57 ( dotted line ) and snr @xmath77 . the 80 objects that meet both selection criteria are shown in green . about 26 of these 80 neas are eliminated by the sparse test and orbital coverage requirements ( see section [ sec - select]).,width=480 ] after our process of selection and elimination ( section [ sec - select ] ) , we were left with 54 neas that exhibit some of the most reliable and strongest drift rates . although we report objects with @xmath57 , we have the most confidence in objects with highest yarkovsky sensitivity , and we show objects in order of decreasing @xmath36 value in our figures . we examined the impact of various choices of reject / recover thresholds when rejecting outlier observations ( fig . [ fig : rejectrecover ] ) . at moderate values of the rejection threshold ( i.e. eliminating less than @xmath25% of observations ) , best - fit values are consistent with one another . in this regime , results are fairly robust against the choice of rejection thresholds . however results do become sensitive to rejection thresholds when a larger fraction of observations is rejected . as the reject / recover thresholds become more stringent , astrometry with evidence of semi - major axis drift is preferentially rejected , and the best - fit @xmath21 values approach zero . our adopted reject / recover thresholds ( @xmath78/@xmath79 ) are stringent enough that they eliminate obvious outliers , but not so stringent as to suppress the yarkovsky signal . in 52 out of 54 cases , repeating the outlier rejection step with the best - fit @xmath21 value resulted in no appreciable change to the result . ) on the best - fit @xmath21 values . results from optical - only fits are shown with their @xmath51 error bars for two representative cases , ( 2202 ) pele and ( 2063 ) bacchus . best - fit @xmath21 values are consistent with one another in the left half of the diagram . values to the right of 2.3/2.2 ( pele ) and 2.5/2.4 ( bacchus ) have snr less than unity and would not meet our selection criteria . our adopted reject / recover thresholds are @xmath80 and @xmath81 . [ fig : rejectrecover ] , width=576 ] as a validation step , we compared the semi - major axis drift rates obtained with our procedure ( both optical - only and radar+optical ) to previously published values ( table [ tab - chescompare ] ) . we found good agreement for golevka @xcite and 1992 bf @xcite , and for most , but not all , neas included in a similar study done by @xcite . the differences between our results and those of @xcite can probably be attributed to our use of debiased data , of improved data weights , and of longer observational arcs extending to 2012 . eight objects included in table [ tab - chescompare ] meet our selection criteria for detailed analysis in the rest of this paper : ( 1620 ) geographos , ( 1685 ) toro , ( 1862 ) apollo , ( 1865 ) cerberus , ( 2063 ) bacchus , ( 2100 ) ra - shalom , ( 2340 ) hathor , and ( 152563 ) 1992 bf . lrrrrrrrrrr nea & & & & & & + & rms & rms & rms & @xmath21 & @xmath21 & @xmath21 + ( 1620 ) geographos & 0.393 & 0.356 & 0.55 & @xmath82 & @xmath83 & @xmath84 + ( 1685 ) toro & & & 0.51 & @xmath85 & & @xmath86 + ( 1862 ) apollo & 1.111 & 0.403 & 0.61 & @xmath87 & @xmath88 & @xmath89 + ( 1865 ) cerberus & & & 0.54 & @xmath90 & & @xmath91 + ( 2063 ) bacchus & & & 0.59 & @xmath92 & & @xmath93 + ( 2100 ) ra - shalom & 0.488 & 0.594 & 0.51 & @xmath94 & @xmath95 & @xmath96 + ( 2340 ) hathor & & & 0.67 & @xmath97 & & @xmath98 + ( 6489 ) golevka & 0.879 & 0.387 & 0.61 & @xmath99 & @xmath100 & @xmath101 + ( 54509 ) yorp & 0.796 & 0.260 & 0.55 & @xmath102 & @xmath103 & @xmath104 + ( 85953 ) 1999 fk21 & & & 0.56 & @xmath105 & & @xmath106 + ( 101955 ) 1999 rq36 & 15.694 & 0.127 & 0.39 & @xmath107 & @xmath108 & @xmath109 + ( 152563 ) 1992 bf & & & 0.60 & @xmath110 & & @xmath111 + several conclusions can be drawn from the data presented in table [ tab - chescompare ] . first , the rms values indicate excellent fits to the astrometry . second , the solutions with non - zero @xmath21 values provide a much better match to the radar data than the gravity - only solutions , with typical rms values decreasing by a factor of 2 or more . third , radar+optical estimates have consistently lower error bars than optical - only estimates , sometimes dramatically so , which is typical in nea studies . finally , there is a generally good agreement between the optical - only @xmath21 values and the radar+optical @xmath21 values , indicating that the optical - only technique is a useful tool that can be used even in the absence of radar data . drift rates for the 54 neas that pass our selection criteria are presented in table [ tbl - results ] along with orbital elements and physical properties . if an object has both a optical - only and a radar+optical value , we used the more accurate radar+optical value in the following figures and calculations ( unless specified otherwise ) . we used equation ( [ equ : dadttyp ] ) with a density of 1,200 kg m@xmath20 to compute efficiency factors @xmath1 and found that objects divided roughly into two groups . in the first group of 42 objects with @xmath112 , most observed @xmath21 values are consistent ( within @xmath51 ) with yarkovsky predictions . we refer to these objects as _ yarkovsky - dominated _ ( figs . [ fig : defyark ] and [ fig : defyarkb ] ) . in the second group of 12 objects with @xmath113 the observed @xmath21 values are somewhat larger than yarkovsky predictions , but improvements in the knowledge of physical properties or in yarkovsky modeling could plausibly bring some of the observed rates in agreement with predictions . we refer to these objects as _ possibly yarkovsky - dominated _ ( fig . [ fig : possyark ] ) . figures [ fig : defyark ] and [ fig : defyarkb ] indicate that there is generally agreement between observations and numerical estimates of yarkovsky drift rates for neas with @xmath114 . these data suggest that @xmath115 represents a typical efficiency for the yarkovsky process . predicted values are based on calculations with obliquities of @xmath116 and @xmath117 , therefore , observed rates that are lower than predictions could still be due to the yarkovsky effect . the majority of objects in fig . [ fig : defyarkb ] appear to exceed predictions . this is a consequence of the snr @xmath77 selection criterion , as it eliminates objects with lower @xmath21 values . on the basis of the entire sample of measured drifts for objects with @xmath57 , we can compute average properties for observed yarkovsky rates and efficiencies . the mean , mean weighted by uncertainties , median , and dispersion are shown in table [ tab - meanmedian ] . the aggregate properties are comparable if we restrict objects to the subset with snr @xmath118 1 , except for slightly increased @xmath21 rates ( median rate of @xmath119 au / myr instead of @xmath120 au / myr ) , as expected . the yarkovsky process appears to have an efficiency @xmath1 of order @xmath121 , with a fairly small dispersion . because the yarkovsky efficiency scales with density ( @xmath122 ) some of the observed scatter is due to density variations . error bars . shaded boxes show a range of predicted yarkovsky rates representing different compositions ( table [ tab - thercon ] ) . as predicted yarkovsky values were calculated assuming @xmath116 or @xmath117 obliquity , the shaded boxes represent maximum drifts for the object . therefore , a fit that lies between a shaded box and @xmath34 is considered to have good agreement . objects with a single corresponding shaded box have a known diameter ( table [ tbl - results ] ) . objects with two shaded boxes did not have known diameters , and were modeled using diameters derived from assumed albedos ( 45% in light blue , larger predicted drift magnitudes , and 5% in dark blue , smaller predicted drift magnitudes ) . the vertical extents of the shaded boxes represent the range of compositional types described in table [ tab - thercon ] , with the larger absolute values representing the rubble pile " composition , and the lower absolute values representing the rock chunk " composition . ] . the observed rates for the majority of objects shown in this figure appear to exceed predicted values . this is a consequence of the snr @xmath77 selection criterion which eliminates objects with lower @xmath21 values . ] exceeding 2 @xmath123 . symbols are as in fig . [ fig : defyark ] . most objects in this figure have measured drifts that lie outside of the range of values expected on the basis of yarkovsky models . this could be due to inaccuracies in our knowledge of physical properties , faulty astrometry , or modeling errors . ] in this section we examine several consequences of our results . first we discuss how the yarkovsky drifts can inform us about asteroid physical properties , spin states , and trajectories . then we discuss binary asteroid ( 1862 ) apollo and the curious case of asteroid ( 1036 ) ganymed . finally we discuss the possible mechanisms for non - yarkovsky driven rates , including association with meteoroid streams and rock comet phenomenon . because a clear connection exists between asteroid physical properties and yarkovsky drifts , we explored the constraints that can be placed on bulk density and surface thermal conductivity for seven objects with well - known diameters and ( excepting one case ) spin periods : ( 1620 ) geographos , ( 1862 ) apollo , ( 2100 ) ra - shalom , ( 2062 ) aten , ( 2340 ) hathor , ( 1566 ) icarus , and ( 3361 ) orpheus . we compared the measured yarkovsky rates to numerical estimates obtained with a range of physical parameters . for these estimates , we assumed a constant heat capacity @xmath124 j kg@xmath65 k@xmath65 ( table [ tab - thercon ] ) and a single value of the bulk density of the surface @xmath125 kg m@xmath20 , but we explore a wide range of bulk density and surface thermal conductivity values . because the obliquities are uncertain or ambiguous in many cases , we chose to illustrate outcomes for two obliquity values , typically 180@xmath59 and 135@xmath59 . our results are shown in figures [ fig : rhoka ] and [ fig : rhokb ] , which are similar to fig . 4 in @xcite . the shaded range consistent with the @xmath51 confidence limits on @xmath21 delineates the space of acceptable bulk densities and thermal conductivities , assuming that the yarkovsky effect is being modeled correctly . by acceptable , we mean consistent with observed @xmath21 values , even though some of the @xmath126 values may not be appropriate for asteroids . infrared observations indicate that ( 2100 ) ra - shalom has a thermal conductivity between 0.1 and 1 w m@xmath65 k@xmath65 @xcite . if we assume a minimum bulk density of 1,500 kg m@xmath20 , this conductivity value is consistent with the range suggested by our yarkovsky rate determination . if we make the same minimum density assumption for ( 1620 ) geographos , our measurements suggest that its surface thermal conductivity is greater than 0.002 w m@xmath65 k@xmath65 . for ( 1862 ) apollo , we show the range of physical properties that are consistent with both the optical - only fits and the radar+optical fits . the precision of the radar measurements dramatically shrinks the size of the measured error bars , with correspondingly tighter constraints on density and surface thermal conductivity . this example illustrates that reliable obliquity determinations will be important to extract physical properties from yarkovsky rate determinations . our measurement of ( 2062 ) aten s drift provides some useful insights . if we assume that its bulk density exceeds @xmath127 kg m@xmath20 , then its surface thermal conductivity @xmath69 must exceed 0.3 w m@xmath65 k@xmath65 . furthermore , if we assume that its bulk density exceeds @xmath128 kg m@xmath20 , the @xmath51 confidence region on the measured yarkovsky drift suggests that its obliquity is between @xmath129 . the yarkovsky simulations for ( 2340 ) hathor were computed with an assumed spin period of 4.5 hours . if the actual period is longer , the curves shown would shift to the left , and if the period is shorter , the curves would shift to the right . consequently , we can not make inferences about the @xmath69 value for this object until its spin period is measured . however , looking at the height of the curve , and with an assumption that the object s bulk density is greater than @xmath127 kg m@xmath20 , we can conclude that ( 2340 ) hathor likely has an obliquity lower than @xmath117 . the assumption of @xmath130 or @xmath117 obliquity for ( 1566 ) icarus restricts this object to low surface conductivity values and low bulk density values , or high surface conductivity values and high bulk density values . although these obliquities do produce physically plausible parameter combinations , it seems likely that the obliquity for this object is @xmath131 . the curves for ( 3361 ) orpheus were calculated with an assumed geometric albedo of 0.15 . as ( 3361 ) orpheus has a positive @xmath21 value , obliquities were assumed to be @xmath116 and @xmath132 . the curve representing an obliquity equal to @xmath116 for this object requires very low ( @xmath133 w m@xmath65 k@xmath65 ) or very high ( @xmath134 w m@xmath65 k@xmath65 ) surface thermal conductivity values for most densities . a more likely scenario is that this object has an obliquity @xmath135 , or perhaps even @xmath136 . an independent measurement of the obliquity could be used to validate obliquity constraints derived from yarkovsky measurements . values . blue ( top ) solid line corresponds to values consistent with best - fit @xmath21 and @xmath117 obliquity , pink ( lower ) solid line corresponds to values consistent with best - fit @xmath21 and @xmath130 obliquity . dashed regions surrounding each solid line encompass the @xmath51 confidence limits on the corresponding @xmath21 determinations . not all values displayed in this k-@xmath13 space are necessarily appropriate for asteroids . infrared observations suggest that ( 2100 ) ra - shalom has a thermal conductivity between 0.1 and 1 w m@xmath65 k@xmath65 @xcite , consistent with the range suggested by our yarkovsky rate determination . for apollo , we show results for both optical - only and radar+optical determinations . the inclusion of radar data greatly reduces the error bars on the measured drift , and therefore the area of the shaded curves . ] . range of bulk densities and thermal conductivities of three yarkovsky - dominated asteroids consistent with their observed @xmath21 values . for ( 2062 ) aten and ( 2340 ) hathor , blue ( top ) solid line corresponds to values consistent with best - fit @xmath21 and @xmath117 obliquity , pink ( lower ) solid line corresponds to values consistent with best - fit @xmath21 and @xmath130 obliquity . the constraints for ( 1566 ) icarus suggest that it may have a lower obliquity than those assumed . ( 3361 ) orpheus has a positive drift , so the blue ( top ) solid line corresponds to values consistent with best - fit @xmath21 and @xmath116 obliquity , pink ( lower ) solid line corresponds to values consistent with best - fit @xmath21 and @xmath132 obliquity . dashed regions surrounding each solid line encompass the @xmath51 confidence limits on the corresponding @xmath21 determinations . not all values displayed in this k-@xmath13 space are necessarily appropriate for asteroids . a period of 4.5 hours was assumed for ( 2340 ) hathor , and a 0.15 geometric albedo was assumed for ( 3361 ) orpheus . ] @xcite and @xcite examined the predominance of retrograde spins and negative yarkovsky drift rates and concluded that they were consistent with the presumed delivery method of neas from the main belt of asteroids . the @xmath137 and 3:1 resonance regions deliver neas to near - earth space @xcite . a main belt asteroid can arrive at the 3:1 resonance at 2.5 au via a positive ( if it originates in the inner main belt ) or negative ( if it originates in the outer main belt ) yarkovsky drift . however , a main belt asteroid can only arrive at the @xmath137 resonance ( at the inner edge of the main belt ) by way of a negative drift . according to @xcite and @xcite , @xmath138 of neas are transported via the @xmath137 resonance , with the rest from other resonances . the net result is a preference for retrograde spins . an observational consequence of this process would be an excess of retrograde rotators in the near - earth asteroid population . @xcite conducted a survey of 21 neas and found the ratio of retrograde / prograde rotators to be @xmath139 . we note that out of the 42 yarkovsky - dominated neas , 12 have a positive @xmath21 value . for this sample , our ratio of retrograde / prograde rotators is @xmath140 , similar to the value found by @xcite . the semi - major axis drifts described in this paper affect nea trajectory predictions . an order of magnitude estimate for the along track displacement due to a non - zero @xmath21 is given in @xcite : @xmath141 where @xmath142 is in units of km , @xmath143 is @xmath21 in @xmath144 au / myr , @xmath145 is the time difference between observations in tens of years , and @xmath146 is the semimajor axis of the object in au . for instance , the estimated along - track displacement due to the observed @xmath21 for ( 1862 ) apollo is 9 km after 10 years . similarly , the estimated along - track displacement for faster - moving ( 1864 ) daedalus is 67 km after 10 years . our data indicate that ( 101955 ) 1999 rq36 , the target of the osiris - rex mission , has a measurable yarkovsky drift of @xmath147 au / myr . although it has a relatively short arc ( 12 years ) it has been observed three times by radar , allowing for an accurate @xmath21 measurement . we estimated the along - track displacement of ( 101955 ) 1999 rq36 over the 6-month duration of the osiris - rex mission to be @xmath148 km , which will be easily detectable by a radio science instrument . \(1862 ) apollo is a binary asteroid @xcite . binary asteroids present a unique opportunity for the determination of physical parameters . if mass and density can be measured from the binary orbit and component sizes , the yarkovsky constraint on thermal conductivity can become much more meaningful . if the orientation of the plane of the mutual orbital can be measured , a plausible obliquity can be assumed , which makes the constraints on thermal properties tighter still . in some cases , actual obliquity measurements can be obtained from shape modeling efforts . @xcite identified a non - gravitational perturbation acting on the orbit of ( 1862 ) apollo , but was not able to determine a drift magnitude . to @xmath51 our observed @xmath21 value for ( 1862 ) apollo agrees with our yarkovsky predictions . \(1036 ) ganymed has by far the largest yarkovsky efficiency value ( @xmath149 ) among the objects presented in table [ tbl - results ] . with a nominal value of @xmath150 au / myr , the measured @xmath21 value is comparable to that of other neas . combined with ganymed s large diameter estimate ( @xmath151 km based on iraf measurements ) , this yarkovsky rate results in an unusually high @xmath1 value . how can this anomaly be explained ? one possibility is that some of the early astrometry , dating back to 1924 , is erroneous . this could be due to measurement errors , timing errors , bias errors , or reference frame conversion errors . we evaluated the semi - major axis drift with various subsets of the available astrometry and found values ranging between @xmath152 and @xmath153 au / myr . on that basis we modified the adopted uncertainties for this object , and our preferred value is @xmath154 au / myr . doing so does not eliminate the possibility of systematic bias in the astrometry , and we are still left with anomalously high @xmath1 values . another possibility is that the diameter of ganymed , an s - type asteroid , is much smaller than reported . this seems unlikely considering the more recent wise albedo measurement of @xmath155 @xcite which suggests a diameter of @xmath156 km . if ganymed s bulk density was especially low , a higher than usual @xmath1 value would be expected , but this would likely explain a factor of 2 or 3 at most , and would not explain the anomalous value . perhaps ganymed departs significantly from a spherical shape , with an effective diameter and mass that are much smaller than those implied by the diameter values reported in the literature . the relatively low lightcurve amplitudes do not seem to support such an argument , unless the asteroid is particularly oblate . in that case one could plausibly arrive at volume and mass estimates that are off by a factor of 5 - 10 . if we can rule out these possibilities ( i.e. ganymed is roughly spherical with no substantial concavities , its diameter estimate is reasonably accurate , and the early astrometry can be trusted ) , and if no other modeling error can be identified , then we would be compelled to accept an anomalously high yarkovsky efficiency for this object . in the course of our study we observed drift values that can not be accounted for easily by yarkovsky drift , because they considerably exceed the predicted yarkovsky rates . in most cases , these can be attributed to poor sensitivity to yarkovsky influences ( fig . [ fig - xmas ] ) . therefore , the high rates can generally be safely discarded . in other cases , the high rates may be due to erroneous optical astrometry or mismodeling of asteroid - asteroid perturbations . however we can not entirely rule out the possibility that some of the high drift rates are secure and will be confirmed by further observation and analysis . if the high rates can not be ascribed to poor yarkovsky sensitivity or faulty astrometry , one would need to invoke other non - gravitational forces . one possibility is that orbits are perturbed when neas are losing gas or dust in an anisotropic manner . to estimate a rough rate of mass loss that would be needed to account for the drifts measured , we used the basic thrust equation @xmath157 where @xmath158 is the force , @xmath159 is the rate at which the mass departs the asteroid , and @xmath160 is the ejection speed . for an asteroid of mass @xmath161 this yields @xmath162 which can be incorporated into gauss form of lagrange s planetary equations @xcite as an acceleration aligned with the velocity of the object . the dependence of the force on heliocentric distance @xmath163 is not known precisely ; we assumed @xmath164 , similar to the yarkovsky dependence , for simplicity , and because the amount of outgassing likely scales with the amount of incident radiation ( as in fig . 4 of @xcite ) . we assumed @xmath165 m s@xmath65 , the value derived by @xcite for 133p / elst - pizarro , and we assumed that the mass is departing in the optimal thrust direction . we quantified the mass loss rates needed to produce the observed drifts of neas with the highest yarkovsky efficiencies . we estimated a rate of 0.16 kg s@xmath65 for ( 154330 ) 2002 vx94 and 2.3 kg s@xmath65 for ( 7889 ) 1994 lx . although these estimates represent the minimum amount of mass loss necessary to account for the observed drifts ( if due to mass loss ) , they are smaller than typical levels from comets . comets have mass loss rates that span a wide range of values . on the high side a rate of @xmath166 kg s@xmath65 was estimated for hale - bopp @xcite . on the low side @xcite measured mass loss rates for three comets , averaged over their orbits : 2p / encke ( @xmath167 kg s@xmath65 ) , 22p / kopff ( @xmath168 kg s@xmath65 ) , and 65p / gunn ( @xmath169 kg s@xmath65 ) . mass loss rates of active asteroids have been estimated to be in the range from @xmath170 kg s@xmath65 ( 113p / elst - pizarro ) to @xmath171 kg s@xmath65 ( 107p / wilson - harrington ) @xcite . mass loss does not seem to be a viable mechanism to explain the semi - major axis drift rate of ( 1036 ) ganymed , as it would require a minimum mass loss rate of @xmath22,500 kg s@xmath65 . this would presumably have left detectable observational signatures , which have not been reported to date . we explore a couple of possibilities for mass loss mechanisms that could cause semi - major axis drifts . to our knowledge , ( 433 ) eros , ( 1566 ) icarus , ( 1620 ) geographos , ( 1685 ) toro , ( 1862 ) apollo , and 1982 ta are the only objects in our sample to have been associated with a meteoroid stream . @xcite found a weak correlation between the first five objects and various streams using the dissimilarity criterion " . however , this metric was later described as not convincing by @xcite , and current literature does not support such associations . in our results , apollo shows good agreement with yarkovsky predictions , with @xmath172 . the yarkovsky force is therefore a plausible cause of apollo s observed semi - major axis drift . the brightening of ( 3200 ) phaethon , the parent body of the geminid meteor shower , has been attributed to a `` rock comet '' phenomenon @xcite . with a perihelion at 0.14 au , ( 3200 ) phaethon s surface temperatures have been estimated by @xcite to be in the range @xmath173 k. the authors propose that these high surface temperatures could create thermal gradients in the body , resulting in thermal fracturing that would release dust . the resulting mass loss would affect the orbit . the combination of mass loss due to decomposing hydrated minerals and thermal fracturing led the authors to term ( 3200 ) phaethon a `` rock comet '' . a moderate amount ( @xmath2 1 kg s@xmath65 ) of mass lost in an anisotropic manner by `` rock comets '' could explain the observed semi - major axis drift rates . modeling of the yarkovsky effect is needed to improve trajectory predictions of near - earth asteroids and to refine our understanding of the dynamics of small bodies . using fits to astrometric data , we identified semi - major axis drifts in 54 neas , 42 of which show good agreement with numerical estimates of yarkovsky drifts , indicating that they are likely yarkovsky - dominated . these objects exhibit yarkovsky efficiencies of @xmath210@xmath3 , where the efficiency describes the ratio of the change in orbital energy to incident solar radiation energy . 12 objects in our sample have drifts that exceed nominal yarkovsky predictions and are labeled possibly yarkovsky - dominated . improvements in the knowledge of physical properties or in thermal modeling could bring these drift rates in better agreement with results from numerical models . however , if the high rates are confirmed by additional observations and analysis , they would be indicative of the presence of other non - gravitational forces , such as that resulting from a loss of mass . cn and jlm were partially funded by nsf planetary astronomy grants ast-0929830 and ast-1109772 . part of this research was conducted at the jet propulsion laboratory , california institute of technology , under a contract with the national aeronautics and space administration . dv was partially supported by the czech grant agency ( grant 205/08/0064 ) and research program msm0021620860 of the czech ministry of education . , j. d. , ostro , s. j. , benner , l. a. m. , chodas , p. w. , chesley , s. r. , hudson , r. s. , nolan , m. c. , klemola , a. r. , standish , e. m. , jurgens , r. f. , rose , r. , chamberlin , a. b. , yeomans , d. k. , & margot , j. l. 2002 , science , 296 , 132 , m. k. , clark , b. e. , nolan , m. c. , benner , l. a. m. , ostro , s. j. , giorgini , j. d. , vilas , f. , jarvis , k. , lederer , s. , lim , l. f. , mcconnochie , t. , bell , j. , margot , j. l. , rivkin , a. , magri , c. , scheeres , d. , & pravec , p. 2008 , icarus , 193 , 20 \(433 ) & eros & 1.46 & 0.22 & 10.83 & 16.84 & 5.270 & 0.25 & 1893 - 2012 & -0.3 & 0.2 & & & 1.81 & 70.56 & 0.38 + ( 152563 ) & 1992 bf & 0.91 & 0.27 & 7.25 & 0.42@xmath174 & & & 1992 - 2011 & -11.6 & 1.0 & & & 11.26 & 40.28 & 0.37 + ( 1862 ) & apollo & 1.47 & 0.56 & 6.35 & 1.50 & 3.065 & 0.25 & 1957 - 2012 & -1.8 & 0.6 & -2.3 & 0.2 & 11.50 & 36.11 & 0.23 + ( 1685 ) & toro & 1.37 & 0.44 & 9.38 & 3.40 & 10.1995 & 0.31 & 1948 - 2010 & -1.4 & 0.7 & & & 2.00 & 24.06 & 0.34 + ( 2062 ) & aten & 0.97 & 0.18 & 18.93 & 1.10 & 40.77 & 0.26 & 1955 - 2012 & -7.5 & 2.4 & & & 3.17 & 19.94 & 0.65 + ( 1620 ) & geographos & 1.25 & 0.34 & 13.34 & 2.56 & 5.22204 & 0.3258 & 1951 - 2012 & -2.4 & 0.7 & -2.5 & 0.6 & 3.85 & 18.15 & 0.48 + ( 2340 ) & hathor & 0.84 & 0.45 & 5.85 & 0.30 & & & 1976 - 2012 & -14.5 & 3.5 & & & 4.11 & 15.32 & 0.31 + ( 1580 ) & betulia & 2.20 & 0.49 & 52.11 & 5.80 & 6.1324 & 0.08 & 1950 - 2010 & -1.4 & 2.0 & -1.3 & 0.9 & 1.46 & 13.45 & 0.53 + ( 3361 ) & orpheus & 1.21 & 0.32 & 2.69 & 0.30 & 3.58 & & 1982 - 2009 & 5.7 & 2.5 & & & 2.25 & 13.04 & 0.13 + ( 1566 ) & icarus & 1.08 & 0.83 & 22.83 & 1.00 & 2.273 & 0.51 & 1949 - 2009 & -3.2 & 2.0 & & & 1.62 & 11.86 & 0.14 + ( 2063 ) & bacchus & 1.08 & 0.35 & 9.43 & 1.35@xmath174 & 14.90 & & 1977 - 2007 & -4.2 & 3.3 & & & 1.26 & 10.58 & 0.42 + ( 887 ) & alinda & 2.48 & 0.57 & 9.36 & 4.20 & 73.97 & 0.31 & 1918 - 2008 & 4.0 & 2.5 & & & 1.59 & 9.42 & 1.12 + ( 1865 ) & cerberus & 1.08 & 0.47 & 16.10 & 1.20 & 6.810 & 0.22 & 1971 - 2008 & -5.1 & 2.7 & & & 1.90 & 9.20 & 0.44 + ( 161989 ) & cacus & 1.12 & 0.21 & 26.06 & 1.90 & 3.7538 & 0.09 & 1978 - 2010 & 2.6 & 2.3 & & & 1.12 & 8.94 & 0.39 + ( 3757 ) & 1982 xb & 1.83 & 0.45 & 3.87 & 0.50 & 9.0046 & 0.18 & 1982 - 2008 & -13.7 & 6.8 & & & 2.04 & 8.82 & 0.49 + ( 6037 ) & 1988 eg & 1.27 & 0.50 & 3.50 & 0.65@xmath174 & 2.760 & & 1988 - 2007 & -14.2 & 4.3 & & & 3.34 & 8.51 & 0.64 + ( 2102 ) & tantalus & 1.29 & 0.30 & 64.01 & 2.04@xmath174 & 2.391 & & 1975 - 2008 & -3.8 & 3.6 & & & 1.08 & 8.31 & 0.60 + ( 2100 ) & ra - shalom & 0.83 & 0.44 & 15.76 & 2.30 & 19.797 & 0.13 & 1975 - 2009 & -4.8 & 2.2 & -5.4 & 1.5 & 3.67 & 8.30 & 0.90 + ( 207945 ) & 1991 jw & 1.04 & 0.12 & 8.71 & 0.52@xmath174 & & & 1955 - 2009 & -6.2 & 4.3 & & & 1.42 & 8.00 & 0.26 + ( 67399 ) & 2000 pj6 & 1.30 & 0.35 & 14.69 & 0.96@xmath174 & & & 1951 - 2009 & -19.4 & 7.2 & & & 2.71 & 7.34 & 1.40 + ( 1036 ) & ganymed & 2.66 & 0.53 & 26.70 & 31.66 & 10.31 & 0.2926 & 1924 - 2012 & -6.6 & 1.5 & & & 4.41 & 7.28 & 14.23 + ( 14402 ) & 1991 db & 1.72 & 0.40 & 11.42 & 0.60 & 2.266 & 0.14 & 1991 - 2009 & -5.0 & 4.3 & & & 1.19 & 7.05 & 0.22 + ( 3908 ) & nyx & 1.93 & 0.46 & 2.18 & 1.00 & 4.42601 & 0.23 & 1980 - 2009 & 9.8 & 3.2 & 12.9 & 2.7 & 4.71 & 5.52 & 0.92 + ( 4660 ) & nereus & 1.49 & 0.36 & 1.43 & 0.33 & 15.1 & 0.55 & 1981 - 2010 & 7.3 & 5.6 & 10.9 & 4.8 & 2.29 & 5.46 & 0.27 + ( 5660 ) & 1974 ma & 1.79 & 0.76 & 38.06 & 2.57@xmath174 & 17.5 & & 1974 - 2005 & -20.1 & 10.4 & & & 1.92 & 5.46 & 2.68 + ( 10302 ) & 1989 ml & 1.27 & 0.14 & 4.38 & 0.45@xmath174 & 19 . & & 1989 - 2006 & 35.3 & 7.1 & & & 4.96 & 5.33 & 1.26 + ( 2061 ) & anza & 2.26 & 0.54 & 3.77 & 2.60 & 11.50 & & 1960 - 2012 & 10.7 & 9.0 & & & 1.19 & 5.00 & 1.88 + ( 22753 ) & 1998 wt & 1.22 & 0.57 & 3.20 & 1.02@xmath174 & 10.24 & & 1955 - 2009 & -5.4 & 5.0 & -6.1 & 4.9 & 1.26 & 4.95 & 0.41 + ( 3753 ) & cruithne & 1.00 & 0.51 & 19.81 & 3.39@xmath174 & 27.4 & & 1973 - 2010 & -11.2 & 5.3 & & & 2.12 & 4.84 & 2.61 + ( 4769 ) & castalia & 1.06 & 0.48 & 8.89 & 1.40 & 4.095 & & 1989 - 2011 & -17.2 & 11.7 & & & 1.47 & 4.59 & 1.69 + ( 4947 ) & ninkasi & 1.37 & 0.17 & 15.65 & 0.65@xmath174 & & & 1978 - 2009 & 13.4 & 10.3 & & & 1.30 & 4.23 & 0.69 + ( 241596 ) & 1998 xm2 & 1.80 & 0.34 & 27.10 & 1.41@xmath174 & & & 1952 - 2011 & -14.4 & 10.7 & & & 1.35 & 4.23 & 1.53 + ( 4034 ) & vishnu & 1.06 & 0.44 & 11.17 & 0.42 & & 0.52 & 1986 - 2009 & -13.8 & 12.1 & & & 1.14 & 3.72 & 0.42 + ( 7336 ) & saunders & 2.31 & 0.48 & 7.17 & 0.65@xmath174 & 6.423 & & 1982 - 2010 & 10.3 & 8.3 & & & 1.25 & 3.50 & 0.47 + ( 2059 ) & baboquivari & 2.64 & 0.53 & 11.04 & 2.46@xmath174 & & & 1963 - 2009 & 17.7 & 12.8 & & & 1.38 & 3.42 & 2.96 + ( 15745 ) & 1991 pm5 & 1.72 & 0.25 & 14.42 & 0.98@xmath174 & & & 1982 - 2007 & -13.2 & 9.0 & & & 1.46 & 3.39 & 1.00 + ( 138911 ) & 2001 ae2 & 1.35 & 0.08 & 1.66 & 0.56@xmath174 & & & 1984 - 2012 & -22.9 & 11.2 & & & 2.04 & 3.38 & 1.02 + ( 18109 ) & 2000 ng11 & 1.88 & 0.37 & 0.81 & 1.12@xmath174 & 4.2534 & & 1951 - 2005 & 12.0 & 9.6 & & & 1.25 & 3.21 & 1.00 + ( 2202 ) & pele & 2.29 & 0.51 & 8.74 & 1.07@xmath174 & & & 1972 - 2008 & 29.5 & 21.2 & & & 1.39 & 2.98 & 2.18 + ( 68950 ) & 2002 qf15 & 1.06 & 0.34 & 25.16 & 2.03@xmath174 & 29 . & & 1955 - 2008 & -11.6 & 6.5 & & & 1.80 & 2.96 & 1.78 + ( 4197 ) & 1982 ta & 2.30 & 0.77 & 12.57 & 1.80 & 3.5380 & 0.37 & 1954 - 2010 & 30.9 & 9.2 & & & 3.36 & 2.88 & 2.84 + ( 3362 ) & khufu & 0.99 & 0.47 & 9.92 & 0.70 & & 0.21 & 1984 - 2004 & -20.4 & 13.2 & & & 1.54 & 2.87 & 1.01 + ( 99935 ) & 2002 av4 & 1.65 & 0.64 & 12.76 & 2.46@xmath174 & & & 1955 - 2011 & -9.8 & 8.0 & & & 1.23 & 2.73 & 1.48 + ( 68350 ) & 2001 mk3 & 1.67 & 0.25 & 29.56 & 2.43@xmath174 & 3.24 & & 1955 - 2007 & -14.4 & 11.9 & & & 1.21 & 2.61 & 2.73 + ( 54690 ) & 2001 eb & 1.63 & 0.26 & 35.36 & 1.18@xmath174 & & & 1952 - 2009 & -14.4 & 13.3 & & & 1.08 & 2.56 & 1.31 + ( 4179 ) & toutatis & 2.53 & 0.63 & 0.45 & 5.40 & 176 . & & 1976 - 2011 & -18.4 & 4.3 & -5.0 & 0.6 & 8.33 & 2.44 & 1.68 + ( 1864 ) & daedalus & 1.46 & 0.61 & 22.20 & 3.70 & 8.572 & & 1971 - 2006 & -16.9 & 8.2 & & & 2.06 & 2.44 & 3.97 + ( 154330 ) & 2002 vx94 & 1.48 & 0.41 & 7.16 & 0.90@xmath174 & & & 1986 - 2011 & 70.2 & 20.6 & & & 3.42 & 2.41 & 4.64 + ( 7753 ) & 1988 xb & 1.47 & 0.48 & 3.12 & 0.68@xmath174 & & & 1988 - 2012 & 18.9 & 12.9 & & & 1.46 & 2.39 & 0.90 + ( 10563 ) & izhdubar & 1.01 & 0.27 & 63.46 & 1.48@xmath174 & & & 1991 - 2010 & 32.3 & 14.5 & & & 2.23 & 2.19 & 3.70 + ( 13651 ) & 1997 br & 1.34 & 0.31 & 17.25 & 1.07@xmath174 & 33.644 & & 1980 - 2011 & -12.4 & 11.8 & & & 1.06 & 2.18 & 1.02 + ( 12923 ) & zephyr & 1.96 & 0.49 & 5.29 & 2.14@xmath174 & 3.891 & & 1955 - 2012 & -26.5 & 12.1 & & & 2.19 & 2.05 & 3.97 + ( 3554 ) & amun & 0.97 & 0.28 & 23.36 & 2.48 & 2.530 & 0.1284 & 1986 - 2012 & 9.0 & 8.8 & & & 1.03 & 2.05 & 1.73 + ( 88254 ) & 2001 fm129 & 1.18 & 0.63 & 1.52 & 1.19@xmath174 & & & 1978 - 2008 & -40.6 & 18.9 & & & 2.15 & 2.05 & 3.01 +

we introduce a new family of fountain codes that are systematic and also have sparse parities . given an input of symbols , our codes produce an unbounded number of output symbols , generating each parity independently by linearly combining a logarithmic number of randomly selected input symbols . the construction guarantees that for any accessing a random subset of encoded symbols , asymptotically suffices to recover the input symbols with high probability . our codes have the additional benefit of logarithmic locality : a single lost symbol can be repaired by accessing a subset of of the remaining encoded symbols . this is a desired property for distributed storage systems where symbols are spread over a network of storage nodes . beyond recovery upon loss , local reconstruction provides an efficient alternative for reading symbols that can not be accessed directly . in our code , a logarithmic number of disjoint local groups is associated with each systematic symbol , allowing _ multiple parallel reads_. our main mathematical contribution involves analyzing the rank of sparse random matrices with specific structure over finite fields . we rely on establishing that a new family of sparse random bipartite graphs have perfect matchings with high probability . systematic fountain code , logarithmic locality , availability .

1401.0734

fountain codes @xcite form a new family of linear erasure codes with several attractive properties . for a given set of @xmath0 input symbols , a fountain code produces a potentially limitless stream of output symbols , each created independently of others as a random combination of input symbols according to a given distribution . ideally , given a randomly selected subset of @xmath4 encoded symbols , a decoder should be able to recover the original @xmath0 input symbols with high probability ( w.h.p . ) for some small overhead @xmath5 . further , fountain codes typically emphasize on efficient encoding and decoding algorithms . in this work , we design a new family of fountain codes that combine multiple properties appealing to distributed storage . one property that is highly desired for distributed storage codes is _ systematic form _ : the original information symbols must appear in the encoded sequence . their presence enables the reading of source data without decoding and is a practical requirement for most storage applications . another important property of distributed storage codes is efficient _ repair _ @xcite : when a single encoded symbol is lost it should be possible to reconstruct it without communicating too much information from other encoded symbols . a related property is that of _ locality _ of each symbol : the number of encoded symbols that need to be accessed to reconstruct that particular symbol @xcite . a key observation is that in a systematic linear code , locality is strongly connected to the sparsity of parity symbols @xcite , _ i.e. _ , the maximum number of input symbols combined in a parity symbol . a parity symbol along with the systematic symbols covered by it form a _ local group_. any symbol in this group can be reconstructed via an appropriate linear combination of the remaining member symbols . the smaller the size of the local group , the lower the locality of the symbols in it . in existing fountain codes , such as lt or raptor codes , each encoded symbol is a linear combination of @xmath6 or even constant number of input symbols on average . however , these codes are not systematic and the low degree of the encoded symbols does not imply low locality . certain classes , such as raptor codes , can be transformed into a systematic form @xcite via a preprocessing of the input . unfortunately , due to the additional step , parity symbols are no longer sparse in the original input symbols . standard fountain codes support computationally efficient encoding and decoding algorithms as a result of a meticulously designed encoded symbol degree distribution . fast decoding algorithms , however , do not translate to efficient repair : these algorithms aim at retrieving the entire source message from a set of available symbols , and are not tailored to the needs of a single symbol reconstruction . on the contrary , single erasures the most frequent scenario in a distributed storage setting can be efficiently repaired by an erasure code featuring low locality . if additionally the code is in systematic form , full scale decoding is invoked only in the unlikely event of multiple erasures that can not be locally repaired . in other words , the presence of the source data in the encoded sequence in conjunction with low locality renders decoding an infrequently used operation , downgrading the need for efficient decoding algorithms . the significance of locality is not limited to the repair problem , _ i.e. _ , the reconstruction of a symbol upon its loss . it extends to the closely related use case of _ degraded reads_. in a distributed setup each symbol is stored on a different storage node , which may be temporarily unavailable due to a variety of reasons . although not permanently lost , a systematic symbol may not be directly accessible , and its local group provides an efficient alternative for reading it . the _ availability _ of a systematic symbol naturally extends the notion of locality , measuring the number of disjoint local groups the symbol belongs to . we define the _ availability _ of a systematic symbol as the number of disjoint sets of encoded symbols which can be used to reconstruct that particular symbol . in effect , it characterizes the number of read requests for a particular systematic symbol that can be simultaneously served . * our contribution : * we introduce a new family of fountain codes that are systematic and also have parity symbols with logarithmic sparsity . we show that this is impossible if we require the code to be mds , but is possible if we require a near - mds property similar to the probabilistic guarantees provided by lt and raptor codes . more concretely , for any @xmath7 we construct codes that guarantee that a random subset of @xmath2 symbols suffices to recover the original @xmath0 input symbols w.h.p . our codes produce an unbounded number of output symbols , creating each parity independently by linearly combining a logarithmic number of randomly chosen input symbols . we show that this structure also provides logarithmic locality : each symbol in our codes is repairable by accessing only @xmath3 other coded symbols . we further define the notion of symbol availability and show that the systematic symbols in our codes feature logarithmic availability : with high probability , for each systematic symbol there exist @xmath6 disjoint sets of symbols that can be used to reconstruct it . this means that multiple parallel jobs can read this symbol concurrently , each by accessing one disjoint set . this new property is motivated by the straggler performance bottlenecks observed recently in distributed storage systems @xcite . one disadvantage of our construction is higher decoding complexity . our codes can be decoded by solving a system of linear equations over @xmath8 , which corresponds to maximum likelihood decoding for the erasure channel . this naive decoding can be implemented using gaussian elimination and requires @xmath9 steps . fortunately , the matrices we construct are sparse , allowing faster decoding : wiedemann s algorithm @xcite can be used to decode in @xmath10 time . standard fountain codes create linear equations that can be solved just by back - substitution which amounts to decoding complexities of @xmath11 for lt @xcite and @xmath12 for raptor @xcite but offer no locality . it remains open to construct fountain codes that have locality and near - linear decoding complexity . our main technical contribution is a novel random matrix result : we show that a family of random matrices with non independent entries have full rank with high probability . the analysis builds on the connections of matrix determinants to flows on random bipartite graphs , using techniques from @xcite . our key result is showing that a new family of sparse random graphs have matchings w.h.p . our random graph contribution is explained in section [ sec : graph - perspective ] . given @xmath0 input symbols , elements of a finite field @xmath8 , we want to encode them into @xmath13 symbols using a linear code . linear codes are described by a @xmath14 _ generator _ matrix @xmath15 over @xmath8 , which when multiplied by an input vector @xmath16 produces a codeword @xmath17 . ideally , we would like @xmath15 to have the following properties : * _ systematic form _ , _ i.e. _ , a subset of the columns of @xmath15 forms the identity matrix , @xmath18 , which implies that the input symbols are reproduced in the encoded sequence . * _ rateless property _ , _ i.e. _ , each column is created independently . the number @xmath13 of columns does not have to be specified for the encoder a priori . equivalently , encoded symbols can be created or removed dynamically upon request , without recreating the entire encoded sequence . * _ mds property _ , _ i.e. _ , any @xmath0 columns of @xmath15 have rank @xmath0 , implying that any subset of @xmath0 encoded symbols suffices to retrieve the input . * _ low locality_. @xmath15 has _ locality _ @xmath19 if each column can be written as a linear combination of at most @xmath19 other columns . if the code is systematic , then sparse parities suffice to obtain good locality @xcite . * _ high availability_. a systematic symbol has _ @xmath20 if it can be written as a linear combination of @xmath20 disjoint sets of symbols , of cardinality @xmath19 . the code has availability @xmath21 , equal to that of the least available systematic symbol . for any code , any sufficiently large subset of encoded symbols should allow recovery of the original data . the size of such a set is tightly related to the reliability of the code . for optimal reliability , _ i.e. _ , in the case of mds codes , an information theoretically minimum subset of @xmath0 encoded symbols suffices to decode . when equipped with systematic form , the generator matrix of an mds code affords no zero coefficient in the parity generating columns . to verify that , consider a parity column with a zero coefficient in the @xmath22-th position : that parity column along with any @xmath23 systematic columns excluding the one corresponding to the @xmath22-th systematic symbol form a singular matrix . if parities are deliberately sparse in the input symbols , seeking to improve the code s _ locality _ , the property that any @xmath0 encoded symbols suffice to retrieve the original data has to be relaxed . in this work , we require that for @xmath1 , a set of @xmath24 randomly selected encoded symbols suffice to decode with high probability ; the decoder may fail , but with a probability vanishing polynomially in @xmath0 . we refer to a code with this property as _ near - mds_. under this constraint , we seek codes that achieve optimal locality , which translates into determining how sparse the parities can be without violating the decoding guarantee . we will show that it is impossible to recover the original message with high probability if the parities are linear combinations of fewer than @xmath25 input symbols . furthermore , we will design codes that achieve logarithmic sparsity in the parities and hence , order optimal locality . we conclude the paper investigating the availability of our construction , showing that with high probability every systematic symbol belongs to a logarithmic number of disjoint local groups . in lt codes , the first practical realizations of fountain codes invented by luby @xcite , the average degree of the output symbols , _ i.e. _ , the number of input symbols combined into an output symbol , is @xmath26 . note , however , that sparsity in this case does not imply good locality , since lt codes lack systematic form . building on lt , shokrollahi in @xcite introduced raptor codes , a different class of fountain codes . the core idea is to precode the input symbols prior to the application of an appropriate lt code . by virtue of the two layer encoding , the per symbol encoding cost which corresponds to average degree of encoded symbols is reduced to a constant , while the @xmath0 input symbols can be retrieved in linear time by a set of @xmath2 encoded symbols , with probability of failure at most inversely polynomial in @xmath0 . however , the original raptor design does not feature the highly desirable systematic form . further , similar to the lt codes , the constant average degree of the encoded symbols does not imply good locality . in the same work @xcite , shokrollahi provided a construction that yields a systematic flavor of raptor codes . the raptor encoding is not applied directly on the input symbols , rather on the output of a preprocessing step of complexity @xmath27 . the source symbols appear in the encoded stream , but due to the preprocessing step the parity symbols are no longer sparse in the original input symbols , despite their constant average degree . gummadi in his thesis @xcite was the first to consider the design of fountain codes explicitly oriented for storage applications , _ i.e. _ , codes that feature systematic form and efficient repair . the latter is quantified by _ repair complexity _ : the average number of symbol operations performed to repair a set of erased symbols . gummadi proposes systematic variants of lt and raptor codes that feature low ( even constant ) expected repair complexity . however , the overhead @xmath5 required for decoding is suboptimal : it can not be made arbitrarily small . our main result is the analysis of the rank of a new family of sparse random matrices over @xmath8 . in particular , we investigate the probability that a @xmath28 matrix comprising any number @xmath29 of systematic columns and @xmath30 random @xmath3-sparse columns has full rank . there is a long line of work on the distribution of the rank of sparse random matrices over a finite field ( _ e.g. _ , work by karp @xcite , kovalenko @xcite , balakin @xcite , cooper @xcite and references therein ) . that line of work , however , typically focuses on random matrices whose entries are independently distributed . in our case , systematic columns carry exactly one nonzero entry , while the number of nonzero entries in the remaining columns is strictly upper bounded , rendering column entries dependent . we introduce a new family of fountain codes that are systematic and also have sparse parities . each parity symbol is a random linear combination of up to @xmath31 randomly chosen input symbols . due to their randomized nature , our codes provide a probabilistic guarantee on successful decoding . in particular , we require that a set of @xmath32 randomly selected encoded symbols , for arbitrarily small @xmath7 can be decoded successfully with high probability , _ i.e. _ , with probability of failure vanishing like @xmath33 . we show that under this constraint , @xmath31 must be of at least logarithmic order in @xmath0 , _ i.e. _ , @xmath34 . surprisingly , however , a logarithmic sparsity level for the parity symbols is also achievable , hence @xmath35 . the sparsity of the parity columns corresponds to the _ locality _ of the code family . our main result , which is asymptotic in @xmath0 , is established in theorem [ thm : main - theorem ] , at the end of this section . we conclude the section with a study of the _ availability _ of our construction . given a vector @xmath36 of @xmath0 input symbols in @xmath8 , the code is a linear mapping of @xmath36 to a vector @xmath37 of higher dimension @xmath13 through a @xmath14 matrix @xmath15 . the encoded sequence comprises an un - encoded copy of the @xmath0 input symbols augmented by parity symbols , hence the _ systematic _ form . without loss of generality , we may assume that @xmath36 lies in the first indices of @xmath37 followed by the parity symbols . a single parity symbol is constructed in a two step process . first , @xmath31 input symbols are successively selected uniformly at random , independently , with replacement . then , a coefficient is uniformly drawn from @xmath8 for each symbol previously selected . the parity is the linear combination of the symbols selected in the first step , weighted with the coefficients drawn in the second step . the same procedure is repeated independently for subsequent parity symbols . the independent construction of parities is the hallmark of a fountain code . ( u1 ) at ( 0,0 ) @xmath38 ; ( u2 ) [ below of = u1 ] @xmath39 ; ( u3 ) [ below of = u2 ] @xmath40 ; ( u4 ) [ below of = u3 ] @xmath39 ; ( u5 ) [ below of = u4 ] @xmath41 ; ( v1 ) at ( 3 , 1 ) @xmath42 ; ( v2 ) [ below of = v1 ] @xmath39 ; ( v3 ) [ below of = v2 ] @xmath43 ; ( v4 ) [ below of = v3 ] @xmath39 ; ( v5 ) [ below of = v4 ] @xmath44 ; ( v6 ) [ below of = v5 ] @xmath39 ; ( v7 ) [ below of = v6 ] @xmath45 ; ( v1 ) edge [ bend right , out=-20 , in=-160 ] ( u1 ) ; ( v3 ) edge [ bend right , out=-20 , in=-160 ] ( u3 ) ; ( v5 ) edge [ out=130 , in=0 ] node[above , near start ] @xmath46 ( u3 ) ; ( v5 ) edge [ out=140 , in=22.5 ] ( 1.5 , -2.5 ) ; ( v5 ) edge [ out=150 , in=45 ] ( 1.5 , -3 ) ; ( v7 ) edge [ out=160 , in=-45 ] ( 1.5 , -4 ) ; ( v7 ) edge [ out=210 , in=315 , looseness=1 ] ( 1.5 , -5 ) ; at ( 1.5 , -4.5 ) @xmath39 ; at ( 1.5 , -4.5 ) @xmath47 ; it is useful to describe our randomized construction through a correspondence to a bipartite graph @xmath48 , depicted in figure [ fig : bipartite_near_mds ] . the set @xmath49 of vertices on the left side corresponds to the @xmath0 input symbols , and the set @xmath50 on the right corresponds to the @xmath13 symbols of the encoded sequence . an edge @xmath51 if the input symbol @xmath52 is one of the symbols participating in the formation of the encoded symbol @xmath53 . each of the @xmath0 first vertices in @xmath50 has degree equal to one and is connected to a distinct input symbol . these @xmath0 vertices correspond to the deterministically constructed systematic part of the encoded sequence . each one of the remaining vertices corresponds to a parity symbol and forms its neighborhood through the following randomized procedure . node @xmath44 throws an edge to a vertex in @xmath49 selected uniformly at random . this step is repeated a total number of @xmath31 times , independently . at the end of this process , vertex @xmath44 has selected a subset @xmath54 of vertices in @xmath49 , the neighborhood of @xmath44 . the parity symbol corresponding to vertex @xmath44 is a random linear combination of the input symbols in its neighborhood . slightly abusing notation , @xmath44 is used to denote both the vertex and the corresponding entry in the encoded sequence @xmath37 . the @xmath55-th encoded symbol can be written as @xmath56 where @xmath46 s are randomly selected uniformly and independently from @xmath8 . the coefficients @xmath46 s can be embedded in the graph representation as weights on the corresponding edges . the edges of the systematic part have unit weights . the degree of @xmath53 , _ i.e. _ , the size of its neighborhood @xmath57 can be at most equal to @xmath31 , the number of edges thrown . it can be strictly smaller if a vertex in @xmath49 is selected multiple times . however , when @xmath31 is much smaller than @xmath0 , @xmath57 will be equal to @xmath31 with high probability , and a parity symbol will be a linear combination of @xmath31 input symbols . to emphasize that @xmath31 is allowed to grow as a function of @xmath0 , we will hereafter denote it by @xmath47 . returning to the matrix representation , the code construction corresponds to a family of generator matrices @xmath15 of the form @xmath58 $ ] . every encoded symbol corresponds to a column of @xmath15 . the identity part confers the systematic form . @xmath59 , the part responsible for the construction of the parity symbols , is a random matrix whose columns are sparse , each bearing at most @xmath47 nonzero entries . any @xmath0 encoded symbols corresponding to linearly independent columns of @xmath15 suffice to retrieve the input @xmath36 . conversely , reconstructing @xmath36 from a randomly chosen set of @xmath60 encoded symbols is possible only if @xmath0 symbols among them correspond to linearly independent columns . therefore , the key property required for successful decoding of a set @xmath61 of @xmath62 randomly selected encoded symbols is that @xmath63 , the @xmath64 matrix formed by the corresponding columns of @xmath15 , including any combination of systematic and parity parts , has full rank w.h.p . the probability that the input can be recovered from @xmath62 randomly selected encoded symbols increases with @xmath47 . equivalently , fewer encoded symbols suffice to attain a certain probability of successful decoding . we have highlighted the extreme case of systematic mds codes : for the optimal guarantee that any @xmath0 symbols suffice to recover the input , @xmath47 can be no less than @xmath0 . to gain further insight , note that for a set of @xmath62 encoded symbols to be successfully decoded , it is necessary that all input symbols are covered by that set . as @xmath47 decreases , so does the probability that a particular symbol is covered by the parities in a set of @xmath62 encoded symbols , impacting the decoding guarantees . on the other hand , as noted section [ sec : problem_description ] , a systematic code with sparse parities has good locality . the relation between @xmath47 and locality is straightforwardly quantifiable : any parity symbol @xmath65 is a linear combination of at most @xmath47 systematic symbols . also rearranging the terms , any systematic symbol @xmath66 covered by a parity @xmath65 can be written as a linear combination of @xmath65 and the remaining systematic symbols covered by @xmath65 . under the assumption that there exist at least one parity symbol covering every systematic symbol @xmath66 , the code has locality at most @xmath47 . in summary , decreasing @xmath47 improves the locality of the code , with a toll on the probability of successful decoding of a random set of @xmath67 encoded symbols , where @xmath5 is a positive constant denoting the decoding overhead . our primary contribution , portrayed in theorem [ thm : main - theorem ] , is identifying how _ small _ @xmath47 can be to ensure that a randomly selected set of @xmath67 symbols is decodable , or equivalently that a @xmath68 submatrix @xmath63 of @xmath15 is full rank , with high probability . [ thm : main - theorem ] consider a matrix @xmath69 $ ] , where each column of @xmath59 is independently constructed as follows : @xmath70 @xmath71 out of the @xmath0 entries are selected uniformly at random with replacement , and @xmath72 a value drawn uniformly at random over @xmath8 is independently assigned to each entry selected in step @xmath70 . then , for constant @xmath73 , and @xmath74 , a randomly selected @xmath75 submatrix @xmath63 of @xmath15 containing any number of systematic columns is full rank with probability at least @xmath76 . theorem [ thm : main - theorem : converse ] establishes a converse result stating that the _ sufficient _ value of @xmath47 of theorem [ thm : main - theorem ] is order - optimal for our construction . [ thm : main - theorem : converse ] ( converse ) if each column of @xmath59 is generated independently as described with at most @xmath47 nonzero entries , then @xmath77 is necessary for a random @xmath78 submatrix @xmath63 of @xmath79 $ ] to be full rank w.h.p . from the two theorems , it follows that our codes achieve optimal locality with a logarithmic degree for every parity symbol . original data is reconstructed in @xmath9 using maximum likelihood ( ml ) decoding , which corresponds to solving a linear system of @xmath62 equations over @xmath8 . note , however , that the wiedemann algorithm @xcite can reduce complexity to @xmath80 on average , exploiting the sparsity of the linear equations , with negligible extra memory requirement . finally , we note a drawback of our analysis : in order to achieve vanishingly small probability of failure as @xmath0 grows , the size of the field must grow accordingly . it suffices , however , that the number of bits per symbol grows logarithmically in @xmath0 : a symbol size of @xmath81 bits , @xmath82 , implies that @xmath83 . ( u1 ) at ( 0,0 ) @xmath38 ; ( v1 ) at ( 4 , 3 ) @xmath42 ; ( v2 ) [ below of = v1 ] @xmath84 ; ( v3 ) [ below of = v2 ] @xmath85 ; ( v4 ) [ below of = v3 ] @xmath39 ; ( v5 ) [ below of = v4 ] @xmath86 ; ( v1 ) edge [ looseness=.2 , out=180 , in=45 ] ( u1 ) ; ( v2 ) edge [ looseness=.2 , out=180 , in=22.5 ] ( u1 ) ; ( v3 ) edge [ looseness=.2 , out=180 , in=0 ] ( u1 ) ; ( v5 ) edge [ looseness=.2 , out=180 , in=-45 ] ( u1 ) ; ( u2 ) at ( 8,5.15 ) @xmath87 ; ( u3 ) [ below of = u2 ] @xmath88 ; ( u4 ) [ below of = u3 ] @xmath89 ; ( u5 ) [ below of = u4 ] @xmath90 ; ( u6 ) [ below of = u5 ] @xmath91 ; ( u7 ) [ below of = u6 ] @xmath39 ; ( u8 ) [ below of = u7 ] ; ( u9 ) [ below of = u8 ] @xmath41 ; ( footprintv1 ) ; ( footprintv2 ) ; ( footprintv3 ) ; ( footprintvm ) ; ( v1 ) edge ( @xmath92 ) ; ( v1 ) edge ( @xmath93 ) ; ( v2 ) edge ( @xmath94 ) ; ( v2 ) edge ( @xmath95 ) ; ( v3 ) edge ( @xmath96 ) ; ( v3 ) edge ( @xmath97 ) ; ( v5 ) edge ( @xmath98 ) ; ( v5 ) edge ( @xmath99 ) ; ( @xmath100 ) ( @xmath101 ) ; at ( @xmath102 ) @xmath103 : the _ footprint _ of @xmath104 . ; ( disjointfooprint ) at ( @xmath105 ) ; ( disjointfooprint ) edge ( @xmath106 ) ; ( disjointfooprint ) edge ( @xmath107 ) ; at ( @xmath108 ) @xmath109 and @xmath110 are disjoint . @xmath85 and @xmath86 are _ isolated_. ; ( availabilityset ) at ( @xmath111 ) ; ( v1 ) edge ( availabilityset ) ; ( v3 ) edge ( availabilityset ) ; ( v5 ) edge ( availabilityset ) ; at ( @xmath112 ) @xmath113 and @xmath86 are pairwise _ isolated_. the _ availability _ of @xmath38 is at least @xmath114 . ; thus far , we have seen that our randomized construction achieves logarithmic locality : every encoded symbol belongs to at least one _ local group _ of cardinality @xmath115 . there is a one - to - one correspondence between the local groups @xmath66 belongs to and the parity symbols that cover @xmath66 . let @xmath116 be the subset of parities that cover @xmath66 . its cardinality is a binomial random variable since every generated parity independently covers @xmath66 with some probability . if the total number of parities generated is @xmath117 for some constant @xmath118 , _ i.e. _ , proportional to the length of the input , then for @xmath119 of theorem [ thm : main - theorem ] every systematic symbol is in fact covered by a logarithmic number of parities w.h.p . [ thm : ub - prob - exist - u - few - parities ] let @xmath117 be the total number of parities generated , for a constant @xmath120 , with each parity symbol constructed as a linear combination of @xmath121 independently selected symbols uniformly at random with replacement . the expected number of parities covering a systematic symbol @xmath66 is @xmath122 \le rc\log(k ) . \end{aligned}\ ] ] further , @xmath123 \right ) \nonumber\\ & \le \frac{1}{k^{\frac{rc\epsilon^2}{2}-1 } } \exp\left ( \frac{\epsilon^2}{2 } \frac{rc\log^2(k)}{k } \right ) . \label{eq : ub - prob - exist - u - few - parities } \end{aligned}\ ] ] for any @xmath5 , an appropriate choice of @xmath124 , and @xmath125 achieves a vanishing bound in . the above result states that with high probability all input symbols are covered by at least @xmath126 symbols for some @xmath1 . in the following , we will omit the constant @xmath127 for simplicity , and assume that every systematic symbol is covered by at least @xmath128 . the _ availability _ of the input symbol @xmath66 is the cardinality of the largest subset of _ local groups _ containing @xmath66 whose only common element is @xmath66 . more formally , consider a parity @xmath43 that covers the systematic symbol @xmath66 , _ i.e. _ , @xmath129 . then @xmath43 is a linear combination of the symbols in @xmath130 , where @xmath131 contains the remaining symbols covered by @xmath43 . the set @xmath131 is referred to as the _ footprint _ of @xmath43 with respect to @xmath66 . note that the _ footprint _ of a parity symbol @xmath43 is a concept relative to the systematic symbol @xmath66 under consideration . two parities @xmath132 are _ isolated _ if their footprints are disjoint . an example is depicted in figure [ fig : footprints ] . the cardinality of the largest subset of @xmath133 such that parities are pairwise isolated corresponds to the availability of the symbol @xmath66 . under the assumption that every systematic symbol is covered by at least a logarithmic number of parities , theorem [ thm : availability - of - the - code ] states that every systematic symbol has a logarithmic availability with high probability . [ thm : availability - of - the - code ] assuming that every input symbol @xmath66 is covered by at least @xmath134 parity symbols created independently as described in section [ sec : our - construction ] , for sufficiently large @xmath0 , @xmath135 and @xmath136 , @xmath137 the randomized construction of our erasure code is naturally mapped to a family of random bipartite graphs @xmath138 depicted in figure [ fig : bipartite_near_mds ] . the correspondence , established early in section [ sec : our - construction ] , lays the foundation for all subsequent analysis , but also provides an alternative viewpoint for our results as purely structural properties of the random graphs , setting the coding background aside . under the graph perspective , theorem [ thm : main - theorem ] states that a randomly chosen subgraph of @xmath139 has a perfect matching . first , consider a balanced random bipartite graph where @xmath140 and each vertex of @xmath50 is randomly connected to @xmath141 nodes in @xmath49 . a classical result by erds and renyi @xcite shows that these graphs will have perfect matchings with high probability . however , the graphs we consider are unbalanced , with @xmath142 and @xmath143 vertices , for @xmath144 , like the one depicted in figure [ fig : bipartite - graph - random - unbalanced - subgraph ] . out of the @xmath62 vertices in @xmath50 , @xmath145 vertices are special with degree @xmath146 , corresponding to systematic symbols , and @xmath147 are connected to @xmath141 vertices in @xmath49 . in that sense , if we set @xmath148 and @xmath149 we recover the classical result of @xcite . our additional analysis is required because our proof needs to hold for all values of @xmath145 ranging from @xmath150 up to @xmath23 . this corresponds to @xmath145 vertices in @xmath49 being trivially matched with those vertices in @xmath50 that have degree @xmath146 , and the remaining @xmath151 vertices being matched via the random edges . more formally , let @xmath152 be a subgraph of @xmath139 , where @xmath153 is a subset of @xmath62 vertices , and @xmath154 is the subset of edges incident to @xmath155 . theorem [ thm : main - theorem ] states that @xmath156 has a perfect matching , which for the unbalanced bipartite graph is a matching that saturates all @xmath0 vertices in @xmath49 . in fact , this observation is a key component in the proof of theorem [ thm : main - theorem ] . the transition from the perfect matching of a subgraph to the rank of a submatrix which finalizes the proof requires only that the random coefficients are drawn from a large enough field . along the same lines , theorem [ thm : main - theorem : converse ] states that randomly throwing @xmath157 edges on the parity symbols are necessary to guarantee that a vertex in @xmath49 is connected in @xmath156 with high probability . the vertices in @xmath50 , the right hand side of @xmath139 , have by construction degree either equal to one or approximately equal to @xmath158 for some @xmath159 . theorem [ thm : ub - prob - exist - u - few - parities ] states that when @xmath160 increases linearly in @xmath161 , the degree of the vertices in @xmath49 is concentrated around its expectation , which is proportional to @xmath47 . ( v1 ) at ( 0 , 0 ) @xmath42 ; ( v2 ) at ( 2 , -1 ) @xmath84 ; ( v3 ) at ( 3 , 1.5 ) @xmath85 ; ( extra1 ) at ( 5 , -0.5 ) @xmath162 ; ( v5 ) at ( 3.3 , -2.5 ) @xmath86 ; ( extra2 ) at ( 0.5 , -2 ) @xmath162 ; ( v1 ) ( v2 ) ; ( v3 ) ( v2 ) ; ( v1 ) ( extra2 ) ; ( v3 ) ( extra1 ) ; plot [ smooth cycle ] coordinates ( @xmath163 ) ( @xmath164 ) ( @xmath165 ) ( @xmath166 ) ( @xmath167 ) ( @xmath168 ) ( @xmath169 ) ( @xmath170 ) ; finally , the availability of a symbol @xmath66 can be mapped to the independence number of a random graph @xmath171 . @xmath133 denotes the set of parities covering @xmath66 , or equivalently the vertices of @xmath50 in the neighborhood of @xmath66 . the set of edges @xmath172 is constructed as follows : for @xmath132 , @xmath173 if and only if the footprints of @xmath43 and @xmath44 overlap . the availability of @xmath66 is the cardinality of the maximum independent set in @xmath174 , which is shown to be at least a constant fraction of @xmath175 , and hence logarithmic in @xmath0 , with high probability . theorem [ thm : availability - of - the - code ] states that this property holds simultaneously for all @xmath176 . the probability is estimated over multiple @xmath177 randomly generated code instances with @xmath0 input symbols and rate @xmath178 , and @xmath177 independent trials per instance . the degree of parity symbols is equal to @xmath179 , with @xmath180 . a fixed value of @xmath181 , corresponds to an expected decoding overhead @xmath182 . , scaledwidth=49.0% ] in this section we experimentally evaluate the probability that decoding fails when a randomly selected subset of encoded symbols is available at the decoder . since our codes are rateless , we can set any target desired rate and examine the performance under random erasures . in this experiment we set the rate equal to @xmath178 ; the generator matrix comprises the @xmath0 columns of the identity matrix and @xmath0 parity generating columns , constructed randomly and independently as described in section [ sec : our - construction ] . the degree of the parities is upper bounded by @xmath179 , where the pre - log factor is arbitrarily set to a small constant value . decoding fails exactly when the columns corresponding to the encoded symbols available to the decoder form a matrix whose rank is strictly less than @xmath0 . a first series of experiments considers a sequence of random trials in which individual encoded symbol are erased in independently with probability @xmath181 . the ensemble of surviving symbols is available to the decoder . this corresponds to the transmission through an erasure channel with erasure probability @xmath181 . the cardinality of the decoding set is a binomial random variable with expected value equal to @xmath183 , which amounts to an expected decoding overhead @xmath184 . a total of @xmath177 code instances are generated and each is subjected to @xmath177 trials per value of @xmath181 . [ fig : simulations - b ] depicts the probability of decoding failure versus the channel erasure probability , @xmath181 . the experiment is repeated for three values of @xmath0 : @xmath185 , @xmath186 , and @xmath187 . the field size is set to @xmath188 , _ i.e. _ , a single byte per symbol , for all values of @xmath0 . in a second series of experiments , the decoding set of cardinality equal to @xmath189 is selected uniformly at random in each trial from the set of @xmath190 encoded symbols . [ fig : simulations - b2 ] depicts the estimated probability of decoding failure versus the decoding overhead @xmath5 . . the probability is evaluated over multiple @xmath177 randomly generated code instances with @xmath0 input symbols and rate @xmath178 , and @xmath177 independent trials per instance . the degree of parity symbols is equal to @xmath179 , with @xmath191 . , scaledwidth=49.0% ] theorem [ thm : main - theorem ] states that when @xmath15 is constructed as described in section [ sec : our - construction ] , a randomly selected @xmath78 submatrix @xmath63 is full rank w.h.p . equivalently , there exists a set of indices @xmath192 with @xmath193 such that @xmath194 submatrix @xmath195 is nonsingular . more formally , @xmath196 in the following , we exploit a connection between determinants and perfect matchings ( p.m.s ) in bipartite graphs . in section [ sec : our - construction ] , we showed the correspondence of the randomly constructed matrix @xmath15 to an unbalanced bipartite graph @xmath48 . the submatrix @xmath63 corresponds to a subgraph @xmath197 , depicted in figure [ fig : bipartite - graph - random - unbalanced - subgraph ] , where @xmath155 is a subset of @xmath62 vertices of @xmath50 , and @xmath198 is a subset of the edges incident to vertices in @xmath155 . similarly , a @xmath199 submatrix @xmath200 of @xmath63 corresponds to a smaller , balanced bipartite graph , @xmath201 , with @xmath0 vertices on each side . ( u1 ) at ( 0,0 ) ; ( u2 ) [ below of = u1 ] @xmath39 ; ( u3 ) [ below of = u2 ] ; ( u4 ) [ below of = u3 ] ; ( u5 ) [ below of = u4 ] @xmath39 ; ( u6 ) [ below of = u5 ] ; ( v1 ) at ( 3 , 0.5 ) ; ( v2 ) [ below of = v1 ] @xmath39 ; ( v3 ) [ below of = v2 ] ; ( v5 ) [ below of = v3 ] @xmath162 ; ( v6 ) [ below of = v5 ] @xmath39 ; ( v7 ) [ below of = v6 ] @xmath162 ; ( v1 ) edge [ bend right , out=-30 , in=-150 ] ( u1 ) ; ( v3 ) edge [ bend right , out=-30 , in=-150 ] ( u3 ) ; ( v5 ) edge [ out=170 , in=0 ] ( 1.5 , -2 ) ; ( v5 ) edge [ out=180 , in=22.5 ] ( 1.5 , -2.5 ) ; ( v5 ) edge [ out=190 , in=45 ] ( 1.5 , -3 ) ; ( v7 ) edge [ out=160 , in=-45 ] ( 1.5 , -4.5 ) ; ( v7 ) edge [ out=180 , in=315 , looseness=1 ] ( 1.5 , -5 ) ; ( v7 ) edge [ out=200 , in=315 , looseness=1 ] ( 1.5 , -5.5 ) ; ( -0.7,-2.35 ) ( -0.7,0.35 ) node [ midway , above , gray , sloped]@xmath145 ; ( -0.7,-5.35 ) ( -0.7,-2.65 ) node [ midway , above , gray , sloped]@xmath151 ; ( 3.7 , 0.85 ) ( 3.7,-1.85 ) node [ midway , above , gray , sloped]@xmath145 ; ( 3.7,-2.25 ) ( 3.7,-5.85 ) node [ midway , above , gray , sloped]@xmath202 ; ( -2.7 , 0 ) ( u1 ) ; ( groupvs ) ; ( @xmath203 ) ( @xmath204 ) ; ( vsname ) at ( @xmath205 ) @xmath206 , @xmath207 . ; @xmath200 is closely related to the edmond s matrix @xmath208 of the corresponding bipartite graph @xmath209 . by definition , the @xmath210-th entry of @xmath208 is a_(i , j ) = \ { ll a_i , j , & ( u_i , v_j ) e _ + 0 , & ( u_i , v_j ) e _ . , where @xmath211 , and @xmath212 s are _ indeterminates_. [ lemma : connection - between - determinant - and - pm ] the determinant of @xmath208 is nonzero if and only if there exists a perfect matching in @xmath209 , _ i.e. _ , @xmath213 however , @xmath200 is not an actual edmond s matrix ; it is obtained from @xmath208 substituting the indeterminates with randomly drawn elements of a finite field @xmath8 . there are two substantially different cases in which @xmath214 : * the determinant polynomial @xmath215 is identically zero , which by lemma [ lemma : connection - between - determinant - and - pm ] occurs if and only if @xmath209 has no perfect matching , or * it is not identically zero , _ i.e. _ , @xmath209 has a perfect matching , but the selected coefficients correspond to a root of the polynomial . the equivalence property in is not inherited by @xmath200 . in contrast to the use of indeterminates , an unfortunate selection of the random coefficients in @xmath200 can lead to zero determinant even when @xmath209 has a perfect matching . however , if the coefficients are drawn from a sufficiently large field , the probability of this event can be driven arbitrarily low . more concretely , if @xmath209 has a perfect matching @xmath216 , then the determinant of @xmath208 is a nonzero polynomial of degree exactly @xmath0 . the probability that a random assignment of coefficients from @xmath8 yields a zero determinant can be upper bounded by @xmath217 , using the schwartz - zippel lemma @xcite . in summary , @xmath218 the @xmath219 matrix @xmath220 has @xmath221 square submatrices . for successful decoding it suffices that at least one such submatrix @xmath222 is nonsingular , _ i.e. _ , has nonzero determinant . in light of , we ask instead whether there exists a subgraph @xmath223 that has a p.m .. observe that a p.m. in a subgraph @xmath223 is also a p.m. in the larger graph @xmath156 . conversely , if @xmath156 has a perfect matching @xmath216 , _ i.e. _ , a matching saturating all vertices in @xmath49 , then such a @xmath224 subgraph @xmath223 exists : its vertices are the endpoints of the edges in @xmath216 . according to the following lemma , @xmath156 has a perfect matching @xmath216 with high probability . [ lemma : one_sided_pm_nonexistence ] consider the bipartite graph @xmath197 corresponding to the submatrix @xmath63 of @xmath15 . @xmath155 contains any number @xmath145 of vertices with degree @xmath146 connected to distinct vertices in @xmath49 , and @xmath225 vertices that have randomly thrown @xmath119 edges as described in section [ sec : our - construction ] . for appropriate constant @xmath226 , @xmath227 for @xmath228 . the probability of equation can be written as & ( : ( _ ) 0 ) + & = 1- ( : ( _ ) 0 ) + & = 1- [ eq : probability_proof_1 ] . as argued in the previous paragraph , the existence of @xmath216 implies the existence of a subgraph @xmath223 that has a perfect matching . the probability @xmath229 , that all submatrices @xmath200 are singular despite the existence of @xmath216 , is upper bounded by the probability that @xmath230 , which was in turn upper bounded in by @xmath217 . hence , @xmath231 . on the other hand , nonexistence of a perfect matching in @xmath232 , implies that no submatrix @xmath200 can have nonzero determinant , hence , @xmath233 . continuing from , we have : ( . & . : ( _ ) 0 ) + & 1- + & = 1-(1- ( m ) ) - ( m ) + & = 1 - -(1- ) ( m ) . [ eq : ub - on - prob - no - sing - square - submatrix ] finally , satisfying the conditions in lemma [ lemma : one_sided_pm_nonexistence ] , we can guarantee that @xmath234 , for @xmath235 . applying the bound on the right hand side of , we obtain the desired result in . to complete the proof , it remains to prove lemmata [ lemma : connection - between - determinant - and - pm ] and [ lemma : one_sided_pm_nonexistence ] . we use the following expression for the determinant : @xmath236 where @xmath237 is the set of all permutations on @xmath238 , and @xmath239 is the sign of permutation @xmath240 . there is a one - to - one correspondence between a permutation @xmath241 and a candidate perfect matching @xmath242 in @xmath209 . note that if the candidate p.m. does not exist in @xmath209 , _ i.e. _ , some edge @xmath243 then the term corresponding to @xmath240 in the summation is @xmath150 . therefore , @xmath244 where @xmath245 is the set of perfect matchings in @xmath209 . if @xmath246 , _ i.e. _ , if @xmath209 has no p.m. , every term in the sum is equal to zero . if on the contrary @xmath209 has a p.m. , there exists a @xmath247 , and hence the term corresponding to @xmath248 is @xmath249 . additionally , there is no other term in the summation containing the exact same set of variables and this term can not be canceled out . in this case , @xmath250 , which concludes the proof of the lemma . [ [ sec : ospmorubg ] ] proof of lemma [ lemma : one_sided_pm_nonexistence ] : existence of perfect matching in the random subgraph ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ we want to establish an upper bound on the probability that there is no _ perfect matching ( p.m. ) _ between @xmath49 and @xmath155 in the random @xmath64 bipartite graph @xmath156 . in fact , we want to show that @xmath252 in the construction of the bipartite graph , suffices to achieve an upper bound asymptotically decaying with a rate @xmath33 . let @xmath253 , @xmath254 , denote the subset of @xmath155 corresponding to systematic encoded symbols . if a p.m. exists , we may assume that all symbols in @xmath255 participate in it . to see that , consider a vertex @xmath256 , connected to a symbol @xmath257 , and assume that @xmath258 is _ not _ included in the p.m. then , @xmath40 must be paired with some vertex @xmath259 , since @xmath43 was the only systematic symbol connected to @xmath40 . in addition , @xmath40 is the only symbol adjacent to @xmath43 , hence @xmath43 does not participate in the p.m. given such a p.m. , we can construct another one substituting @xmath260 with @xmath261 . therefore , without loss of generality , we may assume that all vertices in @xmath253 participate in the p.m. since @xmath262 , @xmath155 contains a nonempty subset corresponding to parity symbols , denoted by @xmath263 . let @xmath264 denote the subsets of @xmath49 matched with vertices in @xmath255 , and @xmath265 the remaining vertices . a p.m. between @xmath49 and @xmath155 exists if and only if a p.m. exists between @xmath266 and @xmath267 . the probability that a p.m. does not exist equals the probability that there exists a _ set of vertices in @xmath266 , _ i.e. _ , a subset of @xmath266 with a joint neighborhood smaller than its cardinality . let @xmath268 and @xmath269 . denote by @xmath270 the event that there exists a set of @xmath22 vertices in @xmath266 that _ contracts _ , _ i.e. _ , has at most @xmath271 neighbors in @xmath267 . this is equivalent to at least @xmath272 vertices in @xmath267 being only adjacent to vertices in @xmath266 other than the @xmath22 vertices of interest . then , applying on the coefficients of @xmath275 , we obtain @xmath276 } \nonumber \\ & \le ( k - s ) \max_{i}\left ( 2^{\left [ b_{1}(i)+b_{2}(i ) + b_{3}(i)\right ] } \right ) , \label{eq : ub - on - a - with - max}\end{aligned}\ ] ] where our objective is now to show that the right hand side of is @xmath3 . let @xmath286 and @xmath287 denote the numerator and denominator of the right hand side of inequality . for the numerator @xmath286 , we have the following upper bound : a. for the first part , and for @xmath296 we have + & ( k ) , & [ eq : ratio_bound_pt1 ] + where for the second inequality we have used and the fact that @xmath297 . b. for the second part , expanding the entropy we have + & = - + & + ( k ) , & [ eq : ratio_bound_pt2 ] + where the last inequality holds for @xmath298 . c. for the third part , for @xmath299 , @xmath300 . for @xmath301 , first observe that @xmath302 is decreasing in @xmath0 for @xmath303 . since @xmath304 and @xmath305 , exploiting the monotonicity of @xmath306 , we have + & + & ( 1 + ) + & ( k ) , & [ eq : ratio_bound_pt3 ] + where the last inequality holds when @xmath307 . consider the decoding graph @xmath156 corresponding to the @xmath64 submatrix @xmath63 of @xmath15 . @xmath156 is a random bipartite graph between @xmath0 input and @xmath62 encoded nodes , such that each encoded node has degree at most @xmath47 . an input symbol is _ covered _ by the set of @xmath62 encoded symbols , if and only if it participates with a nonzero coefficient in the formation of at least one symbol in the set . in terms of the decoding graph , an input node is covered if and only if it is adjacent to at least one encoded node . the probability of decoding failure is lower bounded by the probability that an uncovered input node exists in @xmath156 : all input nodes being covered is a prerequisite for the @xmath0 input symbols to be retrievable from a set of @xmath62 encoded symbols . the problem is equivalent to throwing @xmath311 balls into @xmath0 bins and requiring that no bin is empty with high probability . it is a standard result in balls and bins analysis that throwing @xmath312 balls is necessary to that end . it is hence imperative that @xmath313 . taking into account that @xmath67 , we obtain the desired result . @xmath251 let @xmath314 $ ] be the subset of systematic symbols covered by a parity @xmath65 . also , let @xmath133 be the set of parity symbols covering a systematic symbol @xmath66 , and @xmath315 . note that @xmath316 . the total number of generated parities is @xmath117 . hence , @xmath216 is a binomial random variable with @xmath117 trials and probability of success equal to @xmath317 , the probability that a parity @xmath65 covers the systematic symbol @xmath66 . every parity @xmath65 throws its @xmath47 edges uniformly at random over @xmath318 $ ] , independently , with replacement . a simple union bound yields @xmath319 similarly , we can obtain a lower bound : @xmath320 using the fact that @xmath321 for @xmath322 , inequality is simplified into @xmath323 the probability of that parity @xmath65 covers @xmath66 lies in the range described by and . based on these bounds , we can calculate a range for the expected value of @xmath216 . taking into account that @xmath324 , we have @xmath325 \le rc\log(k ) . \label{eq : bound - on - expected - m}\end{aligned}\ ] ] since each parity is created independently , the following chernoff bound on the lower tail of the distribution of @xmath216 holds : @xmath326 \right ) & \le \exp\left ( -\frac{\epsilon^2}{2 } e[m ] \right ) . \label{eq : chernoff - ub - lowtail - expected - number - of - cover } \end{aligned}\ ] ] the right hand side of can be further bounded as follows @xmath327 \right ) \nonumber\\ & \le \exp\left ( -\frac{\epsilon^2}{2 } \left [ rc\log(k ) - \frac{rc\log^2(k)}{k } \right]\right ) \nonumber \\ & = \exp\left ( - \log(k^{rc\epsilon^2/2 } ) \right ) \exp\left ( \frac{\epsilon^2}{2 } \frac{rc\log^2(k)}{k } \right ) \nonumber \\ & \le \frac{1}{k^{rc\epsilon^2/2}}\exp\left ( \frac{\epsilon^2rc\log^2(k)}{2k } \right),\end{aligned}\ ] ] which is the desired result . @xmath251 let @xmath133 denote the set of parity symbols that cover systematic symbol @xmath66 . we assume for simplicity that @xmath328 . given that every parity was generated independently , the footprints @xmath329 for @xmath330 are independent random variables . we are interested in the maximum cardinality subset @xmath331 such that any two parity symbols @xmath332 are isolated . consider the graph @xmath333 , where edge @xmath334 if and only if @xmath43 and @xmath44 are not isolated . then , @xmath335 corresponds to the maximum independent set in @xmath174 , @xmath336 . since @xmath174 is a random graph , its independence number @xmath337 is a random variable @xmath338 , which is a function of the @xmath339 independently drawn @xmath131 s . function @xmath340 satisfies the bounded differences condition , _ i.e. _ , for any configuration @xmath341 , substituting a single variable @xmath131 with another variable @xmath342 can not impact the function value arbitrarily . in the graph analogy , substituting @xmath131 with @xmath342 for some @xmath22 , corresponds to removing a vertex from @xmath174 along with its incident edges , and inserting a new vertex arbitrarily connected to other vertices . [ lemma : size - of - independent - set - in - graph ] consider an undirected graph @xmath343 with @xmath344 and let @xmath345 denote its independence number , _ i.e. _ the cardinality of the maximum independent set @xmath346 . construct a graph @xmath347 as follows : remove a node @xmath65 from @xmath348 along with all incident edges and insert a new node @xmath349 connected to an arbitrary set of vertices in @xmath348 . then @xmath350 . regardless of whether @xmath351 or not , the set @xmath352 is common in @xmath348 and @xmath347 and remains an independent set in the latter . therefore , @xmath353 , where equality in the second inequality holds only if @xmath351 . inversely , consider the maximum independent set in @xmath347 , @xmath354 . irrespectively of whether @xmath355 , the set @xmath356 is an independent set in @xmath348 too , implying that @xmath357 . we conclude that @xmath358 , which is the desired result . based on the previous lemma , we have @xmath359 provided that @xmath131 s are independent and @xmath340 satisfies condition , mcdiarmid s inequality @xcite yields @xmath360-t\right)\ ; \le \exp\left(- \frac{2 t^2}{m } \right ) , \label{eq : mcdiarmid - bound - on - maximum - independent - set}\end{aligned}\ ] ] for @xmath361 . the concentration result of holds , even if @xmath362 $ ] remains unknown . a trivial lower bound on @xmath362 $ ] can be obtained using those vertices in @xmath174 that are disconnected components , _ i.e. _ , that have degree equal to zero . such vertices correspond to parity symbols that are isolated from all other symbols in @xmath133 , not only those in @xmath335 , and are always members of the maximum independent set . the probability that a parity symbol @xmath330 is isolated from all other symbols in @xmath133 is @xmath363 multiplying with @xmath339 , the number of symbols in @xmath133 , we obtain a lower bound on the expected number of completely isolated parities in @xmath133 , which in turn is a lower bound on @xmath362 $ ] . in other words , @xmath364,\end{aligned}\ ] ] therefore , we have @xmath365 ^ 2}{k}-t\right ) \le \text{pr}\left(z \le e\left[z\right]-t\right ) . \nonumber\end{aligned}\ ] ] combining the above , with inequality , we conclude that @xmath366 ^ 2}{k}-t\right ) \le \exp\left(- \frac{2 t^2}{m } \right ) . \label{eq : mcdiarmid - bound - on - the - lower - bound - of - expected - z}\end{aligned}\ ] ] let @xmath367 ^ 2}{k}$ ] for some @xmath368 . for sufficiently large @xmath0 , @xmath20 will be nonnegative . substituting @xmath20 in , we obtain @xmath369 ^ 2}{k } \right]^2}{m } \right ) . \label{eq : mcdiarmids - bound - for - z - less - than - alogk}\end{aligned}\ ] ] we are interested in the value of @xmath370 for which @xmath371 decreases faster than @xmath372 , for some @xmath373 . it suffices to require the right hand side of is less than @xmath372 . taking logarithms on both sides , it suffices to find @xmath370 such that @xmath374 ^ 2}{m } \le -(1+\lambda ) \log(k ) \nonumber\\ \leftrightarrow & - 2\left[{\underbrace{(1-\alpha)}_{\overline{\alpha}}}^2m- \frac{m^2d(k)^2}{k } \right]^2 \le -(1+\lambda ) m\log(k ) \nonumber\\ \leftrightarrow & \overline{\alpha}^2m^2 - \overline{\alpha } 2\frac{m^3d(k)^2}{k } + \frac{m^4d(k)^4}{k^2 } - \frac{1+\lambda}{2}m\log(k ) \ge 0 . \nonumber\end{aligned}\ ] ] the last inequality is a quadratic inequality on @xmath375 , satisfied when @xmath376 which for appropriate choice of @xmath124 and @xmath125 , can be a solution that asymptotically lies in @xmath377 . therefore , for @xmath378 we have @xmath379 the probability that there exists a systematic symbol with availability lower than @xmath380 can be bounded with a union bound over all systematic symbols : @xmath381 which completes the proof . @xmath251 a. g. dimakis , p. godfrey , y. wu , m. wainwright , and k. ramchandran , `` network coding for distributed storage systems , '' _ information theory , ieee transactions on _ , vol . 56 , no . 9 , pp . 4539 4551 , sept . 2010 . m. sathiamoorthy , m. asteris , d. papailiopoulos , a. g. dimakis , r. vadali , s. chen , and d. borthakur , `` xoring elephants : novel erasure codes for big data , '' _ proceedings of the vldb endowment ( to appear ) _ , 2013 . c. huang , h. simitci , y. xu , a. ogus , b. calder , p. gopalan , j. li , and s. yekhanin , `` erasure coding in windows azure storage , '' in _ proceedings of the 2012 usenix conference on annual technical conference _ , ser . usenix atc12.1em plus 0.5em minus 0.4em berkeley , ca , usa : usenix association , 2012 , pp . o. khan , r. burns , j. plank , w. pierce , and c. huang , `` rethinking erasure codes for cloud file systems : minimizing i / o for recovery and degraded reads , '' in _ usenix conference on file and storage technologies ( fast ) _ , 2012 . r. motwani and p. raghavan , `` algorithms and theory of computation handbook , '' m. j. atallah and m. blanton , eds.1em plus 0.5em minus 0.4em chapman & hall / crc , 2010 , ch . randomized algorithms , pp . c. mcdiarmid , `` on the method of bounded differences , '' in _ surveys in combinatorics , 1989 , j. siemons ed . , london mathematical society lecture note series 141 , cambridge university press _ , 1989 , pp . 148188 .

* abstract . * this paper considers the problem of estimating the structure of multiple related directed acyclic graph ( dag ) models . building on recent developments in exact estimation of dags using integer linear programming ( ilp ) , we present an ilp approach for joint estimation over multiple dags , that does not require that the vertices in each dag share a common ordering . furthermore , we allow also for ( potentially unknown ) dependency structure between the dags . results are presented on both simulated data and fmri data obtained from multiple subjects . * keywords . * hierarchical models , directed acyclic graphs , non - exchangeability , integer linear programming , joint estimation * author footnote . * chris . j. oates ( e - mail : _ c.oates@warwick.ac.uk_ ) and jim q. smith ( e - mail : _ j.q.smith@warwick.ac.uk_ ) , department of statistics , university of warwick , coventry , cv4 7al , uk . sach mukherjee ( e - mail : _ sach@mrc-bsu.cam.ac.uk_ ) , mrc biostatistics unit and school of clinical medicine , university of cambridge , cambridge , cb2 0sr , uk . james cussens ( e - mail : _ james.cussens@cs.york.ac.uk_ ) , department of computer science and york centre for complex systems analysis , university of york , york , yo10 5ge , uk .

1404.1238

this paper considers joint estimation of multiple directed acyclic graph ( dag ) models using integer linear programming ( ilp ) . graphical models are multivariate models in which vertices in a graph @xmath0 represent random variables with edges between the vertices describing conditional independence statements concerning the variables . in many settings the edge structure of the graph is itself uncertain and then an important challenge is to estimate this structure from data . there has been considerable research into structural inference for graphical models over the last decade , including bayesian networks ( bns ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , gaussian graphical models ( ggms ; * ? ? ? * ; * ? ? ? * ) and discrete graphical models @xcite . many graphical models are based on dags and these are the focus of this paper . in many applications , data @xmath1 are collected on multiple units @xmath2 that may differ with respect to conditional independence structure , such that corresponding dags @xmath3 may be non - identical . at the same time , when the units of study are related , the graphs @xmath3 , while potentially non - identical , are expected to have similarities . it is then natural to ask whether such similarity can be exploited by borrowing strength across the estimation problems indexed by @xmath4 and that is the aim of the present paper . specifically we seek to construct a _ joint _ estimator @xmath5 that estimates a collection of dags together . we contrast such an estimator with the _ independent _ estimator @xmath6 that estimates each dag from the associated data only . the best known class of dag models are bns , but more exotic dag models exist ( e.g. * ? ? ? bns admit an extensive theory of inferred causation that has contributed to their popularity @xcite . structure learning for individual bns is a well - studied problem , with contributions including @xcite . structure learning is np - hard , but an approach that has attracted much recent attention is to cast _ maximum a posteriori _ ( map ) dag estimation as a problem in integer linear programming ( ilp ) , as developed simultaneously by @xcite and subsequently extended in @xcite . in brief , this approach solves a sequence of linear relaxations of the map estimation problem via the introduction of cutting planes and combines this with a branch - and - bound search to produce an optimal solution . if the algorithm terminates , the result is guaranteed to be a global maximum of the posterior distribution and hence inherits theoretical guarantees associated with the map estimator ( see e.g. * ? ? ? ( in this paper , algorithms with this property are termed `` exact '' . ) coupled with powerful techniques from discrete optimisation @xcite , ilps represent an efficient and attractive methodology for structural inference , as demonstrated also by recent empirical results ( e.g. * ? ? ? recent advances in this area are discussed in @xcite . the joint estimation of graphical models has recently received attention , for example @xcite put forward a penalised likelihood formulation that couples together estimation for multiple ( undirected ) ggms . however , joint estimation of multiple dags has so far received relatively little attention . the first discussion of this problem that we are aware of is @xcite ; here a greedy search was used to locate a local maximum of a joint bayesian posterior . @xcite described a markov chain monte carlo ( mcmc ) method for sampling from a joint posterior over graphical structures . however , the generic difficulties associated with stochastic search / sampling in large discrete spaces are well known ; these are exacerbated in the joint case by the size of the joint model space and stochastic search / sampling remains challenging in this setting . the focus of this paper is instead on exact , deterministic algorithms . @xcite proposed an exact algorithm based on bayesian model averaging and belief propagation , under the strong assumption that an ordering of the variables @xmath7 applies simultaneously to all units . at the same time @xcite proposed essentially the same algorithm , applied to the specific class of feed - forward dynamic bns , where an ordering of the variables is implicitly provided by the time index . the algorithmic contributions of the present paper are two - fold . first , we show how to cast exact inference over multiple dags as an ilp problem . we consider map - bayesian estimation for multiple dags and require no restriction on the ordering of the variables . this is done by extending methodology presented in @xcite to the case of multiple dags via a hierarchical bayesian formulation . second , we exploit structural constraints that are imposed by the dags in order to improve computational efficiency . as a illustrative example , our methods currently allow estimation of 10 related dags , each with 10 nodes , in time typically less than one minute on a standard laptop . in addition , we extend previous work by allowing for dependencies between the dags themselves and consider also estimation of this dependency structure . previous work on multiple dags has focused on the special case where the units are exchangeable and all pairs of units undergo an equal amount of regularisation ( including * ? ? ? * ; * ? ? ? * ) . however , in practice , relationships between units ( and their underlying graphical models ) may be complex , e.g. hierarchical , with group and sub - group structure , and such structure may itself be subject to uncertainty . @xcite performed exact inference for non - exchangeable feed - forward dynamic bns for the case where the relationships between units are known _ a priori_. @xcite addressed non - exchangeability in the context of general dag models but did not provide an exact algorithm and , again , assumed that the relationships between units are known at the outset . our approach provides a framework that allows for simultaneous learning of both unit - specific dags and the dependency structure that relates them . however , our empirical results suggest that such simultaneous learning may be extremely challenging in practice . the remainder of the paper is organised as follows : section [ def the model ] introduces a statistical framework for multiple dags and discusses regularisation based on graphical structure . section [ sec estimate ] gives exact , ilp - based estimators for multiple dags . section [ applications ] presents a simulation study and results on fmri data from a multi - subject study . finally we close with a discussion of directions for further research . a companion paper that explores the fmri application in more detail is available as @xcite . we begin by introducing the statistical model , deferring discussion of computation to the next section . throughout the shorthand @xmath8 will be used to denote the list of integers @xmath9 . a ( directed ) graph @xmath0 on vertices @xmath8 is characterised by a collection of sets @xmath10 , such that @xmath11 contains precisely the parents of vertex @xmath12 according to @xmath0 . we say @xmath0 is acyclic if @xmath0 contains no sequence of directed edges that begins and ends at the same vertex . write @xmath13 for the space of all directed acyclic graphs ( dags ) with @xmath14 vertices . in this paper vertices @xmath12 in a dag @xmath0 will be associated with random variables @xmath15 . we will use @xmath16 to denote the @xmath17th observation of variable @xmath15 for unit @xmath4 and @xmath18 to denote the collection @xmath19 of these variables . we present our approach from a map - bayesian perspective , but it could also be described as a penalised likelihood approach . we consider dag models for which the conditional distribution of the variables @xmath20 , given the dag @xmath3 and associated parameters @xmath21 , factorises as @xmath22 here @xmath23 denotes the parents of the @xmath12th variable in the dag @xmath3 and @xmath24 are the observed values of these parent variables in sample @xmath17 . additionally , @xmath25 are parameters associated with the conditional distribution for the @xmath12th variable that may depend on sample index @xmath17 , unit @xmath4 and model @xmath23 , though this latter dependence is suppressed in the notation . the joint likelihood that we consider below follows from eqn . [ factor ex ] and the assumption that the observations @xmath26 for each unit @xmath4 are conditionally independent given the dags @xmath3 and associated parameters @xmath27 . specifically , we have that the full likelihood factorises over units , variables and samples : @xmath28 for expositional simplicity , we take the number @xmath29 of samples to be the same for each unit , though this is not strictly required for our methodology . within a bayesian framework we place a prior distribution @xmath30 over parameters @xmath31 , such that parameter sets corresponding to units @xmath4 and @xmath32 are independent conditional upon the two dags @xmath3 and @xmath33 . integrating out the unknown parameters provides the evidence in favour of the joint model @xmath34 : @xmath35 below we introduce our prior distribution over all dags @xmath34 that encodes the notion of dependency that we wish to exploit during estimation . write @xmath36 for the space of undirected networks on vertices @xmath37 . the hierarchical prior that we propose factorises along edges of a network @xmath38 whose @xmath39 vertices correspond to the individual units : @xmath40 here the first product ranges over all edges @xmath41 in the network @xmath42 . the ( positive , symmetric ) function @xmath43 is interpreted as a measure of regularity ( i.e. similarity ) between the dags @xmath3 and @xmath33 ; specific choices for this function are discussed in section [ choose q ] below , motivated by computational convenience in the sequel . the graph @xmath42 indicates which pairs of units have similar graphical structure ( fig . [ model ] ) . for example , @xmath42 may describe a time - ordering of the units , such that consecutive units are expected to have more similar graphical structures , or may indicate group membership within a mixture model . in most existing literature an exchangeability assumption is placed on @xmath44 that corresponds ( implicitly ) to specifying @xmath42 as the complete network ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * etc . ) , though @xcite considered general ( known ) forms for @xmath42 . in this paper we allow for general and potentially unknown @xmath38 . the remaining terms @xmath45 are necessary for multiplicity correction and are discussed in section [ multi correct ] below , again motivated by computational convenience in the sequel . consider the map estimator under the multiple dag prior : @xmath46 when @xmath42 is the complete network all pairs of graphs are regularised equally and we refer to this case as `` exchangeable learning '' . more generally , we consider the setting where @xmath42 itself is unknown . then , we impose a prior distribution over @xmath38 ( described below in section [ a prior ] ) and estimate both the @xmath3 s and @xmath42 ; we refer to this as `` non - exchangeable learning '' and use the map estimator @xmath47 one of the main contributions of this paper , in section [ sec estimate ] , is to prove that for certain choices of the regularity function @xmath43 discussed below , both @xmath48 and @xmath49 are characterised as the solutions to ilps and hence are amenable to exact computation using advanced techniques such as constraint propagation and cutting plane algorithms . below we discuss choices for the regularity function @xmath43 that forms the basis for the multiple dag prior and captures similarity between dags @xmath3 and @xmath33 . write @xmath50 for equality up to an unspecified additive constant . the default choice that we consider for @xmath51 is @xmath52 . \label{hm}\end{aligned}\ ] ] here @xmath53 is the logical xor operator , @xmath54 \in \{0,1\}$ ] is an indicator associated with its ( logical ) argument and the @xmath55 are ( constant ) penalty terms associated with the inclusion of the edge @xmath56 in exactly one of the dags @xmath3 and @xmath33 . combined with the default multiplicity correction discussed below , the prior defined by eqn . [ hm ] is hyper - markov with respect to any dag @xmath3 when @xmath33 is held fixed @xcite ; we will see in section [ sec estimate ] that such priors permit a particularly simple construction of an ilp for the multiple dag model . for many scientific applications where graph structure represents a physical mechanism , eqn . [ hm ] can often be motivated from physical considerations . for example in neuroscience , an edge in a graphical model may have the interpretation of physical or functional connectivity and transfer of information between regions of the brain ( so - called reification ; e.g. * ? ? ? a special case , that treats both units and edges as exchangeable , is the structural hamming distance ( shd ) obtained by setting @xmath57 @xmath58 in eqn . [ hm ] . shd has previously been used to regularise between graphical models by @xcite and to integrate prior knowledge into bns by @xcite . an extension of shd with two degrees of freedom was also considered in this context by @xcite . the empirical results presented in this paper focus on shd due to its interpretability and simplicity , but the methodology is compatible with the general form eqn . [ hm ] and the @xmath55 s could be used to encode prior information on the similarity between units or the propensity for the presence of a particular edge to be conserved between units . the function @xmath59 in the multiple dag prior ( eqn . [ joint prior ] ) is required to adjust for the fact that the size of the space @xmath13 grows super - exponentially with the number @xmath14 of vertices @xcite . we follow @xcite and control multiplicity using the default binomial correction @xmath60.\end{aligned}\ ] ] here @xmath61 is an upper bound on the ( maximum ) in - degree of @xmath0 ( such upper bounds are widely used to control the computational intensity of structural inference ) . this specification has the desirable property that the collective prior probability of all models with @xmath62 predictors is @xmath63 , which is independent of both @xmath62 and @xmath14 . we note that the methods presented in this paper are compatible with the inclusion of additional terms in eqn . [ joint prior ] that encode specific informative priors on the dags @xmath3 but we do not use such priors here . we use the following prior for the undirected graph @xmath42 that encodes pairwise relationships between the units @xmath64 : @xmath65.\end{aligned}\ ] ] ( we adopt the convention that the adjacency matrix for @xmath42 is upper triangular , hence @xmath66 above . ) here the @xmath67 s encourage inclusion of the corresponding edges @xmath41 in @xmath42 and could in principle be used to encode specific knowledge regarding similarity between the units , although we do not pursue this direction here . in all empirical results below we consider the simplest case of @xmath68 . then , @xmath69 can be viewed as an inverse temperature hyper - parameter , with larger values of @xmath69 encouraging denser networks , which in turn correspond to a greater amount of between - unit regularisation . all joint estimators for graphical models that we are aware of require tuning parameters ( e.g. the fused and group flavours of the joint graphical lasso each have two tuning parameters ; * ? ? ? recalling that the dags @xmath44 give a probability model for the observations @xmath70 , hyper - parameters in the priors above may be set using standard procedures such as cross - validation or information criteria . let @xmath71 denote all hyper - parameters to be set . we recommend the use of information criteria for their computational convenience . in particular , using the akaike information criteria ( aic ) we have @xmath72 where @xmath73 are the number of parameters required to specify the conditional distribution of @xmath74 given @xmath75 . thus the aic requires only that we can obtain the map estimate @xmath76 over a range of values of the hyper - parameters @xmath71 , selecting the value that maximises eqn . in contrast , cross - validation requires the marginal predictive likelihood function @xmath77 for held - out data @xmath78 , which for many models may require nontrivial additional computation . we note that a grid search is necessary in the case of multiple tuning parameters ; sequential optimisation is not possible due to the non - orthogonality induced by the joint prior @xmath79 . for more complex joint priors / penalties this represents a challenge that we do not address here . we now consider the computational aspects of map estimation for the models introduced above . we begin with the simpler case where the network @xmath42 is known _ a priori _ and subsequently consider the more general unknown @xmath42 case . the methodology we present below extends the ilp formulation of @xcite to multiple units via the inclusion of additional state variables that capture the similarities and differences between the units . we begin by computing and caching the terms @xmath80 that summarise evidence in the data for the local model @xmath23 for the @xmath12th variable in unit @xmath4 . these are available in closed form for many models of interest , including but not limited to discrete bns with dirichlet priors @xcite , linear gaussian structural equation models with conjugate priors @xcite and multiregression dynamical models @xcite . since our approach is based upon these pre - computed quantities , they could even be obtained numerically ( for more complex models ) using mcmc and related techniques ( e.g. * ? ? ? * ; * ? ? ? these cached quantities are transformed to obtain `` local scores '' , defined as @xmath81 these are the ( log- ) evidence from eqn . [ evidence ] with an additional penalty term that provides multiplicity correction over varying @xmath82 . we define binary indicator variables @xmath83 $ ] that form the basis of our ilp . here @xmath84 is used to denote a possible parent set for the @xmath12th variable in the dag @xmath3 , so the information in the variables @xmath83 $ ] completely characterises @xmath3 . it will be necessary to impose constraints that ensure these variables correspond to a well - defined dag : @xmath85 = 1 \ ; \ ; \ ; \ ; \ ; \forall i , k\tag{c1 ; convexity}\ ] ] constraint ( c1 ) requires that for each unit @xmath4 , every vertex @xmath12 has exactly one parent set ( i.e. there is a well - defined graph @xmath3 ) . to ensure @xmath3 is acyclic we require further constraints : @xmath86 \geq 1 \ ; \ ; \ ; \ ; \ ; \forall k , \emptyset \neq c \subseteq \{1 : p\}. \tag{c2 ; acyclicity}\ ] ] ( c2 ) states that for every non - empty set @xmath87 there must be at least one vertex in @xmath87 that does not have a parent in @xmath87 . it is not challenging to prove ( by contradiction ) that ( c1 - 2 ) exactly characterise the space @xmath13 of dags . next we define an indicator @xmath88 $ ] of the presence of each specific edge @xmath56 in @xmath3 . this can be related to the parent set indicators as @xmath89 = \sum_{\substack{\pi \subseteq \{1:p\}\setminus\{i\ } \\ j \in \pi } } [ g_i^{(k ) } = \pi ] \ ; \ ; \ ; \ ; \ ; \forall i , j , k \tag{c3 } .\ ] ] to decide whether units @xmath90 agree on the presence or absence of a specific edge @xmath56 , we introduce additional variables @xmath91 $ ] where @xmath53 is the logical xor operator . to encode this definition we use the set of linear inequalities @xmath92 & & -[j \in g_i^{(k ) } ] & & \leq & & 0 \label{xor1 } \tag{c4.1 } \\ - \delta^{(k , l)}(j , i ) & & + [ j \in g_i^{(k ) } ] & & -[j \in g_i^{(k ) } ] & & \leq & & 0 \tag{c4.2 } \\ - \delta^{(k , l)}(j , i ) & & -[j \in g_i^{(k ) } ] & & + [ j \in g_i^{(k ) } ] & & \leq & & 0 \tag{c4.3 } \\ + \delta^{(k , l)}(j , i ) & & + [ j \in g_i^{(k ) } ] & & + [ j \in g_i^{(k ) } ] & & \leq & & 2 \tag{c4.4}\end{aligned}\ ] ] these inequalities are optimal in the sense that they define ( in the most concise way ) the convex hull ( in @xmath93 ) of feasible solutions to the corresponding 3-variable xor constraint @xcite . write @xmath94 for a vector containing the binary variables @xmath95 $ ] , @xmath88 $ ] and @xmath96 , so that @xmath97 where @xmath98 . we can now state our first result : [ fixed n ] the map estimate @xmath48 is characterised as the solution of the ilp @xmath99 - \sum_{(k , l ) \in a } \sum_{i = 1}^p \sum_{j=1}^p \lambda_{j , i}^{(k , l ) } \delta^{(k , l)}(j , i ) \label{ip}\end{aligned}\ ] ] subject to constraints ( c1 - 4 ) . note that under an in - degree restriction @xmath100 the dimensionality of the state space @xmath101 can be reduced from exponential in @xmath14 to polynomial in @xmath14 , specifically @xmath102 . it is seen that the dimensionality of @xmath101 increases by a factor @xmath103 when learning @xmath39 dags jointly , as opposed to individually ( assuming that the first term dominates ) . to illustrate the use of result [ fixed n ] , consider a fixed network @xmath42 that is equal to the complete network ( exchangeable learning ) and regularisation based on shd with hyper - parameter @xmath104 . [ example ] displays the map @xmath48 computed ( exactly ) as a function of the regularity hyper - parameter @xmath104 for a small fmri dataset of 3 subjects containing measurements of activity in 4 neural regions . the first row , with @xmath105 , is the result of performing independent estimation on the 3 subjects to obtain map estimators for subject - specific dags . notice that the dags are quite dissimilar ; the neuroscience context from which these data arise suggests that this is most likely an artefact ( i.e. due to variance in estimation ) . as the regularity parameter @xmath104 is increased , the subject - specific dags become more similar until they are eventually identical at @xmath106 and above . in simulation studies below we demonstrate that this kind of exchangeable learning can lead to improved estimation of unit - specific graphical structure . the previous section addressed the case where the interdependency structure @xmath42 among the units was known . below we formulate an ilp for the more general case where @xmath42 is unknown and must be estimated jointly with the unit - specific dags . we introduce additional binary indicator variables @xmath107 $ ] that directly encode those edges that are present in the network @xmath42 . taking together the variables @xmath96 and @xmath107 $ ] allows us to determine whether or not an edge @xmath56 differs between an unit @xmath4 and its neighbour @xmath32 in @xmath42 . specifically , we represent this information by defining a new binary variable @xmath108 $ ] where @xmath109 is the logical and operator . to encode this definition in an ilp we use the following set of linear inequalities @xmath110 & & & & \leq & & 0 \tag{c5.1 } \\ + \delta^{(k , l)}(j , i ) & & & & -\delta^{(k , l)}(j , i ) & & \leq & & 0 \tag{c5.2 } \\ - \delta^{(k , l)}(j , i ) & & + [ ( k , l ) \in a ] & & + \delta^{(k , l)}(j , i ) & & \leq & & 1\tag{c5.3}.\end{aligned}\ ] ] the linear inequalities are again optimal , in the sense that they define ( in the most concise way ) the convex hull ( in @xmath93 ) of feasible solutions @xcite . in this case we write @xmath94 for a binary vector that additionally contains the variables @xmath111 $ ] and @xmath112 , so that @xmath97 with @xmath113 larger than in the exchangeable setting , but with the same asymptotic scaling . we can now state our second result : [ nonex theo ] the map estimate @xmath49 is characterised as the solution of the ilp @xmath114 \label{ip2 } \\ & & - \sum_{i=1}^p \sum_{j=1}^p \sum_{k=1}^k \sum_{l = k+1}^k \lambda_{j , i}^{(k , l ) } \delta^{(k , l)}(j , i ) + \sum_{k = 1}^k \sum_{l = k+1}^k \eta^{(k , l ) } [ ( k , l ) \in a ] \nonumber\end{aligned}\ ] ] subject to constraints ( c1 - 5 ) . the dimensionality of @xmath101 can again be reduced from exponential in @xmath14 to polynomial in @xmath14 through the adoption of an in - degree restriction . our computational implementation is built on the c package gobnilp that performs inference for individual dags @xcite , available at http://www.cs.york.ac.uk / aig / sw / gobnilp/. gobnilp , originally described in @xcite , computes the map estimate on a per - unit basis ; we therefore created an interface that interacts with gobnilp and enforces additional constraints ( c3 - 5 ) that couple together multiple dags as described above . below we elaborate on how such an interface was constructed and summarise additional optimisation routines that underpin associated computational efficiency in the joint setting . since there are exponentially many acyclicity constraints ( c2 ) , gobnilp does _ not _ construct an integer program ( ip ) containing all of them . instead , initially , only the convexity constraints ( c1 ) are included in an ip . gobnilp then ( exactly ) solves the linear program ( lp ) relaxation of the ip where integer variables are ( temporarily ) allowed to take any real value within their bounds . this produces an lp solution @xmath115 . crucially , the lp solution can be found very quickly . gobnilp then searches for ( c2 ) acyclicity constraints that @xmath115 violates and adds them to the ip . linear constraints added in this way are known as `` cutting planes '' since they ` cut off ' the infeasible solution @xmath115 . gobnilp then solves the linear relaxation of this new ip and looks for cutting planes for the new lp solution . this process of solving linear relaxations and adding cutting planes is then repeated . note that the sequence of lp solutions produced in this way provide increasingly tight upper bounds on the objective value of the map estimator . it can happen that an lp solution represents well - defined dags ( i.e. all variables have integer values and the graphs they represent are each acyclic ) . in this case , the problem is solved exactly , since any lp solution provides an upper bound on the objective function . more typically ( at least on larger problems ) a fractional lp solution is generated that satisfies all the constraints . in such a situation gobnilp selects a fractional variable , e.g. @xmath95 $ ] , to branch on , creating two subproblems , one where @xmath95=0 $ ] ( ruling out parent set @xmath116 for vertex @xmath12 for unit @xmath4 ) and one where @xmath95=1 $ ] ( setting the parent set of vertex @xmath12 for unit @xmath4 to be @xmath116 ) . the gobnilp algorithm is then applied recursively on both branches . when an ip variable @xmath95 $ ] is set to 1 , gobnilp performs constraint propagation by setting to 0 those ip variables that indicate other parent sets for vertex @xmath12 or parent set choices for other vertices in the dag that are no longer possible under the acyclicity constraints . this approach to ip solving is known as `` branch - and - cut '' since both branching on variables and adding cutting planes are used . gobnilp is implemented using the scip framework due to @xcite . the basic gobnilp algorithm described above has been optimised in a number of ways : e.g. extra linear constraints are added to the initial ip and additional cutting plane algorithms ( some built into scip , some gobnilp - specific ) are used . in addition , a heuristic algorithm ( based on ` rounding ' the current lp solution ) is used to generate ` good ' but probably sub - optimal dags . this can help prune the search tree produced by branching on variables ( but does not affect exactness of the algorithm ) . see @xcite for further details . for joint estimation we constructed an interface for gobnilp that interlaces with scip to include additional constraints that couple together estimation problems for multiple dags . in practice ( c4 ) are implemented using scip s xor constraint handler . this handles any constraint of the form @xmath117 where @xmath118 and the constraint is satisfied if either @xmath119 and an odd number of the @xmath120 are 1 , or @xmath121 and an even number of the @xmath120 are 0 . we set @xmath122 \oplus [ j \in g_i^{(l)}]$ ] . scip generates these inequalities internally and also provides a propagator for xor constraints ; in the simplest case , as soon as any two of @xmath96 , @xmath88 $ ] and @xmath123 $ ] have their values fixed then scip will immediately fixed the value of the third appropriately . similarly ( c5 ) are implemented using scip s and constraint handler . this handles any constraint of the form @xmath124 , where @xmath118 . we set @xmath125 \ & \delta^{(k , l)}(j , i)$ ] . scip provides a propagator for and constraints that works analagously as for xor constraints . a naive implementation of our methodology can be quite effective , but with some additional consideration we can reduce computational effort by about an order of magnitude . specifically , we introduce additional ( redundant ) constraints @xmath89 + \sum_{\substack{\pi \subseteq \{1:p\}\setminus\{i\ } \\ j \notin \pi } } [ g_i^{(k ) } = \pi ] = 1 \ ; \ ; \ ; \ ; \ ; \forall i , j , k . \tag{c6 ; efficiency}\ ] ] these do not change the definition of the ilp , since intuitively ( c6 ) states that if the edge @xmath126 is present in @xmath3 then the parent set @xmath23 for vertex @xmath12 must contain the vertex @xmath127 , which is trivially true . nevertheless the inclusion of ( c6 ) can greatly reduce search time in practice since it allows for more effective propagation . this is because when ( c6 ) are included then the edge - indicators @xmath88 $ ] become particularly good variables to branch on , producing a more balanced and consequently smaller search tree . in this work we restricted attention to the map estimator , but the gobnilp software and our interface allow efficient computation of the top @xmath17 solutions to the ilp , that is the @xmath17 best joint estimates of the dags , along with their associated values for the objective function ( and hence bayes factors ) . the interfacing software is written for matlab r2014b and is provided in the supplementary materials . in this section we investigate the performance of the proposed joint estimators , using simulated data and data from a neuroscience study . in all examples , we restrict attention to a single regularity hyper - parameter @xmath104 and a single density hyper - parameter @xmath69 . the performance of structure learning algorithms for graphical models can be quantified in many ways ( e.g. * ? ? ? * ; * ? ? ? structural estimation can be viewed as a binary classification task on individual edges and performance can be measured by the matthews correlation coefficient ( mcc ) which is popular in scenarios where the number of negative samples ( i.e. non - edges ) can out - weigh the number of positive samples ( i.e. edges ) . all experiments below were repeated 10 times based on independently generated datasets . in order to generate multiple related dags we employed a mcmc sampling scheme that targets the multiple dag prior @xmath128 . metropolis - hastings proposals that switch the status ( i.e. present or absent ) of an edge ( selected uniformly at random ) within a unit ( selected uniformly at random ) , subject to dag constraints , achieved an average acceptance probability of 83% . then , conditional on a dag @xmath3 , we simulated data @xmath26 . this could be achieved using essentially any likelihood model and in principle this choice will affect the properties of our estimators . we therefore took the simplest approach of directly simulating the evidence terms ( eqn . [ evidence ] ) from log - normal distributions . full pseudocode and convergence diagnostics are provided in the supplementary materials . we emphasise that the multiple rounds of estimation performed in the simulation study below preclude comparison with more computationally intensive approaches such as @xcite . on the other hand , methods that require a shared ordering of the variables such as @xcite do not apply in this general setting . in all experiments below we select hyper - parameters using a brute - force grid search over the ( coarse ) grid @xmath129 , @xmath130 . that is , we solve up to 25 ilps for each evaluation of the aic - based estimator . these candidate values for hyper - parameters were deliberately chosen to match the scale of the simulated data , as explained in the supplementary material . in practice the combinations of hyper - parameters that yield poor estimates ( i.e. low aic ) tend to correspond to ilps that are quickly solved , so that this grid search does not substantially affect computational complexity . to establish an ( approximate ) upper bound on the computational time required to evaluate our proposed estimators , we generated scores @xmath131 independently from @xmath132 , leading to very many dag models attaining similar posterior probabilities . this is a `` worst case scenario '' for the ilp solver , in the sense that maximum effort is required to distinguish between the models . an indication of these upper bounds on the computational times for a single ilp solve is provided in fig . [ times ] . it is seen that the operating range of our procedure is restricted to approximately @xmath133 , though in practice the worst - case scenario is unlikely to be realised and it may be possible to solve larger problems ( this is highly dependent on the values taken by the local scores ) . we make two further observations : firstly , in the exchangeable case with @xmath134 , @xmath135 the program was terminated due to insufficient ram . however the current best solution up to that point achieved an objective value of 71.49 and the upper bound provided by the ilp was 93.31 . this ability to provide a sure upper bound enables a heuristic map search for more challenging problems ( though in the sequel we considered only exact solutions ) . secondly , although the non - exchangeable case with @xmath136 , @xmath137 required 193,047 seconds to terminate ( about 54 hours ) , the map estimate was actually found after 31,020 seconds ( 16% of the total time ) , with the remaining time used to confirm that this was indeed the map . this suggests that computational time could be reduced if we are willing to sacrifice the guarantee of exactness . * exchangeable learning : * firstly we considered the case where the network @xmath42 ( that describes relationships between the units @xmath4 ) is fixed equal to the complete network ( exchangeable learning ) . we focussed on two distinct data - generating regimes : ( i ) the data - generating process has @xmath42 complete with related but non - identical dags , and ( ii ) the data - generating process has @xmath42 as two disconnected complete components of equal size and therefore dags more similar within each component than between components . regime ( i ) employs exchangeable learning in a favourable setting where information is shared between all dags , whilst regime ( ii ) employs exchangeable learning in an unfavourable setting , where the assumption of exchangeability is violated . results for regime ( i ) in fig . [ sims1 ] ( a - d ) demonstrate that substantially improved performance is achieved by the proposed joint estimators compared with independent estimation . we also compared against the pooled estimator that requires all dags to have identical structure - a crude form of information sharing . results showed that this `` same dag '' estimator performed worse in general than inference based on hyper - parameters selected using aic , as expected . the same datasets were employed for each estimator , indicated by dashed lines , to ensure a fair comparison . results from regime ( ii ) in fig . [ sims1 ] ( e - f ) show that , despite the exchangeability assumption being violated , exchangeable learning remains competitive with independent estimation , but same dag " performs poorly , as expected . * non - exchangeable learning : * secondly we considered the case where the network @xmath42 is unknown and must be learned along with the unit - specific dags @xmath44 . in general the data - generating @xmath42 could have any topology ; we again focussed on the two distinct regimes described above . regime ( i ) aims to examine the loss of statistical efficiency that results from employing non - exchangeable learning when exchangeable learning would be more suitable , whilst regime ( ii ) explores the challenging case where the unknown @xmath42 is highly informative , so that the loss of information that comes from not knowing @xmath42 is greatest . results for regime ( i ) in fig . [ sims3](a - b ) demonstrate that , despite being extremely general , non - exchangeable learning still achieves improved estimation compared to both independent inference and the same dag estimator . for regime ( ii ) , shown in fig . [ sims3](c - d ) , we see that estimation is more challenging . indeed the gains in performance compared with independent estimation were small , perhaps reflecting the increased difficulty of jointly learning both @xmath44 and @xmath42 . ( as before , the same dag procedure performed poorly in regime ( ii ) . ) for regime ( ii ) it is interesting to ask whether the network @xmath42 is itself accurately estimated ; results indicated that @xmath42 is not well estimated , with the map network @xmath138 typically being equal to the complete network . results in fig . [ sim5 ] compare the ability to estimate @xmath44 with the ability to estimate @xmath42 , showing that @xmath42 was typically more difficult to estimate compared with the unit - specific dags @xmath44 . this may explain why we do not see substantive gains over independent estimation in the non - exchangeable case , with performance similar to exchangeable learning . the difficulty in estimating @xmath42 is perhaps unsurprising : @xmath42 is latent for the dags @xmath44 that are themselves latent , so that the `` doubly latent '' @xmath42 may be only weakly identifiable . thus , while dag estimation results in the unknown @xmath42 formulation seem at worst competitive with the simpler exchangeable set - up , we would recommend caution in interpreting the estimate @xmath138 itself . we illustrate the methodology by analysing data from a functional magnetic resonance imaging ( fmri ) experiment . graphical models are widely used to represent neural connectivity in such studies and it is typical for fmri experiments to involve multiple subjects . a statistical analysis should therefore take into account not only the interaction between areas of one brain but also the similarities and differences among subjects @xcite . for instance , using fmri data derived from a writing experiment @xcite showed that the communication pattern among brain regions may be subject - specific , while @xcite argue that the causal relations between brain regions can vary in disease states according to the severity of the disease . we analysed a small fmri dataset consisting of six unrelated subjects from the human connectome project @xcite . data were acquired on each subject while they were in a state of quiet repose and preprocessed to obtain 10-dimensional time series representing the activity levels at 10 regions in each subject . the specific application is discussed in greater detail in the companion paper @xcite . the interactions ( or connections ) among different cerebral areas are usually studied through causal dag models @xcite . here we use the multiregression dynamical model ( mdm ) for fmri data developed by @xcite , in which edges denote direct contemporaneous relationships that might exist between nodes and the connections are represented by parameters that vary over time . thus mdms are a generalisation of classical `` static '' bns but , unlike bns , are fully identifiable ( i.e. the equivalence classes are singletons ) . the mdm is defined on a multivariate time series that aims to identify the causal structure among the variables over time @xcite . in the mdm that we consider , a multivariate model for observable series @xmath139 , for subject @xmath4 at time @xmath17 is characterised by a contemporaneous dag , with information shared across time only through evolution of the model parameters @xmath25 . formally , this model is described by the following observation equations and system equations : @xmath140 where @xmath141 is the concatenated parameter vector and @xmath142 , @xmath143 , @xmath144 must be specified . the equations of the mdm can be viewed as a collection of nested univariate linear models , allowing the parameters to be estimated using well - known kalman filter recurrences over time . we refer the reader to @xcite for further details regarding prior specification and computation of the local scores in this setting . * independent learning : * first , we applied independent estimation to each of the 6 subjects , with results shown in fig . [ neuro1 ] ( top ) . these results are unsatisfactory on neurological grounds since it is anticipated that connectivity does not change grossly between subjects . * exchangeable learning : * next , we inferred dags @xmath44 under an exchangeability assumption , for various values of the regularity hyper - parameter @xmath104 ( fig . [ neuro1 ] ) , subjectively set to @xmath145 . ( alternative approaches to set @xmath104 include aic , as described above , or exploiting technical replicate data and examination of bayes factors , as discussed at length in @xcite ) . * non - exchangeable learning : * based on @xmath146 we performed non - exchangeable learning in order to estimate similarities and differences between the 6 subjects in terms of their neural connectivity . specifically , we estimate a network @xmath42 on these 6 subjects , based on various values of the density hyper - parameter @xmath69 . results , shown in fig . [ n neuro ] , indicate that at @xmath147 , subjects 1 and 4 , 2 and 3 , and 5 and 6 exhibit the most similar graphical structures . non - edges @xmath148 in the network @xmath42 imply differences in the subjects graphical structure ; such information may therefore be used to identify subjects with particularly aberrant connectivity for further scientific investigation . going further , we see that subjects 5 and 6 are perhaps the most unusual , being totally disconnected from subjects 1 to 4 in the network @xmath42 when @xmath149 . however , we recommend caution in interpreting @xmath138 , due to the difficulty in estimation that we observed in simulations . we note also that @xmath69 can also be elicited based on bayes factors , as discussed in @xcite . further analysis and interpretation of these results can be found in the companion paper @xcite . we emphasise that this analysis is exploratory and not confirmatory , but could be used to generate hypotheses for further experimental investigation . this paper introduced a general statistical framework for joint estimation of multiple dags . we considered regularisation based on graphical structure and showed how map estimators in this setting could be characterised as the solutions to ilps that admit efficient exact algorithms based on branching - and - cutting as well as constraint propagation . we believe these are the first exact algorithms for the general case of multiple general dag models ( i.e. with no restriction on ordering of the variables ) . in our general framework for joint learning , we allowed also for dependence between the units themselves , that generalises previous exchangeable formulations . results obtained both from simulations and from an application in neuroscience demonstrate that joint estimation of dags can offer increased statistical efficiency relative to independent estimation . however we observed that the relationships between the units themselves were difficult to estimate from data . in the neuroscience application we illustrated how sophisticated likelihood models such as mdms can be used within our framework . importantly , our methodology requires only that local scores are pre - computed and cached ; it is therefore possible to apply exact algorithms retrospectively , following independent analyses of individual datasets , without the need to recompute local scores . in practice we anticipate that the rationale and appropriateness of joint modelling will need be assessed by domain experts on a case - by - case basis . our computations were performed using the gobnilp package @xcite , that includes sophisticated routines for estimation of individual dags ( see supplementary materials ) . nevertheless our simulation results for large numbers @xmath14 of variables and @xmath39 of units required considerable computational time ( in a worst - case scenario ) . it is becoming commonplace in many applications for the number of units to become very large . it remains to be seen whether our approach can scale up to larger problems . recent work where gobnilp was able to find map dags with 1614 nodes in between 3 and 42 minutes are encouraging @xcite . in practice , as discussed in @xcite , it is very hard to estimate the time required to solve an ilp in advance ; rather this depends in a highly non - trivial way on the details of the problem and is related to the `` phase - transition '' in sat solving . we note that the results of this paper are based on the ilp framework of @xcite for individual dags . an alternative ilp formulation , known as `` characteristic imsets '' , was proposed by @xcite specifically for bns . characteristic imsets are closely related to the essential graph of a bn @xcite and have the property that markov - equivalent bns are score equivalent . in contrast , the approaches we pursued involved a non - unique representation of the essential graph . this flexibility is important when dealing with general dag models including mdms , that are uniquely identifiable from data , but are less natural in the context of bns . at present , the computational performance of characteristic imsets on individual bns is inferior to the approach of @xcite pursued here @xcite . in addition it is currently unclear how prior structural information might be incorporated into that framework . on a related note , it would be interesting to design a joint version of the well - known pc algorithm @xcite that complements score - based estimation . interesting extensions include : 1 . * information sharing for parameters . * in many applications it is reasonable to assume similarity of parameter values @xmath27 between units . at present this appears to be challenging to include within our framework and represents an area for further research . * computation . * the applications discussed in this paper were performed on a single cpu and limited in computational intensity to approximately @xmath133 . since many emerging datasets , not least in biomedical applications , contain many more variables @xmath14 and many more units @xmath39 than we considered here , it would be interesting to investigate strategies for parallel and/or approximate computation that scale better to such regimes . in the other direction , our methodology applies to complex local scores that may be estimated numerically ; it would be interesting to gain a theoretical understanding of how any uncertainty in the estimates for the local scores would impact on the map estimators described here . * new statistical models . * by fixing the network @xmath42 in our general framework we can obtain a number of interesting statistical models : ( i ) taking @xmath42 to be a chain allows us to associate units with a temporal ordering and thereby model time - evolving dag structures . ( ii ) introducing auxiliary dags @xmath150 and enforcing @xmath42 to be bipartite on the partition of auxiliary / non - auxiliary dags produces a mixture model for dags , such that each latent dag is a graphical summary of the various dag structures within its cluster . ( iii ) extending this idea , with @xmath151 we recover a mixture model for dags similar to @xcite . these possibilities are currently being explored , with preliminary results for mixture models presented in @xcite . preliminary indications suggest that imposing additional restrictions on @xmath42 in this way may serve to improve the identification of higher - order structure . more generally , it is straightforward to extend our methods to the case of decomposable undirected models by imposing additional constraints that rule out immoralities . * supplementary materials : * the supplementary materials include additional a discussion of the simulated data and the actual code used to produce the results herein . * acknowledgements : * cjo was supported by the centre for research in statistical methodology ( crism ) epsrc ep / d002060/1 . jc was supported by the medical research council ( project grant g1002312 ) . sm was supported by the uk medical research council and is a recipient of a royal society wolfson research merit award . the authors are grateful to lilia carneiro da costa and tom nichols who collaborated in the analysis of fmri data and to mark bartlett who provided technical support with gobnilp . the authors also thank diane oyen and several other colleagues who provided feedback on an earlier draft . 99 acid , s. , de campos , l.m . : searching for bayesian network structures in the space of restricted acyclic partially directed graphs . _ j. artif . res . _ * 18 * , 445 - 490 ( 2003 ) . achterberg , t. : _ constraint integer programming_. phd thesis , berlin institute of technology ( 2007 ) . achterberg , t. : scip : solving constraint integer programs . _ mathematical programming computation _ * 1*(1 ) , 1 - 41 ( 2009 ) . bartlett , m. , cussens , j. : advances in bayesian network learning using integer programming . _ proceedings of the 29th conference on uncertainty in artificial intelligence _ , 182 - 191 ( 2013 ) . chickering , d.m . : optimal structure identification with greedy search . _ j. mach . res . _ * 3 * , 507 - 554 ( 2003 ) . consonni , g. , la rocca , l. : moment priors for bayesian model choice with applications to directed acyclic graphs . _ bayesian statistics _ * 9*(9 ) , 119 - 144 ( 2010 ) . costa , l. , smith , j.q . , nicholls , t. : on the selection of multiregression dynamic models of fmri networked time series . _ crism working paper , university of warwick _ * 13 * , 6 ( 2013 ) . cowell , r.g . : efficient maximum likelihood pedigree reconstruction . biol . _ * 76 * , 285 - 291 ( 2009 ) . cussens , j. : maximum likelihood pedigree reconstruction using integer programming . _ proceedings of the workshop on constraint based methods for bioinformatics ( wcb-10 ) , edinburgh _ ( 2010 ) . cussens , j. : bayesian network learning with cutting planes . _ proceedings of the 27th conference on uncertainty in artificial intelligence _ , 153 - 160 ( 2011 ) . danaher , p. , wang , p. , witten , d.m . : the joint graphical lasso for inverse covariance estimation across multiple classes . _ j. r. statist . b _ * 76*(2 ) , 373 - 397 ( 2014 ) . dawid , a.p . , lauritzen , s.l . : hyper markov laws in the statistical analysis of decomposable graphical models . stat . _ * 21*(3 ) , 1272 - 1317 ( 1993 ) . dawid , a.p . : beware of the dag ! _ j. mach . w & cp _ * 6 * , 59 - 86 ( 2010 ) . ellis , b. , wong , w.h . : learning causal bayesian network structures from experimental data . assoc . _ * 103*(482 ) , 778 - 789 ( 2008 ) . friedman , n. , koller , d. : being bayesian about network structure : a bayesian approach to structure discovery in bayesian networks . _ * 50*(1 - 2 ) , 95 - 126 ( 2003 ) . friedman , j. , hastie , t. , tibshirani , r. sparse inverse covariance estimation with the graphical lasso . _ biostatistics _ * 9*(3 ) , 432 - 441 ( 2008 ) . friston , k.j . : functional and effective connectivity : a review . _ brain connectivity _ * 1*(1 ) , 13 - 36 ( 2011 ) . he , y. , jia , j. , yu , b. : reversible mcmc on markov equivalence classes of sparse directed acyclic graphs . _ ann . _ * 41*(4 ) , 1742 - 1779 ( 2013 ) . heckerman , d. , geiger , d. , chickering , d.m . : learning bayesian networks : the combination of knowledge and statistical data . _ _ * 20*(3 ) , 197 - 243 ( 1995 ) . hill , s. , lu , y. , molina , j. , heiser , l.m . , spellman , p.t . , speed , t.p . , gray , j.w . , mills , g.b . , mukherjee , s. : bayesian inference of signaling network topology in a cancer cell line . _ bioinformatics _ * 28*(21 ) , 2804 - 2810 ( 2012 ) . jaakkola , t. , sontag , d. , globerson , a. , meila , m. : learning bayesian network structure using lp relaxations . _ proceedings of the 13th international conference on artificial intelligence and statistics _ , 358 - 365 ( 2010 ) . li , j. , wang , z.j . , palmer , s.j . , mckeown , m.j . : dynamic bayesian network modeling of fmri : a comparison of group - analysis methods . _ neuroimage _ * 41*(2 ) , 398 - 407 ( 2008 ) . l . , wainwright , m.j . : structure estimation for discrete graphical models : generalized covariance matrices and their inverses stat . _ * 41*(6 ) , 3022 - 3049 ( 2013 ) . mahajan , a. : presolving mixed - integer linear programs . _ wiley encyclopedia of operations research and management science _ ( 2010 ) . mechellia , a. , penny , w.d . , pricea , c.j . , gitelman , d.r . , friston , k.j . : effective connectivity and intersubject variability : using a multisubject network to test differences and commonalities . _ neuroimage _ * 17*(3 ) , 1459 - 1469 ( 2002 ) . meinshausen , n. , bhlmann , p. : high - dimensional graphs and variable selection with the lasso . stat . _ * 34*(3 ) , 1436 - 1462 ( 2006 ) . nemhauser , g.l . , wolsey , l.a . : _ integer and combinatorial optimization . _ wiley , new york ( 1988 ) . niculescu - mizil , a. , caruana , r. : inductive transfer for bayesian network structure learning . _ j. mach . workshop and conference proceedings _ * 27 * , 339 - 346 ( 2007 ) . oates , c.j . , mukherjee , s. : joint structure learning of multiple non - exchangeable networks . _ proceedings of the 17th international conference on artificial intelligence and statistics _ * 33 * , 687 - 695 ( 2014 ) . oates , c.j . , korkola , j. , gray , j.w . , mukherjee , s. : joint estimation of multiple networks from time course data . _ * 8*(3 ) , 1892 - 1919 ( 2014a ) . oates , c.j . , carneiro da costa , l. , nichols , t. : towards a multi - subject analysis of neural connectivity . _ neural compt . _ * 27 * , 1 - 20 ( 2014b ) . oates , c.j . , papamarkou , t. , girolami , m. : the controlled thermodynamic integral for bayesian model comparison . _ crism working paper series , university of warwick _ , * 14 * , 9 ( 2014c ) . oates , c.j . , girolami , m. , chopin , n. : control funtionals for monte carlo integration . _ crism working paper series , university of warwick _ , * 14 * , 22 ( 2014d ) . oates , c.j . , amos , r. , spencer , s.e.f . : quantifying the multi - scale performance of network inference algorithms . mo . b. _ * 13*(5 ) , 611 - 631 ( 2014e ) . oyen , d. , lane , t. : leveraging domain knowledge in multitask bayesian network structure learning . _ proceedings of the 26th aaai conference on artificial intelligence _ ( 2012 ) . oyen , d. , lane , t. : bayesian discovery of multiple bayesian networks via transfer learning . _ ieee 13th international conference on data mining ( icdm ) _ , 577 - 586 ( 2013 ) . pearl , j. : _ causality : models , reasoning and inference ( 2nd ed.)_. cambridge university press ( 2009 ) . penfold , c.a . , buchanan - wollaston , v. , denby , k.j . , wild , d.l . : nonparametric bayesian inference for perturbed and orthologous gene regulatory networks . _ bioinformatics _ * 28*(12 ) , i233-i241 ( 2012 ) . perrier , e. , imoto , s. , miyano , s. : finding optimal bayesian network given a super - structure . _ j. mach . * 9*(10 ) , 2251 - 2286 ( 2008 ) . peters , j. bhlmann , p. : structural intervention distance ( sid ) for evaluating causal graphs . _ neural compt . _ , to appear . poldrack , r.a . , mumford , j.a . , nichols , t.e . : _ handbook of fmri data analysis _ , cambridge university press ( 2011 ) . queen , c.m . , albers , c.j . : intervention and causality : forecasting traffic flows using a dynamic bayesian network . _ j. am . _ * 104*(486 ) , 669 - 681 ( 2009 ) . queen , c.m . , smith , j.q . : multiregression dynamic models . _ j. r. statist . b _ * 55*(4 ) , 849 - 870 ( 1993 ) . scott , j.g . , berger , j.o . : bayes and empirical - bayes multiplicity adjustment in the variable - selection problem . stat . _ * 38*(5 ) , 2587 - 2619 ( 2010 ) . sheehan , n.a . , bartlett , m. , cussens , j. : maximum likelihood reconstruction of very large pedigrees . _ theoretical population biology _ , in press . silander , t. , myllymkki , p. : a simple approach to finding the globally optimal bayesian network structure . _ proceedings of the 22nd conference on artificial intelligence _ , 445 - 452 ( 2006 ) . spirtes , p. , glymour , c. , scheines , r. : causation , prediction , and search . the mit press , 2nd edition ( 2000 ) . studen , m. , vomlel , j. , hemmecke , r. : a geometric view on learning bayesian network structures . j. approx . reason . _ * 51*(5 ) , 578 - 586 ( 2010 ) . studen , m. , haws , d. : on polyhedral approximations of polytopes for learning bayesian networks . _ j. algebraic stats . _ * 4*(1 ) , 59 - 92 ( 2013 ) . sugihara , g. , kaminaga , t. , sugishita , m. : interindividual uniformity and variety of the writing center " : a functional mri study . _ neuroimage _ * 32*(4 ) , 1837 - 1849 ( 2006 ) . thiesson , b. , meek , c. , chickering , d. m. , heckerman , d. : learning mixtures of bayesian networks . _ proceedings of the 14th conference on uncertainty in artificial intelligence _ , 504 - 513 ( 1998 ) . tsamardinos , i. , brown , l.e . , aliferis , c.f . : the max - min hill - climbing bayesian network structure learning algorithm . _ * 65*(1 ) , 31 - 78 ( 2006 ) . van essen , d.c . , smith , s.m . , barch , d.m . , behrens , t.e . , yacoub , e. , ugurbil , k. : the wu - minn human connectome project : an overview . _ neuroimage _ * 80 * , 62 - 79 ( 2013 ) . werhli , a.v . , husmeier , d. : gene regulatory network reconstruction by bayesian integration of prior knowledge and/or different experimental conditions . _ journal of bioinformatics and computational biology _ * 6*(3 ) , 543 - 572 ( 2008 ) . wolsey , l.a . : _ integer programming_. wiley , new york ( 1998 ) . yuan , c. , malone , b. : learning optimal bayesian networks : a shortest path perspective . _ j. artif . res . _ * 48 * , 23 - 65 ( 2013 ) . * supplement to `` exact estimation of multiple directed acyclic graphs '' * chris . j. oates ( e - mail : _ c.oates@warwick.ac.uk_ ) and jim q. smith ( e - mail : _ j.q.smith@warwick.ac.uk_ ) , department of statistics , university of warwick , coventry , cv4 7al , uk . sach mukherjee ( e - mail : _ sach@mrc-bsu.cam.ac.uk_ ) , mrc biostatistics unit and school of clinical medicine , university of cambridge , cb2 0sr , uk . james cussens ( e - mail : _ james.cussens@cs.york.ac.uk_ ) , department of computer science and york centre for complex systems analysis , university of york , yo10 5ge , uk . in simulations we sampled the network @xmath42 directly from the prior @xmath152 and then sampled dags @xmath44 from the prior conditional @xmath128 using discrete - state - space mcmc . lines 2 - 12 in algorithm [ mcmc ] provide pseudo - code for the metropolis - hastings scheme that was employed . fig . [ diagnose ] contains supporting convergence diagnostics . as explained in the main text , the local likelihood @xmath153 is arbitrary and its choice may influence the properties of the joint estimators that we wish to study . we therefore took the most default approach of directly simulating the evidence terms @xmath154 in eqn . [ evidence ] from log - normal distributions . in this way we hope to obtain results that apply quite generally , without the potential for likelihood - specific anomalies . for the aic we assumed that @xmath155 so that each edge is associated with one free parameter as in , for example , a linear structural equation model . specifically , we appealed to the intuition that in many applications only a subset of models will be supported by data . we therefore generated the log - evidence terms independently according to the mixture model @xmath156 , whilst the log - evidence of the true data - generating model was always simulated from @xmath132 . this ensures that the score @xmath131 of the data - generating model belongs on average to the top @xmath157 of all scores , whilst allowing us to discard @xmath158 of all models and thereby reduce the computational burden in this simulation study . all simulations were based on a value @xmath159 . to further mediate computational complexity we imposed an in - degree restriction @xmath160 for all simulation experiments ; this could be relaxed at additional computational effort . ( note that in the `` worst case '' simulations in the main text , all models were assigned log - scores from @xmath132 . ) in principle , suitable values for the hyper - parameters will scale with both @xmath14 and @xmath39 . to mitigate this effect and to allow us to employ the same candidate values @xmath129 , @xmath130 in each experiment , we rescaled the local scores @xmath131 by a factor of @xmath161 . full pseudo - code for the simulation of the local scores @xmath131 is provided in lines 13 - 24 of algorithm [ mcmc ] .

we fit axisymmetric 3-integral dynamical models to ngc 3379 using the line - of - sight velocity distribution obtained from hst / fos spectra of the galaxy center and ground - based long - slit spectroscopy along four position angles , with the light distribution constrained by wfpc2 and ground - based images . we have fitted models with inclinations from 29 ( intrinsic galaxy type e5 ) to 90 ( intrinsic e1 ) and black hole masses from 0 to . the best - fit black hole masses range from , depending on inclination . the preferred inclination is 90 ( edge - on ) ; however , the constraints on allowed inclination are not very strong due to our assumption of constant _ m / l_ . the velocity ellipsoid of the best model is not consistent with either isotropy or a two - integral distribution function . along the major axis , the velocity ellipsoid becomes tangential at the innermost bin , radial in the mid - range radii , and tangential again at the outermost bins . the rotation rises quickly at small radii due to the presence of the black hole . for the acceptable models , the radial to tangential ( ( dispersion in the mid - range radii ranges from , with the smaller black holes requiring larger radial anisotropy . compared with these 3-integral models , 2-integral isotropic models overestimate the black hole mass since they can not provide adequate radial motion . however , the models presented in this paper still contain restrictive assumptions namely assumptions of constant _ m / l_ and spheroidal symmetry requiring yet more models to study black hole properties in complete generality .

astro-ph9912026

the dynamical state and mass distribution in the central regions of elliptical galaxies provide clues to the formation and evolutionary histories of these galaxies ( see merritt 1999 for a review ) . consequently , the existence of central black holes has been the target of intense scrutiny ( see kormendy & richstone 1995 for a review ) . in this paper , we study the elliptical galaxy ngc 3379 using both _ hubble space telescope _ and ground - based photometric and kinematic data in order to understand its central dynamics . at ground - based resolution ngc 3379 is a prototypical elliptical ( e1 ) galaxy with m@xmath6 ( faber _ et al . _ 1997 ) at a distance of 10.4 mpc ( ajhar _ et al . _ lauer _ et al . _ ( 1995 ) classify it as a `` core '' galaxy a galaxy that has a break in the surface brightness profile but still maintains a rising profile into the smallest measured radius . faber _ et al . _ ( 1997 ) hypothesize that core galaxies are associated with massive black holes . however , few core galaxies have strong black hole detections : m87 ( harms _ et al . _ 1994 ) , m84 ( bower _ et al . _ 1998 ) , and n4261 ( ferrarese _ et al . _ . these detections result from gas dynamics , not stellar dynamics . stellar dynamical evidence for central black holes in core galaxies is difficult to obtain since ( 1 ) core galaxies have low surface brightnesses , making kinematic observations difficult , ( 2 ) the lack of rotation complicates the dynamical modeling since velocity anisotropies potentially govern the dynamical support , and ( 3 ) stars near the center travel out to the core radius so their contribution to the central velocity profile is de - weighted ( kormendy 1992 ) . ngc 3379 was chosen for this study because of its relatively high central surface brightness among core galaxies , and because of its otherwise normal morphological and dynamical structure . the proximity , brightness , and normalcy of ngc 3379 make it the object of many ground - based photometric ( de vaucouleurs & capaccioli 1979 , lauer 1985 , capaccioli _ et al . _ 1987 , capaccioli _ et al . _ 1991 ) and kinematic ( kormendy 1997 , kormendy 1985 , van der marel _ et al . _ 1990 , bender _ et al . _ 1994 , statler 1994 , statler & smecker - hane 1999 ) studies . however , previous models for ngc 3379 have yielded inconsistent results . assuming that the departure of the light profile from the @xmath7 law signifies the existence of a stellar disk , capaccioli _ et al . _ ( 1991 ) find that the most likely model is nearly face - on with an inclination of 31@xmath0 ( intrinsic e5 galaxy ) . their conclusion was that ngc 3379 and ngc 3115 have the same intrinsic shape but are seen from different viewing angles . in their model for ngc 3379 the central region is nearly oblate , but at about @xmath8 ( @xmath9 ) the model becomes triaxial . no dynamical information was used in this model . a major uncertainty is that slight departures from a @xmath7 law , on which their model is based , may be due to tidal effects ( kormendy 1977 ) instead of a disk . furthermore , there is no reason to expect galaxies to follow a @xmath7 law exactly . kinematic information is required to determine unambiguously the inclination of a spheroidal system . van der marel _ et al . _ ( 1990 ) , using the ground - based velocity data of davies & birkinshaw ( 1988 ) and franx _ et al . _ ( 1989 ) , conclude that the inclination of ngc 3379 is 60@xmath0 ; however the two datasets are not consistent and give slightly different velocity anisotropy parameters . van der marel _ et al . _ used 2-integral flattened models with a parametrized form for the anisotropy . no goodness - of - fit criteria were given , and thus we are not able to compare their results with those of others . statler ( 1994 ) , using bayesian statistics on the same datasets , prefers models that have inclination greater than 45@xmath0 . statler rejects capaccioli s conclusion that the galaxy is flattened and triaxial at the 98% confidence level , but his results could accommodate those of van der marel _ et al . _ statler uses only the mean velocity measured along four position angles ; the radial variation in the velocity or the dispersion are not considered . furthermore , statler uses data only outside 15 . however , statler has obtained new , high s / n data ( statler & smecker - hane 1999 ) , which he will present in a future paper . the studies above use low s / n , sparse kinematic data , or no kinematic data at all . in this paper , we report results from observations of the full line - of - sight velocity distribution ( losvd ) at various position angles from high s / n ground - based data and data from hst . the data are fit to axisymmetric , three - integral orbit - based models . richstone _ et al . _ ( 1999 ) describe the modeling technique in a companion paper . we exploit the full losvds _ directly _ in the modeling . alternatively , we could use the moments of the losvds ; however , the full shape of the losvd contains important information on the underlying dynamics , and it is desirable to incorporate the full losvd when the spectra have sufficient signal - to - noise to determine this reliably . in fact , our result for ngc 3379 critically depends on the shape of the central losvd . cretton _ et al . _ ( 1999 ) and van der marel _ et al._(1998 ) present a very similar modeling technique , where the only difference from ours is that we fit the full losvd as opposed to their moment fitting procedure . there are other programs that also use the higher moments of the losvd as constraints ( dejonghe , 1987 ; gerhard 1993 ; rix _ et al . _ 1997 ) . obtaining the full losvd at many position angles is the maximum obtainable kinematic information and has now become routine ( _ e.g. _ bender _ et al . _ 1994 , carollo _ et al . _ 1995 , gebhardt & richstone 1999 ) . in 2 , we describe the photometric and kinematic data for ngc 3379 . in 3 , we describe the models and present results in 4 . discussion is given in 5 . we obtained hubble space telescope ( hst ) observations of ngc 3379 , consisting of four 500s f555w ( @xmath10-band ) and four 400s f814w ( @xmath11-band ) wfpc2 images , on 1994 november 19 . centering the galaxy in the high - resolution pc1 camera ( 0.0455 pixel@xmath12 ) provided a total signal in the central pixel of @xmath13 in both @xmath10 and @xmath11 . we deconvolved the summed images with 40 lucy - richardson iterations . the surface brightness , ellipticity , and position angle of the major axis were obtained using the same procedure as in lauer _ et al . _ ( 1995 ) and are presented in table 1 . 0.2 cm 0.3 cm the hst profile reaches only about 12 from the center and must be extended to larger radii using ground - based photometry . peletier _ et al . _ ( 1990 ) have obtained ground - based @xmath14 photometry for ngc 3379 over the radius range 4170 , and capaccioli _ et al . _ ( 1990 ) have @xmath15 photometry out to 500 ; we must therefore estimate the transformations to @xmath10 and @xmath11 using the overlap region with the hst data . the data from peletier _ et al . _ and capaccioli _ et al . _ agree well , and since peletier _ et al . _ provide several colors , we use their results . the colors in the overlap region , based on the average differences there , are @xmath16 and @xmath17 . we have ignored possible variations in the colors as a function of radius . using these colors , we combine the hst and ground - based data to provide the full profile . the left panel in fig . 1 shows the surface brightness profile in @xmath10 along the major axis from 0.02170 . the three panels in fig . 2 show the ellipticity , the position angle of the major axis ( measured from north through east ) , and the color @xmath18 as functions of radius along the major axis . the random uncertainties can be estimated from the local scatter . the dynamical modeling described below assumes axisymmetry . we must therefore determine the best axisymmetric density distribution consistent with the observed surface brightness distribution . strictly speaking , there is none , because , as shown in figure 2 , there are abrupt changes in the ellipticity and position angle at radii less than 1 and larger than 100 . pa changes in particular are normally attributed to variation in the axis ratios with radius in a triaxial distribution , or to real three - dimensional twists in locally axisymmetric objects in which surfaces of constant density , while axisymmetric , are not co - axial . 0.2 cm 0.3 cm however , for several reasons we believe that , at radii less than 100 , an axisymmetric model will suffice for our purposes . first , in the middle range of radii , @xmath19 , the ellipticity and pa are fairly constant at 0.1 and 72 @xmath20 5@xmath0 . second , in the central region there is a dust ring with radius 1 ( discussed below and seen in fig . 5 ) . the absorption by the dust ring 1% of the galaxy surface brightness is enough to affect estimates of both the ellipticity and the position angle . furthermore , the shallow central slope ( dlog @xmath21/dlog @xmath22 ) of the inner profile and the small ellipticity ( @xmath23 ) render the position angle at small radii ( @xmath24 ) highly uncertain . even if the pa gradient seen in fig 2 is real and well determined , the amount of light , and therefore the gravitational impact , of the associated departure from axisymmetry is small . finally , at large radii , departures from axisymmetry have little impact on the central gravitational field . the data thus suggest that ngc 3379 can be reasonably well represented by a spheroid of constant ellipticity . the deprojection of the surface brightness is then unique for a given inclination . the deprojection is based on a non - parametric estimate of the density using smoothing splines ( gebhardt _ et al . _ the right panel of fig . 1 plots the luminosity density for three different inclinations ( 90@xmath0 is edge - on ) . for different inclinations , and hence different flattenings , the luminosity density scales as the ratio of the apparent to the true axis ratio . we did not examine models in which the galaxy is assumed to be axisymmetric but not spheroidal ( _ i.e. _ , with varying ellipticity ) , in part because in this case the deprojection is not unique . romanowsky & kochanek ( 1997 ) determine the uncertainties in the deprojection of an axisymmetric system for various inclinations , and the uncertainties can be quite large for low inclinations ( close to face - on ) . we could test the effect of our spheroidal assumption by modeling the most extreme situation : a galaxy which is seen nearly face - on and containing an embedded disk ( similar to the suggestion of capaccioli _ et al . _ the most likely effect is to change to inclination and _ m / l_@xmath3 constraints , and relaxing the spheroidal assumption should have little effect on the measured black hole mass . this issue , however , will not be addressed here . ngc 3379 was observed on 1995 , january 23 27 with the 0.21square aperture ( `` 0.25-pair '' ) of the fos . the spectrum was taken in two visits ; the total integration time was 215.5 minutes . the wavelength range , 45666815 , includes the mg i _ b _ lines at @xmath25 and the na i d lines at the spectrum was electronically quarter - stepped , giving 2064 1/4-diode pixels . the reciprocal dispersion was 1.09 pixel@xmath12 . the instrumental velocity dispersion was measured to be @xmath26fwhm/2.35@xmath27 pixels = @xmath28 @xmath29 ( internal error ) . this width is intrinsic to the instrument and is not strongly affected by how the aperture is illuminated ( keyes _ et al . _ 1995 ) . we therefore make no aperture illumination corrections to the measured velocity dispersion . flat fielding and correction for geomagnetically induced motions ( gim ) were done as in kormendy _ et al . _ ( 1996 ) . the flat - field image was taken with the same 0.25-pair aperture used for ngc 3379 . the galaxy exposure was divided into 3 and 4 subintegrations for the two visits . because gim shifts between subintegrations are small , unlike the large shifts between galaxy exposures and the flat field , we first averaged all the exposures and used cross - correlation to determine the shift between this average and the flat field . the measured shift was 2.27 pixel . each subintegration was multiplied by the flat - field frame shifted by the above amount . one dead diode was corrected by dividing the affected pixel intensities by 0.8 . then gim shifts between subintegrations were determined ; most shifts were @xmath30 pixels , with the largest being 0.8 pixels . each subintegration was then shifted to agree with the subintegration taken closest in time to the comparison spectrum exposure . finally , the spectra were added , weighted by the exposure times . obtaining the internal kinematic information requires a deconvolution of the observed galaxy spectrum with a representative set of template stellar spectra . we deconvolve the spectrum using a maximum - penalized likelihood ( mpl ) estimate that obtains a non - parametric line - of - sight velocity distribution ( losvd ) . the fitting technique proceeds as follows : we choose an initial velocity profile in bins . we convolve this profile with a weighted - averaged template ( discussed below ) and calculate the residuals to the galaxy spectrum . the program varies the velocity profile parameters bin heights and the template weights to provide the best match to the galaxy spectrum . the mpl technique is similar to that used in saha & williams ( 1994 ) , and merritt ( 1997 ) . however , in constrast to these authors we simultaneously fit the velocity profile and template weights . we impose smoothness on the losvd by the addition to the @xmath31 of a penalty function , which is the integrated squared second derivative ( details are in merritt 1997 and gebhardt & richstone 1999 ) . we estimate the best smoothing from bootstrap simulations ( described below ) . two templates , dsa107 - 442 ( a field k0iii star ) and n188-i69 ( a k2iii star in ngc 188 ) , are used , both observed with the fos . star n188-i69 is the preferred template , as the fitting routine gives essentially all of the weight to this star . however , the losvd determined using only the other star is similar , suggesting that the measurement of the losvd is robust . the results from this maximum likelihood technique were compared to results of a fourier correlation technique ( bender 1990 ) , and no significant differences were found . the losvds obtained in the two visits by hst agree to within their uncertainties , and also the pointings from the two visits agree to within 0.01 . therefore , we sum the two spectra weighted by exposure time to derive a combined losvd . a parameterization of the losvd by gauss - hermite moments ( van der marel & franx 1993 , gerhard 1993 ) gives for the first four moments @xmath32 @xmath29 , @xmath33 @xmath29 , @xmath34 , and @xmath35 . the velocity zero - point is arbitrary and will be discussed in 4 . all error bars correspond to the 68% confidence band . in the lower part of figure 3 we plot the template spectrum ; the noisy upper line is the averaged fos spectrum for ngc 3379 ( shifted by its redshift ) , and the smooth upper line is the template spectrum convolved with the derived losvd . the continuum has been divided out of both template and galaxy spectra . the spectrum shown here is a region around the mg i b triplet ( 5175 ) and not the full spectrum obtained by the fos . 0.2 cm 0.3 cm the 68% confidence band for the losvd is determined using a bootstrap approach . we convolve the template star with the measured losvd to provide the initial galaxy spectrum ( the upper smooth line in fig . 3 ) ; from that initial galaxy spectrum , we then generate 100 realizations and determine the losvd each time . each realization contains randomly chosen flux values at each wavelength position drawn from a gaussian distribution , with the mean given by the initial galaxy spectrum and the standard deviation given by the root - mean - square of the initial fit . the average root - mean - square for the continuum - divided spectrum is 0.033 . the 100 realizations of the losvds provide a distribution of values for every losvd bin , from which we estimate the confidence bands . 0.2 cm -0.8 cm 0.3 cm figure 4 presents the losvd with the 68% confidence band ( solid and dotted lines ) . the losvd is significantly skewed with a tail to positive velocities a feature also detected using the fourier correlation technique . the skewness is crucial for the following analysis and must be considered carefully . in an axisymmetric and even a triaxial system , the _ central _ losvd is symmetric . various attempts were made to get rid of the skewness by reducing the data differently ; we used both templates individually , we used flattened and un - flattened spectra , we used the mg b lines and the na doublet separately and combined , different shifts were used to combine the spectra , and the spectra from both peak - ups were reduced separately . in every case , the skewness remained . since the skewness appears to be real , we must conclude that the fos was not centered , that part of the galaxy light is being obscured , that the galaxy is not symmetric near the center ( possibly because of an offset nucleus ) , or any combination of these affects . if the nucleus is offset , we can not successfully model ngc 3379 as an axisymmetric system . however , there is no evidence to suggest that ngc 3379 has an offset nucleus since the isophotes , except for the dust ring , are concentric . figure 5 is an image of the central 5 of the galaxy with the spheroidal component subtracted . we include in fig . 5 the exact position of the fos aperture . to determine the exact position of the aperture , we have two independent checks : from the values used to peak - up on the galaxy center , and from the acquisition image taken immediately before the galaxy observations . for the peak - up procedure , hst steps by 0.065 and points the fos at the highest pixel value . examination of the peak - up numbers reveals that the likely miscentering is 0.033 , since the center of the peak - up numbers lies roughly between two bins in the peak - up array . this miscentering is also confirmed from the acquisition image which places the fos aperture the same distance from the galaxy center . 0.2 cm -0.8 cm 0.3 cm in order to explain the losvd skewness , the direction of the aperture offset must correspond appropriately . hst placed the aperture on the eastern side of the galaxy , along the major axis . thus we acquire more light from the eastern side relative to the western side . the ne side of the galaxy has negative rotation velocities ( davies & birkinshaw 1988 ) . if one side contributes more light and there is significant rotation , the _ mode _ and mean of the distribution shifts towards that side in this case the ne side with negative velocity and the other side still contributes an extensive tail toward positive velocities , thus generating the observed skewness in the losvd . a further source for the losvd skewness may be that part of the galaxy light is obscured . a dust ring is clearly visible , inclined by about 45@xmath0 from the major axis ( the solid line in fig . 5 ) . from the relative depths of the obscuration on opposite sides , it appears that the sw side of the ring is on the near side of the galaxy , while the ne side is on the far side . this configuration means that the crucial signal from stars near the nucleus is preferentially obscured on the sw side , thus leading to the same effect as the miscentering ; the western side contributes less light than the eastern side causing positive velocity skewness . the miscentering of the aperture results from two effects , errors in the peak - up procedure and errors due to the dust obscuration . to calculate the miscentering from dust , we find the best center by using the galaxy light beyond the dust ring and compare this with the fos center . hst points fos to the highest position in a 0.21aperture on a @xmath36 raster with spacing of 0.065 . the dust does not appear to have a significant effect on the location of the brightest pixel ; it causes a miscentering of only 0.01 . however , errors in the peak - up procedure are more significant , and as mentioned above , the miscentering can be 0.033 . we conclude that the maximum miscentering is 0.04 ( peak - up errors + dust miscentering ) . we then place the fos aperture in the wfpc2 image at a position 0.04 east from the galaxy center and determine the relative amount of light in the aperture from opposite sides of the galaxy . at this position , the light from the ne side of the galaxy is twice as strong as from the sw side . thus , in the central bin of our model , we include all of the light from the ne side and only half from the sw side . an alternative is to use the best symmetrized losvd in the modeling and assume equal contributions from both sides of the galaxy . the dashed line in fig . 4 is the maximum penalized likelihood estimate of the best symmetric losvd . results from both assumptions are discussed below . ground - based spectroscopic data are available from gebhardt & richstone ( 1999 ) ( along four position angles ) , statler & smecker - hane ( 1999 ) ( along four position angles ) , bender _ et al . _ ( 1994 ) ( major axis only ) , gonzalez ( 1994 ) ( major and minor axes ) , franx _ et al . _ ( 1989 ) ( major and minor axes ) , and davies & birkinshaw ( 1988 ) ( major , minor , and 30@xmath0 axes ) . the data from franx _ et al . _ and davies & birkinshaw have lower s / n and will not be considered here . the data from the first four sources all have high signal - to - noise ( s / n ) and in general agree with one another ( see gebhardt & richstone for a detailed comparison ) , with the largest disagreement coming from the large radii data . for the data of gebhardt & richstone , the position angles observed and the radial extraction of the galaxy spectra were designed to correspond exactly with the binning used in the modeling ( described in the next section ) , and we use their data . we use 3-integral axisymmetric models ( richstone _ et al . _ the models are orbit - based ( schwarzschild 1979 ) and constructed using maximum entropy ( richstone & tremaine 1988 ) . we run a representative set of orbits in a specified potential , and determine the non - negative weights of the orbits to best fit the available data . maximizing the entropy helps to provide a smooth phase - space distribution function . rix _ et al . _ ( 1997 ) present a similar code which has been applied to the spherical case ; cretton _ et al . _ ( 1999 ) and van der marel _ et al . _ ( 1998 ) developed a fully - general axisymmetric code that is also orbit - based . the difference between our code and theirs is due to how we impose smoothness and how we fit to observational data . smoothness in our code comes from maximizing entropy , whereas the other authors enforce smoothness of the distribution function directly by minimizing its variation . the most important difference is that we fit the losvd directly , and both rix _ et al . _ and cretton _ et al . _ fit moments of the velocity profile . using the losvd precludes the need to approximate the velocity profile with a parametric function both in the observations and in the orbital libraries . the surface brightness yields an estimate of the luminosity density distribution as explained above ( 2.1 ) , assuming that the luminosity density is axisymmetric on similar spheroids and that the galaxy has a given inclination . to obtain the mass density distribution , we have made one of two assumptions about the mass - to - light ratio ( _ m / l_@xmath3 ) : ( _ i _ ) constant _ m / l_@xmath3 , ( _ ii _ ) _ m / l_@xmath3 that varies according to the @xmath18 color shown in fig . 2 . ngc 3379 is redder in the center by @xmath37 . the analysis of gonzalez ( 1994 ) indicates that this color change is due mainly to metallicity ( _ i.e. _ , no age effects ) . in this case , models of worthey ( 1994 ) predict that _ m / l_@xmath3 may be as much as 25% higher in the center . the @xmath18 variation in fig . 2 is approximately log - linear , so we vary _ m / l_@xmath3 by 25% from the outer to inner radii following the same functional form . finally , we add a central point mass and compute the potential . using the derived potential , we follow orbits which sample the available phase space in energy ( @xmath38 ) , angular momentum ( @xmath39 ) , and the third integral ( @xmath40 ) ( richstone _ et al . _ each orbit crosses the equatorial plane 40 times . binning the orbital distributions both in real space and velocity , and in projected distance and line - of - sight velocity , provides an estimate of the contribution to the losvd from each orbit at each spatial bin . both the real and projected spatial binning are approximately logarithmic in radius and linear in cos @xmath41 , where @xmath41 is measured from the pole . this binning scheme provides approximately equal - mass bins . since we are using axisymmetric modeling , we follow orbits only in @xmath42 and @xmath41 . we therefore have to choose the azimuthal angle for the projection , and , since all angles are equally likely , we choose that angle at random . for every time step , 100 azimuthal angles are chosen . we measure an orbit s contribution on a grid of 80 radial bins , 20 angular bins , and 13 velocity bins compressed to @xmath43 when comparing to the data . the size of the central bin equals the aperture used in the fos ( 0.21 ) . the outermost bin ranges from 149 to 200 . the centers of the five angular bins up from the major axis are at 6@xmath0 , 18@xmath0 , 30@xmath0 , 45@xmath0 , and 72@xmath0 . the velocity bins are about 100 @xmath29 wide , approximately equal to the spectral resolution of the ground - based observations . we follow @xmath44 orbits per model depending on inclination and black hole mass . this number is only for one sign of the angular momentum ; consequently , it must be doubled before comparison to the data . we flip the losvd at every position in the galaxy to provide the appropriate reversed orbit . thus , our final model contains @xmath45 orbits . varying the relative weights of these orbits allows us to determine the best match to the data . it is desirable to constrain every model bin with an observed losvd . for ngc 3379 , we have ground - based observations of the losvd along four of the five position angles used in the modeling . in fact , the observations were designed to mimic the model angular bins listed above . furthermore , the spectra were extracted from the data using the same radial binning scheme as the modeling . therefore , most model bins correspond _ exactly _ to an observation ; _ i.e. _ , no interpolation of the data or the model is necessary for comparison . this correspondence breaks down for the ground - based data in the innermost bins because of seeing . the ground - based data extend out to 64 , about two effective radii . the total number of bins for which we have an observation is 54 . the model losvds are calculated at 13 velocity positions with a spacing of 100 @xmath29 , providing at most 13 independent measurements of the losvd per position . we therefore have a _ maximum _ of 702 ( 54 positions @xmath46 13 losvd bins ) independent observables . however , the actual number is smaller than this , mainly due to the smoothing used in the measurement of the losvds ( discussed in 5 and gebhardt & richstone 1998 ) , which correlates the 13 velocity bins . thus , we will overestimate the goodness - of - fit since we use the 68% confidence bands of the losvds to calculate it . the model incorporates seeing by convolving the light distribution for every orbit with the appropriate psf before comparing to the data . for the ground - based data , the estimated psf , approximated as gaussian , has a fwhm of 1.5 , including both atmospheric seeing and slit size . the hst data point was taken in a square aperture of 0.21 . the psf for hst has a fwhm of around 0.07 , and hence smearing due to the psf is negligible compared to the aperture used . therefore , for the hst data , only the aperture size was considered with no psf convolution . we have checked in a number of models that this approximation has no effect . for each model , we match both the light in each of the @xmath47 real - space bins , and the losvds in the 54 locations where we have data . the technique minimizes the @xmath31 between the model and data losvds . the light from the model matches the deconvolved spatial brightness profile to a tolerance of better than 1% in each bin , which also ensures that the surface brightness matches better than 1% in any projected light bin . the match to the total light in each bin has to be done using the real spatial profile and not the surface brightness profile . if the matching were done in projection , it would admit different density distributions than the one for which the orbits were computed since the deprojection of a flattened , inclined , axisymmetric system is not unique . for each model , the 702 observables ( 13 losvd bins at the 54 positions ) determine the total @xmath31 , given by @xmath48 where the @xmath49 s are the losvd bin heights for the model and the data and @xmath50 is the uncertainty from the 68% confidence band . as an example , figure 6 plots the losvds for both the data and the best model at four positions : the center , a major - axis position near to the center , a position close to one effective radius on the major axis , and a minor - axis position near to the center . judging the goodness - of - fit from the @xmath31 values of the various models is not straightforward because the number of the degrees of freedom is not easy to determine ; the derivation of the losvds includes a smoothing parameter which correlates the values and uncertainties of the velocity bin heights ( gebhardt & richstone 1999 ) . with higher s / n data than present , we could reduce the smoothing parameter to lessen this problem . we do not attempt to estimate directly the actual degrees of freedom and therefore do not have an overall goodness - of - fit measure , but instead we calculate the change in @xmath31 as a function of the three variables black - hole mass , inclination , and _ m / l_@xmath3 . the lowest @xmath31 value thus provides the best model , with the uncertainties given by the classical estimators for the @xmath51 s . 0.2 cm -0.6 cm 0.3 cm 0.2 cm 0.3 cm the total @xmath31 s for most of the models have a value near 200 , giving a reduced @xmath31 of 0.3 , assuming one observable equals one degree of freedom . a reduced @xmath31 near unity would require either that every three losvd bins are correlated by the smoothing , therefore reducing the number of degrees of freedom by three , or that the uncertainties of the bin heights are overestimated by a factor of @xmath52 . in the former case , using @xmath51 as a discriminator for the error of the best - fitting models will not be affected since the number of degrees of freedom has no influence on the calculation of the @xmath31 differences ; @xmath53 will still correspond to 1-@xmath50 . however , in the latter case , both the @xmath31 and @xmath51 values would increase by a factor of three , creating a smaller confidence band on the variables ; @xmath54 would then correspond to 1-@xmath50 . it is likely , however , that the dominant reason for the low reduced @xmath31 is the underestimation of the degrees of freedom since we know _ a priori _ that the smoothing must have an effect . to provide a quick check on smoothing effects , we refit the losvds for a subset of positions using a much smaller smoothing length . as expected , the losvd becomes noisier . measuring the @xmath31 between this new losvd and the model that we fit using the the older losvd , we find that the increase in @xmath31 ranges from 3 - 5 . thus , we conclude that the dominant effect of the smoothing is to correlate the variables and reduce the number of degrees of freedom . in retrospect , it would have been better not to have smoothed the losvds and preserve their statistical independence . note however that our decision to accept @xmath53 as the 1-@xmath50 error is conservative and , if anything , overestimates the errors . we examined models with inclinations from 29@xmath0 to 90@xmath0(intrinsic e5 to e1 galaxy ) , black holes from zero to @xmath1 , and _ m / l_@xmath3 from 4.0 to 8.0 in solar units . the @xmath31 is a function of these three variables . for each inclination , fig . 7 plots @xmath31 as a function of black - hole mass ( zero black - hole mass is shown as @xmath55 ) , using that _ m / l_@xmath3 which provides the smallest @xmath31 at that black - hole mass . all inclinations show a minimum @xmath31 at black - hole masses in the range @xmath56 . the fits clearly prefer the edge - on models ( 90@xmath0 ) . 8 presents contours of @xmath51 as a function of black - hole mass and _ m / l_@xmath3for each inclination . the points represent locations of the modeled values . the contours use a two dimensional smoothing spline ( wahba 1980 ) to estimate the @xmath31 values at positions where no models occur . as in wahba , generalized cross - validation determines the smoothing value ; however , the modeled values are relatively smooth and little smoothing is necessary . 0.4 cm -1 cm 1.0 cm since we have such a large parameter space minimizing several thousand variables we must ensure that the modeling program is finding a true minimum . furthermore , we require that the numeric noise caused by the minimizer be smaller than the quoted uncertainties of the output parameters ( e.g. , the black hole mass ) . the two conditions that govern these checks are the sampling densities that we use in parameter space , and the tolerance used to determine convergence in the minimizer . 8 demonstrates the sampling densities for the black - hole mass and the _ m / l_@xmath3 , where each point represents one model . 7 shows the range and number of modeled inclinations . the smoothness of the @xmath31 contours in both figures demonstrate that we have found the global minimum . as a further check , we have run models with both larger and smaller _ m / l_@xmath3 than those presented , however the fits are significantly worse and we do not include them in the plots . thus , we conclude that the sampling density provides adequate coverage around the global minimum . our second concern is whether the tolerance used for the stopping criteria in the minimizer creates significant noise for the parameter estimation . we must use a stopping tolerance in the program or the time it takes for the @xmath31 to asymptote becomes impractically long . both figures 7 and 8 provide an estimate of the numeric noise . from the soothness of plots , the change in @xmath31 between two adjacent points is much smaller than the global change in @xmath31 . also , in a handful of models , we iterate till there is no change in @xmath31 relative to machine precision , and find that this asymptotic value is within @xmath57 of the stopping value . therefore , the noise in @xmath31 generated from the minimization routine is not significant . the smoothness is the solution is also seen in fig . 9 , where we plot best - fit black hole mass as a function of inclination . although , the sampling of points is not dense , the fact that there are no abrupt changes signifies we are not subject to numeric noise in the solution . 0.2 cm 0.4 cm 0.2 cm 0.3 cm as described in 3.1 , we have also run models with _ m / l_@xmath3 increasing 25% log - linearly with radius towards the galaxy center . in this case , the required _ m / l_@xmath3 is systematically lower than when using the constant _ m / l_@xmath3 assumption ; however , the results for the required black hole and inclination are unchanged . the best model has an inclination of 90@xmath0 , a black hole mass of @xmath58 , and _ m / l_@xmath3@xmath59 . the allowed mass of black holes range from @xmath2 . the internal dynamics of the best - fitting model are shown in fig . the top plot shows the internal velocity dispersions of the three components along the major axis . the model is tangentially biased ( in the @xmath60 direction ) in the smallest bin , radially biased in the mid - range radii , and tangentially biased again at the largest radii . independent of inclination , fig . 10 well characterizes the orbital distribution for each of the best - fitting models . similarly , the mean velocity in the equatorial plane ( bottom plot of fig . 10 ) has a general shape for the best models ; it shows no systematic trends from @xmath61 ( 20 - 1500 pc ) , increasing significantly in the center due to the presence of the massive black hole . beyond about 60 the model is unconstrained since there is no kinematic information , but we plot the model results there since they represent the maximum entropy configuration . for each inclination the trend with black - hole mass is as follows : increasing the black - hole mass requires an increased amount of tangential anisotropy in the central bin , a decrease of the radial bias in the mid - range radii , and the orbits at the largest radii to become more isotropic . the coupling between the black - hole mass and the amount of radial anisotropy in the mid - radii stems from the need to match the same central losvd ; a large central dispersion can be modeled with a large black - hole mass or radial orbits . therefore , as the black - hole mass increases , the need for radial orbits decreases . in the very central bin , a large black - hole mass induces large rotational velocities and also large dispersion in the @xmath60 direction . at the largest radii , the situation is more complex since we have not included a dark halo ; if the projected dispersion is larger than the dispersion obtained assuming isotropy and the stellar mass profile then one can either invoke a dark halo or include tangential anisotropy . in any event , it is not clear whether meaningful analysis at large radii can be obtained from models which do not include a dark halo . to demonstrate why the data seem to prefer an edge - on model that contains a black hole , we must look at the model fits to the data . figure 11 plots gauss - hermite polynomial estimates for the data and model losvds for all four position angles . the difference between the edge - on model and the nearly face - on model is almost entirely at large radii ; the edge - on model matches the large - radial data better , specifically for the velocity and velocity dispersion . in contrast , the difference between the no - black - hole and black - hole model is so subtle that one can barely discriminate those two models in fig . 11 . only in the central bin does the the black - hole model slightly better estimate the skewness in the profile . however , fig . 11 is not the optimal way to compare models since the fit uses the full losvd and not moments . instead of plotting all 54 losvds for various models , fig . 12 plots the difference in @xmath31 at each spatial position . this difference represents the @xmath31 measured for the losvd of a particular model minus that of the best - fit model ( edge - on , black - hole model ) . the two comparison models are the same as in fig . 11 , the best no - black - hole model ( an edge - on model ) and a nearly face - on black - hole model ( at 29@xmath0 ) . as expected , the discrimination between black - hole / no - black - hole comes mainly from the hst / fos data point . a no - black - hole model fails to match the shape of the central losvd ; however , there is a limit to size of the central black hole since very high black - hole masses models over shoot the dispersions in the central few ground - based measurements . without the hst data point , we would not have been able to determine whether a black hole exists in ngc 3379 . the discrimination for the inclination comes from the data at larger radii ; here , the edge - on model is much better able to match the projected kinematics . as we discuss later , this inclination constraint is uncertain due to our assumption of constant _ m / l_@xmath3 . 0.2 cm 0.3 cm there is an uncertainty in the absolute wavelength calibration of the hst spectrum . since we do not assume that the central hst losvd is symmetric , we must determine the velocity zeropoint of this profile relative to the rest frame of the galaxy ( @xmath62 @xmath29 , borne _ et al . _ unfortunately , our original observing strategy did not include an accurate wavelength calibration , since we were concerned mainly with the central dispersion , and thus did not obtain the necessary flanking arc - lamp spectra . we therefore tried a range of velocity zeropoints , models with velocity offsets from @xmath63 @xmath29 . as the offset increases , the lower limit on the black - hole mass grows , whereas the upper limit remains the same . we are not able to fit acceptable models with offsets greater than 15 @xmath29 . therefore , our allowed range of black hole masses is largest using the 0 @xmath29velocity offset . all of the figures contain the results using this value . the models in this paper depend primarily on the input black - hole mass , inclination , and m / l assumption . a difficulty is that different combinations of these variables can have similar appearances in the projected kinematics . for example , the inclination and the m / l variation both affect the fit at large radii ; one can easily see how differently inclined models of the same flattened galaxy will have markedly different projected dispersion profiles , and also the dispersion profile is obviously influenced by the inclusion and shape of a dark halo . thus , by varying the shape and amount of dark halo one may be able to mimic the effect of changing the inclination . since we have not explored a full range of halo models , we will not make strong claims as to our ability to constrain the inclination of the galaxy . however , the black hole mass is mainly driven by the data at small radii , and primarily by the hst / fos data point . fig . 9 demonstrates this independence from the inclination , as our best - fit black hole mass does not depend strongly on the inclination . an even more robust measurement is the orbital structure . as either the inclination or the black - hole mass varies , we find a similar velocity ellipsoid ; at the smallest bin the velocity ellipsoid is mainly tangential , becoming radial in mid - range radii , and tangential again at the largest radii . 13 plots the radial motion relative to the tangential motion for four models at different inclinations . determination of the inclination , black - hole mass , and orbital structure is discussed in detail below . 0.3 cm -0.7 cm 0.3 cm the inclination of ngc 3379 has been the subject of some debate . capaccioli _ et al . _ ( 1991 ) conclude that ngc 3379 is similar to ngc 3115 but nearly face - on , with some degree of triaxiality . capaccioli base this conclusion only on the surface brightness profile and use no kinematic data . in contrast , statler ( 1994 ) concludes that nearly face - on , triaxial models are inconsistent with the data at the 98% confidence limit , strongly contradicting capaccioli s result . he prefers less - inclined models . statler has obtained more recent data ( statler & smecker - hane , 1999 ) and a fuller analysis is in preparation . as seen in sect . 4 , our best models prefer an edge - on configuration . preference for edge - on models arises from two observations ; first , the dispersion profiles along different axes are slightly different , and , in an axisymmetric case , this configuration is not possible for a face - on system . the difference mainly arises from the inability to match the large radial data , however the face - on model also can not match the shoulder in the dispersion profile at 1020 ( previously noted by statler & smecker - hane , 1999 ) ; the shoulder is more pronounced in the central position angles , and the edge - one models match this configuration well ( see fig . second , major - axis rotation exists which is not allowed for a face - on configuration of an oblate axisymmetric system . thus these two restrictions are mainly a result of our axisymmetry assumption . more general models can be constructed than those presented here . we could in principle extend the modeling technique to include triaxial distributions but then the parameter space becomes too large for us to search ( each axisymmetric model takes approximately 20 hours on an ultrasparc ) . second , our assumption that the galaxy is spheroidal could be relaxed . romanowsky & kochanek ( 1997 ) present a technique which provides a range of different density distributions which similarly project . third , we have assumed that _ m / l_@xmath3 is independent of radius ( _ i.e. _ , there is no dark halo ) . ciardullo _ et al . _ ( 1993 ) , using radial velocity measurements of planetary nebulae , find no evidence for a dark halo in ngc 3379 out to 3.5 effective radii ( 120 ) ; however their uncertainties are large and could , in fact , allow for some dark halo ( tremblay _ et al . _ 1995 ) . in all of our models , the last velocity dispersion measurement ( at 64 ) along each position angle is higher than the model velocity dispersion , possibly suggesting the need for an increase in _ m / l_@xmath3 . at one effective radius ( 34 ) , there is no need for an increase in _ m / l_@xmath3 , which is consistent with previous studies of other galaxies ( kormendy & westphal 1989 , kormendy & richstone 1992 , kormendy _ et al . _ thus , assumptions of axisymmetry , spheroidal distribution , and constant _ m / l_@xmath3 all lead to uncertainties in the best - fit inclination . detailed modeling including dark halos has been carried out by rix _ et al . _ ( 1998 ) for ngc 2434 and saglia _ et al . _ ( 1999 ) for ngc 1399 , but only as spherical systems . only by allowing for axisymmetric or triaxial shapes with a dark halo can we set more realistic limits on inclination . even though we are not able to place limits on the inclination , we can measure black - hole limits . the presence of a dark halo will have no effect on the need for a central black hole , or on its mass . the need for the black hole is directly a result of the shape of the central losvd observed with hst . the large dispersion and the skewness drive the models to require a black hole . we have also run models assuming a symmetrized central losvd ( the dashed line shown in fig . the allowable range of black hole masses remains the same for the inclined models in this case . however , for the edge - on case the model with no black hole fits as well . the symmetrized losvd has a dispersion of 275 @xmath29 , compared with 289 @xmath29 for the actual losvd , and has less tail weight ( as can be seen in fig . these differences permit a no - black - hole model . the central bin most influences the model ; most of the orbits ( @xmath64% for inclined models and @xmath65% for edge - on models ) contribute light to this bin , and , furthermore , many have their pericenter there . therefore , the characteristics of the central losvd , in particular the high - velocity components , have a strong effect on orbits throughout the galaxy . in the edge - on case with the symmetrized losvd , the orbits redistribute themselves in such a way as to create a slightly larger central dispersion ( 275 @xmath29 ) without the need for a black hole , and at the same time maintain a fairly flat projected dispersion profile ; _ i.e. _ , at small radii , the orbits are mainly radial , whereas at large radii they become primarily tangential . however , the actual losvd warrants matching the skewness and the high - velocity tails . with no black hole present , the model must use stars at larger radii to create the high - velocity components , thereby limiting its ability to redistribute the orbits at large radii and match the kinematics there . with a black hole , the black hole provides much of the high - velocity tail of the losvd for orbits locally and gives the model more freedom to better match the larger - radii data . the symmetrized and skewed losvds are inconsistent with each other with greater than 95% probability , both using an estimate from the binned values and using the gauss - hermite moments . given the arguments in 2.2.1 , we must include the skewness of the central losvd and thus do not consider the symmetrized losvd models further . relaxing the other two assumptions axisymmetry and spheroidal distribution is unlikely to change the black - hole limits . the models with no black holes all are extremely poor fits to the data ; therefore , even if we allow for the variation in deprojected densities given by romanowsky & kochanek , the results are unlikely to change . of course , since the deprojection is unique for the edge - on models , our results are general in the edge - on case . triaxial models will allow even greater freedom to match the data , however it is not clear whether triaxiality is a viable configuration in the centers of galaxies ( merritt & fridman 1996 , merritt 1999 ) , and we do not consider these models . the position angle of the dust ring in fig . 5 is 45@xmath0 away from the major axis of the galaxy . a dust ring in an axisymmetric system with an offset position angle is only neutrally stable but can be stabilized by a small amount of triaxiality . if the galaxy is nearly face - on , then triaxiality is not necessary to stabilize the dust ring ; in the face - on configuration , since the dust ring is close to edge - on ( about 75@xmath0 ) , it must be in an orbit close to polar , and thus can have arbitrary azimuthal angle . velocity information along the dust ring could provide an important check on the enclosed mass derived from our models . pastoriza _ et al . _ ( 1999 ) , using ground - based gas kinematics , assume that the gas disk in ngc 3379 is nearly face - on . their mass estimate inside 1.3 is @xmath66 , strongly contradicting our result . however , their large mass is due mainly to their disk inclination assumption . if one uses the inclination as measured from the dust ring , then the enclosed mass closely equals our value of @xmath58 ; only with hst can we precisely determine the gaseous disk configuration and determine whether the gas is even in a disk at all . the hst / fos spectrum includes the [ nii ] emissiom , although since it is a single pointing , we do not have the radial information to extract a meaningful dynamical analysis from the gas . however , the width of the line does provide some information on the underlying mass . the measured @xmath50 is 200@xmath29 , in agreement with the mass deduced here ; using the mass from pastoriza _ et al . _ would imply a @xmath50 around 700@xmath29 . to further understand the gas kinematics , we must have spatial information . van der marel _ et al . _ ( 1990 ) apply 2-integral flattened models for ngc 3379 using constant anisotropy . they find an acceptable fit with an inclination of 60@xmath0 and anisotropic orbits . they suffer from the same problem as we do for the inclination estimate ; without inclusion of a dark halo it is not possible to obtain adequate inclination constraints . furthermore , since our axisymmetric models are fully general while their models specify a form for the distribution function , we can not directly compare our results to theirs . van der marel _ et al . _ conclude that there is no evidence to support the need for a third integral using data from the first and second moments of the losvd from davies & birkinshaw ( 1988 ) . results presented here , which are based on higher s / n data and are determined from the full losvd , show that the best model requires three integrals and is not consistent with a 2-integral distribution function ; the internal dispersions along the long axis in the radial and @xmath41 directions are significantly different , in direct conflict with a 2-integral distribution function . as seen in fig . 13 , the ratio of the @xmath41 to the radial dispersion in our best model have a radial variation from 0.6 to 1.2 , with an average around 0.75 . for comparison , this ratio is around 0.6 for the spheroid component of our galaxy using rr lyrae halo stars ( layden 1995 ) . in addition to the measured black - hole mass , the orbital structure appears to be a robust feature as well . the radial anisotropy profile along the major axis in fig . 13 shows that this profile is nearly independent of input inclination . for all acceptable models , the velocity ellipsoid is tangentially biased in the central bin and radially biased in the mid - range radii . these orbital properties are generally true along any position angle . this central tangential bias appears in other 3-integral models studied ( gebhardt _ et al . _ 2000 ) and may be a common feature of early - type galaxies . tangential bias in the central regions around black holes appear to be a common feature in simulations ( quinlan _ et al . _ 1995 , quinlan & hernquist 1997 , merritt & quinlan ( 1998 ) , and nakano & makino 1999 ) . the radial bias in the mid - range radii has also been seen before ; using 3-integral models , both gerhard _ et al . _ ( 1999 ) for ngc 1600 and saglia _ et al . _ ( 1999 ) for ngc 1399 find radial motion throughout the main body of the galaxy . thus , we may be beginning to measure common features of the orbital properties in ellipticals tangential motion in the central regions and radial motion at mid - range radii . at large radii , we must include dark halos since they make a significant contribution there . the next step is to determine dominant evolutionary effects from theoretical models and will require detailed comparisons with n - body simulations . magorrian _ et al . _ ( 1998 ) , using the same ground - based data , fit two - integral models to the observed second - order velocity moment profiles and require a bh of @xmath67 . our more general three - integral models would allow an even larger range of bh masses had we similarly used only the losvd s second moments . when we fit to the full losvds , however , their shape requires that our models be mildly radially anisotropic , reducing the required bh mass by a factor of at least two over magorrian _ et al . _ s result . the models presented in this paper still contain restrictive assumptions namely constant _ m / l_@xmath3 and spheroidal symmetry requiring yet more models to study black hole properties in complete generality . kormendy ( 1993 ) and kormendy & richstone ( 1995 ) find that there is a correlation between the black hole mass and the mass of the spheroid . whether this relation is a real correlation or an upper envelope that extends to smaller black - hole masses remains to be seen . as the data quality and the modeling techniques improve , we should be better able to constrain the relationship between black hole and spheroid mass , and begin to measure correlations with the orbital characteristics . we thank gary bower , richard green , luis ho , and jason pinkney for detailed discussions about the results . we thank an anonymous referee who found an error in the previous version of this paper . we are grateful to c. d. keyes , j. christensen , and j. hayes for help with the data analysis . this work was supported by hst data analysis funds through grant go02600.0187a and by nserc . kg is supported by nasa through hubble fellowship grant hf-01090.01 - 97a awarded by the space telescope science institute , which is operated by the association of the universities for research in astronomy , inc . , for nasa under contract nas 5 - 26555 . rrrrr 0.023 & 14.625 & 1.302 & 0.121 & 92.3 0.046 & 14.739 & 1.314 & 0.121 & 92.3 0.091 & 14.931 & 1.340 & 0.040 & 92.3 0.136 & 14.978 & 1.311 & 0.232 & 92.3 0.182 & 15.047 & 1.322 & 0.181 & 101.8 0.227 & 15.101 & 1.307 & 0.181 & 87.6 0.273 & 15.147 & 1.311 & 0.177 & 87.6 0.318 & 15.184 & 1.313 & 0.149 & 85.8 0.364 & 15.227 & 1.311 & 0.141 & 84.4 0.409 & 15.276 & 1.325 & 0.119 & 84.1 0.455 & 15.288 & 1.311 & 0.119 & 78.8 0.500 & 15.326 & 1.318 & 0.119 & 76.4 0.546 & 15.351 & 1.313 & 0.110 & 64.9 0.566 & 15.368 & 1.326 & 0.083 & 58.1 0.665 & 15.412 & 1.297 & 0.119 & 65.7 0.783 & 15.485 & 1.300 & 0.095 & 58.7 0.921 & 15.578 & 1.299 & 0.089 & 69.6 1.084 & 15.692 & 1.300 & 0.085 & 67.5 1.275 & 15.812 & 1.305 & 0.081 & 70.8 1.500 & 15.933 & 1.299 & 0.101 & 70.9 1.764 & 16.072 & 1.297 & 0.104 & 72.1 2.076 & 16.213 & 1.293 & 0.115 & 73.3 2.442 & 16.365 & 1.295 & 0.117 & 74.0 2.873 & 16.528 & 1.291 & 0.119 & 73.5 3.380 & 16.702 & 1.286 & 0.118 & 73.8 3.977 & 16.895 & 1.284 & 0.111 & 74.0 4.678 & 17.110 & 1.278 & 0.105 & 74.1 5.504 & 17.352 & 1.275 & 0.094 & 74.0 6.475 & 17.596 & 1.272 & 0.088 & 73.8 7.618 & 17.830 & 1.267 & 0.089 & 73.2 8.962 & 18.064 & 1.262 & 0.089 & 73.2 10.544 & 18.313 & 1.257 & 0.090 & 72.1 12.405 & 18.572 & 1.246 & 0.092 & 71.1 14.594 & 18.855 & 1.237 & 0.089 & 69.3 ajhar , e. , lauer , t. , tonry , j. , blakeslee , j. , dressler , a. , holtzman , j. , & postman , m. 1997 , , 114 , 626 bender , r. , 1990 , , 229 , 441 bender , r. , saglia , r.p . , & gerhard , o.e . 1994 , , 269 , 785 borne , k.d . , balcells , m. , hoessel , j.g , & mcmaster , m. 1994 , , 435 , 79 bower , g. _ et al . _ 1998 , , 492 , 111 capaccioli , m. , held , e.v . , & nieto , j.l . 1987 , , 94 , 1519 capaccioli , m. , held , e.v . , lorenz , h. & vietri , m. 1990 , , 99 , 1813 capaccioli , m. , vietri , m. , held , e.v . , & lorenz , h. 1991 , , 371 , 535 carollo , c.m . , de zeeuw , p.t . , van der marel , r.p . , danziger , i.j . , & qian , e.e . 1995 , , 441 , l25 ciardullo , r. , jacoby , g.h . , & dejonghe , h.b . 1993 , , 414 , 454 cretton , n. , de zeeuw , p.t . , van der marel , r.p . , & rix , h .- w . , 1999 , , accepted davies , r.l . , & birkinshaw , m. 1988 , , 68 , 409 dejonghe , h.b . 1987 , , 224 , 13 de vaucouleurs , g. , & capaccioli , m. 1979 , , 40 , 699 faber , s.m . _ et al . _ 1997 , , 114 , 1771 ferrarese , l. , ford , h.c . , & jaffe , w. 1996 , , 470 , 444 franx , m. , illingworth , g.d . , & heckman , t.m . 1989 , , 344 , 613 gebhardt , k. , richstone , d. _ et al . _ 1996 , , 112 , 105 gebhardt , k. , richstone , d. 1999 , in preparation gerhard , o.e . 1993 , , 265 , 213 gonzalez , j. 1994 , phd thesis , ucsc harms , r.j . _ et al . _ 1994 , , 435 , l35 keyes , c. d. , _ et al . _ 1995 , fos instrument handbook ( baltimore : stsci ) kormendy , j. 1977 , , 218 , 333 kormendy , j. 1992 , in `` testing the agn paradigm '' , eds . holt , s.g . neff , & c.m urry ( ny : aip ) , 23 kormendy , j. 1985 , , 292 , l9 kormendy , j. 1993 , in `` the nearest active galaxies '' , eds . j. beckman , l. colina , & h. netzer ( madrid ) , 197 kormendy , j. , & westphal , j. 1989 , , 338 , 752 kormendy , j. , & richstone , d. 1992 , , 393 , 559 kormendy , j. , & richstone , d. 1995 , ara&a , 33 , 581 kormendy , j. _ et al . _ 1996 , , 459 , l57 kormendy , j. _ et al . _ 1997 , , 482 , l139 kormendy , j. , bender , r. , evans , a. , & richstone , d. 1998 , , 115 , 1823 lauer , t.r . 1985 , , 216 , 429 lauer , t.r . _ et al . _ 1995 , , 110 , 2622 layden , a.c . 1995 , , 110 , 2288 magorrian , j. _ et al . _ 1998 , , 115 , 2285 merritt , d. 1997 , , 114 , 228 merritt , d. 1999 , , 111 , 129 merritt , d. , & fridman , t. 1996 , , 460 , 136 nakano , t. , & makino , j. 1999 , , 510 , 155 pastoriza , m. , winge , c. , ferrari , f. , & macchetto , d. 1999 , , accepted peletier , r.f . , davies , r.l . , illingworth , g.d . , davis , l.e . , cawson , m. 1990 , , 100 , 1091 qian , e.e . , de zeeuw , p.t . , van der marel , r.p . , & hunter , c. 1995 , , 274 , 602 quinlan , g. , hernquist , l. , & sigurdsson , s. 1995 , , 440 , 554 quinlan , g. , & hernquist , l. 1997 , newa , 2 , 533 richstone , d. , gebhardt , k. _ et al . _ 1999 , in preparation richstone , d. , & tremaine , s. 1988 , , 327 , 82 rix , h .- w . , de zeeuw , p.t . , carollo , c.m . , cretton , n. , & van der marel , r.p . 1997 , , in press romanowsky , a. , kochanek , c. 1997 , , 287 , 35 rybicki , g.b . 1987 , in structure and dynamics of elliptical galaxies , iau symposium 127 , ed . de zeeuw , p.t . , ( reidel , dordrecht ) , p.397 ryden , b. 1996 , , 461 , 146 saha , p. , & williams , t.b . 1994 , , 107 , 129 schwarzschild , m. 1979 , , 232 , 236 statler , t.s . 1994 , , 108 , 111 statler , t.s . , & smecker - hane , t. 1999 , , feb . 1999 tremblay , b. , & merritt , d. , & williams , t.b . 1995 , , 443 , l5 van der marel , r.p . , binney , j.j . , & davies , r.l . 1990 , , 245 , 582 van der marel , r.p . 1991 , , 253 , 710 van der marel , r.p . , & franx , m. 1993 , , 407 , 525 van der marel , r.p . , cretton , n. , de zeeuw , p.t . , & rix , h .- w . 1998 , , 493 , 613 van dokkum , p.g . , & franx , m. 1995 , , 110 , 2027 wahba , g. 1980 , technical report no . 595 , univ . of wisconsin worthey , g. 1994 , , 95 , 107

we introduce a comprehensive and statistical framework in a model free setting for a complete treatment of localized data corruptions due to severe noise sources , e.g. , an occluder in the case of a visual recording . within this framework , we propose i ) a novel algorithm to efficiently separate , i.e. , detect and localize , possible corruptions from a given suspicious data instance and ii ) a _ maximum a posteriori _ ( map ) estimator to impute the corrupted data . as a generalization to euclidean distance , we also propose a novel distance measure , which is based on the ranked deviations among the data attributes and empirically shown to be superior in separating the corruptions . our algorithm first splits the suspicious instance into parts through a binary partitioning tree in the space of data attributes and iteratively tests those parts to detect local anomalies using the nominal statistics extracted from an uncorrupted ( clean ) reference data set . once each part is labeled as anomalous vs normal , the corresponding binary patterns over this tree that characterize corruptions are identified and the affected attributes are imputed . under a certain conditional independency structure assumed for the binary patterns , we analytically show that the false alarm rate of the introduced algorithm in detecting the corruptions is independent of the data and can be directly set without any parameter tuning . the proposed framework is tested over several well - known machine learning data sets with synthetically generated corruptions ; and experimentally shown to produce remarkable improvements in terms of classification purposes with strong corruption separation capabilities . our experiments also indicate that the proposed algorithms outperform the typical approaches and are robust to varying training phase conditions . localized corruption , occlusion , map based imputation , anomaly detection .

1409.8576

in many applications from a wide variety of fields , the data to be processed can partially ( or even almost completely ) be affected by severe noise in several phases , e.g. , occlusions during a visual recording or packet losses during transmission in a communication channel . the partial , i.e. , localized , corruptions resulted in the data due to such problems often severely degrade the performance of the target application ; for instance , face recognition or pedestrian detection under occlusion @xcite . in order to reduce the impact of this adverse effect , we develop a complete and novel framework , which efficiently detects , localizes and imputes corruptions by identifying the local anomalies in a given suspicious data instance . we emphasize that neither the existence nor , if exists , the location of a corruption is known in our framework . moreover , the proposed algorithms do not assume a model but operate in a data driven manner . we consider the local corruptions as statistical deviations from the nominal distribution of the uncorrupted ( clean ) observations . to detect and localize corruptions , i.e. , such statistical deviations , we model a corruption as an anomaly due to an external factor ( communication failure in a channel or occluder object in an image ) , which locally overwrites a data instance and moves it outside the support of the nominal distribution . however , corruptions that we consider as examples of anomalies have further specific properties such that ( a ) the corruptions in an instance are confined to unknown intervals along the data attributes , i.e. , localized , and ( b ) not only a corrupted part but also all of its subparts are anomalous . thus , a corruption does not provide an anomaly due to an incompatible combination of normal subparts . based on these properties that accurately model a wide variety of real life applications , we characterize the event of corruption and formulate the corresponding detection / localization as an anomaly detection problem , cf . @xcite . the introduced algorithm applies a series of statistical tests with a pre - specified false alarm rate to the parts of the suspicious instance after extracting the nominal statistics from a reference ( training ) data set of uncorrupted ( clean ) observations . as a result , each part is labeled as anomalous / normal and the local anomalies are identified . these parts are generated and organized through a binary tree partitioning of the data attributes , each node of which corresponds to a part of the suspicious instance . once the nodes ( or parts ) are labeled as anomalous / normal on this tree , the patterns of corruption are identified using the aforementioned characterization to detect and localize corruptions . we point out that this localization procedure transforms the nominal distribution into a multivariate bernoulli distribution with a success probability that precisely coincides with the constant false alarm rate of the local anomaly tests . considering the hierarchy among the binary labels implied by the tree as a directed acyclic graph , the resulting multivariate bernoulli distribution achieves a certain dependency structure . under this condition , we derive the false alarm rate of the proposed framework in detecting the corruptions and show that it is a constant rate , namely , no parameter tuning is required even if the data change . if a corruption is localized , then we impute / replace the affected attributes with the estimates of the underlying unknown true attributes . for this purpose , we additionally develop a novel _ maximum a posteriori _ ( map ) estimator using the score function " defined in @xcite . our estimator exploits the local dependencies among the data attributes , where the locality is encoded in the binary partitioning tree . we point out that the implementation of this map estimator does not load extra computational cost since it utilizes the outputs of our anomaly detection approach , which are computed prior to the imputation phase . furthermore , we also propose a novel distance measure named ranked euclidian distance " as a generalization to the standard euclidean distance , which is used in the course of the labeling of each part as anomalous / normal . the proposed distance measure is compared with the standard euclidean distance in the experiments and shown to be superior in terms of detecting and localizing corruptions . we conduct tests over several well - known machine learning data sets @xcite , which are exposed to severe data corruptions . our experiments indicate that the proposed framework achieves significant improvements after imputation up to @xmath0 in terms of the classification purposes and outperforms the typical approaches . the proposed algorithms are also empirically shown to be robust to varying training phase conditions with strong corruption separation capabilities . in this study , the corrupted attributes are considered to be statistically independent with the underlying unobserved true data , i.e. , corrupted attributes are of no use in estimation of the uncorrupted counterparts . hence , if one knows which attributes are corrupted in an instance , then those attributes can readily be treated as missing data , cf . @xcite . for example , classification and clustering with missing data is a well - studied problem in the machine learning literature . the corresponding studies such as @xcite are related to inference with incomplete data @xcite and generative models @xcite , where bayesian frameworks @xcite are used for inference under missing data conditions . alternatively , pseudo - likelihood @xcite and dependency network @xcite approaches solve data completion problem by learning conditional distributions . in @xcite , the probability density of the missing data is modeled conditioned on a set of introduced latent variables and thereafter , a map based inference is used . however , all of the studies @xcite either assume the knowledge of the location of the missing attributes or impose strong modeling constraints ; as opposed to the model free solutions in this paper . on the other hand , imputation is commonly used as a pre - processing tool @xcite . the mixture of factor analyzers @xcite approach replaces the missing attributes with samples drawn from a parametric density , which models the distribution of the underlying true data . the proposed imputation techniques in @xcite are , whereas , both non - parametric and based on the inference of the posterior densities via certain kernel expansions . on the contrary , the map estimator in this study does not even attempt to estimate the posterior density either in a parametric or non - parametric manner . instead , the introduced method is only based on the sufficient rank statistics . we emphasize that unlike our approach , the incomplete data approaches generally assume the knowledge of the missing attributes , i.e. , they are precisely localized and provided beforehand . for example , the occluded pixels in the event of occlusion of a target object in an image can not be known a priori , which requires a detection and localization step . since the existing studies do not have such a step , an exhaustive list of the occluded pixels as the result of a manual inspection of the missing attributes is required as an input to the algorithms proposed in the corresponding literature . in this regard , our study is the first to jointly handle the issues of detecting / localizing missing attributes , i.e. , corruptions , as well as their imputation in one complete framework . hence , the generic local corruption detection and imputation algorithm of our framework complements the missing data imputation approaches as an additional merit . data imputation and completion is also essential in image processing for handling corrupted images , e.g. , @xcite . generally , a corrupted image is restored by explicitly learning the image statistics @xcite or by using neural networks @xcite . these denoising studies do not attempt to localize corruptions in an image , but treat them as a noise and filter it out using statistical approaches applied to the image globally . even though this is a valid approach for image enhancement , an attempt to correct / enhance an image globally in case of only a localized corruption might be even detrimental since the uncorrupted parts are also affected by global operations . additionally , it is not usually possible to locally impute corrupted portions using denoising approaches . there exist several studies that aim localization as well . studies such as @xcite indicate that occlusion , as an example of corruption , is a common phenomenon and detrimental in pedestrian detection as well as face recognition applications . in this regard , detection of occluded , i.e. , corrupted , visual objects had been previously investigated in a number of studies @xcite . in these studies , occlusion detection is performed using domain specific knowledge ( visual cues ) or external information ( object geometry ) , which , however , are not always available in general data imputation setting . from the machine learning perspective , descriptors are extracted from various parts of the occluded object in @xcite and similarly ; part - based descriptors are weighted with the occlusion measure in @xcite to relieve the corresponding degrading effects . since these approaches do not directly target handling occlusions , i.e. , corruptions , they only provide partial or limited solutions . several other studies propose solutions via extracting occlusion maps , e.g. , @xcite . in @xcite , hog based classification errors ; and in @xcite , template based reconstruction errors are used to generate such an occlusion map . however , both studies assume rigid models and significantly rely on domain specific knowledge ; and in general fail to remain applicable if the data source belongs to another domain . in this study , we assume that data is generic and no domain information is available , yet detection and imputation of corruption is necessary for improving the subsequent processing stages , such as classification . 1 . this study is the first to jointly handle localized data corruptions in one statistical framework that is designed completely model free for the goal of separating a corruption and imputing the affected data attributes . we also provide a false alarm rate ( in detecting corruptions ) analysis of the framework via directed acyclic graphs . 2 . a novel map estimator for imputation and a novel distance measure for corruption localization purposes is proposed . the proposed framework is computationally efficient in the sense that ( i ) it effectively utilizes a binary search for corruption separation , and ( ii ) the computational load due to our map based imputation is insignificant . we propose a characterization for anomalies , e.g. , rarities , incompatible combinations and corruptions , which is a novel notion . + in section [ sec : pd ] , we provide the problem description . we then present our algorithm in section [ sec : thethemethod ] and the associated computational complexity in section [ sec : complexity ] . we report the corruption detection / localization performance of the proposed algorithm as well as the improvement in classification tasks achieved by the imputation in section [ sec : experiments ] . the paper concludes with a discussion in section [ sec : conclusion ] . we have a possibly corrupted test instance @xmath1 along with a set of uncorrupted ( clean ) independent and identically distributed observations @xmath2 as the nominal training ( reference ) data , where @xmath3 \in \mathbb{r}^d \sim f_{0}(\mathbf s)$ ] , @xmath4 is data dimensionality and @xmath5 is the unknown nominal density . the test instance @xmath6 is considered to be corrupted with probability @xmath7 by severe noise in multiple non - overlapping intervals along its dimensions ( attributes ) , which are completely unknown . suppose that for such an interval , the corruption is localized and confined to the attributes @xmath8 for some @xmath9 and @xmath10 in @xmath11 $ ] with @xmath12 . we assume that the corrupted attributes are uniformly and independently distributed , @xmath13 , where @xmath14 is the uniform distribution defined in a finite support . moreover , @xmath15 is also statistically independent with the true data and hence , the knowledge of @xmath16 is irrelevant to the uncorrupted counterparts . note that this corruption model implies a total erasure of data in several unknown portions due to an independent source overwriting the attributes in those portions , e.g. , an occluder in computer vision applications @xcite . typically , since no information is provided about the independent source in such applications , we consider that the uniformity assumption draws a worst case scenario and it is realistic . on the other hand , @xmath6 is considered to be uncorrupted with probability @xmath17 . therefore , whether a test instance @xmath6 includes a corruption is unknown ; and it is generally modeled to be drawn from the mixture @xmath18 @xcite , where @xmath19 is the probability density of the corrupted instances . the density @xmath19 can be derived from the unknown nominal density @xmath5 using the described corruption model , if the distributions of @xmath9 , @xmath10 and the number of corrupted intervals are further specified ; which is unnecessary in the context of this paper . hypothetically , if one can correct an instance @xmath6 drawn from the density @xmath19 by replacing all the corrupted attributes , e.g. , @xmath16 , with the underlying true attributes , e.g. , @xmath20 , and obtain @xmath21 , then @xmath21 should follow the nominal density @xmath22 . similarly , if the corruptions in @xmath6 can be localized , then the corresponding portions would follow the multivariate uniform density @xmath23 of the appropriate dimensionality . on the other hand , this corruption model potentially creates significant statistical deviations from the reference data since a corrupted observation @xmath24 and @xmath25 , in general , increasingly diverges from @xmath22 as the corruption strength increases . here , the corruption strength can be considered as the number of corrupted attributes and/or the variance of the corruption @xmath26 that overwrites the true data . furthermore , our modeling of corruptions poses a missing ( incomplete ) data problem since the unknown true attributes @xmath20 in a corrupted interval are statistically irrelevant to the corrupted attributes @xmath16 . in this paper , by exploiting the statistical deviations from the nominal distribution of observations , we aim to detect and localize the possible corruptions in a given instance @xmath6 and impute the corrupted or missing attributes . to this end , we formulate an anomaly detection approach to define this framework in section [ sec : thethemethod ] , where we draw the distinctions among several examples of anomalous observations and separate the event of corruption . then , we propose our algorithm and analyze the associated false alarm probability in detecting corruptions as well as the computational complexity . in this section , we develop a novel framework for a complete treatment of possible corruptions in the input data @xmath6 . for presentational clarity and without loss of generality , we assume that the input data @xmath6 can be corrupted only in a single interval throughout this section . note that the generalization to the case of corruptions spread onto several intervals is immediate and indeed , we present a corresponding detailed experiment in section [ sec : experiments ] . since the corruptions are modeled as local statistical deviations within this framework , we give a brief description of the anomaly detection approach that we work with in section [ sec : corruptionsandanomalies ] . based on the characterization of corruptions through their distinctive properties in section [ sec : treelozalization ] , we present algorithm tcs ( tree - based corruption separation ) . after we derive a novel map estimator for imputation in section [ sec : theimp ] , we derive the false alarm rate of the proposed framework in detecting the corruptions in section [ sec : acf ] . a localized corruption is considered to affect an instance in a certain part(s ) such that the affected attributes statistically deviate from the vast majority of the data . the proposed algorithm in this paper localizes the corrupted attributes by identifying the local anomalies through a series of statistical checks of the test instance with the reference data . in this section , we briefly describe the anomaly detection approach that we work with and present a novel distance measure for the corruption localization purpose . the probability density of a possibly corrupted test instance @xmath6 can be modeled as @xmath27 @xmath28 is the null hypothesis from which the nominal data are drawn , @xmath29 is the hypothesis representing the corrupted observations , and @xmath30 $ ] is the corresponding mixing coefficient . within the framework of anomaly detection approaches , the nominal distribution @xmath5 is usually assumed unknown or hard to estimate ; and instead , a set of nominal observations is provided . then for a given test instance @xmath6 , the task in @xcite is to decide whether the null hypothesis @xmath31 was realized or the alternative @xmath32 such that the detection rate ( of anomalies ) is maximized with a constant false alarm rate @xmath33 . for this purpose , the score function @xcite @xmath34 is proposed , where @xmath35 is the indicator function and @xmath36 is the euclidean distance from @xmath6 to its nearest @xmath37th neighbor in @xmath38 , if @xmath39 ; and to its nearest @xmath40th neighbor in @xmath38 , otherwise . based on this score function , the test instance @xmath6 is declared as anomalous @xcite , if @xmath41 when the mixing distribution @xmath19 is assumed uniform , it is shown in @xcite that @xmath42 is an asymptotically consistent estimator of the density level of the test instance @xmath43 under certain smoothness conditions . remarkably , @xmath44 provides the minimum volume set at level @xmath33 , which is the most powerful decision region for testing @xmath31 vs @xmath32 with a constant false alarm rate @xmath33 @xcite . we note that the precision of the test defined in degrades faster with the dimensionality than it improves with the size of the training data . as a result , we here point out several practical issues about detecting the existence of a corruption with this approach . briefly , i ) a direct test of an instance @xmath6 does not localize a possible corruption for imputation , ii ) on the contrary , a truly corrupted instance , i.e. , an instance of hypothesis @xmath32 , does not necessarily test positive due to the limited training data , high dimensionality as well as that the corruption might not be sufficiently strong ; and iii ) corruptions have further specific properties in addition to that they provide anomalies , which must be incorporated to achieve a better false alarm rate compared to @xmath33 . + _ ranked euclidean distances _ : to address the first issue in this list , we propose a novel distance measure ( not a metric in the mathematical sense ) , which is sensitive to only a certain @xmath45 fraction of the attributes for a given pair of instances @xmath6 and @xmath46 . for instance , a corruption of only a single attribute in a given test instance @xmath6 might be significantly strong such that the whole instance turns anomalous with the test in used with the standard euclidean distance . in this case , any part of the instance @xmath6 including the corrupted attribute would test positive , which creates an ambiguity in terms of the localization , i.e. , separation , of the corrupted attribute , and in turn requires an exhaustive search over all possible subsets in the space of the attributes . to overcome such ambiguities , we propose a distance measure so that the test in results positive only when the corruption has a sufficiently large support , which disregards a pre - specified fraction of the attributes that are most responsible for a possible corruption . we define this measure for an @xmath47 $ ] as @xmath48 where @xmath49 is a permutation of the attributes with @xmath50 is the floor operator . since this distance measure depends only on the @xmath45 fraction of the least deviated attributes between @xmath6 and @xmath46 , a corruption must have a support of at least @xmath51-length to make an instance anomalous with respect to the reference data . here , @xmath52 can be seen as the precision of the localization when an anomalous instance is checked with the test in using the distance measure defined in . this precision obviously can not be made arbitrarily large since as @xmath53 approaches @xmath54 , the distance @xmath55 becomes more prone to noise and the correlation structure between the attributes is less exploited . we investigate this trade - off further in our simulations . the distance measure @xmath55 recovers the standard euclidean distance when @xmath56 and will be named in the rest of the paper as ranked euclidean distance " . we note that for the cases @xmath57 , @xmath55 fails to be a metric in the mathematical sense , i.e. , @xmath58 is not satisfied , which requires to specify a nominal density model on @xmath5 to derive the same asymptotic consistency in @xcite for the score values @xmath59 in estimating the density levels @xmath60 with @xmath55 . however , we do not assume -in this work- any density model for @xmath5 or do not take any stochastic assumptions regarding the data source . in the following section , we characterize the corruptions by presenting their specific properties and propose an algorithm to localize and impute corruptions . if a test instance is subject to corruption in a small part only , the corruption might not be detectable when it is checked using an anomaly detection algorithm without a detailed analysis in its parts . on the other hand , an anomalous observation does not necessarily contain a corruption since it might simply be a false alarm , in fact an uncorrupted observation . to address these two issues , we propose a statistical analysis of a test instance through its parts using a binary partitioning tree in the space of data attributes on which , we also provide a characterization to separate the event of corruption among possible anomaly scenarios . suppose that an instance @xmath61\in \mathbb{r}^d$ ] corresponds to the root node @xmath62 on a binary tree . using half - way splits for presentational simplicity , let the set of attributes @xmath63 be assigned to the left child node @xmath64 of the root and @xmath65 assigned to the right child node @xmath66 , fig . [ fig : example1 ] . note that @xmath67 with @xmath68 and @xmath69 . based on this strategy for generating subparts of an instance , we propose algorithm tcs ( tree - based corruption separation ) to separate and impute corruptions , which recursively expands a depth-@xmath70 binary tree to partition the space of attributes . for each node @xmath71 created in the course of this expansion , the corresponding attributes / part of the test instance , e.g. , @xmath72 with @xmath73 , is checked whether it is consistent with the reference data restricted to those attributes , e.g. , @xmath74 with @xmath73 , using the test defined in . we here use the ranked euclidean distance @xmath75 in this testing with a pre - specified @xmath45 . therefore , each node @xmath71 encountered in this expansion is assigned a binary label as anomalous / normal and a fully labeled ( possibly unbalanced ) tree is obtained for the test instance @xmath6 . we emphasize that algorithm tcs does not completely construct this depth @xmath70-binary tree at the beginning but instead expands it by creating the nodes and the edges as needed to achieve an efficient implementation , which continues until that each data attribute is decided to be corrupted or uncorrupted . we consider several scenarios where the observation @xmath76 at a node @xmath71 can be anomalous . in fig . [ fig : simplecorruption ] , the nodes are illustrated as circles , if the corresponding part is found to be anomalous ; and squares otherwise . an anomaly can be wide - spread onto the attributes and consist of anomalous subparts as illustrated in fig . [ fig : simplecorruption]a , which is regarded as a conclusive pattern since a corruption is characterized and defined in section [ sec : pd ] by that all of the subparts of a corrupted instance are also corrupted . hence , a corruption at the starred node in fig . [ fig : simplecorruption]a is declared , unless it is the root node . note that a global corruption at the root is disregarded in this work since it is not localized . in another case , an anomalous observation could be non anomalous in its parts as illustrated in fig . [ fig : simplecorruption]b , which simply happens due to an incompatible or rare combination of attributes in its subparts . this is a typical situation , where an anomalous observation is not corrupted . hence , this case also provides a conclusive pattern in our consideration such that a corruption is rejected at the anomalous node . on the contrary , the case in fig . [ fig : simplecorruption]c is an inconclusive pattern , which suggests a corruption at the right child , however , whether the corruption is spread in attributes of that child or localized is unknown . hence , the attributes of the right child is further split and explored similarly . then , if the conclusive pattern in fig . [ fig : simplecorruption]a ( or fig . [ fig : simplecorruption]b ) is realized , then the corruption is accepted and localized ( or rejected ) at the starred node in fig . [ fig : simplecorruption]d . otherwise , the search continues . on the other hand , if a significantly small subset of the corrupted attributes are left at the left child node in fig . [ fig : simplecorruption]c , it might not be detectable and labeled as normal . then the corresponding attributes should further be split as illustrated in fig . [ fig : simplecorruption]d . this process recursively defines a corruption localization with an improved false alarm rate as several anomalies are rejected as they are false alarms , i.e. , non corrupted anomalies . the introduced algorithm tcs then searches in a breadth - first - search fashion the described binary tree for a corruption . when the conclusive ( or terminating ) pattern shown in fig . [ fig : simplecorruption]a ( fig . [ fig : simplecorruption]b ) is found in the course of this expansion , the search is stopped at the parent node of the found pattern , i.e. , the tree is pruned on that branch , and corruption is declared ( or no corruption is found and no action is necessary ) for the corresponding attributes . this search of corruption at each branch starting from the root node continues to the corresponding leaf node , unless a terminating pattern is found . finally , if a conclusive pattern is not encountered at a branch from the root to an anomalous leaf , we opt to accept the corruption at the leaf to favor a better detection at a cost of an increased corruption false alarm rate . an illustration of the progress of the algorithm is given in fig . [ fig : example1 ] , where the corrupted attributes are successfully located . note that a small set of the attributes are mislabeled as corrupted , i.e. , false alarms , in the region 3 , which can be corrected if the partitioning resolution is improved by increasing the depth @xmath70 . = 3.5 cm we emphasize that in most of the detection and estimation applications , the posterior density , e.g. , @xmath77 in , of the target is too complicated to assume realistic parametric models so that the nonparametric approaches are often favored in such situations @xcite . in accordance , we introduce an algorithm that works under a completely model free setting regarding both the localization of the corruptions and the imputation . furthermore , we point out that map based estimators are generally known to generate more plausible results when the posterior density is multi - modal , compared to mmse based estimators , i.e. , simple ( possibly weighted ) averaging , which can even generate infeasible solutions @xcite . this is often the case especially for the computer vision and machine learning applications such as edge preserving image denoising @xcite . for instance , the gradients in an occluded pedestrian image would get too smoothed in an mmse based imputation , which might cause the gradient based feature extractors , e.g. , hog @xcite , to fail or not perform satisfactory in case of an pedestrian detection application @xcite . for these reasons , we propose a novel map based imputation technique , which always generates feasible and likely estimates and approximates the true map estimator as the size of the reference data increases . once a corruption is localized for an instance @xmath6 at a node @xmath71 , then our task is to estimate the original attributes @xmath78 using the training data set @xmath38 as well as the instance @xmath6 and impute accordingly , i.e. , replace the corrupted attributes in @xmath6 with the estimates . since we assume the corrupted attributes @xmath79 to be statistically independent with the underlying true data @xmath78 , we treat the corrupted attributes as missing data , which then should have no effect in estimation of the true attributes . hence , we condition this estimation of the data @xmath78 on the remaining attributes in @xmath6 . on the other hand , we note that in most of the applications such as the image compression @xcite , the data attributes being in sufficiently close proximity are usually modeled to manifest high correlation . in accordance , we propose to estimate the unknown data @xmath78 conditioned on the attributes @xmath80 associated with its nearest neighbor on our tree , i.e. , the sibling node @xmath81 of @xmath71 . note that due to the localization of corruptions by algorithm tcs ( tree - based corruption separation ) , the attributes at the sibling node @xmath81 are certainly detected to be uncorrupted in case of the standard euclidean distance ; and detected to be uncorrupted with significantly high probability in case of the ranked euclidean distance ( cf . section [ sec : acf ] ) . in the following , we introduce a novel _ maximum a posteriori _ ( map ) estimator of the true data underlying the corrupted attributes based on the standard euclidean distance ( @xmath75 with @xmath56 ) and then discuss the generalization over @xmath45 for the ranked euclidean distance measure . we also stress that the implementation of this estimator is only based on the outputs of our corruption localization algorithm , which are computed before the imputation phase in the course of algorithm tcs . therefore , computationally , the imputation phase that we develop is efficient such that it does not require further computations . * input : * @xmath82 ; @xmath83 initialize @xmath84 : set of corrupted attributes initialize @xmath85 : imputed test data create the root node @xmath86 and label create nodes @xmath87 and @xmath88 ; and label declare corruption at @xmath71 : @xmath89 impute attributes @xmath90 in @xmath46 declare corruption at @xmath91 : @xmath92 impute attributes @xmath93 in @xmath46 and * return:*@xmath94 and @xmath46 since the only relevant part of the test instance @xmath6 to the proposed map estimator is @xmath80 , we have @xmath95 where @xmath96 represents a realization of the conditional probability density of the true data underlying the corrupted attributes @xmath90 . then , the map estimator of @xmath96 maximizes the posterior distribution as @xmath97 for any @xmath98 , and under certain smoothness constraints on @xmath5 with @xmath99 , let @xmath100 hold with some probability @xmath101 , where @xmath102 ( w.r.t . the standard euclidean distance ) is the @xmath103-ball around @xmath96 in @xmath104 and @xmath105 . then we point out that @xmath106 hence , since @xmath103 can be made arbitrarily small , we obtain @xmath107 and by the bayes rule @xmath108 with probability @xmath54 , where the denominator is dropped since it does not depend on the maximizer , i.e. , @xmath96 . to approximate the map estimator given in , we adapt the nonparametric k - nearest neighbor ( knn ) based density estimation approach @xcite . let us define a small neighborhood around @xmath80 in @xmath109 as @xmath110 where @xmath111 is the euclidean distance and @xmath112 is the @xmath111 distance from @xmath113 to its nearest @xmath114th neighbor in @xmath115 for some @xmath116 . note that as @xmath117 , @xmath118 , where @xmath119 is the lebesgue measure . then yields @xmath120 with probability @xmath54 . when @xmath121 is sufficiently large with @xmath122 for some @xmath123 or @xmath124 is sufficiently small , we assume that @xmath125 is subject to negligible variations only . then , we ( with probability @xmath54 ) obtain the approximation : @xmath126 where , in order to obtain the corresponding maximum in the reference set @xmath38 , knowing the rank statistics in @xmath127 is enough , i.e. , explicitly estimating / computing the density is unnecessary . therefore , using the density function defined in , we obtain @xmath128 for sufficiently large @xmath121 , note that @xmath129 approximates @xmath130 @xcite , i.e. , @xmath131 @xmath132 based on which we replace , i.e. , impute , the corrupted attributes @xmath79 in the instance @xmath6 with @xmath133 and obtain the imputed data as @xmath46 . this estimator is implemented in algorithm tcs ( tree - based corruption separation ) at every node in the tree , where a corruption is detected . namely , we i ) obtain the @xmath37 neighbors of the test instance in the reference data set @xmath38 with respect to the attributes associated with the node @xmath81 ; ii ) for those neighbors in @xmath38 , find the one , say @xmath134 , attaining the largest score value defined in using the attributes associated with the parent node @xmath135 ; then iii ) impute the instance @xmath6 , which is detected to be corrupted at the node @xmath71 , using @xmath134 for the attributes @xmath136 . in the realistic case of high dimensional and limited data , when the standard euclidean distance is used as in our derivations , @xmath80 might include corrupted attributes even though it is detected as normal , which clearly adversely affects the calculation of the neighborhood @xmath137 in . in addition , @xmath76 might only include a small support of corruption , and then we would not like to impute @xmath76 completely . to overcome these two issues , we propose to use the ranked euclidean distance defined in . to this end , the neighborhood @xmath137 is defined using @xmath55 with an appropriate @xmath138 in . this cancels the adverse effect , up to a certain degree , of a possible corruption in @xmath80 as desired . nevertheless , recalling that @xmath55 only uses the @xmath45 fraction of the attributes @xmath139 and set the others free , @xmath75 is not a metric in the mathematical sense and then , as @xmath117 , @xmath118 does not hold . as a result , the correlation structure given in is less exploited in imputation as @xmath45 decreases . meanwhile , as @xmath45 decreases , the support of the detected corruption in @xmath140 increases , i.e. , localization improves . therefore , we obviously have a trade - off between the imputation quality and the localization , which is sensitive to the choice of @xmath45 and investigated in the experiments in greater detail . however , @xmath45 should be set typically around @xmath141 since we use half - way splits . finally , note that the imputation brings almost no further computational complexity , since these steps do computationally only depend on the anomaly detection results , cf . and , at the corrupted node , its sibling node as well as its parent node , which are all generated prior to the imputation steps . in the following section , the proposed framework is shown to achieve a constant false alarm rate in terms of the corruption detection . moreover , this false alarm rate is precisely calculated under a certain dependency structure among the anomalous / normal labels on the partitioning tree . since the imputation is an overwriting " operation , whether or not to impute a suspicious instance is certainly a critical " decision . in case of a false decision if the suspicious instance is in fact uncorrupted , i.e. , a false alarm in detecting corruptions " , the imputation would correspond to data loss . in this section , we study the rate of such occurrences and analyze the false alarm rate of the proposed algorithms in detecting corruptions . the anomaly detection test applied at every node in algorithm tcs ( tree - based corruption separation ) operates with a constant false alarm rate @xmath33 , whereas the proposed approach is able to reject corruptions at anomalous nodes . for example , when the terminating pattern in fig . [ fig : simplecorruption]b is encountered , all the anomalies that can be present in the tree rooted from the terminating pattern are rejected , i.e. , they are not counted as corruptions . for this reason , the false alarm rate of the proposed approach must be defined in the sense of corruptions as opposed to anomalies . to analyze this false alarm rate in detecting corruptions , one also must account for that the anomaly detection test at a node could be strongly correlated with the outputs of the previous tests in the course of algorithm tcs , since the data attributes are in general correlated . in this section , we first model the labeling of the nodes , i.e. , anomalous vs normal , on the partitioning tree , cf . [ fig : example1 ] , as a directed acyclic graph @xcite achieving a certain dependency structure and then derive the false alarm rate of algorithm tcs . under this modeling , we also show that the constant false alarm rate in detecting the local anomalies at each node globally maps to also a constant false alarm rate in detecting the corruptions . recall that algorithm tcs expands the binary tree in fig . [ fig : example1 ] for a given uncorrupted test instance @xmath142 and declares a corruption only if the conclusive pattern in fig . [ fig : simplecorruption]a is encountered or a leaf node is found anomalous in the described breadth - first search . in addition to the corruption localization as well as the imputation capabilities of the proposed algorithm tcs , let us denote the corruption detection in algorithm tcs by @xmath143 , if @xmath142 is detected to be corrupted and @xmath144 , otherwise . then our task is to find the false alarm probability in detecting the corruptions , which is given by @xmath145 where @xmath33 is the constant false alarm rate of the detection at each node and @xmath5 is the nominal density . next , we observe that algorithm tcs maps every data instance to a binary observation such that the nominal distribution @xmath5 is transformed into a multivariate bernoulli distribution @xmath146 , i.e. , @xmath147 where @xmath148 , @xmath70 is the depth and @xmath149 is the anomaly decision at the root node such that @xmath150 , if an anomaly detected ; and @xmath151 , otherwise . similarly for the others such as @xmath152 is the decision at the left hand child of the root and @xmath153 is the decision at the right hand child . note that the proposed algorithm does not completely construct the binary tree but expands , i.e. , the nodes and the edges are created as needed . therefore , we do not completely observe the binary vector @xmath154 that an instance @xmath142 maps to , however , we temporarily suppose that all the labels are available for ease of exposition . once @xmath142 is mapped to @xmath154 , since algorithm tcs declares a corruption based on only the vector of binary labels @xmath154 , we equivalently have @xmath155 where @xmath156 is the corruption decision ( with abuse of notation ) , @xmath157 is the complement , i.e. , @xmath158 and @xmath146 is the corresponding nominal probability mass function such that @xmath159 in order to calculate the probability mass function @xmath146 , we model the binary tree , where each node corresponds to a binary random variable , as a directed acyclic graph @xcite such that the binary random variables at any two sibling nodes are independent conditioned on the knowledge of the label at the parent node . namely , for any non leaf node @xmath71 and its children @xmath160 and @xmath161 on the binary partitioning tree , we assume the following conditional independency for the associated random labels : @xmath162 , from which we obtain @xmath163 cf . fig . [ fig : dag ] . here , we emphasize that @xmath142 ( or @xmath154 ) is assumed to be uncorrupted in the false alarm analysis to calculate the probability given in , i.e. , it does not have any localized corruptions by definition . then , if @xmath142 is declared , at the root node without loss of generality , as anomalous then this anomaly is not due to a corruption but simply a rarity " as the test in is based on density levels . on the contrary to the case of corruption , since a rarity " at a node is not a localized phenomenon , we expect that the children inherit the parent label independently . therefore , we assumed the conditional independency in as a generating dependency structure for the simplest graph presented in fig . [ fig : dag ] , which straightforwardly generalizes to the binary tree of the anomalous vs normal labels from root to the leaves . based on this , we obtain @xmath164 where @xmath165 is the collection of the binary variables associated with the nodes in the tree rooted from node @xmath62 that excludes @xmath149 , and the last equation follows from and the bayes rule . we observe that the starred factors in the expression are of similar forms such that the last equation can be expanded further using similar lines of derivations up until the leaves appear . thus , the calculation of @xmath166 requires the calculation of the probabilities of the form @xmath167 or @xmath168 , e.g. , @xmath169 in . let us denote any child of the node @xmath71 by @xmath81 for generalization . note that if @xmath170 and @xmath171 were independent then we would have @xmath172 when @xmath173 . however , we anticipate a statistical dependency between @xmath170 and @xmath171 generating a positive covariance . that is , conditioned on the knowledge of @xmath170 , we would like to impose that @xmath171 is more likely to attain the value @xmath174 compared to the prior conditions , i.e. , @xmath81 is likely to inherit the label of its parent . on the other hand , provided that @xmath170 and @xmath171 are identically dependent , we would have that @xmath175 , where @xmath176 is the indicator function . to introduce this into the derivations , we parameterize the probability mass function @xmath177 as the weighted average between @xmath178 and @xmath179 as @xmath180 where @xmath181 $ ] is a parameter defining the degree of dependency which generates an increasing covariance as @xmath182 increases in the interval @xmath183 $ ] such that @xmath184 implies the statistical independency of @xmath170 and @xmath171 ; and @xmath185 implies identical dependency . then , the probability mass function @xmath166 can be calculated using this parametrization based on the recursion in @xmath186 . hence , exhaustively enumerating all possible @xmath154 s and running algorithm tcs for each of them , one can calculate the false alarm rate @xmath187 in , which is not a practical choice . instead , through the conditional factorization in , we opt to simplify the expression and obtain an efficient recursion . to this end , for a given node @xmath71 with depth @xmath188 , let us define the probability conditioned on @xmath174 that algorithm tcs does not declare a corruption in the tree rooted from @xmath71 denoted by @xmath189 as @xmath190 here , @xmath189 solely depends on the depth variable @xmath191 due to the symmetric factorization by the conditional independency from parents to children . therefore , the notation simplifies to @xmath192 or @xmath193 . using the @xmath194 possible configurations for @xmath195 , we can calculate @xmath192 as a function of @xmath196 . noting that two of those configurations are the conclusive patterns , termination and corruption patterns , we obtain @xmath197 where @xmath198 as a short hand notation ; the second term corresponds to the continuation of algorithm tcs and the first term corresponds to the terminating pattern . unlike the second term , the first term does not have a multiplier since the search stops at such a node . note that the corruption pattern is disregarded by definition . similarly , we also have @xmath199 recalling that we declare corruptions at leaf nodes on the basis of local anomalies , we can further define @xmath200 and provide the initialization to the recursion @xmath192 and @xmath193 . on the other hand , we never declare corruptions at root since we are focused only on localized corruptions , which is an exception and can straightforwardly incorporated in our recursions . in terms of the recursions regarding @xmath192 , the only change is that the corruption pattern should not be disregarded which does not lead to a corruption detection and so does not stop the search . then , we simply have @xmath201 and the recursion @xmath193 stays valid for @xmath202 . now that we have the recursion equations defined for all depth levels on the binary tree , we can efficiently calculate the false alarm rate of algorithm tcs as follows . letting @xmath62 represent the root node , we obtain from @xmath203 then , recalling that @xmath204 , the false alarm rate @xmath187 is given by @xmath205 which is equivalent to first calculating the probability that algorithm tcs never declares a corruption and then subtracting this probability from @xmath54 . we point out that the false alarm rate @xmath187 of algorithm tcs ( tree - based corruption separation ) in detecting the corruptions as found in is a data - independent quantity . therefore , under the simplification through the conditional independency , we conclude that the false alarm rate @xmath33 of the anomaly detection at each node maps to a constant false alarm probability of our corruption detection @xmath187 . secondly , even though the dependency parameter @xmath182 does not appear , i.e. , hidden , in the expression , @xmath187 is clearly affected by @xmath182 . for example , if @xmath185 , i.e. , if the binary label of a child node is identically dependent on the parent label and hence @xmath206 , then it can be shown that @xmath207 . if @xmath184 , i.e. , if the binary label of a child node is independent with the parent label and hence @xmath208 , then obviously @xmath209 . in fig . [ fig : dagfig ] , we plot the hypothetical curves resulted from mapping the constant false alarm rate @xmath33 in detecting the local anomalies to the corruption false alarm rate @xmath187 via the described model of conditional independency for several degree of dependencies @xmath182 . we experimentally discuss the efficacy of this model in representing the relation between @xmath33 and @xmath187 in section [ sec : experiments ] . moreover , the parameter @xmath182 between @xmath174 and @xmath171 can also be chosen depth dependent , i.e. , @xmath210 , instead of a uniform choice over the partitioning tree . an example of a depth dependent modeling is given in section [ sec : experiments ] . finally , note that the directed acyclic graph modeling of the anomalous vs normal labeling uniformly holds for all @xmath45 s in choosing the ranked euclidean distance . we also discuss the impact of various @xmath45 s on the fitness of the described dependency structure in section [ sec : experiments ] . in the following section , we explain the important points of our implementation and discuss the corresponding computational complexity . computationally , the main building block in algorithm tcs ( tree - based corruption separation ) is the application of the anomaly test defined in , which computes the train - to - train distance matrix @xmath211 and the test - to - train distance vector @xmath212 . operating on these distances , the score function defined in for the test instance must be computed , which then requires the computation and sorting of @xmath213 . in addition , since we label each node as anomalous or not in our tree expansion , these distances must actually be computed at each node with respect to the corresponding attributes , e.g. , @xmath214 and @xmath215 at a node @xmath71 . for this purpose , we adapt the integral image " approach in case of using the standard euclidean distance . namely , let us define the volume @xmath216 , @xmath217 with @xmath218 ; and @xmath219 , @xmath217 ( similarly for @xmath220 ) . then , we simply have @xmath221 at a node @xmath71 , where @xmath136 corresponds to the set of attributes in positions between @xmath222 and @xmath223 . the volume @xmath224 and the sorting of @xmath213 can be computed offline once the training set is provided , which defines a training phase complexity @xmath225 , where sorting is the dominant contributor . for a given test instance , we compute @xmath226 and sort at each node @xmath71 in the expansion of our tree , which defines the test phase complexity @xmath227 for our algorithm , where sorting is the dominant contributor . in case of using the ranked euclidean distances , since it is not possible to utilize the integral image approach anymore , the computational load is multiplied by constant factors . next , we illustrate the efficacy of the proposed framework in separating , i.e. , detecting and localizing , corruptions and imputing . in this section , we test the introduced algorithm tcs ( tree - based corruption separation ) over several well - known machine learning data sets subject to synthetically generated data corruptions to demonstrate the performance of the proposed approach . we first discuss the efficacy of the false alarm rate estimation method explained in section [ sec : acf ] in terms of the corruption detection and evaluate the performance of the critical steps in algorithm tcs , which are the corruption detection , localization and imputation . then , we report the improvements achieved by the proposed framework in several classification tasks . in the first set of experiments , we adapted a @xmath228 digit classification task consisting of a training set of @xmath229 samples and a test set of @xmath230 samples based on the usps data @xcite . each of these samples is a @xmath231 gray scale image of either a 0 " image or a 1 " image , where each pixel has a real intensity value in [ 0,1 ] . we synthetically generate a corruption as described in section [ sec : pd ] and apply to each instance in the test set with probability @xmath232 . to be more precise , for a test instance chosen to be corrupted , we ( uniformly ) randomly specify a square region of size between @xmath233 of the total area , i.e. , the number of pixels in the chosen region is not less than @xmath234 and not more than @xmath235 , overwrite each pixel in this region with a value randomly ( using the uniform distribution @xmath26 ) drawn from the interval @xmath183 $ ] . then , after the training and test instances are vectorized column wise such that @xmath236 , the proposed algorithm tcs is provided with the clean training data and run over the test set . we emphasize that by this vectorization scheme , the corrupted square region corresponds to multiple corrupted intervals in the vectorized observation . hence , this example also illustrates that algorithm tcs can handle multiple corruptions . ideally , the neighborhood size parameter @xmath37 for both imputation and corruption separation purposes should be optimized at every node of our binary tree since the data dimensionality from node to node varies . however , we opt not to optimize @xmath37 for presentational clarity and set as @xmath237 near the midpoint of @xmath238 $ ] , which is empirically found appropriate . using the @xmath228 digit usps data , we investigate the response of the algorithm tcs to the local anomaly detection false alarm rate @xmath239 and the ranked euclidean distance parameter @xmath240 . as for the depth parameter , we use the deepest possible tree with @xmath241 such that the leaves are associated with @xmath194 pixels and hence , @xmath54 pixel at least is then used in the distance calculation with @xmath242 . = 6.4 cm in fig . [ fig : dagfig ] , we compare the hypothetical false alarm rate @xmath187 we derive in section [ sec : acf ] with the corresponding experimental realizations with respect to varying local anomaly detection false alarm @xmath243 . the hypothetical map from @xmath33 to @xmath187 is generated with several choices for the dependency parameter , @xmath244 , whereas the realizations correspond to several choices for the distance parameter , @xmath245 . our experiments indicate that when the statistical dependency @xmath182 in from a parent node to one of its children nodes , cf . [ fig : dag ] , is chosen around @xmath246 , the relationship between the local anomaly false alarm rate @xmath33 and the corruption detection false alarm rate @xmath187 is accurately modeled . this experimentally shows that the labeling of local anomalies over a binary partitioning tree shown in fig . [ fig : example1 ] can be considered as a directed acyclic graph . we also observe that in the case of euclidean distance , i.e. , @xmath55 with @xmath56 , while @xmath247 is more accurate for small @xmath33 s , @xmath182 tends to approach @xmath248 as @xmath33 increases for a better modeling . this small deviation mainly happens since the conditional independency assumption explained in fig . [ fig : dag ] do not hold in case of euclidean distance for a certain pattern . namely , although the labeling for a parent and its children nodes as @xmath249 ( a normal parent node with anomalous children nodes ) is not possible with the standard euclidean distance due to the test defined in , the directed acyclic graph modeling assigns it a positive probability , which then overestimates the corruption detection false alarm rate . nevertheless , this positive probability is the smallest among the ones assigned to the all possible patterns of three nodes as desired and hence , the ordering of the patterns in terms of their probabilities is still reasonable even in the case of the euclidean distance . on the contrary , since this pattern is also possible in case of the ranked euclidean distance , the accuracy of our hypothetical results improves as @xmath45 decreases . = 6.1 cm next , we study the corruption detection and localization performance of our algorithm on the @xmath228 digit usps data . in fig . [ fig : roc ] , we plot the empirical false alarm rates versus the empirical true detection rates in terms of both corruption detection and corruption localization with respect to @xmath243 . here , the true detection rate is the empirical probability , i.e. , relative frequency , of a truly corrupted data instance ( data attribute in case of localization ) to be declared corrupted and , the false alarm rate is the empirical probability of a truly uncorrupted data instance ( data attribute in case of localization ) to be declared corrupted . as we discuss in section [ sec : thethemethod ] , the ranked euclidean distance is experimentally shown to produce a better detection as well as localization performance on the usps data as @xmath45 decreases . recall that for a small @xmath45 around @xmath250 , we enforce a corruption to be widely spread for algorithm tcs to detect it at a node , which then clearly improves the localization . similarly , the corruption detection performance also improves as @xmath45 decreases . since the ranked euclidean distance disregards a certain fraction of largest attribute - wise deviations , algorithm tcs behaves conservatively in declaring corruptions . this reduces the false alarms in terms of the local anomalies and in turn , reduces the false encounters of the terminating pattern shown in fig . [ fig : simplecorruption]b . hence , the corruption search is not stopped mistakenly , and algorithm tcs does not miss certain corruptions , which leads to a better detection rate with the ranked euclidean distance using a small @xmath45 around @xmath250 . we emphasize that the local anomaly detection false alarm rate @xmath33 can be set independently for detection and localization to precisely determine the operating point on the roc curves in fig . [ fig : roc ] . however , in this study , we use one single @xmath33 in all phases of algorithm tcs . note that when the false alarm rate is set around @xmath251 , our algorithm is able to provide a detection rate around @xmath252 and a localization rate around @xmath248 . = 9.0 cm = 5.0 cm on the other hand , the ranked euclidean distance parameter @xmath45 can not be made arbitrarily small since , as @xmath45 decreases , the determination of the neighbors of a test instance in the reference set degrades and hence , the imputation quality is adversely affected . observe that with small @xmath45 , only a small fraction of attributes are used in determination of @xmath137 in despite that the rest of the attributes might be informative through the local correlations and hence , the imputation quality degrades . we illustrate this effect in fig . [ fig : impqual ] , where we use the improvements in the distance wise deviations after imputation to measure the imputation quality . for this purpose , we define @xmath253 as the distance wise imputation quality , where @xmath254 is the number of the corrupted test instance ( which is approximately @xmath255 ) , @xmath256 is a corrupted test instance , @xmath257 is the uncorrupted original instance and @xmath258 is the corresponding instance after imputation . note that this quality metric measures on average that how much of the distortion by the corruption is recovered after imputation . the average imputation quality defined in is plotted versus the local anomaly detection false alarm @xmath243 in fig . [ fig : impqual ] . we first observe that for large @xmath33 , since the false alarm rate is also large , the imputation even further disturbs the data . secondly , for small @xmath33 around @xmath259 , the proposed imputation technique is able to correct a corrupted instance up to @xmath260 in case of @xmath261 . moreover , our experiments also indicate that for @xmath45 less than @xmath250 , the ranked euclidean distance is not able to produce desirable results despite its superiority in terms of detection and localization , which reinforces our discussion about that @xmath45 can not be made too small . unlike an mmse based approach , our map based imputation does not target minimizing the reconstruction error but the most likely replacement for a corruption . therefore , the distance - wise imputation quality measure in is not fair to use for a comparison . indeed , an mmse based estimator for imputation would produce visually blurry results , for instance on the usps data . in this regard , we present several visual examples that the proposed framework generates on the usps data with @xmath262 in fig . [ fig : visualexamples ] . note that the presented visual examples tend to generate image gradients that are naturally aligned with the image statistics , since our imputation method is not based on an averaging to minimize the reconstruction error but filling in the missing part with the most likely candidate extracted from the reference data . we also observe some cases , where the corruption along a border between the cells of our partitioning tree remains after the imputation , cf . the second last column in fig . [ fig : visualexamples ] . the residual corruptions in such cases can be handled by increasing the depth of the tree or using m - ary trees as opposed to binary splits , which is not in the scope of this study . in addition to the visual comparisons , we also evaluate the performance of the introduced framework in terms of the classification purposes . on the described @xmath228 digit usps data , we report the data scatter plots of the test instances in fig . [ fig : thesep ] , where we project the original , corrupted and imputed test data onto the two eigen vectors of the training set with the largest eigenvalues for visualization . we clearly observe a better class separation between two classes after the imputation , when compared to the class separation in the corrupted data . the distance between the class means is calculated to be @xmath263 in the original uncorrupted data , @xmath264 in the corrupted data and @xmath265 after the imputation . this approximately corresponds to a @xmath266 performance improvement achieved by the proposed framework both in terms of the mean separation and classification accuracy , cf . [ fig : thesep ] . [ cols="^,^,^ " , ] [ table : impute ] our algorithms , tcs - map " ( tree - based corruption separation with map imputation ) , perform significantly better than the method m - nn " and comparably with the method tcs - nn " . we point out that the method tcs - nn " strongly relies on the proposed algorithm tcs , which is the reason underlying the success of tcs - nn " since the method m - nn " is outperformed by tcs - nn " ( both of the methods m - nn " and tcs - nn " uses the nn imputation but m - nn " does not have the proposed tree - based corruption separation step ) . the method m - nn " occasionally even further corrupts the data , cf . the negative improvements for sonar or g241c or g241n . moreover , it is difficult to choose between the methods 4-nn " and 16-nn " since there is no clear superiority . namely , there is definitely a scaling issue for the method if the corruption in an instance covers a small portion , e.g. , @xmath267 , then choosing @xmath268 too large would leave several undetectable corruptions in addition to imputing large chunks of clean data due to the false alarms ( negative improvements ) . similarly , if the corruption in an instance covers a large portion , e.g. , @xmath269 , then choosing @xmath268 too small would not only again harden the anomaly detection ( due to the use of insufficient data in detecting corruptions ) but also complicate the imputation since on what to condition the imputation becomes ambiguous . for a good imputation in this case , one would need to identify ( with significant imperfections due the use of insufficient data because of a small @xmath268 ) all the corrupted and uncorrupted segments ; and then condition the imputation on uncorrupted ones , which leads to a non - homogenous different evaluation for every instance that requires computationally a very high load . therefore , choosing an appropriate @xmath268 is in general hard since it must depend on the amount of the corruption , which might be unknown and random . the proposed framework resolves this scaling issue in a computationally efficient way via the binary searches and fast imputations . moreover , our algorithm tcs - map " is experimentally shown to be also robust to corruptions in training data in the sense that it strongly preserves its corruption separation / imputation capabilities even after including @xmath270 corruptions in training . lastly , our estimator that is used for imputation asymptotically ( as the data size increases ) recovers the true map estimator ; and the nn estimator is certainly asymptotically sub - optimal both in the mean square error ( mse ) and the likelihood maximization sense . however , the map imputation and the nn imputation ( only when combined with the proposed tree based corruption separation ) performs comparably in our experiments , which is due to the sparsity of the data compared to the dimensionality . we first note that the map estimator yields the nn estimator , if the neighborhood size is set @xmath271 in the imputation phase , which can easily be achieved via cross validation for @xmath37 . clearly , with such a cross validation , the map imputation can only perform better than the nn imputation and we opt not to optimize @xmath37 for presentational clarity . additionally , the nn imputation is definitely sensitive to corruptions in the training data since the attributes of the nearest neighbor that are used for imputation can also be corrupted with a certain probability , e.g. , with probability @xmath272 in our experiments . on the other hand , that possibly ( with probability @xmath272 ) corrupted nearest neighbor would achieve a lower score value @xmath59 if it was truly corrupted and it would not be picked by our map estimator for imputation . thus , the proposed map estimator can handle such situations and is robust to corruptions , where the nn imputation performance potentially degrades more . in fact , in our experiments , the map imputation either enhances its superiority or becomes superior or approaches nn imputation for most of the data sets after including corruptions in training , e.g. , image or ringnorm data sets in table [ table : nonlin ] . to further demonstrate the power of the proposed estimator , we devise a separate experiment , where we use a data set consisting of two gaussian components with unitary covariances . we use the means @xmath273 $ ] and @xmath274 $ ] for positive and negative classes , respectively . we generate @xmath275 samples as the training data , ( @xmath276 for each class ) , and next suppose that the second attribute of each sample is corrupted / missing ; and therefore imputed by our map estimator with varying neighborhood size @xmath37 and the nn estimator . note that in this part , we use standard euclidean distance only . after repeating this @xmath277 times , the resulting imputed data is compared to the original training data in the mse sense and in terms of the classification accuracy . we summarize our findings in table [ table : impute ] , where the map imputation with @xmath271 coincides with the nn imputation . the map imputation is consistently better than the nn imputation as expected . for instance , when @xmath278 , we obtain @xmath279 improvements in the mse sense ; and @xmath267 improvements in terms of the classification ( original classification accuracy is around @xmath280 ) . in this paper , we proposed a comprehensive framework for handling localized and severe data corruptions . the novel contributions of the proposed framework includes ( i ) a first algorithm to jointly detect and localize such corruptions by identifying the local anomalies , ( ii ) a _ maximum a posteriori _ based estimator for imputation ; and a distance measure for corruption separation purposes , ( iii ) computational efficiency via the binary searches and the fast imputations , and ( iv ) a characterization for anomalous observations , e.g. , rarities , incompatible combinations and corruptions . we point out that our algorithm does not assume prior information or a model for the input data and instead , works in a completely data driven way . furthermore , we conducted a false alarm rate analysis and showed that the desired false alarm rate in detecting corruptions can be set independently with the input data . our algorithm is tested against the synthetically generated corruptions in several well - known machine learning data sets and experimentally shown to provide signicant improvements in terms of classication purposes with strong corruption separation capabilities . the proposed algorithms outperform the typical approaches and are robust to varying training phase conditions . stefanos zafeiriou , georgios tzimiropoulos , maria petrou , and tania stathaki , `` regularized kernel discriminant analysis with a robust kernel for face recognition and verification , '' , vol . 3 , pp . 526534 , 2012 . roberto perdisci , guofei gu , and wenke lee , `` using an ensemble of one - class svm classifiers to harden payload - based anomaly detection systems , '' in _ sixth international conference on data mining . ieee icdm06 . _ , 2006 , pp . 488498 . david heckerman , david m. chickering , christopher meek , robert rounthwaite , and carl kadie , `` dependency networks for inference , collaborative filtering , and data visualization , '' , vol . 1 , pp . 4975 , 2001 . nelson h. yung and andrew h. lai , `` detection of vehicle occlusion using a generalized deformable model , '' in _ proceedings of the ieee international symposium on circuits and systems _ , 1998 , . 4 , pp .

we study the plaquette valence bond solid phase of the spin antiferromagnet heisenberg model on the square lattice within the bond operator theory . we start by considering four spins on a single plaquette and determine the bond operator representation for the spin operators in terms of singlet , triplet , and quintet boson operators . the formalism is then applied to the model and an effective interacting boson model in terms of singlets and triplets is derived . the effective model is analyzed within the harmonic approximation and the previous results of zhitomirsky and ueda [ phys . rev . b * 54 * , 9007 ( 1996 ) ] are recovered . by perturbatively including cubic ( triplet triplet triplet and singlet triplet triplet ) and quartic interactions , we find that the plaquette valence bond solid phase is stable within the parameter region , which is narrower than the harmonic one . differently from the harmonic approximation , the excitation gap vanishes at both critical couplings and . interestingly , for , the excitation gap corresponds to a singlet triplet excitation at the point while , for , it is related to a singlet singlet excitation at the point of the tetramerized brillouin zone .

1309.6490

two dimensional frustrated quantum antiferromagnets have been receiving a lot of attention in recent years . here the interplay between frustration ( dynamic or geometric ) and quantum fluctuations may destroy magnetic long range order ( lro ) yielding to quantum paramagnetic ( disordered ) phases , such as valence bond solids ( vbss ) with broken lattice symmetries or spin liquids , where lattice symmetries are preserved.@xcite an interesting example of a frustrated quantum magnet is the spin@xmath0 @xmath1@xmath2 antiferromagnet ( afm ) heisenberg model on the square lattice:@xcite @xmath11 here @xmath12 is an spin@xmath0 operator at site @xmath13 and @xmath14 and @xmath15 are , respectively , the nearest neighbor and next nearest neighbor exchange couplings as illustrated in fig . [ fig : model](a ) . several different theoretical approaches have been employed to study the @xmath1@xmath2 model in the last few years . @xcite it is now well established that the model has semiclassical nel magnetic lro with ordering wave vector @xmath16 for @xmath17 , collinear magnetic lro with @xmath18 or @xmath19 for @xmath20 , and a quantum paramagnetic ( disordered ) phase within the intermediate parameter region @xmath21 . however , the nature of such a disordered phase and the quantum phase transition at small @xmath2 are still under debate . these two issues are mainly associated with the fact that large scale quantum monte carlo simulations can not be used here due to the so - called sign problem.@xcite different proposals have been made for the ground state of the disordered phase of the @xmath1@xmath2 model : a columnar vbs [ fig . [ fig : dimer](a ) ] , where both translational and rotational lattice symmetries are broken,@xcite a plaquette vbs [ fig . [ fig : model](b ) ] , where only the translational lattice symmetry is broken,@xcite a mixed columnar plaquette vbs,@xcite and gapless spin liquids.@xcite more recently , evidences for a gapped @xmath22 spin liquid@xcite have also been found . 1.5 cm 1.5 cm about the quantum phase transitions : while there are strong indications@xcite that a first order quantum phase transition takes place at @xmath23 ( the boundary between the quantum paramagnetic and the collinear phases ) , it is still not clear whether a first order@xcite or a continuous@xcite quantum phase transition occurs at @xmath24 ( the boundary between the nel and the quantum paramagnetic phases ) . if a vbs phase sets in within the magnetic disorder region , the former scenario is in agreement with the landau ginzburg framework ( the nel and the vbs phases are characterized by two different order parameters ) while the latter is in favor of the so - called deconfined quantum criticality.@xcite a candidate theory for a possible continuous quantum phase transition between a @xmath22 spin liquid and a nel phase is recently proposed in ref . . ( open circle ) and @xmath25 ( filled circle ) form a singlet state . @xmath26 and @xmath27 are the primitive vectors of the dimerized lattices . , title="fig:",width=151 ] 0.5 cm ( open circle ) and @xmath25 ( filled circle ) form a singlet state . @xmath26 and @xmath27 are the primitive vectors of the dimerized lattices . , title="fig:",width=151 ] we should also mention that , more recently , the @xmath1@xmath2 model on the _ honeycomb _ lattice has also been studied . here the main motivation are quantum monte carlo results@xcite for the half filled honeycomb hubbard model which provide some evidences for a gapped spin liquid phase within intermediate values of the on site repulsion @xmath28 . density matrix renormalization group ( dmrg ) calculations have been performed on the honeycomb lattice @xmath1@xmath2 model and it is found that as @xmath29 increases , a nel phase , a plaquette and a dimerized vbs phases set in.@xcite similar results are reported in ref . , where the coupled cluster method is employed . an useful approach to describe vbs phases of a heisenberg model is the bond operator theory introduced by sachdev and bhatt.@xcite such a formalism can be seen as the analog of the holstein primakoff representation , but here we consider fluctuations above a quantum paramagnetic ground state instead of a ( semiclassical ) state with magnetic lro . the formalism developed in ref . is appropriate to describe dimerized phases , such as the columnar [ fig . [ fig : dimer](a ) ] and staggered [ fig . [ fig : dimer](b ) ] vbss.@xcite a generalized method suitable for describing tetramerized phases , such as the plaquette vbs [ fig . [ fig : model](b ) ] , was later introduced by zhitomirsky and ueda.@xcite however , here only a partial bond operator representation for the spin operators [ in terms of the lowest energy singlet and the triplet ( boson ) operators ] was considered : the high energy singlet and the quintet operators ( see below ) were neglected . in this paper , we revisit the work of zhitomirsky and ueda@xcite and study the plaquette vbs phase of the @xmath1@xmath2 model within the bond operator theory . we derive the full bond operator representation ( in terms of singlet , triplet , and quintet boson operators ) for spin@xmath0 operators on a single plaquette and apply such a generalized formalism to the @xmath1@xmath2 model . our study is not only restricted to the analysis at the harmonic ( mean field ) level of an effective boson model in terms of the lowest energy singlet and the triplet operators as done in ref . , but we also include the high energy singlet operator and go beyond the harmonic approximation : cubic ( singlet triplet triplet and triplet triplet triplet ) and quartic interactions are perturbatively considered . our main motivations are a series of results@xcite which indicates the stability of the plaquette vbs phase and a recent study@xcite concerning a dimerized phase of a triangular lattice heisenberg afm , where we show that cubic ( triplet triplet triplet ) interactions have an important role in the determination of the excitation spectrum of such a frustrated quantum magnet . @xmath2 model as obtained from bond operator theory within ( a ) the harmonic and ( b ) the cubic quartic approximations . , width=245 ] we calculate the ground state energy [ fig . [ fig : egs](b ) ] and the dispersion relation of the singlet and triplet excitations ( fig . [ fig : disp ] ) of the plaquette vbs phase within the ( mean field ) harmonic and the cubic quartic approximations . in the latter , cubic and quartic interactions are perturbatively added to the harmonic results . our main findings are the following : + ( a ) harmonic approximation . the plaquette phase is stable within the parameter region @xmath30 , see fig . [ fig : phase - diag](a ) . the excitation gap [ fig . [ fig : gap](a ) ] is always finite and it is related to a singlet triplet excitation ( triplet gap ) for @xmath31 and a singlet singlet one ( singlet gap ) for @xmath32 . + ( b ) cubic quartic approximation . the region of stability of the plaquette phase is @xmath33 [ fig . [ fig : phase - diag](b ) ] with the excitation gap vanishing at both critical couplings @xmath34 and @xmath35 [ fig . [ fig : gap](b ) ] . for @xmath36 , the excitation gap is no longer associated with a singlet triplet excitation at the @xmath8 point , but with a singlet singlet one at the @xmath10 point of the tetramerized brilluoin zone [ see fig . [ fig : model](c ) ] . the decay rates of the singlet and triplet excitations are also obtained [ see figs . [ fig : decay ] and [ fig : decay2 ] ] . the reader not interested in the technical details may skip secs . ii v and go straight to sec . our paper is organized as follows : in sec . [ sec : bondop ] , we generalize the ( dimer ) bond operator formalism@xcite for the case of four spins @xmath37 on a single plaquette . in sec . [ sec : boson - model ] , we apply the generalized bond operator representation to the @xmath1@xmath2 model and derive an effective model in terms of singlet and triplet boson operators . sec . [ sec : harmonic ] is devoted to the analysis of the effective boson model in the harmonic approximation . the ground state energy and the dispersion relations of the singlet and triplet excitations are calculated . in sec . [ sec : cubic ] , we consider cubic ( singlet triplet triplet and triplet triplet triplet ) interactions in second order perturbation theory and quartic ones in the ( no self consistent ) hartree fock approximation and calculate the corrections to the harmonic results ( cubic quartic approximation ) . we compare our results with previous ones and discuss their implications for the @xmath1@xmath2 model in sec . [ sec : discussion ] . our findings are summarized in the last section . some details of the calculations discussed in the main part can be found in the five appendixes . in ref . , a bond operator representation for two spins @xmath3 in a dimer is introduced . in this section , we consider the case of four spins @xmath3 in a plaquette and develop a bond operator representation for the spin operators in terms of singlet , triplet , and quintet ( boson ) operators . we should mention that such a formalism was already discussed in refs . and but , in that case , the high energy singlet state @xmath38 and the quintet states @xmath39 , @xmath40 , and @xmath41 ( see below ) were not considered . as far as we know , this is the first time that the complete bond operator representation for spins in a plaquette is derived . @xmath1@xmath2 afm heisenberg model on a single plaquette , eq . . ( b ) eigenvalues of the hamiltonian as a function of @xmath29 : @xmath42 , @xmath43 , and @xmath44 are , respectively , the energies of the singlet , triplet and quintet states.,width=170 ] 0.7 cm @xmath1@xmath2 afm heisenberg model on a single plaquette , eq . . ( b ) eigenvalues of the hamiltonian as a function of @xmath29 : @xmath42 , @xmath43 , and @xmath44 are , respectively , the energies of the singlet , triplet and quintet states.,width=268 ] let us consider the heisenberg model restricted to four spins in a single plaquette as illustrated in fig . [ fig : plaquette](a ) : @xmath45 it is easy to show that the eigenvalues of the hamiltonian are given by @xmath46 the behaviour of the spectrum as a function of @xmath29 is shown in fig . [ fig : plaquette](b ) . for @xmath47 , the ground state is given by the singlet state @xmath48 whose energy is @xmath49 . there are four excited energy levels : @xmath50 is the eigenvalue related to the singlet state @xmath38 . @xmath51 is the energy of the six triplet states @xmath52 and @xmath53 with @xmath54 while @xmath55 is the energy of the three triplet states @xmath56 . finally , @xmath44 is the eigenvalue associated with the five quintet states @xmath39 , @xmath40 , and @xmath41 . note that the excitation gap is associated with a singlet triplet transition ( triplet gap ) for @xmath57 and with a singlet singlet one ( singlet gap ) for @xmath58 . we refer the reader to appendix [ ap : hilbert ] for the explicit expressions of the singlet , triplet and quintet states in terms of the 16 states @xmath59 , @xmath60 , @xmath61 , @xmath62 , @xmath63 . as discussed in the previous section , the hilbert space of four spins @xmath37 ( @xmath64 , @xmath65 , @xmath66 , and @xmath67 ) in a single plaquette is made out of 16 states : two singlet , nine triplet , and five quintet states . we can introduce a set of boson operators which creates these states out of a fictitious vacuum @xmath68 , namely , @xmath69 with @xmath70 and @xmath54 . in order to remove unphysical states from the enlarged hilbert space , the constraint @xmath71 should be introduced . following the ideas of ref . for the dimer case , we calculate the matrix elements of each component of the four spins operators within the basis @xmath72 , i.e. , we determine @xmath73 , @xmath74 , @xmath62 , with @xmath75 . based on the obtained results , one concludes that the three components of the four spin operators @xmath76 can be written in terms of boson operators @xmath77 , @xmath78 , and @xmath79 as @xmath80 with @xmath75 and @xmath81 . here , the upper and lower signs refer respectively to @xmath82 and @xmath83 , @xmath84 and @xmath85 respectively for @xmath86 and @xmath87 , @xmath88 is the completely antisymmetric tensor with @xmath89 , @xmath90 is a symmetric tensor , @xmath91 , @xmath92 , and @xmath93 , and summation convention over repeated indices is implied . similarly , one shows that the hamiltonian assumes the form @xmath94 since the bond operator representation is quite involved , it is useful to consider an approximate expansion for the spin operators @xmath95 . in particular , neglecting the high energy quintet states , eq . reduces to @xmath96 where @xmath75 and the coefficients @xmath97 , @xmath98 , and @xmath99 are given by @xmath100 and zero otherwise . is quite similar to the bond operator representation for two spins @xmath3 in a dimer , see e.g. , eqs . ( 2.2 ) and ( 2.3 ) from ref . . the bond operator representation can be generalized to the lattice case and the corresponding heisenberg model can be expressed in terms of the boson operators @xmath101 , @xmath102 , @xmath103 , @xmath104 , @xmath105 , and @xmath106 . 1.7 cm in this section , we apply the bond operator formalism developed above to study the plaquette vbs phase of the @xmath1@xmath2 model . the idea is to map the heisenberg model into an effective boson model in terms of the singlet @xmath107 and the triplet @xmath108 operators . we start by rewriting the hamiltonian in terms of the underline ( tetramerized ) square lattice defined by the plaquettes as shown in fig . [ fig : model](b ) : @xmath109 here , the numbers @xmath110 and @xmath111 in the site indices @xmath112 , @xmath113 , etc respectively indicates the nearest - neighbor vectors @xmath114 with @xmath115 being the lattice spacing of the _ original _ square lattice ( in the following we set @xmath116 ) . note that the unit cell of the underline square lattice has four spins : @xmath117 , @xmath118 , @xmath119 and @xmath120 . we then substitute eq . generalized to the lattice case into eq . , i.e. , we consider the approximate bond operator representation where the high energy quintets are neglected , and , after some algebra , find that the hamiltonian assumes the general form : @xmath121 here @xmath122 is a constant , @xmath123,\ ] ] the terms @xmath124 contain @xmath125 triplet @xmath108 and @xmath126 singlet @xmath107 operators , and the constraint is taking into account by adding to the hamiltonian the term @xmath127 with @xmath128 being a lagrange multiplier . within the bond operator formalism , the plaquette vbs state shown in fig . [ fig : model](b ) can be seen as a condensate of the lowest energy singlets @xmath129 . in order to implement such a ( reference ) state , we replace @xmath130 in eq . . we then end up with an effective hamiltonian solely in terms of the triplet @xmath103 and the singlet @xmath102 boson operators . both @xmath128 and @xmath131 will be self - consistently determined later . 2.0 cm 1.0 cm 2.0 cm finally , performing a fourier transform , i.e. , @xmath132 @xmath133 where @xmath134 with @xmath135 being the number of sites of the original square lattice and the momentum sums run over the tetramerized brillouin zone [ fig . [ fig : model](c ) ] , we find that in momentum space the @xmath124 terms in eq . read @xmath136 @xmath137 @xmath138 @xmath139 @xmath140 , \label{h21}\end{aligned}\ ] ] @xmath141 , \label{h22}\end{aligned}\ ] ] with @xmath142 and @xmath81 . the coefficients @xmath143 , @xmath144 , @xmath145 , @xmath146 , @xmath147 , and @xmath148 can be found in appendix [ ap : coef ] . let us now study the effective boson model in the lowest order approximation , the so called harmonic approximation . in this case , we neglect @xmath149 , @xmath150 , @xmath151 , and @xmath152 and consider @xmath153 note that the hamiltonian is quadratic in the singlet @xmath154 and the triplet @xmath155 boson operators . moreover , the singlet sector is already diagonalized and decoupled from the triplet one . 2.0 cm in order to diagonalize the triplet sector @xmath156 , it is useful to introduce the six component vector @xmath157 which allow us to rewrite eq . in matrix form : @xmath158 here @xmath159 and the @xmath160 matrix @xmath161 reads @xmath162 with @xmath163 and @xmath164 being @xmath165 hermitian matrices whose elements are @xmath143 and @xmath144 respectively . although the diagonalization of the @xmath160 problem is quite involved ( we briefly outline the analytical procedure in appendix [ ap : diag ] ) , it is possible to show that , after the diagonalization , eq . acquires the form @xmath166 where @xmath167 is the ground state energy , the @xmath160 matrix @xmath168 reads @xmath169 and the six component vector @xmath170 is given by @xmath171 the relation between the two set of boson operators @xmath78 and @xmath172 is @xmath173 with @xmath174 and @xmath175 being @xmath165 matrices whose elements are the bogoliubov coefficients @xmath176 and @xmath177 . the explicitly expressions of the triplet excitation energies @xmath178 and the the bogoliubov coefficients @xmath176 and @xmath177 in terms of the @xmath143 and @xmath144 functions can be found in appendix [ ap : diag ] . finally , from the saddle points conditions @xmath179 and @xmath180 , self - consistent equations for @xmath128 and @xmath131 follow , namely @xmath181 , \nonumber \\ & & \label{self - eq } \\ n_0 & = & 1 + \frac{3}{2n'}\sum_{a,{{\bf k } } } \left [ 1 + \frac{\partial \omega_{a,{{\bf k}}}}{\partial\mu } \right ] . \nonumber\end{aligned}\ ] ] once @xmath128 and @xmath131 are numerically calculated , the triplet @xmath178 and the singlet @xmath182 excitation energies are completely determined . we numerically solve the self consistent equations and find solutions within the range @xmath183 as indicated in fig . [ fig : phase - diag](a ) . the behaviour of the parameters @xmath131 and @xmath128 and the ground state energy as a function of @xmath29 are respectively shown in figs . [ fig : egs](a ) and ( b ) . one sees that @xmath131 has a maximum at @xmath184 and that @xmath185 monotonically increases with @xmath29 . for comparison , we include the ground state energy of the dimerized columnar [ fig . [ fig : dimer](a ) ] and staggered [ fig . [ fig : dimer](b ) ] vbs phases as obtained from the ( dimer ) bond operator theory at the harmonic level ( see appendix [ ap : col - stag ] for details ) . note that the plaquette vbs state is the lowest energy one and that it extends over a region of the parameter space much larger than the dimerized vbss : the columnar vbs is stable for @xmath186 while the staggered vbs only for @xmath187 . figures [ fig : disp](a ) and ( b ) shows the energy of the triplet @xmath178 ( solid and dotted dashed lines ) and the singlet @xmath188 ( dashed line ) excitations for @xmath189 and @xmath190 , respectively . recall that @xmath188 is dispersionless in the harmonic approximation . one sees that for @xmath191 , the minimum ( gap ) of the triplet dispersion relation occurs at the center of the tetramerized brillouin zone [ @xmath8 point , see fig . [ fig : model](c ) ] while for @xmath192 , at the @xmath193 point . as shown in fig . [ fig : gap](a ) , such a changing in the momentum associated with the excitation gap takes place at @xmath194 . interestingly , the gap changes from a triplet gap to a singlet one at @xmath195 . finally , note that the excitation gap is always finite within the parameter region @xmath183 , i.e. , there is no indication of a continuous quantum phase transition at any critical coupling @xmath2 . since the energy of the singlet @xmath196 and the triplet @xmath197 excitations are quite close for @xmath198 [ see figs . [ fig : disp](a ) and ( b ) ] , it is interesting to consider the effects of the cubic interaction @xmath151 [ eq . ] . moreover , we have recently shown that cubic ( triplet triplet triplet ) interactions provide important renormalizations to the harmonic ( mean field ) excitation spectrum of a dimerized vbs phase in a frustrated quantum magnet.@xcite motivated by these two points , in this section we consider both cubic terms @xmath149 and @xmath151 within second order perturbation theory and calculate the corrections to the harmonic results determined in the previous section . we also consider the quartic terms @xmath150 [ eq . ] and @xmath152 [ eq . ] within the ( no self - consistent ) hartree fock approximation . although the quartic terms provide very small corrections to the harmonic results , they are important in the determination of the critical couplings . such a procedure constitutes the so called cubic quartic approximation . the first step is to express @xmath149 and @xmath151 in terms of the bosons @xmath172 . with the help of eq . , it is possible to show that @xmath199 \label{h30-v2}\end{aligned}\ ] ] and @xmath200 . \label{h21-v2}\end{aligned}\ ] ] here @xmath142 ( summation over repeated indices is assumed ) , the sum over @xmath201 has only three terms , @xmath202 , and the expressions of the renormalized cubic vertices @xmath203 and @xmath204 [ see fig . [ fig : vertices](a ) ] in terms of the bogoliubov coefficients @xmath176 and @xmath177 are given in appendix [ ap : cubic ] . 2.0 cm ( cubic quartic approximation ) . @xmath205 ( solid black line ) and @xmath206 ( dashed red line ) are respectively the decay rates of the lowest energy triplet excitation at the @xmath207 and @xmath10 points , see figs . [ fig : decay ] and [ fig : model](c ) . @xmath208 ( dotted dashed green line ) and @xmath209 ( dashed orange line ) are respectively the decay rates of the singlet excitation at the @xmath8 and @xmath210 points . , width=245 ] figures [ fig : vertices](b ) and ( c ) show the lowest order diagrams that contribute to the ( normal ) triplet self energies @xmath211 with @xmath212 , while fig . [ fig : vertices](d ) shows the ones related to the singlet self energy @xmath213 . the solid line in each diagram corresponds to the bare ( harmonic ) @xmath172 triplet propagator , @xmath214 and the dashed line denotes the bare @xmath196 singlet propagator , @xmath215 with @xmath182 . hereafter , we omit the @xmath216 index in the triplet green s functions and self energies since the @xmath217 triplet branches for each @xmath115 are degenerate . note that there are no bare anomalous @xmath172 propagators . although they can be generated in perturbation theory , we neglect them in the following ( for details , see note 40 from ref . ) . using standard diagrammatic techniques for bosons at zero temperature , we find that only the diagrams @xmath218 , @xmath219 , @xmath220 , @xmath221 , @xmath222 , and @xmath223 shown in fig . [ fig : vertices ] are finite , and therefore , @xmath224 and @xmath225 the expressions of the different components of the self energies and are shown in appendix [ ap : cubic ] . turning to the quartic terms @xmath150 and @xmath152 , it is possible to show that @xmath226 + \mathcal{o}(b^4 ) \label{h40-v2}\end{aligned}\ ] ] and @xmath227 where the constant @xmath228 and the coefficients @xmath229 and @xmath230 can be found in appendix [ ap : cubic ] . note that eq . is not diagonal in the @xmath115 and @xmath172 indices . the ( normal ) triplet and singlet self energies are then respectively given by @xmath231 the renormalized singlet @xmath232 and triplet @xmath233 excitation energies and the decay rates @xmath234 are given by the poles of the corresponding green s function @xmath235 : @xmath236 note that in addition to the anomalous hartree - fock self energies , the normal ones with @xmath237 are also neglected , since it significantly simplifies the determination of the poles of the green s function . the above equation is solved within the on - shell approximation,@xcite where the self energy is evaluated at the bare ( harmonic ) single particle energy : @xmath238 such a procedure , which is less involved than the off shell approximation adopted in ref . , provides reasonable results for the excitation spectra ( see below ) without the discontinuities and logarithmic singularities reported in ref . . finally , the ground state energy reads @xmath239 where @xmath185 is the harmonic term and the expressions of the corrections due to cubic [ @xmath240 and quartic [ @xmath228 ] interactions are presented in appendix [ ap : cubic ] . the renormalized singlet @xmath242 and triplet @xmath233 excitation spectra for @xmath189 and @xmath190 are respectively shown in figs . [ fig : disp](c ) and ( d ) while the corresponding decay rates @xmath234 , in figs . [ fig : decay](a ) and ( b ) . one sees that the excitation energies decrease as compared to the harmonic ones , an effect similar to what we have found in the triangular lattice quantum magnet.@xcite in particular , the singlet excitation branch , which now acquires a dispersion , is the lowest energy excitation for both configurations . we find that the contributions of @xmath243 and @xmath244 to the renormalized triplet spectra @xmath233 are much larger than the ones associated with @xmath245 and @xmath246 . moreover , we also find that the renormalizations due to the cubic vertices are stronger than the ones associated with the quartic interactions . the most important contributions of @xmath247 to the triplet excitation spectra occurs around @xmath248 . the behaviour of the excitation gap as a function of @xmath29 is shown in fig . [ fig : gap](b ) . note that the gap vanishes at the critical couplings @xmath249 and @xmath250 , indicating that the plaquette vbs phase is stable only within the parameter region @xmath33 . such a result sharply contrasts with the ones obtained within the harmonic approximation [ see figs . [ fig : phase - diag](a ) and ( b ) ] . moreover , as @xmath29 increases , the excitation gap changes from a triplet gap to a singlet one : for @xmath251 , the gap is associated with a singlet triplet excitation at the @xmath8 point while , for @xmath252 , with a singlet singlet excitation at the @xmath10 point . recall that such a change in the nature of the excitation gap for @xmath253 is similar to the behaviour of the @xmath1@xmath2 model on a single plaquette [ see fig . [ fig : plaquette](b ) ] . in addition to renormalize downward the excitation energies , the cubic vertices may also enable two particle decay of the singlet and triplet modes [ figs . [ fig : decay](a ) and ( b ) ] . in particular , note that for @xmath189 and @xmath190 , the triplet decay rate @xmath254 while the singlet one @xmath209 is finite . indeed , while the former is constant , the latter has an almost monotonic behaviour , decreasing with @xmath29 , see fig . [ fig : decay2 ] . such a result indicates that the excitation gap acquires a finite decay rate for @xmath36 , that decreases and ( almost ) vanishes closes to the critical coupling @xmath35.@xcite finally , we should note that cubic and quartic vertices provide very small corrections to the harmonic ground state energy , see fig . [ fig : egs](b ) . according to the harmonic bond operator theory ( sec . [ sec : harmonic ] ) , the plaquette vbs phase has lower energy than the dimerized columnar [ fig . [ fig : dimer](a ) ] and staggered [ fig . [ fig : dimer](b ) ] ones . moreover , the ground state energy of the plaquette phase monotonically increases with @xmath29 while , for the dimerized phases , @xmath185 is a convex function with a minimum around @xmath255 . the behaviour of the plaquette ground state energy qualitatively agrees with exact diagonalization data , which show that @xmath185 monotonically increases with @xmath29 , reaches a maximum around @xmath256 , and then decreases.@xcite such an agreement could be seen as a further indication that the plaquette phase might set in within the disordered region of the @xmath1@xmath2 model . a similar behaviour for the ground state energy is also observed in coupled cluster,@xcite hierarchical mean field,@xcite and tensor network states@xcite calculations . as mentioned in the introduction ( sec . [ sec : intro ] ) , zhitomirsky and ueda@xcite studied the plaquette vbs phase of the @xmath1@xmath2 model within the bond operator theory at the harmonic level without including the high energy singlet state @xmath38 . they found that the plaquette phase is stable for @xmath257 and that it has lower energy than the dimerized columnar vbs . in particular , for @xmath194 , they found that the excitation gap @xmath258 while the ground state energy @xmath259 . although the region of stability of the plaquette phase that we arrive at [ see fig . [ fig : phase - diag](a ) ] differs from their results , both harmonic ( mean field ) calculations show that the plaquette vbs phase extends over a region much larger than the @xmath1@xmath2 model paramagnetic one ( @xmath21 , see sec . [ sec : intro ] ) . our mean field results are in reasonable agreement with ref . : recall that we also find that the plaquette vbs state is more stable than the dimerized columnar state [ fig . [ fig : egs](b ) ] . moreover , for @xmath194 , the gap @xmath260 and the ground state energy @xmath261 . as described in sec . [ sec : cubic ] , cubic and quartic vertices strongly modify the harmonic singlet and triplet excitation spectra of the @xmath1@xmath2 model , similar to what we have recently found for a triangular lattice afm.@xcite one important consequence is that the region of stability of the plaquette vbs phase ( @xmath33 ) is reduced as compare with the harmonic one [ see figs . [ fig : phase - diag](a ) and ( b ) ] and it is now quite close to the disordered region of the @xmath1@xmath2 model found in previous calculations , see sec . [ sec : intro ] . such a result shows that cubic and quartic interactions are indeed relevant for a proper description of the plaquette vbs phase within the bond operator approach . we should note that although the cubic corrections to the harmonic results are much larger than the quartic ones , the latter has an important role in the determination of the lower critical coupling : including only the cubic vertices , we find that the region of stability of the plaquette phase is @xmath262 . although the region of stability derived within the cubic quartic approximation almost agrees with the paramagnetic region of the @xmath1@xmath2 model , the lower critical coupling @xmath34 is smaller than the ones reported in the literature , i.e , @xmath263 , see sec . [ sec : intro ] . in particular , it is even smaller than the one derived within linear spin wave theory , @xmath264 , ( corrections up to second order in the @xmath265 expansion of the sublattice magnetization even increase the lower critical coupling , i.e. , the region of stability of the nel phase increases when @xmath265 corrections are added to the linear spin wave results , see ref . for details ) . differently from spin wave theory , where @xmath265 can be taken as a small parameter,@xcite the bond operator formalism lacks such a quantity ( in principle , the density of excited triplets can be considered as a small parameter , see ref . for details ) and therefore , it is difficult to systematically determine corrections to the mean field results . we believe that the results derived here could be improved once : ( a ) the full singlet and triplet propagators , instead of the bare ones , are employed in the calculation of the normal triplet [ eqs . ] and singlet [ ] self energies ; ( b ) the anomalous cubic and quartic self energies are considered ; ( c ) the influence of the quintet excitations are taken into account ; and/or ( d ) an alternative treatment of the constraint is employed ( see sec . ii.c from ref . for details ) . however , it is difficult to say which one is the most relevant contribution to the determination of the phase boundary . the nature of the excitation gap of the plaquette vbs phase is also affected by cubic and quartic vertices : for @xmath7 , we find a triplet gap while for @xmath36 , a singlet one . it should be contrasted with the harmonic approximation : the gap changes from a triplet gap to a singlet one at @xmath266 . interestingly , one of the first exact diagonalization data@xcite for the @xmath1@xmath2 model indicates that the excitation gap is associated with a singlet singlet excitation for @xmath267 . we should also note that : ( i ) the hierarchical mean field approach@xcite also indicates that the excitation gap changes from a triplet to a singlet one , but at @xmath268 ; ( ii ) the dmrg calculations recently reported in ref . , which find some evidences for a @xmath22 spin liquid phase , point to a singlet gap smaller than the triplet one within the whole disordered region . cubic and quartic vertices also influence the nature of the phase transitions at small and large @xmath2 . recall that ( sec . [ sec : cubic ] ) for @xmath34 , a triplet gap vanishes , indicating a continuous quantum phase transition either to an ordered phase or to a mixed phase@xcite ( nel phase with plaquette modulation ) . as discussed in the introduction , the former scenario is in favor of the deconfined quantum criticality theory@xcite for the nel vbs transition while the latter scenario is in agreement with the landau ginzburg framework . on the other hand , for @xmath35 , a singlet gap vanishes which , in principle , points to a continuous quantum phase transition to a dimerized columnar vbs phase : note that a suitable linear combination of @xmath48 and @xmath38 [ see eq . and fig . [ fig : singlets ] ] yields a ( columnar ) dimer state . here , a continuous transition to a mixed phase ( columnar vbs with plaquette modulation ) should not be excluded either.@xcite such a result is in contradiction with previous ones ( see sec . [ sec : intro ] ) which indicate that a first order quantum phase transition takes place at @xmath269 from a quantum paramagnetic phase to a collinear ( ordered ) one ( see discussion below ) . finally , in order to check the accuracy of our results , it is interesting to compare the ground state energy and the excitation gap for @xmath255 , which is deep in the disordered phase , with the available data . within the cubic quartic approximation ( sec . [ sec : cubic ] ) , we find that @xmath270 , which is in reasonable agreement with ( plaquette ) series expansion results,@xcite coupled cluster calculations,@xcite the latest exact diagonalization data for @xmath271 sites,@xcite and a very recent dmrg ( ref . , see also note at the end of sec . [ sec : summary ] ) that respectively indicate that @xmath272 , @xmath273 , @xmath274 , and @xmath275 . in this case , one notices that the different methods agree fairly well . on the other hand , there is no consensus about the value of the excitation gap . for instances , we find for the triplet excitation gap @xmath276 , @xmath277 , @xmath278 , and @xmath279 respectively derived within dmrg,@xcite green function monte carlo,@xcite hierarchical mean field [ see also note ( i ) above],@xcite and exact diagonalization@xcite approaches . recall that [ fig . [ fig : gap](b ) ] we arrive at @xmath280 and @xmath281 for the singlet and triplet excitation gaps , respectively , which are larger than the values reported in the literature . as discussed above , such excitations gaps could decrease if , for instance , cubic and quartic interactions are self - consistently considered . the results that we have derived within the bond operator theory ( cubic quartic approximation ) allow us to state that if a plaquette vbs phase sets in for @xmath253 , then such a phase displays a _ excitation gap . this is the same feature of a possible spin liquid phase described by recent dmrg simulations.@xcite therefore , the determination of the nature of the excitation gap is not enough to make a distinction between the plaquette vbs phase and a @xmath22 spin liquid for @xmath253 . the fact that a singlet gap vanishes at @xmath35 disagrees with previous calculations ( see sec . [ sec : intro ] ) . such a result could indicate that : ( a ) the plaquette columnar vbs transition is indeed a true quantum phase transition and a first order columnar vbs collinear quantum phase transition takes place at a larger @xmath2 , ( b ) a first order quantum phase transition to the collinear phase pre emptes the plaquette columnar vbs transition , or ( c ) a mixed columnar plaquette phase@xcite may set in within the disordered region . it should be mentioned that the possibility of a series of intermediate paramagnetic phases between the nel and the collinear phases [ scenario ( a ) ] is discussed in ref . and that the plaquette columnar quantum phase transition was studied by kotov _ et al._,@xcite who showed that such a quantum critical point belongs to the @xmath282 universality class ( equivalent to 3d ising ) . we intend to investigate the above scenario ( c ) within the bond operator theory in a future publication . in this paper , we revisited the work of zhitomirsky and ueda@xcite and studied the plaquette vbs phase of the square lattice @xmath1@xmath2 afm model within the ( tetramerized ) bond operator theory . we improved the previous analysis by including the high energy singlet state within the description and perturbatively taking into account the effects of cubic ( singlet triplet triplet and triplet triplet triplet ) and quartic vertices above the harmonic ( mean field ) results . we showed that cubic and quartic interactions play an important role in the determination of the singlet and the triplet excitation spectra . as a consequence , the region of stability of the plaquette phase is smaller than the harmonic one . interesting , we found that at @xmath191 , the excitation gap of the plaquette vbs phase changes from a triplet gap to a singlet one , which vanishes at @xmath35 . we would like in the near future to apply the formalism discussed here to study the stability of the plaquette vbs phase in some extensions of the @xmath1@xmath2 model . for instance , the square lattice @xmath1@xmath2@xmath283 afm model , where there are some evidences@xcite that the inclusion of a next next nearest neighbor afm coupling @xmath283 favors the stability of the plaquette phase . we also believe that effects of anisotropy in the plaquette vbs phase can also be addressed . in this case , one candidate is the square lattice @xmath284@xmath285 afm model recently considered in ref . . as a final remark , we would like to mention that it would also be interesting to consider the afm @xmath1@xmath2 model on the honeycomb lattice within the procedure developed here . there are numerical evidences that a plaquette vbs phase may set in within the zero temperature phase diagram not only in the @xmath1@xmath2 model@xcite but also in the @xmath1@xmath2@xmath283 model.@xcite however , this is a much more involved task since the hilbert space of six spins @xmath37 on a hexagon has 64 states : five singlet , 27 triplet , 25 quintet , and seven septet states . in this case , it is very difficult to determine the bond operator representation , i.e. , the equivalent of eq . , for the spin operators . _ note added_. we recently became aware of dmrg calculations@xcite which indicates that the plaquette vbs phase is stable for @xmath286 . the authors also found that the nel order vanishes for @xmath287 and that a possible gapless spin liquid phase may set in for @xmath288 . we thank m. vojta , e. miranda , and a. o. caldeira for helpful discussions and fapesp , project no . 2010/00479 - 6 , for the financial support . and @xmath38 [ eq . ] . the symbols are the same as in figs . [ fig : model ] and [ fig : dimer].,width=207 ] in this section , we provide the expansion of the eigenvectors of the hamiltonian ( the two singlet , @xmath48 and @xmath38 , nine triplet , @xmath289 with @xmath70 and @xmath54 , and five quintet , @xmath39 , @xmath40 , and @xmath41 , states ) in terms of the 16 states @xmath59 , @xmath60 , @xmath61 , @xmath62 , etc . it is possible to show that @xmath290 where the upper and lower signs respectively refer to @xmath291 and @xmath292 , @xmath293 , and @xmath294 in particular , the singlet states can also be written as@xcite @xmath295[4,3 ] + [ 1,4][2,3 ] \right ) , \nonumber \\ & & \label{s0-s1 } \\ |s_1\rangle & = & [ 1,2][4,3 ] - [ 1,4][2,3 ] , \nonumber\end{aligned}\ ] ] where @xmath296 $ ] with @xmath297 denotes that the spins @xmath298 and @xmath299 form a singlet , see figs . [ fig : model ] and [ fig : singlets ] . note that @xmath48 is even while @xmath38 is odd under a @xmath300 rotation . here , we quote the explicitly expressions of the coefficients @xmath143 , @xmath144 , @xmath145 , @xmath146 , @xmath147 , and @xmath148 [ see eqs . ] : @xmath301 , \nonumber \\ & & \nonumber \\ \xi^{abc}_{{\bf k}}&= & -in^{1/2}_0\sum_n \left [ g^{abc}_3(n)e^{-i{{\bf k}}\cdot{{\bf n } } } + \bar{g}^{bca}_3(n)e^{i{{\bf k}}\cdot{{\bf n } } } \right ] , \nonumber \\ & & \nonumber \\ \chi^{abcd}_{{\bf k}}&= & -\frac{1}{2}\sum_n \left [ g^{abcd}_4(n)e^{i{{\bf k}}\cdot{{\bf n } } } + g^{cdab}_4(n)e^{-i{{\bf k}}\cdot{{\bf n } } } \right ] , \nonumber \\ & & \nonumber \\ \bar{\xi}^{ab}_{{\bf k}}&= & n^{1/2}_0\sum_n \left [ g^{ab}_{so}(n)e^{-i{{\bf k}}\cdot{{\bf n } } } + g^{ba}_{os}(n)e^{i{{\bf k}}\cdot{{\bf n } } } \right ] , \nonumber \\ & & \nonumber \\ \bar{\chi}^{ab}_{{\bf k}}&= & \sum_n \left [ g^{ab}_{ss}(n)e^{-i{{\bf k}}\cdot{{\bf n } } } + g^{ba}_{ss}(n)e^{i{{\bf k}}\cdot{{\bf n } } } \right ] , \label{coef - hboso01}\end{aligned}\ ] ] with @xmath302 , and @xmath303 . @xmath304 and @xmath55 are the triplet eigenvalues of the single plaquette hamiltonian and the @xmath305 coefficients are given by @xmath306 @xmath307 @xmath308 @xmath309 @xmath310 with the replacements @xmath311 , @xmath312 with the replacements @xmath313 , and @xmath314 with @xmath315 . here @xmath316 corresponds to the nearest - neighbor vectors and the @xmath317 , @xmath318 , and @xmath319 coefficients are shown in eq . . in this section , we briefly summarize the analytical procedure used to diagonalize the triplet sector of the harmonic hamiltonian . in order to deal with such a @xmath160 problem , we follow the procedure described in refs . and . it should be mentioned that we have recently employed this scheme to diagonalize a similar @xmath320 problem.@xcite since we are considering a bosonic system , instead of @xmath161 [ see eq . ] , we should diagonalize @xmath321 where @xmath322 is the @xmath165 identity matrix . it is easy to show that the ( positive ) eigenvalues of the matrix are ( roots of a cubic polynomial ) @xmath323^{1/2 } , \nonumber \\ & & \\ \omega_{3,{{\bf k } } } & = & \left [ -\frac{1}{3}a_{2,{{\bf k } } } + 2{\rm re}(s_{{\bf k } } ) \right]^{1/2 } , \nonumber\end{aligned}\ ] ] where @xmath324 the coefficients @xmath325 read @xmath326 , \nonumber \\ & & \nonumber \\ a_{1,{{\bf k } } } & = & \omega^2_{1,{{\bf k}}}\omega^2_{2,{{\bf k } } } + \omega^2_{1,{{\bf k}}}\omega^2_{3,{{\bf k } } } + \omega^2_{2,{{\bf k}}}\omega^2_{3,{{\bf k } } } \nonumber \\ & & \nonumber \\ & - & 4(b^{12}_{{\bf k}})^2\left ( a^{11}_{{\bf k}}- b^{11}_{{\bf k}}\right ) \left ( a^{22}_{{\bf k}}- b^{22}_{{\bf k}}\right ) \nonumber \\ & & \nonumber \\ & - & 4(b^{23}_{{\bf k}})^2\left ( a^{22}_{{\bf k}}- b^{22}_{{\bf k}}\right ) \left ( a^{33}_{{\bf k}}- b^{33}_{{\bf k}}\right ) \nonumber \\ & & \nonumber \\ & - & 4(b^{13}_{{\bf k}})^2\left ( a^{11}_{{\bf k}}- b^{11}_{{\bf k}}\right ) \left(a^{33}_{{\bf k}}- b^{33}_{{\bf k}}\right ) , \nonumber \\ & & \nonumber \\ a_{2,{{\bf k } } } & = & - \left(\omega^2_{1,{{\bf k } } } + \omega^2_{2,{{\bf k } } } + \omega^2_{3,{{\bf k } } } \right),\end{aligned}\ ] ] where @xmath327 with @xmath328 , and @xmath303 , and @xmath143 and @xmath144 [ see eq . ] are respectively the elements of the @xmath165 hermitian @xmath163 and @xmath164 [ see eq . ] . the determination of the bogoliubov coefficients @xmath176 and @xmath177 , the elements of the @xmath165 matrices @xmath174 and @xmath175 , is quite involved . using the properties of the matrix @xmath329 ( see sec . 5 from ref . ) and after some lengthy algebra , it is possible to show that @xmath330 , \nonumber \\ & & \nonumber \\ v^{jb}_{{\bf k}}&= & \frac{\mu_{jb,{{\bf k}}}}{g_{b,{{\bf k } } } } \left(\omega_{b,{{\bf k } } } - a^{jj}_{{\bf k}}+ b^{jj}_{{\bf k}}\right ) \left(\omega_{b,{{\bf k } } } + a^{33}_{{\bf k}}- b^{33}_{{\bf k}}\right ) , \nonumber \\ & & \nonumber \\ v^{3b}_{{\bf k}}&= & \frac{i}{g_{b,{{\bf k } } } } \left [ b^{33}_{{\bf k}}\nu_{b,{{\bf k } } } + 2\lambda_{b,{{\bf k}}}b^{23}_{{\bf k}}\left ( a^{22}_{{\bf k}}- b^{22}_{{\bf k}}\right ) \right . \nonumber \\ & & \nonumber \\ & + & \left . 2\mu_{b,{{\bf k}}}b^{13}_{{\bf k}}\left ( a^{11}_{{\bf k}}- b^{11}_{{\bf k}}\right ) \right],\end{aligned}\ ] ] where @xmath331 , @xmath332 and @xmath333 \nonumber \\ & & \nonumber \\ & + & \nu_{b,{{\bf k}}}\left ( \omega_{b,{{\bf k } } } + a^{33}_{{\bf k}}- b^{33}_{{\bf k}}\right)\left [ \nu_{b,{{\bf k}}}\left ( \omega_{b,{{\bf k } } } + a^{33}_{{\bf k}}+ b^{33}_{{\bf k}}\right ) \right . \nonumber \\ & & \left . \nonumber \\ & + & \left . 4\lambda_{b,{{\bf k}}}b^{23}_{{\bf k}}\left ( a^{22}_{{\bf k}}- b^{22}_{{\bf k}}\right ) + 4\mu_{b,{{\bf k}}}b^{13}_{{\bf k}}\left ( a^{11}_{{\bf k}}- b^{11}_{{\bf k}}\right ) \right ] , \nonumber \end{aligned}\ ] ] with @xmath334 , and @xmath303 . in this section , we study the dimerized columnar [ fig . [ fig : dimer](a ) ] and staggered [ fig . [ fig : dimer](b ) ] vbs phases of the @xmath1@xmath2 model within the ( dimer ) bond operator formalism@xcite at the harmonic approximation . we only quote the main results and refer the reader to secs . ii and iii from ref . for more details . ( thick solid black line ) and @xmath335 ( thin solid red line ) are the energy of the triplet excitations [ eq . ] of the columnar vbs at the @xmath336 and @xmath207 points of the dimerized brillouin zone . similar for the staggered vbs , @xmath335 ( thick dashed black line ) and @xmath337 ( thin dashed red line ) are the energy of the triplet excitations at the @xmath207 and @xmath338 points of the dimerized brillouin zone . the dotted lines are the gap of the plaquette valence bond solid ( triplet excitations , harmonic approximation ) as shown in fig . [ fig : gap](a ) . , width=268 ] the effective model [ the equivalent of eq . ] in terms of the boson triplet operators @xmath339 with @xmath54 reads @xmath340 \nonumber\\ & & \nonumber \\ & + & \frac{1}{2\sqrt{n'}}\epsilon_{\alpha\beta\lambda}\sum_{{{\bf p}},{{\bf k}}}\xi_{{{\bf k}}-{{\bf p } } } \ ; t^\dagger_{{{\bf k}}-{{\bf p}}\alpha}t^\dagger_{{{\bf p}}\beta}t_{{{\bf k}}\lambda } + { \rm h.c . } \nonumber \\ & & \nonumber \\ & + & \frac{1}{2n'}\epsilon_{\alpha\beta\lambda}\epsilon_{\alpha\mu\nu } \sum_{{{\bf q}},{{\bf p}},{{\bf k } } } \gamma_{{\bf k}}\ ; t^\dagger_{{{\bf p}}+{{\bf k}}\beta}t^\dagger_{{{\bf q}}-{{\bf k}}\mu}t_{{{\bf q}}\nu}t_{{{\bf p}}\lambda}. \label{boso - dimer}\end{aligned}\ ] ] here , @xmath341 with @xmath135 being the number of sites of the original square lattice , the momentum sum runs over the dimerized brillouin zone , @xmath342 , \nonumber \\ & & \nonumber \\ \xi_{{\bf k}}&= & -\sqrt{n_0}\left[j_1\sin(2k_x ) + j_2\sin(2k_x + k_y ) \right . \nonumber \\ & & \left . \nonumber \\ & + & \left . j_2\sin(2k_x - k_y ) \right ] , \nonumber \\ & & \nonumber\\ \gamma_{{\bf k}}&= & -\frac{1}{2}\left [ j_1\cos(2k_x ) + 2(j_1 + j_2)\cos k_y \right . \nonumber \\ & & \left . \nonumber \\ & + & \left . j_2\cos(2k_x + k_y ) + j_2\cos(2k_x - k_y ) \right]\end{aligned}\ ] ] for the columnar vbs , and @xmath343 , \nonumber \\ & & \nonumber \\ \xi_{{\bf k}}&= & \sqrt{n_0}j_1\left[\sin k_y + \sin(k_y - k_x ) - \sin(k_x ) \right ] , \\ & & \nonumber \\ \gamma_{{\bf k}}&= & -\frac{1}{2}\left [ ( 2j_2 + j_1)(\cos k_x + \cos k_y ) + j_1\cos(k_x - k_y ) \right ] \nonumber \end{aligned}\ ] ] for the staggered vbs . in deriving eq . , we considered the following nearest neighbor vectors : @xmath344 and @xmath345 ( columnar ) and @xmath346 and @xmath345 ( staggered ) , see fig . [ fig : dimer ] , and we set @xmath116 . similar to the plaquette phase , the parameter @xmath347 is the average value of the singlet operator @xmath348 while @xmath128 is the lagrange multiplier that enforce ( on average ) the constraint on the total number of bosons per site ( dimerized lattice ) . within the harmonic approximation , the hamiltonian can be diagonalized , and therefore one finds that the ground state energy is given by @xmath349 while the energy of the triplet excitations assume the form @xmath350 after self consistently calculating @xmath131 and @xmath128 , we find the behaviour of the ground state energy [ fig . [ fig : egs](b ) ] and the excitation gaps ( fig . [ fig : gap - col - stag ] ) in terms of @xmath29 . recall that for the columnar vbs phase , the @xmath336 and @xmath207 vectors correspond to the @xmath351 and @xmath352 vectors of the original ( nondimerized ) square lattice.@xcite the renormalized cubic vertices @xmath203 and @xmath353 [ see fig . [ fig : vertices](a ) and eqs . and ] in terms of the bogoliubov coefficients @xmath176 and @xmath177 are given by @xmath354 @xmath355 with the replacement @xmath356 , @xmath357 and @xmath358 with the replacement @xmath359 . finally , it is possible to show@xcite that only the cubic vertices @xmath365 and @xmath366 in fig . [ fig : vertices](a ) contribute to the ground state energy and therefore , we have @xmath367 the correction to the ground state energy due to the quartic terms read @xmath368 the fact that @xmath205 , @xmath206 , and @xmath208 are not strictly zero and that @xmath209 has small oscillations in fig . [ fig : decay2 ] might be related to the numerical procedure employed to evaluate the integrals in eqs . and .

galaxies with stellar bulges are generically observed to host supermassive black holes ( smbhs ) . the hierarchical merging of galaxies should therefore lead to the formation of smbh binaries . merging of old massive galaxies with little gas promotes the formation of low - density nuclei where smbh binaries are expected to survive over long times . if the binary lifetime exceeds the typical time between mergers , then triple black hole ( bh ) systems may form . we study the statistics of close triple - smbh encounters in galactic nuclei by computing a series of 3-body orbits with physically - motivated initial conditions appropriate for giant elliptical galaxies . our simulations include a smooth background potential consisting of a stellar bulge plus a dark matter halo , drag forces due to gravitational radiation and dynamical friction on the stars and dark matter , and a simple model of the time evolution of the inner density profile under heating and mass ejection by the smbhs . we find that the binary pair coalesces as a result of repeated close encounters in% of our runs , and in% of cases a new eccentric binary forms from the third smbh and binary remnant and coalesces during the run time . in about 40% of the runs the lightest bh is left wandering through the galactic halo or escapes the galaxy altogether , but escape of all three smbhs is exceedingly rare . the triple systems typically scour out cores with mass deficits - 2 their total mass , which can help to account for the large cores observed in some massive elliptical galaxies , such as m87 . the high coalescence rate , prevalence of very high - eccentricity orbits , and gravitational radiation `` spikes '' during close encounters in our runs , may provide interesting signals for the future laser interferometer space antenna ( lisa ) . [ firstpage ] black hole physics cosmology : theory galaxies : elliptical and lenticular , cd galaxies : interactions galaxies : nuclei methods : numerical

astro-ph0612517

in the favored cold dark matter cosmology , present - day galaxies were assembled hierarchically from smaller building blocks at earlier cosmic times . since all nearby galaxies with stellar spheroids are observed to host nuclear smbhs @xcite , hierarchical merging leads inevitably to the formation of smbh binaries @xcite . if the binary lifetime exceeds the typical time between mergers , then some galactic nuclei should contain systems of three or more smbhs . these systems are particularly interesting as they often lead to the ejection of one of the bhs at a speed comparable to the galactic escape velocity @xcite . in massive elliptical galaxies the typical speeds are @xmath2 , far greater than attainable through gravitational radiation recoil @xcite . spatially resolved pairs of nuclei have been observed in a few active galaxies . the most famous example is ngc 6240 , an ultraluminous infrared galaxy ( ulirg ) in which two distinct active galactic nuclei ( agn ) are clearly seen in hard x - rays at a projected separation of @xmath01 kpc @xcite . @xcite observed a variable uv source , possibly a second active nucleus , at a projected separation of @xmath3 pc from the primary nucleus in the spiral galaxy ngc 4736 , which shows signs of a recent merger . @xcite have detected what is thought to be an smbh binary at a projected separation of just 7.3 pc in the radio galaxy 0402 + 379 , through multi - frequency radio observations using the very long baseline array ( vlba ) . we begin by discussing the theory of how such systems evolve , and the conditions under which they might acquire a third bh . when two galaxies merge , their dense nuclei sink to the center of the merger product by dynamical friction . as the nuclei spiral in , tidal forces gradually strip the two smbhs of their surrounding stars and dark matter . in mergers between galaxies of comparable mass , the bhs are able to come together and form a bound smbh binary on a timescale of order @xmath4 yrs . the binary continues to harden by dynamical friction until it reaches a separation of order @xmath5 known as the `` hardening radius '' ( e.g. @xcite ) . here @xmath6 is the reduced mass of the two bhs with masses @xmath7 and @xmath8 , @xmath9 is the velocity dispersion of the stars beyond the binary s sphere of gravitational influence , @xmath10 is the binary mass ratio @xmath11 , and @xmath12 is the total mass of the binary . for smaller separations the binary looks like a point mass to the distant stars contributing to dynamical friction , but close stellar encounters preferentially harden the binary and so dominate further energy loss . only stars on nearly radial orbits , with periapsis distances of order the binary separation , can extract energy from ( `` harden '' ) the binary in this stage . these stars undergo strong 3-body interactions with the binary and escape its vicinity with speeds comparable to the black holes orbital speed . in the low - density nuclei of large elliptical galaxies , the total mass in stars on such `` loss cone '' orbits is small compared to the mass of the binary . furthermore the two - body stellar relaxation time is long compared to a hubble time , so once the stars initially on loss cone orbits are cleared out , the loss cone remains empty @xcite . since the binary must eject of order its own mass per @xmath13-folding in its semi - major axis , the system stops hardening around @xmath14 unless some other mechanism causes sufficient mass flux through the binary . if the binary reaches a separation around @xmath15^{1/4 } \hspace{2pt } \cdot \notag \\ & \left ( \frac{\tau_{gw}}{10^{10 } \hspace{2pt } { \rm yrs } } \right)^{1/4 } f^{-1/4}(e ) \hspace{6pt } { \rm pc},\end{aligned}\ ] ] where @xmath13 is the orbital eccentricity of the binary and @xmath16 , then it can coalesce on a timescale @xmath17 through gravitational radiation @xcite . to get from @xmath14 to @xmath18 it must bridge a gap @xmath19^{3/4}\ ] ] by some mechanism other than stellar - dynamical friction or gravitational radiation . the question of whether and how it crosses this gap has become known as the `` final parsec problem '' @xcite . in many galaxies there probably are alternative mechanisms for crossing the gap . when gas - rich galaxies merge , tidal torques channel large amounts of gas into the central @xmath20 pc @xcite . the gas may lose energy through radiation and angular momentum through viscous torques , and is therefore not subject to a loss cone problem . using smoothed particle hydrodynamics simulations @xcite compute a merger time of order @xmath21 yrs in an environment typical of the central regions of ulirgs , which are thought to be gas - rich galaxies caught in the act of merging @xcite . the nuclei of galaxies are also observed to contain numerous massive perturbers ( mps ) such as star clusters , molecular clouds , and possibly intermediate - mass black holes ( imbhs ) . these objects scatter stars into the loss cone much more efficiently than other stellar mass objects , since the relaxation rate scales as the perturber mass for a fixed mass density of perturbers . @xcite extended the fokker - planck loss cone formalism @xcite to accomodate a spectrum of perturber masses and account for relaxation by rare close encounters with mps . they show that the population of known mps in the nucleus of the milky way is sufficient to bring a @xmath22 m@xmath23 bh binary to @xmath18 in @xmath24 yrs , and it is reasonable to expect similar perturber populations in other star - forming spiral galaxies . the final parsec problem is often mentioned as a caveat when predicting the smbh coalescence signal in low - frequency gravitational wave detectors such as the upcoming laser interferometer space anntena ( lisa ) . however the lisa event rate is expected to be dominated by small galaxies at high redshift @xcite , where the gas content and central densities tend to be high and the relaxation times short . for this reason _ the stalling problem is probably not a significant concern for the lisa smbh coalescence signal_. on the other hand bh ejections by gravitational radiation recoil @xcite may play an important role in the high - redshift coalescence rate . the long - term survival of smbh binaries is likewise unlikely in the gas - rich cores of quasars and ulirgs . however none of the gap - crossing mechanisms discussed so far are likely to reduce the coalescence time below a hubble time in mergers between giant , gas - poor elliptical galaxies . @xcite show that a significant fraction of stars on `` centrophilic '' orbits in a triaxial potential can greatly increase the mass flux into the loss cone . some non - axisymmetric potentials can also excite bar instabilities that cause rapid mass flow through the binary and efficient coalescence @xcite . however a central smbh can disrupt box orbits and induce axisymmetry in the inner regions of a triaxial galaxy @xcite , and it is uncertain how often these geometry - specific mechanisms bring the coalescence time below a hubble time . one can naively assess the likelihood of coalescence by considering the `` full '' and `` empty '' loss cone hardening times , @xmath25 and @xmath26 , in the nuclei of various galaxies assuming a spherical and isotropic distribution function . @xmath25 is the hardening time assuming every star kicked out of the loss cone is instantly replaced , while @xmath26 is the time assuming stellar two - body relaxation to be the only replenishing mechanism . in small , dense galaxies @xmath27 yrs and @xmath28 yrs while in the lowest - density cores of giant ellipticals and cd galaxies @xmath29 yrs and @xmath30 yrs @xcite . while the empty loss cone rate is difficult to believe in any galaxy given at least some clustering on scales larger than 1 m@xmath23 , it also seems difficult to approach the full loss cone rate if there is no gas around and no strong radial bias in the stellar distribution . from this point of view the stalling of binaries seems unlikely in small galaxies but probable in low - density , gas - poor ellipticals . if some binaries do survive for around a hubble time , then the hierarchical buildup of galaxies will inevitably place three or more smbhs in some merging systems . an inspiralling satellite affects the evolution of a binary smbh even long before it sinks to the center , by perturbing the large - scale potential and scattering stars into the loss cone . we may estimate the extent of this effect as a function of satellite mass and distance from the center of the host galaxy using a rough but simple argument due to @xcite . the change in velocity necessary to deflect a star at radius @xmath10 into the loss cone is @xmath31 , where @xmath32 is the characteristic specific angular momentum of stars on loss cone orbits @xcite , @xmath33 is the smbhs radius of influence , and @xmath34 is the binary separation . the dynamical time at this radius is @xmath35 , so the acceleration required to scatter a star into the loss cone is roughly @xmath36 . equating this with the tidal acceleration caused by the satellite , @xmath37 , where @xmath38 is the satellite s radius , yields @xmath39 , or with @xmath40 , @xmath41^{1/3 } r.\ ] ] the @xmath38-dependence of @xmath42 reflects the tidal stripping of the satellite as it spirals inward . equation defines a critical radius @xmath10 , outside of which the satellite can deflect stars into ( and out of ) the loss cone in one dynamical time . the mass flux through the binary induced by the satellite is then approximately @xmath43 where @xmath44 is the density of the host galaxy at radius @xmath10 and @xmath45 is the geometrical factor accounting for the fraction of stars on loss cone orbits as a function of radius @xmath38 , assuming an isotropic distribution function @xcite . for a fixed satellite mass and distance , we can then define a `` binary feeding '' timescale by @xmath46 to determine whether the scattering of stars into the loss cone by the satellite is sufficient to harden the binary enough to prevent a close 3-body encounter before the intruder arrives at the galactic center , we must compare @xmath47 with the timescale on which the satellite spirals in by dynamical friction . in the approximation of slow inspiral we may write the dynamical friction timescale as @xmath48},\ ] ] where @xmath49 . @xmath50 in equation is computed from @xmath51 , where @xmath52 is the mass of the host galaxy enclosed within radius @xmath38 . @xmath53 is the satellite mass contained within the tidal truncation radius obtained from a simple point mass approximation , @xmath54^{1/3 } r$ ] ( this slightly underestimates @xmath55 as the satellite approaches the center of the host ) . in fig . 1 we plot @xmath56 and @xmath47 as a function of @xmath38 for a satellite with one third the stellar mass of the host , which contains a binary with @xmath57 m@xmath23 . both host and satellite are modelled as hernquist profiles @xcite , with their masses and effective radii set by observed scaling relations . the details of the galactic model are described further in 2.1 - 2 . , on which an inspiralling satellite scatters mass into the loss cone of an smbh binary , with the dynamical friction timescale , @xmath56 , on which the satellite spirals in . upper ( red dashed ) line : @xmath47 computed from equation ; lower curve : @xmath56 computed from equation . both timescales are plotted as a function of the satellite s distance from the center of the host galaxy . for this plot we chose a binary mass of @xmath58 m@xmath23 and merger mass ratio of 3:1 in the stars . the galactic model is discussed in the text . ] since @xmath47 remains about an order of magnitude above @xmath56 throughout the inspiral , this simple calculation makes it plausible that the binary survives the merger process and undergoes close triple interactions with the infalling smbh . the tidal approximation ( as well as our treatment of dynamical friction ) breaks down as the satellite approaches @xmath59 , so the plot is cut off at a separation of @xmath20 pc , when the satellite still has @xmath04 @xmath13-foldings to go to reach @xmath14 . however this final stage of the inspiral is found to proceed very rapidly in n - body simulations @xcite . the merger s effect on the binary may be dominated by violent relaxation or collective effects such as a bar instability @xcite , in which case our two - body approach does not capture its essence . the evolution of the core distribution function under the influence of a major merger is an intriguing open problem for simulators . after the third bh becomes bound to the binary ( but still before the onset of close 3-body interactions ) another hardening mechanism may become important . if the angle of inclination @xmath60 of the outer binary ( formed by the intruder and the inner binary center - of - mass ( com ) ) exceeds a critical angle @xmath61 , then the quadrupolar perturbation from the intruder induces eccentricity oscillations through a maximum @xcite @xmath62 since the gravitational radiation rate increases sharply toward high eccentricities , these `` kozai oscillations '' can greatly enhance the radiation , possibly causing the binary to coalesce before it can undergo strong 3-body interactions with the intruder @xcite . general relativistic precession can destroy the kozai resonance ( e.g. @xcite ) , but @xcite find that this does not happen for @xmath63^{1/3 } \sqrt{\frac{1-e_{in}^{2}}{1-e_{out}^{2}}},\ ] ] where @xmath64 and @xmath65 are the semi - major axes of the inner and outer binaries , @xmath66 is the outer binary mass ratio , and @xmath67 and @xmath68 are the inner and outer eccentricities . this leaves a window of about a factor of 10 in @xmath69 in which the kozai mechanism can operate before unstable 3-body interactions begin . the actual enhancement of the gravitational radiation rate of course depends on the amount of time spent at high eccentricity , but one may place an upper limit on the importance of kozai oscillations by computing the radiation timescale if the inner binary spends all of its time at @xmath70 . the orbit - averaged power radiated by gravitational radiation is given by @xmath71 @xcite , where @xmath72 is the semi - major axis and @xmath13 is the eccentricity . in fig . 2 we plot contours of the gravitational radiation time @xmath73 in the @xmath72-@xmath60 plane by putting @xmath70 into equation , for an equal - mass @xmath74 m@xmath23 binary . m@xmath23 ) . the cosine of the initial inclination angle is plotted on the horizontal axis , and the inner binary semi - major axis is plotted on the vertical axis . contours are plotted for gravitational radiation timescales of @xmath75 , @xmath76 , @xmath77 , and @xmath78 yrs , if the binary were to stay at maximum eccentricity throughout the whole oscillation cycle . the horizontal lines indicate the hardening radius and the separation such that a circular binary would coalesce on a @xmath79 yr timescale . ] this may seem like a gross overestimate of the gravitational radiation rate , especially since the shape of the kozai oscillations is in fact such that the binary spends more time near @xmath80 than near @xmath70 . however since @xmath17 is so strongly dominated by periapsis passages at @xmath81 , the shift in the contours for a realistic high-@xmath13 duty cycle is only modest . see @xcite for comparison with a detailed study of radiation enhancement by kozai oscillations in binaries with initial @xmath82 yrs . for a binary at @xmath14 , kozai oscillations can induce coalescence within @xmath76 yrs in @xmath83% of cases assuming @xmath84 is uniformly distributed . in the remainder of cases the inner binary may survive until the outer binary shrinks to the point of unstable 3-body interactions . if the intruder comes close enough before it causes sufficient hardening of the ( inner ) binary , then a strong 3-body encounter takes place . strong encounters are characterized by a significant transfer of energy between the binary s internal degrees of freedom and the com motion of the binary and third body . when the intruder is slow relative to the binary s orbital speed @xmath85 , energy typically flows from the inner binary to the outer components , so that the binary is more strongly bound after the encounter . this is one manifestation of the negative specific heat characteristic of gravitationally bound systems . the encounter ends in the escape of one of the three bodies , usually the lightest , from the system at a speed comparable to @xmath85 . when the lightest body @xmath86 escapes , momentum conservation requires that the binary com recoil in the opposite direction with a speed smaller by a factor @xmath87 . it is instructive to compare the expected ejection velocities of the binary and @xmath86 with the typical galactic escape velocity . for a circular binary with @xmath88 , the binding energy at the hardening radius is @xmath89^{3/2}$ ] erg . the binding energy at the radius where @xmath90 yrs is @xmath91^{5/4}$ ] erg . the mean energy @xmath92 harvested from the binary in close encounters with slow intruders is about @xmath93 , though the median @xmath92 is somewhat lower @xcite . energy conservation implies that the escaper leaves the system with kinetic energy @xmath94 $ ] while the binary leaves with @xmath95 $ ] in the system com frame . for an equal mass binary with @xmath96 m@xmath23 , this gives ejection velocities of @xmath97 and @xmath98 for the binary at @xmath14 , and @xmath99 and @xmath100 for the binary at @xmath18 . any nonzero eccentricity of the binary will increase the semi - major axis corresponding to a fixed @xmath17 , lowering the ejection velocities for the binary at @xmath18 . also any deviation from equal masses will result in a smaller fraction of the extracted energy being apportioned to the binary and a smaller binary recoil velocity . the typical escape velocity for galaxies hosting @xmath101 m@xmath23 bhs is around 1500 @xmath102 , accounting for both the stars and the dark matter . from these numbers , it appears that single escapes will be fairly common as repeated encounters harden the binary to @xmath103 . however accounting for realistic mass ratios and eccentricities ( the first 3-body encounter tends to thermalize the eccentricity even if it starts off circular ) , binary escapes should be rare . since the binary must come near the escape velocity to remain outside the nucleus for a significant amount of time , we do not expect triple interactions to empty many nuclei of bhs . we will quantify these statements with our triple - bh simulations . the formation of triple smbh systems through inspiral of a merging satellite leads to a rather specific initial configuration . the three bhs start off as a bound `` hierarchical triple , '' consisting of an inner binary with @xmath104 and a more widely separated outer binary with semi - major axis @xmath65 . for very large @xmath69 we expect hierarchical triples to exhibit very regular behavior ; in this case the third body sees the inner binary as a point mass and the system essentially consists of two independent ( inner and outer ) binaries . however as @xmath69 approaches unity , secular evolution gives way to chaotic 3-body interactions in which the orbits diverge and the system becomes subject to escape of one its components . @xcite derive a criterion for the stability of 3-body systems based on an analogue with the problem of binary tides . the most distant intruder orbit at which unstable interactions can begin is reliably estimated by @xmath105^{2/5},\ ] ] where @xmath106 is the periapsis separation of the outer binary , @xmath64 is the semimajor axis of the inner binary , @xmath107 is the outer binary mass ratio , and @xmath68 is its eccentricity . this criterion has great practical importance due to the high numerical cost of unnecessarily following weak hierarchical systems . it specifies an optimal starting point for our simulations , which aim to study strong interactions in 3-body systems starting off as hierarchical triples . naively one might expect a strong 3-body encounter following a merger with a galaxy hosting a binary , so long as the intruder does not induce coalescence of the binary before it reaches the center . however the stability criterion implies a condition for close interactions much more stringent than this . to undergo a chaotic encounter with the inner binary , the intruder must reach the stability boundary before the outer binary hardens and stalls . a triple system covers somewhat more stellar phase space than a binary of the same size , but not by much for a stable hierarchical system . this means that the merger process can not cause the binary to harden by more than around an @xmath13-folding for a nearly circular , equal - mass system before the intruder arrives at the center . though the order - of - magnitude estimates in the previous section make this plausible , further study is needed to determine the likelihood of unstable triple interactions in realistic merger situations . an eccentric outer binary relaxes the criterion somewhat , but dynamical friction tends to circularize the orbits of satellites with moderate initial eccentricities before they reach the nucleus @xcite . we therefore assume near - circular initial orbits and begin each simulation from a weakly hierarchical configuration . triple smbh systems in galactic nuclei were first considered by @xcite , who computed an extensive series of newtonian 3- and 4-body orbits , and compared the slingshot ejection statistics to the observed structure of extragalactic radio sources . @xcite included a gravitational radiation drag force in the 3-body dynamics . he showed that this perturbing force could in some cases yield much higher ejection velocities than would be possible in newtonian gravity , with associated bursts of gravitational waves . the more complex problem of three or four smbhs coming together in the hierarchical merging process and interacting in a galactic potential was first addressed by @xcite and @xcite , who experimented with a variety of initial bh configurations . @xcite studied binary - binary scattering in galactic nuclei using initial conditions ( ics ) based on extended press schechter theory @xcite . @xcite followed the formation of triple bh systems in halo merger trees tracking the hierarchical buildup of smbhs from @xmath108 m@xmath23 seeds in high-@xmath9 peaks at @xmath109 . using a simple analytic prescription for the ejection velocities , they inferred the presence of a large population of smbhs and imbhs wandering through the halos of galaxies and intergalactic space . @xcite performed the first full n - body simulations of equal - mass triple bh systems embedded in stellar bulges , an important contribution to our understanding of galactic nuclei . because of the large computation time required for each run , they could not statistically sample the highly varied outcomes of the 3-body encounters as the previous authors did . in this paper we study the dynamics of repeated triple - smbh interactions in galactic nuclei . between close encounters we follow the wandering bhs through the galaxy as their orbits decay by dynamical friction . we use physically - motivated initial bh configurations and mass distributions , and updated galactic models characteristic of the low - density , massive elliptical galaxies in which smbh binaries are most likely stall . we include both a stellar and a dark matter component , with the stellar spheroid fixed to lie on the observed @xmath110 and @xmath111 relations @xcite . the close encounters are treated using a ks - regularized bulirsch - stoer integrator provided by sverre aarseth @xcite . the inner density profile is updated throughout the simulations to roughly account for core heating by dynamical friction and stellar mass ejection . gravitational radiation losses are modelled as a drag force determined by the relative coordinates and velocities of each pair . each simulation takes only a few minutes to run , so we can try a variety of distributions of ics and statistically sample the outcomes for each . we use this algorithm to study a variety of consequences of the ongoing encounters , such as the merging efficiency of bh pairs , the time spent wandering at various distances from the galactic center , the distribution of final sizes and eccentricities of the binaries remaining in the galaxy after a steady state has been reached , and the extent of the core scouring caused by the triple smbh systems . aside from the motivating order - of - magnitude calculations in previous sections , this paper does not address the question of _ whether _ close triple smbh systems form in galactic nuclei . we start our simulations from a state that the system must reach shortly before the onset of unstable 3-body interactions _ assuming _ that they occur , and proceed to derive the subsequent evolution . our results may be used to argue for or against the occurrence of triple systems in real galaxies , as observations support or disfavor the signatures that we derive . in 2 we describe our model and code methods . in 3 we present the results of our study , and in 4 we discuss these results and conclude . to get a physically motivated distribution of bh mass ratios , we associate the formation of the inner and outer binaries with the last two major mergers in the history of the galactic halo hosting the triple system . we use extended press - schechter theory @xcite to calculate the probability distributions of the halo formation times and progenitor masses , and randomly select the parameters of the previous two mergers from these distributions . we then assign a bh to each progenitor halo using a simple prescription based on the assumption of a flat galactic rotation curve . @xcite derive the instantaneous halo merger rate , @xmath112}{[1-\sigma^{2}(m_{f})/\sigma^{2}(m_{1})]^{3/2}}.\end{gathered}\ ] ] this equation gives the probability , per unit time per unit mass of @xmath113 , of a given halo of mass @xmath114 merging with another halo of mass @xmath113 to form a product of mass @xmath115 @xmath116 @xmath117 at time @xmath118 . here @xmath119 is the present - day variance of the linear density field on mass scale @xmath120 , @xmath121 where @xmath122 is the power spectrum of density fluctuations today , @xmath123 is a tophat window function , and @xmath38 is related to @xmath120 through @xmath124 , the volume times the present - day matter density . @xmath122 is related to the primordial power spectrum through the transfer function @xmath125 , which encapsulates the suppression of perturbations on small scales due to radiation pressure and damping over the history of the universe . for @xmath125 we adopt the standard fitting formulae of @xcite . for the linear growth function @xmath126 we use the approximation @xmath127 \left [ 1+\frac{\omega_{\lambda}(z)}{70 } \right ] } , \ ] ] good to within a few percent for all plausible values of @xmath128 and @xmath129 @xcite . @xmath130 is the growth function for an einstein - de sitter universe , @xmath131 $ ] is the matter density ( normalized to the critical density ) as a function of redshift , and we take @xmath132 assuming the rest of the density is in the form of a cosmological constant . @xmath133 has the weak redshift dependence @xcite @xmath134.\ ] ] we adopt the cosmological parameters obtained from three years of data collection by the wilkinson microwave anisotropy probe ( wmap ) , @xmath135 , @xmath136 , @xmath137 , @xmath138 , and @xmath139 @xcite . since the merger rate diverges as @xmath140 @xmath141 0 , applications of the formula that track individual merging halos must employ a cutoff mass ratio @xmath140 @xmath142 @xmath143 , such that all mergers below @xmath143 are treated as smooth _ accretion _ rather than as discrete mergers ( see @xcite for further discussion ) . the instantaneous rate of accretion onto a halo of mass @xmath120 at redshift @xmath144 is @xmath145 to get the growth history ( `` accretion track '' ) of a halo of mass @xmath146 at time @xmath147 due to accretion since the last merger , one need only solve the differential equation @xmath148 @xmath116 @xmath149 $ ] , subject to the initial condition @xmath150 @xmath116 @xmath146 . we integrate this equation backward in time using a 4th - order runge - kutta method to get the accretion tracks of the halos in our simulations . since we are interested in bh binary formation , we loosely associate @xmath143 with the halo mass ratio such that tidal stripping of the satellite would prevent the eventual merging of the two nuclei . n - body simulations of galaxy mergers place this mass ratio in the range @xmath143 @xmath0 @xmath151 , depending on the density and orbital parameters of the satellite @xcite . hence our canonical choice is @xmath143 @xmath116 0.3 , and we also try values of @xmath152 ( runs d1 ) and 0.5 ( runs d5 ) , the latter being the halo mass that corresponds to a stellar mass ratio of @xmath0 3:1 in our prescription . following @xcite , we write the probability , per unit time , of a halo with mass @xmath115 at time @xmath118 arising from a merger with a smaller halo of mass between @xmath120 and @xmath153 ( the `` capture rate '' ) as @xmath154 \hspace{2pt } \frac{n(m , t)}{n(m_{f},t ) } \hspace{2pt } dm,\end{gathered}\ ] ] the eps merger rate excluding halos below the threshold @xmath143 , and weighted by the number of mass @xmath120 halos per unit halo of mass @xmath115 . the rate at which halos of mass @xmath115 form through all mergers at time @xmath118 is @xmath155 the probability distribution function ( pdf ) of formation times of halos with mass @xmath146 at time @xmath147 is @xmath156 \hspace{2pt } e^{-\int_{t}^{t_{0 } } r_{f}[m(t'),t ' ] dt'}.\ ] ] given a formation time @xmath157 and corresponding mass @xmath158 along the past accretion track of @xmath146 , the mass of the larger progenitor @xmath114 is distributed according to @xmath159 = \frac{2g(m_{1},m)}{\int_{m\delta_{m}/(1+\delta_{m})}^{m/(1+\delta_{m } ) } g(m',m ) dm'},\ ] ] where @xmath160^{-3/2}.\ ] ] by choosing formation times and progenitor masses randomly according to and , we capture the stochasticity of the intervals between mergers above @xmath143 , but treat merging below this threshold only in the mean . see @xcite , @xcite , and @xcite for further details and derivations of and . 3 shows the distribution of formation times , progenitor masses , and accretion tracks for a present - day @xmath161 m@xmath23 halo for @xmath162 0.1 , 0.3 , and 0.5 , and the accretion tracks for 1 , 2.3 , and 5 @xmath163 m@xmath23 halos with @xmath143 fixed at 0.3 . all accretion tracks are normalized to the present - day mass @xmath146 . note the insensitivity of the shape of these tracks to @xmath146 , as expected for masses above the critical mass @xmath164 . halo at @xmath165 . _ upper right : _ pdf of masses of the larger progenitor of the same halo given by equation , normalized to the mass of the merger product , @xmath166 . _ lower left : _ past accretion tracks of a present - day @xmath167 halo back to @xmath168 , normalized to the mass at @xmath165 . _ lower right : _ normalized accretion tracks for three different halo masses . ] our algorithm for generating the bh masses is illustrated schematically in fig . 4 . for each run we begin with a halo of mass @xmath146 @xmath116 @xmath161 @xmath169 at time @xmath147 @xmath142 @xmath170 , choose it s formation time @xmath171 randomly according to equation , and find the mass @xmath172 along its accretion track at that time . the mass @xmath173 is assigned to the dark matter halo hosting the triple bh system , and the physical time for the run to end if other termination conditions are not met first is set to @xmath174 . to explore the dependence of the results on the absolute mass scale , we also try beginning with a @xmath175 @xmath169 halo ( runs h1 ) . we model the halo as a hernquist profile @xcite , which is identical to an nfw profile @xcite in its inner regions if the scale radius @xmath176 is related to the nfw scale radius by @xmath176 @xmath116 @xmath177}$ ] , where @xmath178 is the halo concentration defined by @xmath179 . the hernquist model falls off as @xmath180 instead of @xmath181 far outside @xmath176 @xcite . the virial radius @xmath182 is given by @xmath183^{1/3 } h^{-1 } { \rm kpc},\end{gathered}\ ] ] where @xmath184 @xmath116 @xmath185 - 39[\omega_{m}(z)-1]^{2}$ ] @xcite , and @xmath178 roughly follows the median relation from the @xmath186cdm simulations of @xcite , @xmath187^{0.13}/(1+z)$ ] . the @xmath144 dependence of @xmath188 and @xmath178 nearly cancel to make @xmath176 depend only weakly on redshift , so we simply use the @xmath165 relation between @xmath189 and @xmath176 in our simulations . we choose the mass @xmath114 of the larger progenitor of @xmath173 randomly according to equation , and assign a mass @xmath113 @xmath116 @xmath190 to the smaller progenitor . before the merger the larger progenitor is assumed to have hosted a bh binary , while the smaller one hosted a single bh . repeating the procedure used for @xmath146 , we assign formation times @xmath191 and @xmath192 to @xmath114 and @xmath113 using equation , and choose progenitor masses @xmath193 , @xmath194 , @xmath195 , and @xmath196 according to equation . having constructed a set of progenitor halos , we now need a bh - halo relation @xmath197 to complete our algorithm . we obtain such a relation by equating the halo virial velocity @xmath198 to the circular velocity @xmath199 of the stellar spheroid , and using empirical @xmath200 and @xmath201 correlations to connect @xmath199 to @xmath202 , similar to the approaches in @xcite and @xcite . combining @xmath203^{1/6 } { \rm km \hspace{3pt } s^{-1 } } \end{gathered}\ ] ] @xcite with @xmath199 @xmath204 @xmath205^{0.84}$ ] @xmath102 @xcite and @xmath206 @xmath204 @xmath207^{1/4.02}$ ] @xcite , we arrive at the relation @xmath208 where @xmath209^{-1/2}$ ] . in our canonical runs we set the masses of the inner binary members to @xmath210 and @xmath211 , and that of the intruding bh to @xmath212 . note that in this prescription the intruder is usually lighter than the heavier binary member , so that most of the 3-body interactions result in an exchange . to examine the effect of more interactions without exchange , we try choosing @xmath213 $ ] for the intruder mass in runs ( mx ) . as there is neither a direct causal relationship between @xmath202 and @xmath189 predicted by theory @xcite nor a tight correlation directly observed between these two variables , and we know that identical halos may host galaxies of different morphologies and occupation numbers , @xmath197 should be taken with something of a grain of salt . nevertheless it is a useful way to generate simple but physically - motivated bh mass distibutions when no information other than the halo mass is available . for the halo @xmath146 hosting the bh triple system randomly from equation . given @xmath214 and @xmath215 , we select two progenitor masses @xmath114 and @xmath113 according to equation , assign the binary to the larger one and the third bh to the smaller one . we repeat this process going back one step further in the `` merger tree '' to get the masses of the binary constituents . ] we make one final modification to the set of bh masses used in our simulations . if the outer binary s hardening radius lies outside the stability boundary given by equation with @xmath216 , then the decay of the outer orbit is expected to stall before a strong encounter can begin . to roughly account for this we exclude all ics where @xmath217 @xmath218 @xmath219 . the final distribution of bh mass ratios is shown in fig . 5 for @xmath1620.1 , 0.3 , and 0.5 . . _ lower panel : _ @xmath220 , where @xmath221 is the mass of the lightest bh and @xmath222 is the sum of the masses of the other two bhs . ] in the upper panel we plot the inner binary mass ratios , while the lower panel shows the distribution of @xmath220 , where @xmath221 is the mass of the lightest bh and @xmath222 is the sum of the masses of the other two bhs . this ratio determines the binary recoil speed when the lightest bh is ejected from the system . the total bh mass is typically @xmath24 m@xmath23 in our canonical runs . to complete the galactic model we surround the bh system by a smooth stellar potential superimposed on the dark matter halo . the stars are modelled using the `` @xmath223-models '' of @xcite , with a sharp break to shallower slope @xmath224 added at @xmath225 : @xmath226 our canonical model is the @xmath227 ( hernquist ) profile , and we also try @xmath228 ( runs sc ) to explore the effect of a steeper inner profile and higher central density ( @xmath229 for @xmath228 vs. @xmath230 for the hernquist profile at the bh radius of influence ) . @xmath231 and @xmath232 were initialized to reflect the cusp destruction caused by the inspiralling bhs in reaching their initial configuration , and were updated throughout the simulation to account for the continued core heating and mass ejection . our algorithm for updating the core is described further in 3.5 . the parameters @xmath120 and @xmath72 in the @xmath223-models were set based on the tight correlations observed between smbh mass @xmath202 and stellar bulge mass @xcite and velocity dispersion @xcite . @xcite found the relation @xmath233 between @xmath202 and the virial mass @xmath234 of the stellar bulge , where @xmath235 is the half - light radius and @xmath236 is the effective bulge velocity dispersion . they set @xmath237 ( @xmath238 would be 8/3 for an isothermal sphere ) to get an average ratio of unity between @xmath239 and the dynamically measured masses @xmath240 of galaxies with more direct stellar - dynamical mass determinations @xcite . @xmath236 is typically measured over either a circular aperture of radius @xmath241 @xcite or a linear aperture out to @xmath235 @xcite @xcite discuss the essential agreement between the velocity dispersions measured in these two ways . thus for each model ( @xmath242 ) we compute the projected radius @xmath243 containing half the integrated surface brightness ( assuming a constant mass - to - light ratio ) , and velocity dispersion @xmath244 at radius @xmath241 . here @xmath245 and @xmath246 are constants depending on the density profile ( see @xcite for relevant formulae ) . the parameter @xmath120 is then chosen to satisfy @xmath247 @xmath248 @xmath249 . for the hernquist model with @xmath250 and @xmath251 , @xmath252 . the scale radius is then obtained from @xmath253 , where @xmath254 is the velocity dispersion computed from the @xmath110 relation of @xcite . in each of these relations @xmath202 is set to the total mass of the triple bh system . in a perfectly smooth , spherically symmetric galactic potential , bhs ejected on distant radial orbits return directly to the center to interact strongly with any other nuclear black holes . since real galaxies are clumpy and triaxial , the interaction will more realistically be delayed until the orbit of the ejected bh decays by dynamical friction . to mitigate this problem we flattened the @xmath223-models by adding two low - order spherical harmonic terms to the spherical potential @xcite : @xmath255 where @xmath256 is the potential of the spherical @xmath223-model , @xmath257 , and @xmath258 . since near - sphericity in the inner regions is probably a necessary prerequisite for the survival of the inner binary for of order a hubble time until the next merger @xcite , the parameters @xmath259 were chosen to give a spherical profile near the galactic center , and axis ratios approaching 1.3 and 1.5 for @xmath260 . a similar triaxial modification was applied to the dark matter halo , and the relative orientation of the stellar and halo potentials was chosen randomly . by misaligning their axes we eliminate any artificial stable orbits ( e.g. along the long axis of an ellipsoid ) near which ejected bhs tend to return on a perfectly radial orbit to the center . this triaxial modification had the desired effect of preventing frequent strong encounters at periapsis on distant orbits , but had little influence on the global outcome statistics . we assume that the 3-bh system starts off as a hierarchical triple on the verge of unstable 3-body interactions . in our canonical runs we initialize the inner binary semi - major axis @xmath64 to @xmath14 . to study the effect of varying @xmath64 we also try runs with @xmath261 ( runs ba ) and @xmath262 ( runs sa ) . the outer binary semi - major axis @xmath65 is set by the stability criterion of @xcite , equation the initial eccentricity of the inner ( outer ) binary was chosen uniformly between 0.0 and 0.2 ( 0.3 ) , in accordance with the low eccentricities found in galaxy merger simulations where dynamical friction tends to circularize the orbits of satellites as they spiral inward @xcite . the three euler angles of the intruder s orbital plane were chosen randomly relative to the reference plane of the binary orbit , as was the phase of the initial periapsis of the binary . both orbits were always started at periapsis ; since many orbital periods elapse before unstable interactions begin , the relative phase is effectively randomized in any case . having defined an initial configuration of three bhs embedded in a stellar+dark matter potential , we next describe how we evolve the system forward in time . we treat the close 3-body encounters using sverre aarseth s _ chain _ code , an implementation of the n - body regularization technique of mikkola and aarseth @xcite . the masses are first ordered so that neighbors in the chain are the dominant two - body interactions , then the ks - transformation @xcite is applied to neighboring pairs . this transformation eliminates the singularity at @xmath263 in newtonian gravity and transforms the equations of keplerian motion to the simple harmonic oscillator equation @xcite . external perturbing forces of arbitrary strength depending on the coordinates , velocities , and/or time are simply incorporated into the formulation ( though of course singularities in these perturbing forces need not be eliminated by the change of variables ) . we use this to add a galactic potential ( 2.1 - 2 ) , a gravitational radiation back - reaction force , and a stellar - dynamical friction force on the intruding bh . the regularized equations of motion are integrated using the bulirsch - stoer ( bs ) method @xcite based on romberg extrapolation . for unperturbed sinusoidal motion , the bs integrator requires only two or three timesteps per orbital period ! when the binary and third body are far apart we switch to two - body motion ( of the single bh and binary com ) using a 4th - order runge - kutta ( rk4 ) method . we simultaneously evolve the binary semi - major axis and eccentricity using orbit averaged equations , @xmath264 dt$ ] and @xmath265 dt$ ] , where @xmath266 and @xmath267 are the contributions from stellar interactions and gravitational radiation . the timesteps are adaptively controlled with a simple step - doubling scheme : at each step the 14 numbers @xmath268 are all required to remain the same to within an error @xmath269 @xmath116 @xmath270 under doubling of the step size . to avoid wasting computation time when any of these values approach zero , we accept agreement to @xmath271 decimal places as an alternative criterion for convergence . for the calculations reported in this paper we set @xmath272 . the relative perturbation to the binary from the third body at apoapsis , @xmath273 is used to decide which integration method to use at any given time . here @xmath274 is the apoapsis distance between @xmath275 and @xmath276 , @xmath86 is the intruder mass , and @xmath277 is the distance of the intruder from the binary com . we switch to two - body rk4 integration each time @xmath278 falls below @xmath279 and call the _ chain _ code again when @xmath278 reaches @xmath280 . we choose different @xmath278 thresholds for beginning and ending close encounters to prevent overly frequent toggling between the two methods . when @xmath281 bhs remain in the simulation ( after coalescence of the inner binary or escape of one or more bhs from the galaxy ) , we primarily use the rk4 integrator , but call the _ chain _ code to treat very close two - body encounters . since chain regularization is defined only for three or more bodies , we add a light and distant `` dummy '' particle when using this method for two - body motion . during the two - body motion we declare the single bh or binary ( remnant ) to have escaped if its distance from the galactic center exceeds 500 kpc and its specific energy @xmath282 exceeds @xmath283 , the energy needed to escape from @xmath284 to infinity . we declare the binary to have coalesced during a close encounter when ( i ) @xmath285 , where @xmath286 is the schwarzchild radius of the larger member of the pair , or ( ii ) @xmath287 @xmath288 @xmath289 and @xmath287 @xmath288 50 yrs while @xmath290 , where @xmath291 is the current outer binary dynamical time . during the rk4 integration we require that @xmath287 @xmath288 50 yrs or @xmath292 at periapsis , where @xmath293 are the schwarzchild radii of the two binary members . upon coalescence we replace the pair with a single body of mass @xmath222 and the com position and velocity . a run ends when ( a ) only one smbh remains in the galaxy and it has settled to the center of the potential by dynamical friction ; ( b ) two bhs remain and have formed a hard binary at the galactic center ; ( c ) the physical time exceeds @xmath294 , the current age of the universe minus the halo formation time ; ( d ) all bhs have escaped the galaxy ; or ( e ) the physical time spent in a call to _ chain _ exceeds a maximum allowed time @xmath295 . the last condition is added to avoid spending too much computation time on very long close encounters . during the two - body evolution we apply a dynamical friction force given by chandrasekhar s formula @xcite , @xmath296}{v^{2 } } \hat{v},\ ] ] where @xmath297 , to the single bh and binary com . the factor in square brackets @xmath204 @xmath298 for @xmath299 and @xmath204 @xmath300 for @xmath301 . we take @xmath302 , 1 \right\}\ ] ] for the coulomb logarithm , where @xmath38 is the bh s distance from the galactic center . for @xmath303 in equation we use @xmath304 $ ] , effectively capping the density at its value at the bh radius of influence , @xmath305 , when the bhs pass through the core . the semi - major axis @xmath72 of the binary also evolves under stellar - dynamical friction as it wanders through the galaxy . however chandrasekhar s formula applied separately to the binary constituents does not give a good description of this evolution , since the hard binary loses energy through close 3-body encounters with stars , while equation relies on the assumption that the energy loss is dominated by weak two - body encounters . we approximate the evolution of @xmath72 using a formulation for the decay rate of a hard , massive binary in a uniform and isotropic sea of stars developed in @xcite and @xcite . the formulation was calibrated with an extensive series of 3-body scattering experiments in @xcite and tested against n - body simulations in @xcite . the binary decay rate is given by @xmath306 where the hardening rate @xmath307 can be approximated by the empirical fitting function @xcite @xmath308^{1/2}}.\ ] ] here @xmath309 is the characteristic velocity distinguishing the hard binary regime stars with @xmath310 can not be easily captured into bound orbits and preferentially harden the binary in close encounters . in our simulations the binary com is often speeding through the stellar medium at @xmath311 after an energetic ejection , so the stellar medium looks `` hotter '' in its frame of reference . to account for this we replace @xmath9 in equations and with @xmath312 , a good approximation since @xmath307 is not very sensitive to the shape of the distribution function ( e.g. @xmath313 for a maxwellian vs. @xmath314 for a uniform velocity distribution ) . for @xmath303 in equation we took @xmath304 $ ] as we did for the drag on the com . we ignored the mild eccentricity evolution @xmath315 , which is shown in @xcite to be far weaker than that predicted by chandrasekhar s formula for hard eccentric binaries . when the amplitude of oscillation of one of the two masses falls below @xmath316 , we stop integrating its motion and place it at rest at the galactic center until the second body returns to within a distance of twice the break radius , @xmath317 . if the settled mass is the binary , then we also stop updating its semi - major axis for stellar hardening , assuming that it clears out its loss cone and stalls once it stops moving about the nucleus and encountering new stars . since the total mass in loss cone stars is small compared to the bh mass in the low - density galaxies that we consider , to good approximation the binary stalls as soon as the replenishing mechanism ( motion ) shuts off . during close encounters between the three bhs an orbit - averaged prescription for stellar - dynamical friction is not feasible . however the triples are still marginally stable at the boundary given by equation , so we apply a drag force given by chandrasekhar s formula with @xmath318 to the intruder at the beginning of each run . at the onset of chaotic interactions in the first encounter ( defined loosely by the first time the closest pair is not formed by the original binary members ) this perturbation is shut off , and it remains off in all later close encounters . fortunately the chaotic interactions occur on timescales very short compared to a dynamical friction time , so it is valid to neglect stellar dissipation during close encounters . gravitational radiation is modelled using the @xmath319 $ ] post - newtonian ( 2.5pn ) back - reaction acceleration computed by @xcite , evaluated in the two - body com frame ( e.g. gultekin et al . 2006 ) , @xmath320 } + \vec{v } { \big [ } -\frac{6g(m_{1}+m_{2})}{r}-2v^{2 } { \big ] } { \big \}}.\end{gathered}\ ] ] @xmath321 and @xmath322 are the relative positions and velocities of the two masses . we sum the force linearly over all pairs , a valid approximation provided the perturbations from the third body and other external tidal forces are instantaneously small at periapsis . when averaged over a complete orbit , equation is equivalent to the @xcite equations for the binary semi - major axis and eccentricity , [ pet ] @xmath323 however when @xmath324 comes close to one on hyperbolic orbits , so that @xmath325 , @xmath326 as given by equation becomes _ positive _ , though we know physically that gravitational waves can only carry energy away from the system . to give the correct answer averaged over an orbit , this positive contribution must be cancelled by extra energy loss near periapsis , making the equation potentially sensitive to numerical error . this effect is much less pronounced in the @xcite form than in other expressions derived for the radiation back - reaction acceleration - they derived the formula specifically for practical use on the problem of two point masses ( see appendix of @xcite and references therein ) . for computational ease we neglect the lower - order 1 - 2pn terms ( precession of the periapsis ) in the post - newtonian expansion . though much larger in magnitude than the radiation reaction force , these terms are unimportant in the statistical sense because they conserve the intrinsic properties of the system , such as energy @xcite . we need not concern ourselves with relativistic precession destroying the kozai resonance since the semi - major axis ratio given by equation is much smaller than that of equation . one way to establish the reliability of our integration methods is to test them on problems with known solutions . 6 shows an example on the two - body problem with gravitational radiation . @xmath116 @xmath21 yrs at the beginning of each integration . _ left : _ semi - major axis evolution for initial eccentricities of @xmath327 @xmath116 0.0 , 0.5 , 0.9 , and 0.99 ( bottom to top ) . _ right : _ eccentricity evolution for @xmath327 @xmath116 0.5 , 0.9 , and 0.99 . the curves indistinguishability demonstrates the reliability of all three methods . _ lower panels : _ hyperbolic orbits with impact parameters @xmath328 set to 80% and 120% of the critical value for gravitational radiation capture , computed using the dd acceleration in the _ chain _ code . _ left : _ 0.8@xmath329 ; bh is captured . _ right : _ 1.2@xmath329 ; bh is not captured . the blue asterisks are points along the newtonian trajectory ( without gravitational radiation ) . the deviation from the newtonian trajectory after periapsis can be seen in both plots , even though the energy remains positive in the latter . ] the upper panels show the evolution of the semi - major axis @xmath72 and eccentricity @xmath13 of four decaying elliptical orbits , computed using ( a ) our rk4 integrator and equation , with an error tolerance of @xmath330 , ( b ) the _ chain _ code and equation , with @xmath331 , and ( c ) the @xcite equations . in each case the initial semi - major axis @xmath332 was chosen to give a gravitational radiation timescale of @xmath333 yrs , and the four curves ( from bottom to top ) are for eccentricities of 0.0 , 0.5 , 0.9 , and 0.99 . the agreement of the three computation methods demonstrates the reliability of both the rk4 integrator and our implementation of the _ chain _ code in handling dissipative forces . the lower panels show two hyperbolic orbits with periapsis distances around 30 times the schwarzchild radius @xmath286 of the larger bh , computed using equation in _ chain_. the rk4 integrator was found to fail some tests on very close approaches from hyperbolic orbits with gravitational radiation , so we treat all such approaches using the regularized _ chain _ code in our runs , even during the unperturbed binary evolution . the blue asterisks are points along the newtonian orbits while the red solid lines show the trajectories with gravitational radiation . there is a simple analytic expression for the maximum periapsis distance for gravitational radiation capture from a hyperbolic orbit , @xmath334^{2/7},\ ] ] where @xmath275 and @xmath276 are the masses of the two bodies and @xmath335 is their relative velocity at infinity . the orbit on the lower left begins at 80% of the critical impact parameter and the incoming bh is captured . on the right the intruding bh starts at 120% of the critical impact parameter and is not captured , though the deviation from the newtonian trajectory due to the energy radiated at periapsis can be seen on the way out . we tried iterating over impact parameters close to the critical value and found that the code reproduces equation to within a part in @xmath78 for periapsis distances @xmath336 , and to within a part in @xmath337 for @xmath338 . we also evaluated the performance of the code by repeating our canonical set of 1005 runs with a static inner profile to check the precision of energy conservation . in fig . 7 we histogram the energy errors , computed as @xmath339 - e_{f}}{e_{0}-e_{f } } \right|,\ ] ] where @xmath340 and @xmath341 are the initial and final energies , and the two terms in the sum under the integral are the work done by dynamical friction and gravitational radiation during the current stage of the code . in the upper panel we separately plot the errors for close 3-body encounters , rk4 integration of the unperturbed binary motion ( `` far '' ) , and close two - body encounters computed with _ chain _ during the unperturbed binary evolution . the plot includes all code stages where the energy dissipated was at least @xmath342 in code units , or about a part in @xmath343 of the initial binding energy of the system . the black ( heavy ) histogram shows the errors for close encounters normalized to the _ initial _ energy instead of the dissipated energy in the denominator of equation , since the energy dissipated was very small in many close encounters . in the lower panel we combine the energy errors from the various code stages to get an effective energy error for each entire run , @xmath344 we had to combine the separate errors to obtain @xmath345 since the galactic potential is handled slightly differently during different stages of the code , e.g. the triaxial modification is applied only during the rk4 integration . in a large majority of cases @xmath345 falls between @xmath346 and @xmath347 , and energy is conserved to better than a part in @xmath348 in every run . the excellent energy conservation gives us confidence in the robustness of our integration methods . we begin with an overview of the outcomes of our 3-body simulations . in subsequent sections we focus on various effects in more detail . our data consists of eight sets of 1005 runs , each sampling a different distribution of the ics . a set of 1005 runs took anywhere from @xmath04 to @xmath030 hours to finish on five 2.0 ghz opteron processors , depending on the ics . in our canonical runs ( cn ) , we chose @xmath349 for the threshold merger mass ratio , modelled the stellar bulge as a hernquist ( @xmath350 ) profile , started off the inner binary at @xmath14 , and generated the ics from a @xmath161 m@xmath23 halo at @xmath165 . in each of the remaining runs we varied one of these assumptions . runs d1 and d5 used @xmath351 and @xmath352 to explore the effects of widening or narrowing the range of bh mass ratios . in runs mx we assigned a mass @xmath213 $ ] instead of @xmath212 to the intruding bh , as discussed in 2.1 . in runs ba and sa we started off the inner binary at @xmath353 and @xmath354 instead of at @xmath14 . we initialized the stellar bulge to an @xmath228 profile in runs sc , to explore the effect of a steeper inner cusp . finally in runs h1 we generated the ics from a @xmath175 m@xmath23 halo at @xmath165 , for total bh masses of @xmath355 m@xmath23 , about an order of magnitude lower than in our canonical runs . table 1 summarizes the outcomes . .summary of outcomes for eight different distributions of the ics . see text for explanation of the entries . [ cols="<,<,<,<,<,<,<,<,<,<",options="header " , ] [ obsmbh ] for the rest of the galaxies we used the @xcite @xmath356 relation to estimate the bh masses in galaxies with @xmath357 , and the @xcite @xmath110 relation for those with @xmath358 . showing both plots gives an idea of how much the mass deficits vary with @xmath202 estimator . in set cn @xmath011% of the runs resulted in cores with @xmath359 . in some rare runs where the binary was ejected to a large distance and then brought back by dynamical friction , we obtained even higher mass deficits ( up to @xmath360 ) . the binary is more efficient at core scouring than the single bh since it is more massive . if each independent binary inspiral adds @xmath361 to the mass deficit , then the mean @xmath362 enhancement of @xmath361 due to the triple encounters is equivalent to one extra merger in the system s history . an @xmath362 one standard deviation above the mean is equivalent to two extra mergers . triple - smbh systems in galactic nuclei produce a range of phenomena and signatures rather different from those expected if no more than two smbhs occupy them at a time . we have developed an efficient numerical method for following the evolution of 3-body systems in the centers of galaxies , and used it to explore the outcomes of such encounters in massive elliptical galaxies at low redshift . we find a high efficiency of smbh coalescence due to the encounters , providing a `` last resort '' solution to the final parsec problem . there is , however , a caveat in extending this result immediately to all bh masses . if we define @xmath363 to be the binary semi - major axis where escape of one bh first becomes likely ( @xmath364 where @xmath365 is a factor of order 10 for a hernquist profile ) , then since @xmath366 we have @xmath367 if @xmath222 obeys the @xmath110 relation @xmath368 , so @xmath369 . in other words , at smaller bh masses , the lightest bh is more likely to escape the galaxy before driving the binary to coalescence by gravitational radiation . by focusing on massive galaxies we have chosen the systems where the binary is _ least _ likely to coalesce by other means ( e.g. gas or massive perturbers ) , and _ most _ likely to coalesce in the next merger with the help of 3-bh interactions . we may address the efficiency of triple - induced coalescence in much smaller - mass systems in future studies . we find that close triple encounters can produce a population of high - eccentricity binaries , whose gravitational radiation signal could potentially be observable by lisa . such signals originate from kozai oscillations in hierarchical triples at high initial inclinations and highly eccentric binaries formed following distant ejections . as the eccentricity increases , the radiation spectrum peaks at progressively higher harmonics of the fundamental frequency , approaching a nearly flat spectrum as @xmath370 @xcite . a circular @xmath371 m@xmath23 bh binary remains below the band of frequencies ( @xmath372 hz ) detectable by lisa throughout its inspiral , but the occurrence of high - eccentricity coalescences could extend lisa s sensitivity into this mass range , or lengthen the duration of its sensitivity to @xmath373 m@xmath23 events . a highly eccentric binary produces a `` spiky '' waveform that looks quite different from that of a circular system ( see fig . 7 in @xcite ) . gravitational radiation `` spikes '' at very close approaches during chaotic 3-body interactions could also produce radiation bursts detectable by lisa . if triple encounters are indeed limited to massive systems at low redshift , then the importance of these considerations is limited by the expected event rate in this mass range , assuming efficient coalescence . this rate is highly uncertain , ranging from @xmath374 @xcite to @xmath375 @xcite depending on the merger and bh population model adopted . if 3-bh systems occur in other contexts , e.g. imbhs in galactic nuclei or star clusters , then the phenomena we have discussed may be observationally relevant even if the high - mass smbh event rate is low . a detailed look at the gravitational waveforms expected from 3-body encounters and their expected detection rates is an interesting topic for a future study . the slingshot ejections in triple encounters produce a population of `` wandering '' smbhs in and outside the halos of galaxies . in systems that have undergone several major dry mergers ( e.g. cd galaxy systems ) , one might expect a few such ejected smbhs to be floating in the vicinity . as of yet , no probable way of observing these wandering bhs has been proposed . in principle one can imagine a star bound to the ejected smbh entering a giant phase and overflowing its roche lobe , producing some accretion onto the smbh and an observable flare . single ejections could also in principle affect bh - bulge correlations such as the @xmath110 relation , but since it is the lightest bh that gets ejected this effect would fall well within the observed scatter in the correlations for just one or two ejection events . triple interactions in galactic nuclei can have a large effect on the expected properties of stable smbh binaries in the local universe . while many models of binary formation predict mostly circular binaries around @xmath14 , 3-body encounters produce binaries at all eccentricities . they also create a population of stalled binaries at separations significantly smaller than @xmath14 but still larger than @xmath18 , as does any partial gap - crossing mechanism . better measurements and statistics on the mass deficits in cored elliptical galaxies may provide clues on the history of the nuclear smbh activity in these systems . triple bh encounters produce a highly scattered distribution of core sizes , with mass deficits up to @xmath376 higher than expected for successive binary coalescences . the apparent peak at mass deficits of @xmath377 times the nuclear bh mass in observed cores may very tentatively hint that multiple - bh encounters are not the norm in these systems . this signature of binary or multiple - bh activity is appealling because ( a ) its duty cycle is the lifetime of the galaxy ; ( b ) it is present whether binary pairs stall or coalesce ; and ( c ) it can be observed even in the complete absence of radiative activity , such as disk accretion or jet production . however the interpretation of galaxy cores is complicated by multiple mergers , the possibility of partial stellar cusp regeneration from traces of cold gas , and observational complications such as projection effects in nonspherical galaxies and optimizing the fitting / extrapolation algorithm to best represent the mass deficit . there is a need for theoretical studies on the cores produced by smbh mergers in triaxial galaxies , since triaxility seems to be the most likely candidate for a gap - crossing mechanism in dry mergers between gas - poor , giant ellipticals . inferring the nuclear histories of galaxies from their observed core properties will likely be a topic of much interest in the future . we would like to thank suvendra dutta for technical help at the parallel computing center of the institute for theory and computation ( itc ) , and sverre aarseth and seppo mikkola for making their n - body algorithms and codes available . we are also grateful to scott hughes , michael eracleous , marta volonteri , and fred rasio for useful discussions . this research was supported in part by an fqxi grant and harvard university funds .

we examine the constraints cosmological observations can place on any trans - planckian corrections to the primordial spectrum of perturbations underlying the anisotropies in the cosmic microwave background . we focus on models of trans - planckian physics which lead to a modulated primordial spectrum . rather than looking at a generic modulated spectrum , our calculations are based on a specific model , and are intended as a case study for the sort of constraints one could hope to apply on a well - motivated model of trans - planckian physics . we present results for three different approaches a grid search in a subset of the overall parameter space , a fisher matrix estimate of the likely error ellipses , and a monte carlo markov chain fit to a simulated cmb sky . as was seen in previous analyses , the likelihood space has multiple peaks , and we show that their distribution can be reproduced via a simple semi - analytic argument . all three methods lead to broadly similar results . we vary 10 cosmological parameters ( including two related to the trans - planckian terms ) and show that the amplitude of the tensor perturbations is directly correlated with the detectability of any trans - planckian modulation . we argue that this is likely to be true for any trans - planckian modulation in the paradigm of slow - roll inflation . for the specific case we consider , we conclude that if the tensor to scalar ratio , , the ratio between the inflationary hubble scale , and the scale of new physics has to be on the order of if the modulation is detectable at the 2 level . for a lower value of , the bound on becomes looser .

astro-ph0412613

conventional cosmology is based upon a four dimensional spacetime whose evolution is governed by the einstein field equations . this picture certainly breaks down at the planck scale due to quantum corrections to the gravitational action . furthermore , stringy effects may change our understanding of spacetime at energies one or two orders of magnitude below the planck scale . unfortunately , both the planck and string energy scales are entirely inaccessible to direct experimentation . however , thanks to the dramatic expansion of the universe during the inflationary era , it is possible that cosmology provides an unexpected window into physics at the shortest _ length _ scales @xcite-@xcite . if we assume the universe has always been matter dominated and is now around 13 billion years old , the current horizon size is roughly 40 billion lightyears , or @xmath7 planck lengths . if inflation ended at the gut scale ( @xmath8gev ) with efficient reheating , the universe has grown roughly @xmath9 times larger since reheating , and by at least a similar amount during inflation , in order to ensure that the universe is homogeneous on scales equal to the present horizon size . consequently , all distances significantly smaller than galactic scales in the current universe were necessarily sub - planckian at some point during inflation . if the inflationary period continues for longer than the minimum time necessary to solve the cosmological flatness and horizon problems , all presently visible scales must have crossed the planckian boundary during inflation . conversely , without inflation all astrophysical scales are mapped to lengths larger than the planck length at all times since the age of the universe was of order the planck time . on the face of it , the fact that astrophysical scales in the present universe correspond to sub - planckian scales during the inflationary epoch is simply a mathematical curiosity . however , inflation generates perturbations through the exponential stretching of quantum fluctuations . the textbook calculation of the primordial perturbation spectrum implicitly ignores the consequences of the breakdown of spacetime at some new fundamental lengthscale . for a clear exposition of the standard treatment of perturbations , including a discussion of the effect of boundary conditions , see @xcite . in string theory the effective minimum length can be significantly larger than the planck length , by perhaps one or two orders of magnitude @xcite . in this case , the ratio between the inflationary hubble scale , @xmath1 , and the scale of new stringy physics , @xmath2 can be as large as @xmath10 . dimensional analysis alone suggests that any modifications to the perturbation spectrum induced by including this minimum length will be a function of @xmath11 , where @xmath12 is , of course , unknown . if @xmath12 is unity , then there is at least a corner of parameter space where we might expect potentially observable modifications to the spectrum . however , for larger values of @xmath12 , any modifications to the spectrum are most likely unobservable . string theory and other candidate unified models are not sufficiently mature to permit an _ ab initio _ analysis of perturbation generation , and predict the value of @xmath12 . consequently , all studies of trans - planckian modifications to the inflationary perturbation spectrum require some ansatz which , while motivated by our understanding of stringy / planckian physics , introduces a degree of arbitrariness into the calculation . this ansatz necessarily modifies either the initial conditions or the evolution of the perturbations . generically , modifications to the initial conditions produce changes in the spectrum on the order of @xmath6 , whereas modifications to the evolution result in changes on the order of @xmath13 . it is worth noting , however , that no calculation has predicted that the modifications vanish completely . normally , this lack of theoretical consensus would deter detailed studies of the observational consequences of any given model of trans - planckian physics . however , even a small possibility that the fingerprints of planckian or stringy physics can be found on the sky provides more than enough motivation for us to take a careful look at what constraints can be established with present data or future measurements . in this paper , we focus on one model of trans - planckian physics , first suggested by danielsson in the case of de sitter space @xcite and generalized to non - de sitter backgrounds by easther , greene , kinney and shiu @xcite . there is nothing intrinsically special about this model our analysis is effectively a case study of what might be achieved when one has a specific model to test . in particular , we do not enter into the debate as to whether or not such `` truncated @xmath14-vacua '' are self - consistent with respect to being well - defined in the ultraviolet limit or with respect to their suitability as initial conditions for inflation . we simply consider the model as a reasonable case study for investigating the discriminatory power of forthcoming cmb observations . the one general statement we can make is that it is far more likely that trans - planckian effects will have an observable impact on the spectrum if inflation takes place at a relatively high energy scale . in principle , inflation can occur at almost any scale high enough to allow for baryogenesis and nucleosynthesis in the post - inflationary universe and this can be far below the gut scale . however , since we only expect a detectable signal if @xmath6 is comparatively large , we are pushed into the corner of parameter space where @xmath2 is small ( compared to the planck scale ) while @xmath1 is roughly gut - scale . a small @xmath2 ( two orders of magnitude below the planck scale , say ) is entirely consistent with string theory , whereas @xmath1 is effectively a free parameter . an upper bound on @xmath1 arises from the limits on the tensor component to the primordial perturbation spectrum which is potentially observable via both the temperature anisotropies in the cmb , and the b - mode of the cmb polarization . ( a rough upper limit on @xmath1 from the wmap data is @xmath15 . ) it is very easy for inflation to happen at a low enough scale to ensure that a primordial tensor spectrum is effectively unobservable by any conceivable experiment @xcite . however , if we were to observe a contribution to the cmb that appears to be related to planckian or stringy physics , there is a strong likelihood that we would be in the region of parameter space where @xmath1 was large enough in order to ensure that the tensor contribution was also detectable . there are important consequences to this qualitative correlation between trans - planckian contributions to the cmb and the tensor perturbations . first , it further raises the stakes for high precision measurements of the cmb . we already know that an observable tensor spectrum fixes the inflationary energy scale and rules out some of inflation s competitors , such as the ekpyrotic @xcite and pre - big bang scenarios @xcite , and that single field models of inflation must satisfy a set of consistency conditions @xcite . however , on top of these achievements , determining @xmath1 would let us put direct constraints on the scale of new physics in any specific model of trans - planckian physics , since @xmath2 would be the only remaining free parameter . if @xmath1 turns out to be at the high end of the permitted range , it increases the likelihood that we will be able to see any trans - planckian signal that is there to be found . finally , one us [ re ] recently argued that a high value of @xmath1 during inflation is naturally correlated with a scalar spectrum with a significant running index ( @xmath16 ) @xcite . we do not consider models with substantially broken scale invariance in this paper , but do incorporate a non - zero running into our analysis below . from a practical perspective , if we perform a fit to cmb data with a model that includes trans - planckian effects , it is important to include the tensor modes in any analysis , since the same parameter ( @xmath1 ) that governs their amplitude and scale dependence is also likely to appear in the trans - planckian corrections . conversely , simply observing an oscillatory spectrum would not be enough to prove that planck / string scale physics had left an imprint on the cmb one can produce the same result purely with physics at lower energy @xcite . this problem has been addressed in several previous analyses . the first computation of trans - planckian modifications to the spectrum was @xcite , which used a fisher matrix calculation to predict that the planck mission will put stringent bounds on a oscillatory component in the fundamental spectrum . however , this analysis marginalized over a small subset of cosmological parameters , and the use of a fisher matrix in this context is problematic @xcite . subsequent calculations have been less optimistic . in particular , @xcite argued that any reasonable modulation to the spectrum is most likely unobservable , although we will argue that their pessimism is probably misplaced . okamoto and lim @xcite looked at the constraints on a general superimposed oscillation to the spectrum . they varied several cosmological parameters as well as those describing the oscillation , and concluded that values of @xmath17 might be reached with a cosmic variance limited survey , but ignored the role of tensor modes . martin and ringeval , @xcite have studied models with independent parameters governing the amplitude and frequency of the trans - planckian modulation . their analysis is distinguished by the possibility that very high - frequency oscillations with non - negligible amplitude are considered , with the frequency of the oscillations producing the primary constraint on @xmath6 . in such a case , they conclude that very strong constraints are possible on the value of @xmath6 , even when restricted to _ current _ data . this paper is organized as follows : in sec . [ sec : model ] , we discuss the specific model of trans - planckian physics we consider in this paper . the modulation to the power spectrum is correlated to the tensor / scalar ratio , which we argue is well motivated because of its relationship to a @xmath18-independent boundary condition . in sec . [ sec : quicksurveys ] , we discuss two complementary surveys of the trans - planckian parameter space : first , we consider a grid - based search of the parameter space , which is computationally inefficient but enables a broad sweep of the space . second , we consider a fisher - matrix based analysis , which handles large numbers of parameters in an efficient way , but does not sample the global properties of the likelihood surface . in sec . [ sec : markovchains ] , we describe a markov chain monte carlo analysis , which constitutes a complementary method for exploring the parameter space . all three analyses are in good agreement . comments and conclusions are presented in sec . [ sec : conclusions ] . quantum fluctuations during inflation produce two fluctuation spectra : the curvature perturbation in the comoving gauge @xmath19 , and the two polarization states of the primordial tensor perturbation , @xmath20 and @xmath21 . we parameterize these power spectra by @xmath22 where @xmath23 is a normalization constant , and @xmath24 is some pivot wavenumber . the running , @xmath25 , is defined by the second derivative of the power spectrum , @xmath26 , for both the scalar and the tensor modes . this parameterization gives the definition of the spectral index , [ eq : nsdef ] n_s(k)-1 n_s(k_)-1 + ( ) , for the scalar modes , and [ eq : ntdef ] n_t(k ) = n_t(k_)+ ( ) , for the tensor modes . in addition , we re - parameterize the tensor power spectrum amplitude , @xmath27 , by the `` tensor / scalar ratio @xmath5 '' , the relative amplitude of the tensor - to - scalar modes , given by @xmath28 for a single slowly rolling inflaton , @xmath29 , so we can reduce the number of parameters by fixing @xmath30 once @xmath5 is chosen . we choose the pivot - scale @xmath31 mpc@xmath32 at which to evaluate @xmath33 , @xmath5 and @xmath34 , where @xmath35 and @xmath36 are related ( see * ? ? ? * ) through @xmath37 note that in this definition , the trans - planckian parameter @xmath38 , which appears in eq . [ eq : tpps ] , is simply the first slow - roll parameter ( see e.g. * ? ? ? * ) , @xmath39 where @xmath40 is the reduced planck energy ; it is related to the tensor - to - scalar ratio by @xmath41 . the definition in terms of @xmath42 , called the _ hubble slow roll _ expansion , is exact , and the expression in terms of the inflationary potential @xmath43 , called the _ potential slow roll _ expansion , is an approximation valid in the limit of a slowly rolling field , @xmath44 . the primordial power spectra @xmath45 and @xmath46 are the underlying spectra which are modulated by effects from planck - scale physics , described below . the specific model of trans - planckian physics we are working with @xcite assumes that rather than reducing to the minkowski mode function @xmath47 in the infinite past , the perturbations are matched at some finite time , corresponding to the moment ( different for each comoving perturbation ) when their physical wavelength exceeds the minimum length , @xmath48 . following @xcite , we can compute the modified spectrum ; @xmath49 where @xmath50 refers to the spectrum obtained with the conventional bunch - davies vacuum . in this case the scalar and tensor spectra are both modified by the same amount . the @xmath51 are @xcite @xmath52 here @xmath53 is a rescaled `` time '' , defined by @xmath54 @xcite , so that in the absence of a new scale @xmath53 decreases from @xmath55 to @xmath56 , with horizon crossing taking place ( by definition ) when @xmath57 . in this language , @xmath58 is the critical time for the @xmath18-th mode , at which its physical scale becomes larger than the minimum length . this is a @xmath18-dependent quantity , since in all slow roll models , @xmath1 is a ( slowly varying ) function of time , and in general @xcite , @xmath59 where @xmath38 is the usual slow roll parameter . this relationship is exact when @xmath38 is expressed in the hubble slow roll expansion , and approximate when the potential slow roll form is used . consequently , y_c(k ) = ( ) ^ where @xmath2 is the scale of new physics , and the starred quantities again refer to some fiducial value of @xmath18 . the quantity @xmath60 is the value of the quantum mode function ( given in the short - wavelength limit by @xmath61 ) evaluated when the wavelength of the mode is equal to a fixed physical cutoff length @xmath62 , or equivalently @xmath63 . @xcite for a detailed discussion . keeping only the lowest order term in @xmath64 , we can write the modified spectrum in a more transparent form , @xmath65^{1/4 } \delta^{bd}_{{\cal r},h},\ ] ] where @xmath66 is a phase factor quantifying our lack of _ a priori _ knowledge about the physical `` pivot - scale '' @xmath24 . we note that the functional form of the modification above is determined not by the details of the boundary condition , but by the behavior of the _ background _ , in particular the changing ratio of the inflationary horizon size to the ( fixed ) fundamental scale . such a sinusoidal modulation likely to be a generic feature of any trans - planckian ansatz that changes the initial conditions for a given perturbation , so long as the boundary condition itself has no intrinsic dependence on @xmath18 . since this effect depends on @xmath1 , which gradually decreases during the course of the inflationary epoch , the original spectrum is multiplied by a ( small ) modulation , with amplitude of order @xmath6 and , as a consequence , the amplitude of the modulation decreases with increasing @xmath18 . a plot for the specific case of power - law inflation is to be found in @xcite , and in fig . [ fig : modulation ] . there are several ways to estimate parameters , and their likely uncertainties , in cmb data . in addition to the markov chain monte carlo ( mcmc ) methods discussed below , two widely used approaches have been grid searches in parameter space and fisher matrices , which provide theoretical estimates of the likely accuracy that can be obtained with a given measurement . a key observation that applies to all methods is that we can not simply estimate trans - planckian parameters after the other cosmological parameters are determined . if we add new parameters ( from trans - planckian physics or elsewhere ) we must re - estimate all the other parameters in our model . failing to do so will result in unphysically small error ellipses . grid methods have no intrinsic theoretical shortcomings , but with more than three or four variables they become far too time consuming for a thorough search , since the number of points on the grid increases exponentially with the number of parameters . fisher matrix techniques scale well with the total number of parameters , but they can not cope with a likelihood surface that contains multiple peaks , and we will see that this is precisely the situation we are confronting here . that said , we use both methods to make a rough survey of the parameter space before discussing our markov chain results . in all cases , we use a modified version of the code cmbfast 4.5.1 with the high - precision option @xcite to calculate the power spectra . the @xmath67space and @xmath18space resolution of cmbfast was significantly increased in order to resolve the trans - planckian modulations with sufficient accuracy . we do not , however , calculate every @xmath67mode , as suggested in @xcite this would render the computational cost of exploring such a high - dimensional parameter space prohibitively high , since the computation of a single model takes @xmath68 minutes of computing time at the resolution used , roughly an order of magnitude longer than needed on the same hardware with an unmodified version of cmbfast . however , unlike @xcite , we are not testing modulations with independent amplitudes and frequencies ; in our model , very high frequency oscillations either have strongly suppressed amplitudes or enormous tensor / scalar ratios , or both . thus , while we need to sample in @xmath67-space sufficiently to resolve the oscillations we are looking for , sampling every @xmath67-mode is not a requirement for the model at hand . in order to calculate the best possible constraints obtainable for the trans - planckian model above , we wish to simulate an ideal noiseless cmb experiment with full sky coverage . assuming cmb multipoles are gaussian - distributed , the likelihood function for such an experiment is given by @xmath70}{\sqrt{\det { \bf c}}}.\ ] ] here , @xmath71 is the data vector of spherical harmonic coefficients @xmath72,\ ] ] where , for example , the temperature map has been expanded in spherical harmonics as @xmath73 , and * c * is the covariance matrix given by @xmath74 the terms @xmath75 and @xmath76 are zero by global isotropy . in the covariance matrix , @xmath77 denote theoretical power spectra . now we define the estimator for the power spectra from the data as : @xmath78 since the universe is assumed to be isotropic , the likelihood function is independent of @xmath79 , and summing over it , one obtains , up to an irrelevant constant : @xmath80 the likelihood has been normalized with respect to the maximum likelihood , where @xmath81 , so that it behaves like a @xmath82 statistic . given a set of observational data points , the goal is then to compute the likelihood function for various choices of parameters , _ i.e. _ a bayesian analysis . since we are interested in forecasting the performance of future measurements , the `` data '' is synthetic rather than from an actual observation . various methods for performing the likelihood analysis have advantages and disadvantages with respect to computational efficiency and coverage of the parameter space . in the next section , we discuss the results of a grid - based search of the trans - planckian parameter space . grid calculations are a simple `` brute force '' method of exploring a given parameter space . given a set of observational data points ( for example a cmb multipole spectrum ) , the bayesian best - fit @xmath83-parameter model is found by dividing the @xmath83-dimensional parameter space into a grid of evenly spaced points and computing the likelihood function of each model relative to the data . error bars are assigned by the rate of dropoff of the likelihood function from the best - fit point , which in a coarse grid typically involves interpolation between the likelihoods actually computed on the grid .. since the number of calculations increases exponentially with the number of parameters @xmath83 , grid methods are inefficient for exploring large parameter spaces . since we are interested in forecasting the capabilities of future cmb measurements we do not use current observational data points , but produce `` synthetic '' data by computing a fiducial model representing the assumed underlying cosmology and assign a set of error bars to the points in the fiducial model representing the accuracy of the observation . our hypothetical baseline experiment is cosmic variance limited to @xmath84 in all four cmb spectra : @xmath85 , @xmath86 , @xmath87 , and @xmath88 . while this is an unrealistic assumption , it characterizes the amount of information which is intrinsically contained in the cmb spectra themselves , that is : what is the best we can possibly do ? since grid methods are particularly well suited to performing a broad survey of a small parameter space , we choose our parameter space to be the trans - planckian amplitude @xmath6 , and the phase @xmath66 of the oscillations , which is in principle arbitrary . other cosmological parameters are held fixed . the fiducial model we assume for the grid calculation is approximately the best - fit from the wmap data set , @xmath89 , @xmath90 , @xmath91 , @xmath92 , @xmath93 , @xmath94 , @xmath95 . we assume a tensor / scalar ratio of @xmath96 , and trans - planckian parameters @xmath97 and @xmath98 . ( these apparently odd choices for the trans - planckian parameters are chosen so that the fiducial model lies exactly on a grid point in fig . [ fig : finegrid ] . ) figure [ fig : coarsegrid ] shows the result of a coarse grid covering a large range of parameters , showing the substantial degeneracy in the @xmath6 - @xmath66 plane , in particular many `` islands '' in the likelihood function representing good fits between the test and fiducial models . figure [ fig : finegrid ] is a finer grid run over a smaller region of the same parameter space , showing more detail in the shape of the likelihood function . a coarse grid ( 100 points in the range @xmath99 $ ] by 75 points in the range @xmath100 $ ] ) covering a broad region of the trans - planckian parameter space . the inner ( colored ) contours are drawn at the @xmath101 , @xmath102 , and @xmath103 levels , and the outer ( black ) contours are at a @xmath104 relative to the best fit to better show the shape of the likelihood function . the fiducial model has @xmath97 and @xmath98 . , width=297 ] a fine grid ( 100 points in the range @xmath105 $ ] by 100 points in the range @xmath106 $ ] ) showing detail in a small region of the parameter space from fig . [ fig : coarsegrid ] . the fiducial model is in the center of the plot , and contours are drawn at @xmath101 ( inner , red ) , @xmath102 ( middle , blue ) , and @xmath103 ( outer , green ) levels . , width=297 ] several conclusions can be drawn from this analysis . first , the projected error ellipses for our hypothetical cosmic - variance limited measurement are extremely small in the @xmath6 plane , indicating a clear ability to detect the trans - planckian oscillations in the power spectrum . however , because of the presence of a large number of nearly degenerate `` islands '' in the likelihood function , it will be difficult to pin down the precise values of both the trans - planckian parameters , especially the phase . this is consistent with the results obtained in previous analyses @xcite . however , we do note that for the fiducial model chosen , _ none _ of the best - fit `` islands '' is consistent with zero amplitude , @xmath107 . therefore , although the values of trans - planckian parameters are poorly constrained , the case of zero modulation can be ruled out to high significance . figure [ fig : smallamplitude ] shows the shape of the likelihood surface relative to a fiducial model with @xmath108 , and fig . [ fig : null ] shows the likelihood relative to the null hypothesis , @xmath107 , which gives a measure of the size of the _ upper limit _ one could place on @xmath6 if no signal were detected . ( we emphasize that the @xmath103 upper limit of about @xmath109 in fig . [ fig : null ] does not take into account variations in other cosmological parameters . ) grid results for a fiducial model with @xmath108 and @xmath110 , showing the broadening of the degeneracy for smaller amplitude . ( note that the fiducial model is missed due to finite grid size effects.),width=297 ] grid results for the null hypothesis , @xmath111 , estimating an upper limit on @xmath6 if no trans - planckian signal is detected.,width=297 ] @xmath112 @xmath113 why are these degenerate `` islands '' present in the parameter space ? in order to understand this result in a semi - analytic sense , we consider the effect of the degeneracy on the primordial power spectra . the motivation of the following analysis is that likelihood degeneracies arise from frequency `` beating '' between underlying modulations in the data and the model . parameter combinations where `` beating '' occurs over the largest @xmath18ranges give rise to discrete likelihood `` islands '' or local maxima in the @xmath114 vs @xmath66 plane . defining the trans - planckian modification factor multiplying the standard bunch - davies power spectrum as @xmath115,\ ] ] where @xmath116 , we define the parameter @xmath117 where @xmath118 here , we have performed the integration over the relevant scales for the cmb , @xmath119 @xmath120 . if @xmath121 , there will be a near - perfect degeneracy between the parameter sets @xmath122 and @xmath123 . assuming that @xmath38 can be constrained by detecting tensor modes , a local likelihood maximum is likely to be present at @xmath124 even though the underlying `` data '' has @xmath125 . in the case corresponding to our fiducial models above , @xmath126 , only a simple degeneracy @xmath127 exists . in figures [ fig : deger ] , we compute @xmath117 for several values of non - zero @xmath128 $ ] . while the results in these figures are in the analytically tractable power - spectrum space , figure [ fig : finegridi ] shows a comparison with figure [ fig : finegrid ] above , which was calculated the full @xmath129 formalism . since convolution with the cmb transfer functions will tend to smear out likelihood degeneracies present in primordial power - spectrum space , we expect the likelihood `` islands '' to be less prominent in the full @xmath129 calculation than in the semi - analytic calculation . these results show that the presence of discrete likelihood maxima in parameter space gives some extra information beyond merely being a numerical headache and a barrier to parameter estimation . their presence in a parameter estimation analysis is an indication of an underlying modulation in the data , and the appearance of these degeneracies on the @xmath114 vs @xmath66 plane in this model is a function of the value of @xmath114 of the data . of course , should such an effect be detected in real data , whether or not such a modulation is due to a primordial effect or some systematic effect in the data , is a question which would need to be investigated thoroughly . while the grid method is a good tool for exploring the global shape of the likelihood function , it is an oversimplification to vary only two parameters in the fit . error bars derived from such a two - parameter analysis will be unrealistically optimistic , since a complete analysis must take into account degeneracies between the parameters of interest and a large number of other cosmological parameters . to investigate the effect of including other cosmological parameters in the fit , we apply a fisher matrix technique in sec . [ sec : fishermatrix ] and monte carlo markov chain methods in sec . [ sec : markovchains ] . the fisher matrix technique is a simple and efficient method for estimating the expected measurement errors when marginalizing over a large number of parameters , for which grid methods would be unacceptably time consuming . measurement uncertainty in cosmological parameters is characterized by the fisher information matrix @xmath130 . ( for a review , see ref . @xcite . ) given a set of parameters @xmath131 , the fisher matrix is given by @xmath132 where @xmath133 and @xmath134 is the inverse of the covariance matrix between the estimators @xmath135 of the power spectra . calculation of the fisher matrix requires assuming a `` true '' set of parameters and numerically evaluating the @xmath136 s and their derivatives relative to that parameter choice . the covariance matrix for the parameters @xmath137 is just the inverse of the fisher matrix , @xmath138 , and the expected error in the parameter @xmath139 is of order @xmath140 . the full set of parameters @xmath131 we allow to vary is : 1 . tensor / scalar ratio @xmath5 , 2 . spectral index @xmath33 , 3 . running @xmath141 , 4 . normalization @xmath142 , 5 . baryon density @xmath143 6 . hubble constant @xmath144 , 7 . reionization optical depth , @xmath145 , 8 . cosmological constant density @xmath146 , 9 . trans - planckian amplitude @xmath6 , 10 . trans - planckian phase @xmath66 . the total density is fixed such that the universe is flat , @xmath147 . since one needs only to calculate the derivative of the likelihood function with respect to each parameter , fisher matrix calculations scale linearly with the number of parameters and are therefore very efficient for forecasting errors in fits marginalizing over a large number of parameters . this technique , however , gives no information about the _ global _ shape of the likelihood surface . obviously , since we do not know the values of the trans - planckian parameters in the real universe , only the size of the errors , not the location of the central value , is significant . we perform a fisher matrix analysis for two different forecast data sets : first , in order to understand the best - case scenario , we consider a measurement for which the errors are limited by cosmic variance to @xmath84 , similar to that assumed for the grid method in sec . [ sec : grid ] . second , we consider the more immediately realistic case of a measurement with error bars similar to those projected for the planck surveyor satellite , for which we assume a two - channel experiment measuring both temperature and polarization . the 143 ghz channel is assumed to have an @xmath148-arcminute beam and a pixel sensitivity of @xmath149 , and the 217 ghz channel is assumed to have an @xmath150-arcminute beam and a pixel sensitivity of @xmath151 . we are especially interested in the degeneracy between trans - planckian modulations and the other parameters determining the shape of the primordial power spectrum , @xmath33 and @xmath141 . this degeneracy is highly dependent on the value of @xmath5 in the underlying fiducial model we choose , because for small @xmath5 , the trans - planckian modulation becomes very long - wavelength and thus mimics a change in the spectral index @xmath33 or running @xmath141 ( fig . [ fig : modulation ] ) . ( for the case @xmath152 , the `` modulation '' loses all scale - dependence and simply affects the normalization . the value of @xmath6 thus decouples from the other parameters determining the shape of the power spectrum , and the degeneracy vanishes in this limit . ) @xmath153 we therefore plot error ellipses in the @xmath33 - @xmath6 plane and the @xmath154 - @xmath6 plane for various choices for @xmath5 running from @xmath155 to @xmath156 . the fiducial model assumed in all cases has @xmath157 and phase @xmath158 . other parameters are identical to those used for the grid calculation in sec . [ sec : grid ] . figure [ fig : cosvarfisher ] shows the @xmath33 - @xmath6 plane for the case of the cosmic - variance limited measurement , and fig . [ fig : cosvarfisherdn ] shows the @xmath141 - @xmath6 plane . in this case , the measurement is sufficiently accurate that the degeneracy between the power spectrum parameters @xmath33 , @xmath141 and the trans - planckian amplitude @xmath6 is negligible , although the ability to constrain @xmath6 is significantly degraded for small values of @xmath5 . these error estimates can be directly compared with those generated by a full markov chain monte carlo analysis with similar parameters , ( figs . [ fig : params_bigr ] and [ fig : params_tinyr ] ) , and are in good agreement . @xmath102 error ellipses in the @xmath33 - @xmath6 plane , calculated by fisher matrix , assuming a fiducial model with @xmath159 and @xmath6 = 0.01 and a cosmic - variance limited measurement to @xmath84 . the different ellipses show the effect of changing the choice of tensor / scalar ratio @xmath5 in the fiducial model : larger values of @xmath5 make it easier to distinguish the trans - planckian modulation . the top panel shows ellipses for @xmath160 ( black , outer ) , @xmath161 ( blue , next in ) , @xmath96 ( green , second smallest ) , and @xmath162 ( red , inner ) . the bottom panel shows the same figure zoomed in to show the three inner ellipses . ( negative values for @xmath6 are nonphysical , and are included to show the full error ellipse.),title="fig:",width=297 ] @xmath102 error ellipses in the @xmath33 - @xmath6 plane , calculated by fisher matrix , assuming a fiducial model with @xmath159 and @xmath6 = 0.01 and a cosmic - variance limited measurement to @xmath84 . the different ellipses show the effect of changing the choice of tensor / scalar ratio @xmath5 in the fiducial model : larger values of @xmath5 make it easier to distinguish the trans - planckian modulation . the top panel shows ellipses for @xmath160 ( black , outer ) , @xmath161 ( blue , next in ) , @xmath96 ( green , second smallest ) , and @xmath162 ( red , inner ) . the bottom panel shows the same figure zoomed in to show the three inner ellipses . ( negative values for @xmath6 are nonphysical , and are included to show the full error ellipse.),title="fig:",width=297 ] @xmath102 error ellipses in the @xmath33 - @xmath6 plane , calculated by fisher matrix , assuming a fiducial model with @xmath159 and @xmath6 = 0.01 and an observation with the projected sensitivity of the planck surveyor satellite . the different ellipses show the effect of changing the choice of tensor / scalar ratio @xmath5 in the fiducial model : larger values of @xmath5 make it easier to distinguish the trans - planckian modulation . the top panel shows ellipses for @xmath163 ( black , outer ) , @xmath161 ( blue , next in ) , @xmath96 ( green , second smallest ) , and @xmath162 ( red , inner ) . the bottom panel shows the same figure zoomed in to show the three inner ellipses.,title="fig:",width=297 ] @xmath102 error ellipses in the @xmath33 - @xmath6 plane , calculated by fisher matrix , assuming a fiducial model with @xmath159 and @xmath6 = 0.01 and an observation with the projected sensitivity of the planck surveyor satellite . the different ellipses show the effect of changing the choice of tensor / scalar ratio @xmath5 in the fiducial model : larger values of @xmath5 make it easier to distinguish the trans - planckian modulation . the top panel shows ellipses for @xmath163 ( black , outer ) , @xmath161 ( blue , next in ) , @xmath96 ( green , second smallest ) , and @xmath162 ( red , inner ) . the bottom panel shows the same figure zoomed in to show the three inner ellipses.,title="fig:",width=297 ] @xmath102 error ellipses in the @xmath141 - @xmath6 plane , calculated by fisher matrix , assuming a fiducial model with @xmath159 and @xmath6 = 0.01 and a cosmic - variance limited observation . the different ellipses show the effect of changing the choice of tensor / scalar ratio @xmath5 in the fiducial model : larger values of @xmath5 make it easier to distinguish the trans - planckian modulation . the top panel shows ellipses for @xmath163 ( black , outer ) , @xmath161 ( blue , next in ) , and @xmath96 ( green , smallest ) . the bottom panel shows the same figure zoomed in to show the three inner ellipses.,title="fig:",width=297 ] @xmath102 error ellipses in the @xmath141 - @xmath6 plane , calculated by fisher matrix , assuming a fiducial model with @xmath159 and @xmath6 = 0.01 and a cosmic - variance limited observation . the different ellipses show the effect of changing the choice of tensor / scalar ratio @xmath5 in the fiducial model : larger values of @xmath5 make it easier to distinguish the trans - planckian modulation . the top panel shows ellipses for @xmath163 ( black , outer ) , @xmath161 ( blue , next in ) , and @xmath96 ( green , smallest ) . the bottom panel shows the same figure zoomed in to show the three inner ellipses.,title="fig:",width=297 ] @xmath102 error ellipses in the @xmath141 - @xmath6 plane , calculated by fisher matrix , assuming a fiducial model with @xmath159 and @xmath6 = 0.01 and an observation with the projected sensitivity of the planck surveyor satellite . the different ellipses show the effect of changing the choice of tensor / scalar ratio @xmath5 in the fiducial model : larger values of @xmath5 make it easier to distinguish the trans - planckian modulation . the top panel shows ellipses for @xmath163 ( black , outer ) , @xmath161 ( blue , next in ) , @xmath96 ( green , second smallest ) , and @xmath162 ( red , inner ) . the bottom panel shows the same figure zoomed in to show the three inner ellipses.,title="fig:",width=297 ] @xmath102 error ellipses in the @xmath141 - @xmath6 plane , calculated by fisher matrix , assuming a fiducial model with @xmath159 and @xmath6 = 0.01 and an observation with the projected sensitivity of the planck surveyor satellite . the different ellipses show the effect of changing the choice of tensor / scalar ratio @xmath5 in the fiducial model : larger values of @xmath5 make it easier to distinguish the trans - planckian modulation . the top panel shows ellipses for @xmath163 ( black , outer ) , @xmath161 ( blue , next in ) , @xmath96 ( green , second smallest ) , and @xmath162 ( red , inner ) . the bottom panel shows the same figure zoomed in to show the three inner ellipses.,title="fig:",width=297 ] this situation is very different for the case of planck surveyor , shown in figs . [ fig : planckfishern ] and [ fig : planckfisherdn ] for the @xmath33 - @xmath6 and @xmath141 - @xmath6 planes , respectively . we see that in this case , the inclusion of a trans - planckian modulation in the parameter set strongly degrades the ability to measure the power spectrum parameters . this degeneracy is lifted if the tensor / scalar ratio is large enough , effectively disappearing altogether for a tensor / scalar ratio of order @xmath164 . finally , as expected , the measurement uncertainties on the trans - planckian parameters increase significantly when a full set of other cosmological parameters is taken into account . the fisher matrix analysis is in good agreement with the full markov chain monte carlo analysis discussed in sec . [ sec : markovchains ] , indicating that a trans - planckian modulation of order @xmath108 is only detectable at the @xmath102 level in the most optimistic case of a very large tensor / scalar ratio @xmath164 . in the next section we discuss the mcmc analysis in detail and derive the error ellipses relative to a fiducial model with @xmath107 , testing our ability to distinguish a signal from the null hypothesis . this analysis ( which includes a longer lever arm of @xmath165 ) , indicates that a cosmic - variance limited measurement can place a @xmath102 upper limit on trans - planckian fluctuations of around @xmath166 for the case with a large tensor / scalar ratio ( @xmath167 ) , and @xmath168 for the case with an undetectably small tensor / scalar ratio . in this section , we use a markov chain monte carlo ( mcmc ) technique to evaluate the likelihood function of model parameters . this approach , proposed by @xcite in the context of cosmological parameter estimation , has become the standard tool for such analyses ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? mcmc is a method to simulate posterior distributions . in particular , we simulate observations from the posterior distribution @xmath169 , of a set of parameters @xmath170 given event @xmath171 , obtained via bayes theorem , @xmath172 where @xmath173 is the likelihood of event @xmath171 given the model parameters @xmath14 and @xmath174 is the prior probability density . the mcmc generates random draws ( i.e. simulations ) from the posterior distribution that are a `` fair '' sample of the likelihood surface . from this sample , we can estimate all of the quantities of interest about the posterior distribution ( mean , variance , confidence levels ) . the mcmc method scales approximately linearly with the number of parameters , thus allowing us to perform a likelihood analysis in a reasonable amount of time . a properly derived and implemented mcmc draws from the joint posterior density @xmath175 once it has converged to the stationary distribution . for our application , @xmath14 denotes a set of 10 parameters . the set is almost the same as used for the fisher matrix calculation . we have 4 `` late - time '' parameters : the physical energy density in baryons @xmath176 , the total physical energy density in matter @xmath177 , the optical depth to reionization @xmath145 , the hubble constant in units of @xmath178 km s@xmath32 mpc @xmath179 . there are 6 `` primordial parameters '' : the tensor - to - scalar ratio @xmath5 , the spectral slope @xmath33 , the running of the spectral slope @xmath180 , the power spectrum amplitude @xmath34 , and the parameters of the trans - planckian model defined above , @xmath6 and @xmath66 . event @xmath171 will be the set of observed @xmath181 . the priors on the model are that the universe is flat , so that @xmath182 , and the dark energy is a cosmological constant with equation of state @xmath183 . the definitions used for the unmodified primordial power spectra of the scalar and tensor perturbations @xmath184 and @xmath185 are given in section [ sec : model ] . the experiment in the case study is considered to measure @xmath186 power spectra to @xmath187 ; beyond this , the systematic errors in removing the contamination from secondary cmb anisotropies are likely to be non - negligible . we include the effect of weak lensing on all power spectra , since this is the `` theoretical limit '' in the accuracy of the measurement - weak lensing aliases power between multipoles and tends to smooth out small - scale oscillations in the data . therefore , any of the further uncertainties that must be accounted for in a practical cmb experiment , such as foreground removal , partial sky coverage etc , will degrade the constraints obtainable on the trans - planckian parameters relative to those found in this study . for the simulated `` data '' , we consider two cases : one with significant primordial tensor modes ( denoted case h ) , and one with an undetectably small contribution of primordial tensors ( denoted case l ) . case h is a mass - term inflation model with tensor - to - scalar ratio @xmath188 which is consistent with _ wmap _ data at the @xmath101 level . case l is simply the _ wmap _ lcdm concordance model with @xmath189 ( quadrupole ratio @xmath190 ) . the objective of this study is to see how far from the null case the trans - planckian parameters must be before they are distinguished from the null model by at least @xmath102 . consequently , these fiducial models do not contain any trans - planckian oscillations . our fiducial models also do not contain any running of the primordial scalar spectral index ; however , the effect of this parameter would be degenerate with a very long wavelength trans - planckian modulation . hence we marginalize over this parameter in our analysis . in the practical implementation of the mcmc , we ran 16 chains for case h and 32 chains for case l , each started at random @xmath103 steps away from the fiducial model . by making use of preliminary test chains , the proposal distribution was calibrated using the covariance between the parameters in order to minimize correlations ; the chains were configured to step in the principal directions of the parameter covariance matrix optimal @xmath191% acceptance rates for case h ; for case l , only a few % acceptance rates were achieved even after several recalibrations of the proposal matrix , indicating the complexity of the likelihood surface . the chains were run on a high - performance 64-bit opteron beowulf cluster , and 300,000 and 500,000 total models were computed for case h and case l , respectively . the presence of `` islands '' in the likelihood space meant that the chains converged relatively slowly , compared to what is typical for a fit to cmb observations when the trans - planckian terms are absent . in fact , we stopped the chains for case l slightly short of full convergence , but long after the predicted parameter regions had stabilized . tables [ table : params_bigr ] and [ table : params_tinyr ] show the fiducial models and recovered marginalized 1d 68% probability parameter constraints on the ten model parameters for case h and case l , respectively . figures [ fig : params_bigr ] and [ fig : params_tinyr ] show the marginalized 2d joint 68% and 95% confidence regions for pairs of the six `` primordial '' parameters for the same models . in case h , which has potentially observable primordial tensor modes , a deviation of @xmath6 from null is detectable at greater than @xmath102 level if @xmath192 in an ideal experiment , which renders @xmath193 potentially detectable in a real - world experiment . in case l , which has unobservably small primordial tensor modes , one requires @xmath194 in order to be detectable at greater than the @xmath102 level in an ideal experiment , rendering such a detection implausible in the real world . @xcite discussed a degeneracy between @xmath38 and @xmath114 which appears to be essentially eliminated in our model due to the extra constraint on @xmath38 from tensor modes . .[table : params_bigr ] case h : a fiducial model with @xmath188 and no trans - planckian modulations . the input model , recovered mean parameters and the 68% probability level of the 1d marginalized likelihood . the derived parameter @xmath195 ( the linear theory prediction for the amplitude of fluctuations within 8 mpc / h spheres ) is given in the last row . [ cols="<,^,^",options="header " , ] the key observations to be drawn from these results are : 1 . the constraint on @xmath6 differs by an order of magnitude between case h and case l. this reinforces the point made above that _ if _ our universe happens to be in a favorable part of the parameter space where the energy scale of inflation is high enough to produce an observable amplitude of primordial tensor modes , then one has a significantly improved chance of detecting any trans - planckian modifications to the spectrum whose properties are similar to those of the case tested here . the cosmological parameters are constrained at the @xmath196 level except for the scalar spectral index and the amplitude of fluctuations at @xmath197 , which are only constrained at the @xmath198 level . therefore , another cosmological data - set that aids in constraining these parameters ( for example , the proposed ska survey , @xcite ) will minimize the degeneracies with these parameters shown in figs.[fig : params_bigr ] and [ fig : params_tinyr ] and improve parameter constraints further @xcite . we believe that this analysis is the most thorough investigation of the detectability of a trans - planckian modulation to the primordial power spectrum that has been performed to date . we consider three complementary approaches a simple grid search , a fisher matrix evaluation of the likely error ellipses , and a monte carlo markov chain fit to a simulated cmb spectrum , and find that they are all in broad agreement . moreover , we also explain the `` islands '' seen in the likelihood space in previous papers on this topic @xcite , and show that their distribution and properties can be understood and reproduced via a simple analytic argument . the approach we have taken here is analyze a _ specific _ ansatz for introducing a minimum length into the calculation of the perturbation spectrum in a general model of slow - roll inflation @xcite . we do not claim that this model is a correct description of the trans - planckian contribution to the perturbation spectrum , but intend it as a case study of what might be possible if one has a specific and well - motivated correction to the spectrum that leads to a @xmath18 dependent modulation . there is considerable theoretical uncertainty surrounding this point , and it is entirely possible that a future rigorous calculation of the perturbation spectrum within string theory or some other model of ultra high energy physics will predict that there is _ no _ modulation to be observed , and the bound calculated here will not apply . we believe that the constraint obtained here is applicable to a general class of modulated spectra . in practice , one can imagine adding an arbitrary modulation to the primordial spectrum , as considered by okamoto and lim @xcite , in which case one has parameters which describe the amplitude , wavelength , and phase of the modulation as a function of @xmath18 . in our model there are only two free parameters the phase of the modulation , and @xmath6 which determines the amplitude . the phase will always be arbitrary in the absence of detailed theory that matches specific scales in the present universe to those during inflation , which would require a full understanding of the post - inflationary expansion history and the physics of reheating , while the amplitude is directly related to the scale of new physics , which is what we are trying to measure . however , the wavelength of the oscillation turns out to be given in terms of the slow - roll parameter @xmath38 , which is itself directly related to the tensor / scalar ratio @xmath5 . this dependence is easy to understand , since the form of the modulation is not determined by the ( fixed ) boundary condition , but by the slow variation in the horizon size @xmath199 with time . the precise functional form of the connection between @xmath38 and the modulated spectrum may vary in other specific models . however , it is very likely that the modulation can always be expressed in terms of @xmath6 and the scale dependence of @xmath1 or thus @xmath38 , and perhaps the higher order slow - roll parameters , which are also reachable through their contribution to the scalar spectral index . consequently , we believe that the results we have seen here will generalize to other modulated spectra , even though we have reduced the number of trans - planckian parameters relative to those considered by okamoto and lim . the second general conclusion we draw is that the detectability of any modulated spectrum depends on the value of @xmath1 . since the wavelength of the modulation depends on @xmath38 , detecting the primordial tensors provides an orthogonal constraint on the value of this parameter , and thus constrains the trans - planckian corrections . since @xmath1 fixes the energy scale at which inflation occurs , it is arguably the single most interesting cosmological parameter that is not currently fixed by observations . the analysis here simply adds to the importance of this parameter , since we have shown that measuring @xmath1 will put tighter bounds on any trans - planckian corrections to the spectrum . since the theoretical uncertainty about the form and existence of these corrections is yet to be resolved , we can not advocate mounting an observational campaign solely to look for this kind of signal . however , there is already good and sufficient reason to make an all out effort to measure @xmath1 during inflation , via high precision measurements of the cmb and particularly its polarization . we believe that mission planners will want to be aware of the possibility that precision cmb measurements may potentially probe physics at the string scale , since this will not add to the cost of any mission but represents an exciting and additional use of the data they can be expected to return . in quantitative terms , we see that the with a large ( @xmath200 ) value of the tensor to scalar ratio , a `` perfect '' map of the primordial temperature and polarization anisotropies to the cmb could rule out a trans - planckian modulation to the spectrum at the @xmath201 level for @xmath202 . this is roughly the same as the result reported by okamoto and lim . we vary a larger set of cosmological parameters than they do , which reduces the level of precision we can hope for . however by adding the tensor contribution to the analysis and removing a free parameter from the specification of modulation we can tighten the bound . like okamoto and lim , we consider a perfect measurement of the cmb and any real experiment will be contaminated by imperfectly removed foregrounds . consequently , we believe that a value of @xmath6 on the order of 0.01 represents a reasonable lower bound on what can be detected in practice , provided the amplitude of the tensor contribution is significant . on the other hand , adding other orthogonal datasets , such as a measurement of @xmath179 and the late time expansion history of the universe from a snap - like mission , or constraints on the primordial spectrum from future large scale structure surveys will further constrain the error ellipses relative to those which can be achieved via the cmb alone . at this point , the theoretical uncertainty surrounding this calculation makes it unwise to push this analysis significantly further . however , we believe that considerable progress has been made in understanding the physical issues surrounding the trans - planckian corrections to the primordial spectrum over the last few years . consequently , it is not excessively optimistic to hope that the remaining issues can be resolved on a shorter timescale than it will take to perform a measurement of the cmb that even approaches the precision we have assumed in our analysis . in this light , we are particularly interested in the effective field theory approach taken in @xcite , and extended in the more recent preprint @xcite . in this case , the operators which can contribute to a modification to the spectrum can be cataloged within effective field theory . from the observational perspective , an analogous calculation to the one presented here will constrain the values of the prefactors in front of these operators , and we intend to pursue this in future work . we thank brian greene , gaurav khanna , eugene lim , alexey makarov , jerome martin , koenraad schalm , and uros seljak for useful discussions , and are grateful to the staff at yale s high performance computing facility for their assistance . re is supported in part by the united states department of energy , grant de - fg02 - 92er-40704 . hvp is supported by nasa through hubble fellowship grant # hf-01177.01-a 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 ; she acknowledges the hospitality of the institute of astronomy , cambridge , where part of this work was carried out . whk acknowledges the hospitality of the perimeter institute , waterloo , ontario , where part of this work was carried out . 99 r. h. brandenberger , arxiv : hep - ph/9910410 . a. kempf , phys . d * 63 * , 083514 ( 2001 ) [ arxiv : astro - ph/0009209 ] . j. martin and r. h. brandenberger , phys . d * 63 * , 123501 ( 2001 ) [ arxiv : hep - th/0005209 ] . r. h. brandenberger and j. martin , mod . a * 16 * , 999 ( 2001 ) [ arxiv : astro - ph/0005432 ] . j. c. niemeyer , phys . d * 63 * , 123502 ( 2001 ) [ arxiv : astro - ph/0005533 ] . r. easther , b. r. greene , w. h. kinney and g. shiu , phys . d * 64 * , 103502 ( 2001 ) [ arxiv : hep - th/0104102 ] . r. easther , b. r. greene , w. h. kinney and g. shiu , phys . d * 67 * , 063508 ( 2003 ) [ arxiv : hep - th/0110226 ] . j. c. niemeyer and r. parentani , phys . d * 64 * , 101301 ( 2001 ) [ arxiv : astro - ph/0101451].0 a. kempf and j. c. niemeyer , phys . d * 64 * , 103501 ( 2001 ) [ arxiv : astro - ph/0103225 ] . u. h. danielsson , phys . d * 66 * , 023511 ( 2002 ) [ arxiv : hep - th/0203198 ] . r. easther , b. r. greene , w. h. kinney and g. shiu , phys . d * 66 * , 023518 ( 2002 ) [ arxiv : hep - th/0204129 ] . n. kaloper , m. kleban , a. e. lawrence and s. shenker , phys . d * 66 * , 123510 ( 2002 ) [ arxiv : hep - th/0201158 ] . r. h. brandenberger and j. martin , int . j. mod . a * 17 * , 3663 ( 2002 ) [ arxiv : hep - th/0202142 ] . j. martin and r. brandenberger , phys . d * 68 * , 063513 ( 2003 ) [ arxiv : hep - th/0305161 ] . k. schalm , g. shiu and j. p. van der schaar , arxiv : hep - th/0401164 . s. shankaranarayanan and l. sriramkumar , arxiv : hep - th/0403236 . v. s. kaplunovsky , nucl . b * 307 * , 145 ( 1988 ) [ erratum - ibid . b * 382 * , 436 ( 1992 ) ] [ arxiv : hep - th/9205068 ] . n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . b * 429 * , 263 ( 1998 ) [ arxiv : hep - ph/9803315 ] . i. antoniadis , n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . b * 436 * , 257 ( 1998 ) [ arxiv : hep - ph/9804398 ] . v. bozza , m. giovannini and g. veneziano , jcap * 0305 * , 001 ( 2003 ) [ arxiv : hep - th/0302184 ] . d. h. lyth , phys . lett . * 78 * , 1861 ( 1997 ) [ arxiv : hep - ph/9606387 ] . u. seljak and c. m. hirata , phys . d * 69 * , 043005 ( 2004 ) [ arxiv : astro - ph/0310163 ] . j. khoury , b. a. ovrut , p. j. steinhardt and n. turok , phys . d * 64 * , 123522 ( 2001 ) [ arxiv : hep - th/0103239 ] . m. gasperini and g. veneziano , astropart . * 1 * , 317 ( 1993 ) [ arxiv : hep - th/9211021 ] . j. e. lidsey , d. wands and e. j. copeland , phys . rept . * 337 * , 343 ( 2000 ) [ arxiv : hep - th/9909061 ] . j. e. lidsey , a. r. liddle , e. w. kolb , e. j. copeland , t. barreiro and m. abney , rev . phys . * 69 * , 373 ( 1997 ) [ arxiv : astro - ph/9508078 ] . r. easther , arxiv : hep - th/0407042 . c. p. burgess , j. m. cline , f. lemieux and r. holman , jhep * 0302 * , 048 ( 2003 ) [ arxiv : hep - th/0210233 ] . l. sriramkumar and t. padmanabhan , arxiv : gr - qc/0408034 . l. bergstrom and u. h. danielsson , jhep * 0212 * , 038 ( 2002 ) [ arxiv : hep - th/0211006 ] . o. elgaroy and s. hannestad , phys . d * 68 * , 123513 ( 2003 ) [ arxiv : astro - ph/0307011 ] . t. okamoto and e. a. lim , [ arxiv : astro - ph/0312284 ] . j. martin and c. ringeval , phys . d * 69 * , 083515 ( 2004 ) [ arxiv : astro - ph/0310382 ] . j. martin and c. ringeval , phys . d * 69 * , 127303 ( 2004 ) [ arxiv : astro - ph/0402609 ] . j. martin and c. ringeval , arxiv : hep - ph/0405249 . l. verde _ et al . _ , astrophys . j. suppl . * 148 * , 195 ( 2003 ) [ arxiv : astro - ph/0302218 ] . w. h. kinney , phys . d * 56 * , 2002 ( 1997 ) [ arxiv : hep - ph/9702427 ] . a. ashoorioon , a. kempf and r. b. mann , arxiv : astro - ph/0410139 . c. armendariz - picon and e. a. lim , jcap * 0312 * , 006 ( 2003 ) [ arxiv : hep - th/0303103 ] . u. seljak and m. zaldarriaga , astrophys . j. * 469 * , 437 ( 1996 ) [ arxiv : astro - ph/9603033 ] . m. tegmark , a. taylor and a. heavens , astrophys . j. * 480 * , 22 ( 1997 ) [ arxiv : astro - ph/9603021 ] . n. christensen and r. meyer , arxiv : astro - ph/0006401 . n. christensen , r. meyer , l. knox and b. luey , class . * 18 * , 2677 ( 2001 ) [ arxiv : astro - ph/0103134 ] . a. kosowsky , m. milosavljevic and r. jimenez , phys . d * 66 * , 063007 ( 2002 ) [ arxiv : astro - ph/0206014 ] . a. lewis and s. bridle , phys . d * 66 * , 103511 ( 2002 ) [ arxiv : astro - ph/0205436 ] . l. knox , n. christensen and c. skordis , arxiv : astro - ph/0109232 . s. rawlings , f. b. abdalla , s. l. bridle , c. a. blake , c. m. baugh , l. j. greenhill and j. m. van der hulst , arxiv : astro - ph/0409479 . s. hannestad and l. mersini - houghton , arxiv : hep - ph/0405218 . b. r. greene , k. schalm , g. shiu and j. p. van der schaar , arxiv : hep - th/0411217 .

we report on the results of an-band photometric variability survey of eighteen l dwarfs . we find that seven exhibit statistically significant variations above the 95.4% confidence level with root - mean - square scatter ( including photometric errors ) between 0.010 and 0.083 mag . another five targets have variability probabilities% , suggesting that these are likely variable objects . three of the variable objects display significant peaks in a clean periodogram that are several times higher than the noise . the period for 2mass 0345 + 25 is clearly not intrinsic to the object and can be dismissed . the periods found for 2mass 0746 + 20ab and 2mass 1300 + 19 are unique but longer than those periods likely from rotation velocity measurements and they do not represent periodic behavior in the light curve that persists through the entire data set . these observations suggest that we are not observing the rotation modulation of a long - lived albedo feature . instead , rapid evolution of atmospheric features is likely causing the non - periodic variability . the remaining variable objects show no prominent features in their light curves , suggesting even more rapid evolution of atmospheric features . we argue against the existence of magnetic spots in these atmospheres and favor the idea that non - uniform condensate coverage is responsible for these variations . the magnetic reynolds number in the atmosphere of l dwarfs is too small to support the formation of magnetic spots . in contrast , silicate and iron clouds are expected to form in the photospheres of l dwarfs . inhomogeneities in such cloud decks and the evolution of the inhomogeneities can plausibly produce the observed photometric variations .

astro-ph0205305

large numbers of l dwarfs @xcite are being discovered by surveys such as the 2 micron all sky survey ( 2mass ) and the sloan digital sky survey ( sdss ) . despite their growing number , less than 10% of known l dwarfs have been observed to determine whether or not they are photometrically variable @xcite . photometric monitoring of l dwarfs , and brown dwarfs in general , not only explores why some objects are variable while others are not , but it can also help constrain rotation periods or timescales for the evolution of surface features . multicolor photometric and spectroscopic monitoring could even be used to identify the nature of those features . @xcite presented the first attempt to detect clouds in brown dwarf atmospheres . they observed the m9 brown dwarf lp944@xmath220 and the l5 brown dwarf denis 1228@xmath215 through two narrow - band filters chosen to detect changes in tio absorption . the changes in the tio band strength were presumed to indicate changes in the opacity , which occurs when tio is depleted through condensation . they found that lp944@xmath220 was variable , but denis 1228@xmath215 was not . the authors speculated that the passage of clouds over the disk of lp944@xmath220 produced small changes in the brightness temperature in their narrow - band filters and caused small variations ( @xmath30.04 magnitude ) . however , the lack of variability in their l dwarf does not exclude the possibility of clouds in that object since the variations observed in lp944@xmath220 had smaller amplitudes than the errors for denis 1228@xmath215 . @xcite conducted a variability search in the broad - band @xmath0 filter and found evidence of variability in the l1.5 dwarf 2mass 1145 + 23 . the object displayed @xmath30.04 magnitude variations that repeated with a period of 7.1 hours . in an expanded study of 21 l and m dwarfs @xcite found that over half of their sample exhibited statistically significant variations with root - mean - square ( rms ) scatter between 0.010 and 0.055 magnitude and time scales of 0.4 to 100 hours . they were unable to find periodic light curves for many of the variables . 2mass 1145 + 23 , however , exhibited variability with a period of 11.2 hours . they suggested that evolving surface features , possibly dust clouds or magnetic spots , were responsible for the change in period . a similar varying period has also been observed in the m9.5 dwarf star bri0021 by @xcite . photometric variability has recently been reported for the l2 dwarf kelu-1 @xcite . like @xcite , these authors use a special filter positioned at a region sensitive to changes in tio and crh absorption . kelu-1 displayed small ( @xmath4 ) peak - to - peak variations that phased well to a period of 1.8 hours . either magnetic spots or clouds could plausibly be associated with the observed variability . magnetic fields have been measured for several m dwarfs and estimates of their strength are a few kg @xcite , with surface filling factors generally @xmath5 . periodic photometric variations have been observed in only a few m dwarfs , suggesting that either the surfaces of these objects are completely covered with spots or that the spots are few in number and uniformly distributed @xcite . in addition , @xcite reports that no good data exist to support the scenario of cyclic , organized spots in m dwarfs cooler than m0 . for the l dwarfs , clouds are a reasonable potential source of variability . iron , enstatite , and forsterite are the most abundant species expected to condense at the atmospheric temperatures and pressures characteristic of the l dwarfs @xcite . once condensed , the species likely settle into discrete , optically thick , cloud decks , with optically thicker clouds arising in progressively later l dwarfs @xcite . since the atmospheric circulation pattern of most l dwarfs is likely similar to that of jupiter @xcite , it is not unreasonable to expect that many l dwarfs also have a banded appearance . any large inhomogeneities ( thicker clouds or clearings in the cloud deck ) could then produce a photometric signal . @xcite show that if jupiter were to be observed in thermal emission as an unresolved point source , the great red spot would provide a photometrically detectable signal . if l dwarfs have similar cloud features to jupiter , then it is plausible that they may also exhibit photometric variations . we are conducting a photometric monitoring program of l dwarfs in the @xmath6 band . in this study we present the light curves for several l dwarfs showing statistically significant variability . we discuss which of these display significant periodicity and why the others do not . we provide theoretical calculations that reject the magnetic spot hypothesis for the origin of the variation and argue in favor of clouds . our target list was derived from a compilation of all published spectroscopically determined l dwarfs . we searched the list for all l dwarfs above @xmath7 and brighter than @xmath8 . for the objects whose @xmath0 magnitudes had not been measured we estimated their brightness in that filter based upon the objects spectral types and the @xmath9 colors from @xcite . better estimates of these @xmath0 magnitude were made after observations using instrumental magnitudes and @xmath9 colors . at the start of this project there were 24 objects that fit these criteria . table [ tab - list ] lists the 18 l dwarfs in our sample that were observed . except for an l4 , we have at least one object for each decimal subclass between l0 and l4.5 , inclusive . all are discoveries from the 2mass survey @xcite and are field objects . their @xmath0-band magnitudes range from 15.11 mag ( measured ) to @xmath318.1 mag ( estimated ) . finally , exactly half of the sample has measurable h@xmath10 emission . 2mass 0746 + 20ab and 2mass 1146 + 22ab are close visual binaries as seen with keck and the hubble space telescope @xcite and are unresolved in our observations . @xcite state that keck echelle spectroscopy indicate 2mass 0345 + 25 is a spectroscopic binary . however , additional spectra of 2mass 0345 + 25 can neither confirm nor reject the possibility of a binary system ( reid 2001 , private communication ) . only two other targets were observed for binarity ( 2mass 0036 + 18 and 2mass 1338 + 41 ; reid et al . 2001 ) , but none was found . however , nearly 25% of all l dwarfs observed in binary studies are shown to be binaries . therefore , the probability that other binaries exist in our sample is quite high . we used the new mexico state university 1 m telescope at apache point observatory for our observations . the telescope was equipped with an apogee 512@xmath11512 pixel , thermo - electrically cooled ccd with a pixel scale of 0.8 arcsec / pixel and was operated robotically . for each cycle the telescope was focused on a bright star and the script chose one to three targets that had airmasses between 1.03 and 1.9 . after slewing to a target , the telescope pointing was refined by slewing to a nearby bright standard star and centering the star in the ccd . the telescope reslewed to the target and found a guide star from the usno - sa2.0 catalog @xcite . on most nights three consecutive 5 minute exposures were taken in the cousins @xmath0 band for each target ; no other filters or exposure times were used . this cycle was repeated as many times as possible throughout the night . each target was usually observed every 1 - 2 hours in a given night , depending on the target s hour angle and priority . observations for this program started on 2000 october 30 and ended on 2001 june 20 . during this period the telescope was used as much as possible for this program . observations were taken at all moon phases and under a variety of seeing conditions . over 3700 science images ( not including positional calibration frames and focus runs ) were obtained , resulting in over 300 hours of on - sky time for science observations . as with any observing program , time was occasionally lost due to inclement weather and engineering problems . the longest off - sky period was approximately 1 month from mid - december to mid - january . the images were reduced using procedures in the xvista astronomical image reduction and analysis package . all images were first dark subtracted and then flat fielded . finally , a correction was applied to reduce the effect of night sky emission line fringing in the frames . these steps are discussed in more detail below . dark current is an additive noise that depends strongly on the operating temperature of the ccd and the exposure time . the amount of charge accumulated increases with the exposure time ; the rate of charge production increases with increasing temperature . we limited our thermo - electrically cooled ccd to be set at only three possible temperatures ( @xmath240 , @xmath245 , and @xmath250@xmath12 c ) , dependent upon the ambient temperature . the dark level was typically @xmath13 of the total background for observations taken at @xmath250@xmath12 c and @xmath14 for observations at @xmath240@xmath12 c. consequently , dark current was an important source of background in our images . we constructed a master dark from thirty 5-minute dark exposures obtained in late november taken at a ccd temperature of @xmath250@xmath12 c. ( the resulting error in photometry caused by using this dark frame with science frames obtained at the other temperatures was well below 1% . ) the frames were combined using a median average and divided by the exposure time to yield a dark frame in counts per second . the dark frame multiplied by the exposure time was subtracted from each image . the division of the flat field removes the pixel - to - pixel variations of the quantum efficiency . as with the dark frame , we used the same flat field for all images . this flat was constructed from about 12 individual twilight flat field frames taken around the same dates as the dark frames . no major changes to the telescope or the instrument configuration were made during the entire observing program . nonetheless , we tested the flat obtained in november with a set of flats obtained at the end of the program . there was an average deviation between the two flats of 0.5% across the entire chip and a maximum deviation of about 3% close to the edges . inspection of flat - fielded images did not show any prominent residual features attributable to an improper flat . however , to account for small flat field errors , we added an error of 0.005 mag in quadrature with the photometric errors . night sky emission lines can reflect several times inside a thinned ccd and cause interference . this interference is a source of coherent background in our science images and manifests itself as a fringe pattern present in the raw images . the pattern is stable but the amplitude of the fringes is highly variable and depend on moonlight , the exposure time , cloud cover , and position in the sky . fringes are also an additive effect , unlike the multiplicative effect of the flat field . since many of our targets are faint compared to the background , it is important that we correctly determine the pattern of the fringes and accurately compute the pattern s amplitude in the raw images . incidentally , there is no evidence of fringing in our twilight flats , presumably because the source of the light in the flats is scattered sun light . the flux from this scattered light dominates over the night sky lines responsible for the fringing in the science frames . we assumed that the background of the images consisted of a normalized fringe frame representing the interference component multiplied by some image - dependent factor @xmath15 , a constant continuum sky @xmath16 , and brightness gradients along the x and y axes scaled by @xmath17 and @xmath18 , respectively , to account for any moonlight or scattered light . after a frame was flat fielded and dark subtracted a mask was placed on the frame to remove all pixels 5@xmath19 above the sky . the linear x and y gradients were constructed and best fits to the factors @xmath16 , @xmath15 , @xmath17 , and @xmath18 were found for a 300@xmath11300 pixel box from the central region of the image . the normalized fringe frame was multiplied by @xmath15 and subtracted from the image . we constructed the normalized fringe frame from a large set ( @xmath20 ) of science frames that were either from different target fields or from fields in which the stars had been significantly offset . for a first guess at the fringe frame , we reduced the individual frames with the flat field and dark frame . each frame was then divided by its sky level , creating a set of normalized frames . these frames were median averaged to form a fringe frame . the next iteration reduced the images with the dark , flat field , and the first guess at the fringe frame so that the four components of the background , most importantly @xmath16 and @xmath15 , could be found . next , we reduced the raw frames again , this time using only the flat and dark . these frames were normalized by subtracting the sky continuum @xmath16 and dividing by the fringe level @xmath15 . they were then median averaged to create a new fringe frame . this process was iterated three more times , with each iteration using the fringe frame from the previous iteration . the final result was a fringe frame normalized to a mean value of 0 . this method for the fringe removal worked reasonably well , in the sense that in most frames the fringe patterns were completely removed . for the frames in which the fringes were not completely removed , the residual fringe level was significantly @xmath13 the level of the background . we added an error term , discussed below , to account for sky subtraction errors from an imperfect fringe subtraction . we used aperture photometry to obtain the instrumental magnitudes of the targets and references . a circular aperture with a radius of 4 pixels was used to compute source fluxes ; sky levels were determined from an annulus between 10 and 24 pixels from the source center . we performed differential photometry with the reference being the average brightness of non - variable stars in the field . using a reference that is the average of many stars increased the signal - to - noise ratio in our target light curves . furthermore , any color dependence on telluric extinction due to the difference in color between the targets and references was minimized by obtaining our science frames at low airmass ( nearly 80% of our science frames were obtained at an airmass below 1.4 and 50% were below 1.24 airmasses ) . during the reduction process some frames were thrown out because there was an obvious problem with those frames . the most common problems were streaked stars due to tracking or guiding errors and missing stars from cloudy skies or an automatic dome closure . frames with less obvious problems ( e.g. being slightly out of focus ) were allowed to go through the reduction process , but some were removed later as discussed below . the first step in computing the mean reference is the determination of which reference stars are non - variable . our ability to detect the variability of an object depends strongly on that object s photometric errors . these errors are computed from the poisonian error of the object , the sky noise error , and the error of the mean sky level , which is taken to be 0.15% of the mean sky value . we use this last term to account for errors in the fringe subtraction . we find that its use increases the photometric errors in all objects . the increase is largest for the fainter objects , allowing for a more conservative estimate of variability in these objects . we also add an error of 0.005 mag in quadrature with the photometric errors to account for changes in the flat field that occurred throughout the course of the observations . this additional error is a considerable fraction of the total error in the brightest objects and negligible in the faint ones . consequently , its effect on the detection of variability is only important for the bright objects . after all the data are collected for a target , the data set is put through several filters to remove bad points . the first processing filter removes all frames in which the average fwhm ( full width at half maximum ) of the stellar profiles is @xmath21 . this ensures that we do not use frames where the stars are slightly elongated , out of focus , or subject to poor seeing conditions . in general , these high fwhm frames have larger photometric errors . the data for each reference star , @xmath22 , is analyzed in a method similar to @xcite to determine if the reference is stable and acceptable . this is accomplished by first computing the average flux for all references in frame @xmath23 , excluding reference @xmath22 , and converting it to a magnitude , @xmath24 @xmath25 where @xmath26 is the instrumental magnitude for reference @xmath27 in frame @xmath23 , @xmath28 is the number of references . the difference in magnitude between reference @xmath22 and the mean reference is formed and @xmath29 is computed in order to determine the probability that @xmath22 is variable , @xmath30 where @xmath31 @xmath32 is the number of frames , and @xmath33 is the photometric error associated with @xmath34 . the @xmath29 statistic is computed for each reference and the probability of variability ( @xmath35 ) for the reference with the largest @xmath29 is calculated . if @xmath3695.4% ( a 2@xmath19 detection ) , that reference is flagged as variable and removed from the list of references . this process of computing the mean reference and removing variable references is iterated until no more variable references are found . finally , we calculate a mean reference for each frame based solely on the non - variable references . using all frames we compute the median brightness of the mean reference . we then remove any frame for which the brightness of the mean reference deviates by more than 0.5 mag from the median brightness . this effectively removes frames that are heavily obscured by terrestrial clouds . using this set of good frames , we re - analyze each reference ( even the ones previously flagged as variable ) and check for variability . as before the non - variable references are combined to make a mean reference for a given frame . in most cases the set of good references were the same before and after the removal of faint frames . table [ tab - sv ] presents how many good references were found for each l - dwarf field . differential magnitudes are computed for the target , using the mean reference as the reference `` star . '' the mean brightness of this differential light curve is found and a @xmath29 test is used to determine how much the light curve deviates from a constant value centered on the mean . the value of @xmath29 and the number of degrees of freedom in the computation of @xmath29 are used to calculate the probability that @xmath29 is obtained by chance . if this probability is @xmath37 ( i.e. @xmath38% , a 2@xmath19 detection ) , then the target is flagged as variable . targets suspected of being variable are searched for periodic variations using the method described below . it is important to note @xmath29 for an object and , consequently , the variability probability , depend strongly on the estimate of that object s errors . these errors are difficult to accurately determine and any changes to our estimate of the errors results in a change in @xmath29 . therefore , these results should be regarded as more of a gauge for variability than an absolute classification . a clean periodogram routine @xcite is used to search for periods in the data of the suspected variable objects . this routine computes a `` dirty '' power spectrum , whose components consist of the spectrum of the frequency components convolved with the observing window function derived from the temporal sampling of the data . the power spectrum is `` cleaned '' by subtracting the dominant peak convolved with the window function from the dirty spectrum to produce a residual spectrum . to aide in the stabilization of the routine , only a fraction , called the gain , of the response from the dominant period is removed . next , the second highest peak ( multiplied by the gain ) is removed from the residual spectrum . subsequent peaks are removed in the same fashion until all clean components are obtained . this process is repeated for additional cleanings . after a user specified number of cleanings are performed , the final spectrum is constructed based on the clean components and the final residual spectrum . for consistency in our period computations we use a gain of 0.5 and 100 cleans for all variable targets . the clean routine removes most of the `` problem '' frequencies associated with the window function that are present in a lomb - scargle routine @xcite . the result is a routine that is better able to find significant periods present in the data . the downside to this method is that there is no indicator for the noise in the power spectrum . @xcite provide an algorithm for estimating the noise in a clean power spectrum . this noise depends on the total amount of time in large gaps ( we define a large gap as any time between points @xmath3912 hours ) for the observing run . the amount of time attributed to large gaps was usually considerable for our runs and resulted in very small power spectrum noise values . consequently , numerous peaks could occur in a given power spectrum that are several hundred times higher than the computed noise . this makes it difficult to establish which peaks are significant and which are meaningless . to overcome this shortcoming , we attempt to estimate the noise by calculating the power expected for a synthetic data set with the same temporal sampling and rms scatter as the real data set . the synthetic data set is created by randomly shuffling the magnitudes in the real data set to other times in the data set . the values of the magnitudes do not change , only the time at which they occur . any periodicity in the original data set is erased when the data are randomized . therefore , the results of the clean analysis for the synthetic data provide what powers can be expected for a set a random data with a certain rms scatter . we take the mean value of the primary peak power from 1000 random shuffles as the noise level in the original data set . we calculate the primary peak power to noise ratio ( pnr ) ; data with periodic variations should have a pnr value several times greater than 1 . the variability results for our sample are presented in table [ tab - sv ] . included are the total number of nights each object was observed and the number of nights in the final data set , the number of good images used in the analysis and the total taken , the number of days the data sets cover ( t@xmath40 ) , the number of good references used for the mean reference and the total number initially considered , the probility the l dwarf is variable ( @xmath35 ) , the standard deviation of the points from the mean level ( @xmath41 ) , and the average error ( @xmath42 ) . we have already defined the variable l dwarfs as those with probabilities above 95.4% . this group constitutes 7 out of the sample of 18 targets . the five objects with @xmath4380% are considered possibly variable , since they are likely to be viewed as variable with higher precision photometry . the six objects with @xmath44 clearly show no variability above the detection limits of this study . supporting of our claim for variability in the l dwarfs , figure [ perhist ] shows the fraction of the total number of objects with a given probability of variability . over 60% of the references have probabilities below 5% , whereas only 10% of the l dwarfs do . in addition , about 40% of the l dwarfs are flagged as variable ( @xmath38% ) compared to about 15% for the references . furthermore , figure [ histogram ] shows the number of variable objects as a function of instrumental magnitude . the fraction of references in a given magnitude bin that are variable does not show any indication of being a strong function of object brightness , suggesting that we are not underestimating the errors in these objects . the slight increase between @xmath45=14 - 15 in the references is likely due to small number statistics . given this evidence , it seems clear that the l dwarfs derive from an inherently more variable population than the reference stars . three of the targets in our sample were also monitored by @xcite : 2mass 1146 + 22ab , 2mass 1439 + 19 , and 2mass 0345 + 12 . our results disagree on the status of 2mass 1146 + 22ab . this object is significantly non - variable in our sample , whereas @xcite detected variability with an rms amplitude of 0.015 mag . our average photometric error for this object is almost four times larger than this amplitude . it is quite possible that 2mass 1146 + 22ab varies with such a small amplitude as to be undetectable in this work . therefore , even though we find that this object does not display statistically significant variations , higher signal - to - noise observations could show that it is still a low - amplitude variable . the second object , 2mass 1439 + 19 , is classified as a possible variable here and a non - variable by bailer - jones & mundt . interestingly , their variability probability is larger than the one we find here ( 90% compared to 80% ) . given that the probabilities are similarly high , this object is likely a low - amplitude variable . we agree with the results of @xcite on the classification of 2mass 0345 + 12 as a variable object . the light curve of 2mass 0345 + 25 ( figure [ 2m0345dif ] ) shows quite a bit of scatter , but no periodic trends are evident . the primary peak in the clean power spectrum occurs at 24.1@xmath460.1 hours with a power @xmath18 times higher than the noise . a couple of reference stars have prominent clean peaks at 12 hours , indicating that the period is not intrinsic to any one object . consequently , we list 2mass 0345 + 25 as a variable object in table [ tab - sv ] , but do not present it in table [ tab - pow ] . @xcite presented preliminary results for three l dwarfs in this program . they found 2mass 0036 + 18 and 2mass 0135 + 12 to be variable and 2mass 1412 + 16 to be non - variable . an error was found in the reduction process used to obtain those results . the fwhm used to reject high fwhm frames was only from one star , and not an average of all stars . upon correction of the error , the classification of 2mass 0036 + 18 changed from variable to non - variable . the classifications of the other two objects did not change , but their data sets did change slightly . the new light curve for 2mass 0135 + 12 is shown in ( figure [ 2m0135dif ] ) . this object does not have a significant period ( table [ tab - pow ] ) , contrary to what was reported previously . 2mass 1108 + 68 is another l dwarf displaying statistically significant variations and no significant period . with a baseline of nearly 6 months ( figure [ 2m1108dif ] ) , the data obtained for this object represents the most extensive photometric monitoring of any l dwarf to date . though not as extensive as 2mass 1108 + 68 , 2mass 0746 + 20ab also has very good coverage ( figure [ 2m0746dif ] ) . this binary is the brightest object in our target list and is variable . the most dominant peak in the clean power spectrum ( figure [ 2m0746cln ] ) is quite high compared to other peaks and is present at a period of 31.0@xmath460.1 hours . the power of this period is nearly 5 times higher than the noise and the phased data ( figure [ 2m0746phs ] ) shows a roughly sinusoidal light curve . @xcite present @xmath47 measurements for twelve l dwarfs , including 2mass 1439 + 19 and 2mass 1146 + 22ab from this study . the values they derive span from 10@xmath462.5 to 60@xmath465 km s@xmath48 . for typical brown dwarf radii @xcite , these velocities translate to rotation periods @xmath4910 hours . because the inclination angle of the rotation axis is unknown , this period is an _ upper _ limit to the true rotation period of these objects . consequently , the 31-hour period found here is likely too long to be related to the rotation period . the first 70 days of coverage for 2mass 0746 + 20ab reveal an interesting trend ( figure [ 2m0746zoom ] ) . the data appear to have a rough saw - tooth light curve that repeat with a period of approximately 20 days . after hjd 1950 this pattern is not as prominent . it is unclear what variability source can cause a saw - tooth pattern such as this . spots and clouds should produce more gradual changes in the light curve . a flare would produce a sudden brightening , followed by a gradual dimming , the opposite to what is seen here . furthermore , @xcite estimate an orbital semi - major axis of 3.4 au , giving a period of about 18 years . therefore , both the shape and duration of the feature are not what would be expected for an eclipsing system . 2mass 1300 + 19 also has an interesting feature in its light curve ( figure [ 2m1300dif ] ) . the sequence of points around hjd=2027 ( figure [ 2m1300zoom ] ) show roughly sinusoidal variations . this span of points is approximately the same as the best period for the entire data set , 238 hours , indicating that this feature is the source of the period . the period is much longer than expected for a rotation period . it seems likely that some other cause is responsible for the feature . although the feature is reminiscent of an eclipsing binary light curve , this scenario seems unlikely . the depth of the feature implies a secondary object radius about 18% the radius of the l dwarf , roughly 3 times larger than earth . the duration suggests an orbital separation @xmath50 au . the probability that we should observe an edge - on binary system with an orbital separation @xmath50 au just as it is eclipsing is extremely small ( @xmath51 ) . furthermore , while there are stars with l and t dwarf companions at separations greater than 500 au , there are no known companions to l dwarfs beyond 10 au . it is possible that the 10-day feature in this object s light curve is related only to its surface features . the sudden creation and dissipation of a large storm could possibly produce the changes we see . a similar event happened with saturn several years ago @xcite , but over a longer dissipation timescale . however , features on jupiter and saturn are known to evolve on timescales from hours to years , depending on the features positions and rotation orientations ( beebe 2001 , private communication ) , so the timescale seen here is certainly plausible . a large storm would not only change the brightness of the object , it should also have an effect on the photometric colors . table [ tab - pow ] presents the primary clean periods and their pnr for six variable and five possibly variable l dwarfs . the only periods considered significant are those with pnr values above 3 , 2mass 0746 + 20ab and 2mass 1300 + 19 . we have discussed why we do not believe these periods to be related to the rotation period of these objects . in addition to the objects mentioned above , our sample contains several other objects ( table [ tab - sv ] ) . these objects do not have significant periods , nor do they have any interesting features in their light curves . the light curves for all objects will be found in @xcite . we have discussed the existence of statistically significant photometric variations in seven of eighteen l dwarfs and possible variations in another five . these variable objects cover the entire span of spectral types present in our sample , from l0 to l4.5 . their data were sampled with a minimum timescale of @xmath35 minutes , intra - night sampling @xmath31 - 2 hours , and maximum baselines between 26 and 230 days . most of these variables show no significant periodicity . whatever the source of these variations , it must be present in a variety of spectral types and , hence , effective temperatures . two properties are commonly mentioned in the literature in which l dwarfs can be unusual : h@xmath10 and variability . these may or may not be related since some of the variable l dwarfs have h@xmath10 present in their spectra and others do not . we now discuss possible sources for the variations and their likelihood of being the present in these objects . many l dwarfs exhibit h@xmath10 emission @xcite , which is known to be an indicator of high chromospheric temperatures and magnetic activity in earlier type stars @xcite . if the variability in l dwarfs is caused by magnetic spots , then it is plausible to expect a correlation between h@xmath10 emission and the variable objects . in agreement with the conclusions of @xcite and @xcite , figure [ hafig ] shows no correlation between h@xmath10 emission ( i.e. magnetic activity ) and variability for the sample of l dwarfs from @xcite and this study . this result could imply that either magnetic activity is not the source of the variation or that the h@xmath10 emission is not magnetic in origin for these objects . as with earlier - type stars , wave heating has been suggested as the mechanism responsible for producing hot brown dwarf upper atmospheres @xcite . convection - forced waves propagate and grow as they rise through the upper atmosphere , eventually releasing their energy and heating the gas as they dissipate . the resulting high temperatures combined with magnetic field effects are possibly responsible for the h@xmath10 emission . l dwarfs with weak magnetic fields should exhibit little or no h@xmath10 emission . in light of this it is quite notable that the fraction of objects with h@xmath10 emission peaks at spectral type m7 and decreases at earlier and later spectral types @xcite . no l dwarfs later than l5 show h@xmath10 in emission @xcite , although one t dwarf does @xcite . since the early l population consists of both young brown dwarfs and old stars , this trend could indicate an inability of either substellar objects or cool stars to produce magnetic fields appropriate to maintain the h@xmath10 emission @xcite . using kinematics as a probe for age , @xcite have argued that the substellar late - type dwarfs show less h@xmath10 emission than stellar dwarfs of the same effective temperature . this might imply that the process by which h@xmath10 emission is produced is driven by the mass of the object and not the effective temperature . indeed , the dissipation of acoustic waves in the upper atmosphere could be responsible for the heating of the chromosphere . unfortunately , this process is poorly characterized at the masses and effective temperatures of interest here . in addition to h@xmath10 emission , radio emission can also be a signature of a magnetic field . radio flares as well as quiescent emission have recently been reported for the m - dwarfs lp944@xmath220 @xcite and bri0021 @xcite , and the l3.5 dwarf 2mass 0036 + 18 @xcite , an object in our sample . these authors infer that the radio emission is caused by synchrotron emission and estimate field strengths of @xmath35 g for lp944@xmath220 , 5 - 50 g for bri0021 and 20 - 350 g for 2mass 0036 + 18 . since there are essentially no methods to accurately determine the magnetic field strengths of objects other than the sun @xcite , these values are based upon models and assumptions that might not be correct . regardless of these assumptions , however , @xcite surmises that this data imply a substantial , non - neutral corona in this l dwarf , although the mechanics to create such a corona are poorly understood . if the magnetic field strengths are correct for lp944@xmath220 , bri0021 , and 2mass 0036 + 18 , they are apparently much less than the field strengths of active m dwarfs @xcite . in addition , the substellar nature , old age @xcite , and rapid rotation @xcite of lp944@xmath220 support a weak field strength ; many l dwarfs with spectroscopically determined rotation velocities are rotating quite rapidly and lack significant h@xmath10 emission @xcite , suggesting that the magnetic fields of these presumably old objects are too weak to slow down the rotation . the rotation velocity of 2mass 0036 + 18 is unknown and it has no measurable h@xmath10 emission . the existence of a magnetic field in lp944@xmath220 is also supported by the observation of an x - ray flare @xcite . an x - ray flare in an old , non - accreting object such as this can only be caused by magnetic activity . however , the lack of quiescent x - ray emission suggests that the magnetic field is quite weak . rutledge et al . postulate that because of the rapid rotation in this object , either the turbulent dynamo is being suppressed or the magnetic field is being configured into a more organized form . they also suggest that the lack of ionization in the cool photosphere prevents the magnetic field from coupling with the gas in the atmosphere , causing the field to dissipate . @xcite arrive at the same conclusion with their study of the x - ray flare of the m8 dwarf vb 10 . they go so far as to estimate the ionization fraction in its atmosphere , by extrapolating the ionization fractions in the atmospheres of early dwarfs down to the effective temperature of vb 10 . they estimate that the ionization fraction in late m dwarfs is 2 orders of magnitude lower than in early m dwarfs and 3 orders of magnitude smaller than in the sun , and conclude that magnetic footprints ( i.e. spots ) are unlikely to exist in the photosphere . the cooler m9 dwarf lp944@xmath220 is even less likely to have magnetic spots than vb 10 , supporting the conclusion by @xcite that they were detecting the signature of clouds . this , in turn , implies that magnetically produced spots are unlikely to be found in l dwarfs . however , it is still useful to examine what conditions are needed for the formation of magnetic spots and to determine if these conditions exist in l dwarfs . magnetic spots in the sun ( i.e. sunspots ) are thought to form as magnetic flux tubes rise to the photosphere @xcite . for magnetic buoyancy to be important , the plasma must be a sufficiently good conductor . this criterion is most likely not satisfied in cool l - dwarf atmospheres , especially in the low pressure regions where the temperatures are also low . in these regions the free electron abundance is small , @xmath52 @xmath53 around 1 bar for a 2000 k model . while this density is only an order of magnitude less than coronal densities estimated by @xcite , it is a factor of 10@xmath54 smaller than the densities of neutral species ( e.g. h@xmath55 , and he ) at this atmospheric pressure . the corona is likely a low - density region populated primarily by free electrons , whereas the atmosphere at a pressure of 1 bar is largely neutral . therefore , these two regions should have substantially different electrical conductivities . the small fraction of free electrons relative to neutral species in the atmosphere suggests that this region should not be a good conductor , nor should it be able to support the formation of any magnetic spots . nonetheless , it is instructive to explore the possibility of magnetic spots in more detail . to estimate the strength of coupling between the gas and the magnetic field we compute the magnetic reynolds number @xmath56 for l - dwarf atmospheres . @xmath56 is a dimensionless parameter describing how efficiently a gas interacts with a magnetic field ; @xmath57 @xcite , where @xmath58 is a length scale , @xmath59 is a velocity scale , and @xmath60 is the magnetic diffusivity of the gas . when @xmath61 1 , the magnetic field slips through the gas with no interaction ; for @xmath62 1 , the magnetic field is frozen in the gas . to compute @xmath56 we rely on atmosphere models computed by @xcite for cloudy l dwarfs . the models employ the cloud model of @xcite with the best - fitting sedimentation parameter @xmath63 . the l - dwarf @xmath64 range is still uncertain , but likely lies between about 2200 and 1300 k. we consider models with @xmath64 of 2000 to 1200 k and surface gravity of @xmath65 cm s@xmath66 , appropriate for a @xmath67 jupiter mass brown dwarf . the appropriate length scale @xmath58 to use in the calculation of @xmath56 is not obvious . in the area around sunspots , @xmath58 is usually taken as the size of the sunspot , a small fraction of the solar radius . we choose to set @xmath68 , the pressure scale height . typical values of @xmath69 are around 10@xmath70 cm at the 1 bar level . for the velocity scale we use that predicted by mixing length theory . to place a conservative upper limit on @xmath59 and hence @xmath56 , we assume that the entire thermal flux of the l dwarf is carried by convection . this is reasonably accurate below the photosphere , but overestimates the velocity scale , and @xmath56 , above . typical convective velocities are computed to be 10@xmath71 - 10@xmath72 cm s@xmath48 . the value for @xmath60 is computed from @xcite ; @xmath73 where @xmath74 is a factor to account for the partial ionization of the plasma , @xmath75 is temperature , @xmath76 is the number density of neutral atoms and molecules , @xmath77 is the number density of electrons , and @xmath78 is the coulomb logarithm @xcite . we use the abundance tables of @xcite to calculate @xmath77 and @xmath76 . at a pressure of 1 bar @xmath76 is @xmath79 @xmath53 for all models , whereas values of @xmath77 at this level are @xmath80 @xmath53 for the 1200 k model and @xmath52 @xmath53 for the 2000 k model . equation [ eq : coulomb ] provides a useful relation for computing the coulomb logarithm at temperatures below 10@xmath70 k. this relation might not be valid in the atmospheres of l dwarfs , where the temperatures of interest here do not exceed 4100 k. for a given model , @xmath81 as calculated above varies between 5 and 30 throughout the atmosphere . in order to test the validity of these results , we computed @xmath56 with @xmath81 at the unrealistic values of 1 and 100 . we find that @xmath56 computed at these limits differs by less than 1 part in @xmath82 for our coolest model and @xmath13 for our warmest model and stress that for the region of parameter space covered by our calculation the precise value of @xmath81 is essentially irrelevant . nonetheless , for the results considered here , @xmath81 is computed with equation [ eq : coulomb ] . the value of @xmath56 as a function of pressure in the model atmospheres is shown in figure [ fig - rey ] . @xmath56 is very small throughout the entire upper atmosphere ; only at pressures of @xmath83 and higher does @xmath56 start to approach 1 . by comparison , @xmath56 near sunspots is estimated to be 10@xmath72 - 10@xmath70 @xcite . open circles denote the approximate base of the photosphere ( where @xmath84 ) for the models shown . note that in contrast to the sun , @xmath56 only approaches unity well below the photosphere . at @xmath56=1 , the plasma and the magnetic field will interact with each other only to a small degree . any weak magnetic disturbances deep in these atmospheres are unlikely to affect the surface thermal flux since the winds and weather patterns alluded to earlier will redistribute the upwelling thermal flux before it is radiated . thus , throughout the atmospheres of l dwarfs , spanning the range from roughly l2 to l8 @xcite we expect little or no interaction between the visible atmosphere and the magnetic field . as mentioned above , the choice of @xmath58 and @xmath59 are not obvious . it is easy to imagine the use of other values , both larger and smaller , for these parameters . larger values would increase @xmath56 and smaller values would , of course , decrease it . nonetheless , the values used here are reasonable and meant only to estimate @xmath56 . it is also important to note that we have made no assumptions regarding how magnetic fields in l dwarfs are made or sustained . despite our comparisons to the sun , there is no reason to expect a magnetic field in an l dwarf to be created in the same method as the field in the sun . indeed , the origin of the magnetic field is irrelevant to the calculation above , since it is the characteristics of the gas that regulate the degree of coupling . the thermal fluxes emerging from l - dwarf atmospheres are affected by clouds . early l dwarfs have relatively thin clouds high in the atmosphere @xcite . figure [ fig - rey ] illustrates the trend for the clouds to form progressively deeper in the atmosphere at later spectral types . the base of the photosphere is marked for the models shown . unlike the case for @xmath56 approaching unity , the clouds form in the immediate vicinity of the photosphere and thus are well placed to affect the emitted thermal flux . for objects with @xmath85 ( approximately l2 and cooler ; * ) clouds play an important role in the emitted flux since the condensate opacity is significant . for an arbitrary l dwarf the emitted flux in some spectral regions will be limited by the cloud deck , while in others gaseous opacity reaches optical depth unity above the cloud . @xcite illustrate the effects of gaseous and condensate opacity for a variety of l dwarfs . if there is a transitory clearing in the cloud deck additional flux will emerge from those spectral regions in which the cloud opacity is otherwise dominant . examples are the peaks of flux emerging from the water band windows in _ z , j , h , _ and @xmath86 bands and the optical flux in @xmath0 band . the resulting bright spots on the objects will be similar to the ` 5-@xmath87 m hot spots ' of jupiter @xcite where flux emerges from holes in the ammonia cloud . an atmosphere with such non - uniformly distributed high contrast regions should be quite capable of producing photometric variations . as cloud optical thickness increases with later spectral type , l - dwarf @xmath88 color becomes redder , eventually saturating around 2 @xcite . clouds in cooler objects lie well below the photosphere , and leave the radiating region in the atmosphere relatively clear . the clear atmosphere partially manifests itself in the blue @xmath88 seen in the t dwarfs @xcite . models predict @xcite that a hypothetical l dwarf with no clouds will be substantially bluer ( @xmath89 mag ; * ? ? ? * ) at @xmath88 than a more realistic object with the same effective temperature and a cloudy atmosphere . thus , if the average l dwarf at a given spectral type is entirely covered with clouds , it would not likely be seen as a variable and it would have more typical @xmath88 color . in order for photometric variations to arise by the cloud mechanism there must be non - uniformity in the cloud coverage , such as clearings in the clouds . so , not only would clear sections of the atmosphere ( holes ) provide a source for brightness variations , flux emerging through such holes would cause the @xmath88 color to be somewhat bluer than the average object . the lack of a similar trend for the early l and late m dwarfs might indicate a different mechanism is at work in those atmospheres . of course variability caused by thicker clouds would result in variable objects being redder in @xmath88 than non - variable objects . time - resolved multicolor photometry should be sufficient to determine if variability is indeed connected to color . in the absence of such observations , we examine the previously published @xmath88 colors in search of systematic differences between variable and non - variable l dwarfs . figure [ jkfig ] shows that there is no distinct difference between the variable and non - variable targets . incidentally , plots using @xmath90 and @xmath91 also fail to reveal any significant trends @xcite . perhaps this indicates that the cloud features producing the variations are not as simple as a single , isolated spot . @xcite argue that the atmospheres of l dwarfs likely exist in one of two states , chaotic or banded . since clouds respond to atmospheric motions , they would presumably reflect one of these two morphologies . in general , schubert & zhang expect higher mass objects to have more chaotic and three - dimensional internal dynamics than lower mass objects , meaning that the higher mass objects are less likely to have banded cloud features . it is not obvious which cloud morphology would better produce photometric variations . for this reason , it is useful to speculate on what effects different cloud morphologies can have on an object s photometry . objects with more chaotic atmospheres , for example , might be more likely to have uniformly distributed clouds . when rotating , such objects might show little photometric variation , unless chaotic motion produces rapid evolution of the clouds . if the chaotic atmospheres result in fairly complete cloud coverage , then we might expect that more massive l dwarfs would be redder in @xmath88 and tend not to be variable . conversely , if chaotic atmospheres more often produce large clearings in the clouds , then they may more easily produce photometric signatures than banded atmospheres . in this scenario , the more massive l dwarfs would be bluer in @xmath88 and tend to be variable . furthermore , rapidly evolving , chaotic atmospheres may be responsible for the changes in photometric period observed for some objects . it is easy to imagine similar scenarios for banded clouds . the lack of any trends with color for the few l dwarfs observed for variability could indicate that a variety of cloud morphologies are present in these objects . clearly , more observations and modeling are required to better characterize atmospheric circulation and weather in l - dwarf atmospheres . the presence of clouds can explain the different types of variability seen in l dwarfs : non - variable , non - periodic variable , and periodic variable . the non - variable l dwarfs could be those objects whose atmospheres are either completely covered with clouds or whose clouds are uniformly distributed in spots and bands . such cloud morphologies result in small photometric variations below our detection limit . for an object to be classified as a periodic variable it must have some large feature that produces photometric variations and is stable , both temporally and spatially , over the entire span of the observations . none of the variable l dwarfs exhibit any periodic variations that lasted the entire program , suggesting that these conditions are not satisfied simultaneously in these objects . for example , a large storm could migrate latitudinally or dissipate and reform at a different latitude as does the great dark spot in the atmosphere of neptune @xcite . if wind speed is a function of latitude as on all the giant planets of our solar system , then spots at different latitudes will circle the object with different periods . if the migrations or dissipations occur on timescales much shorter than the baselines of our observations , then we would not be able to find any periodic signal . on the other hand , if the evolutionary timescales are a significant fraction of the baseline , then a prominent feature should produce a temporary signal , though not necessarily periodic . only two objects possibly fall under the latter category , 2mass 0746 + 20ab and 2mass 1300 + 19 . they show features changing on timescales about 10 days and longer . the shortest baseline for the other variable objects is around 25 days . any evolution of clouds in their atmospheres must occur on timescales shorter than this . indeed , @xcite conclude that features they observe must evolve on timescales shorter than their maximum observing duration of around 5 days . if an l dwarf has several large features at different latitudes , the photometric variations produced could be non - periodic . as mentioned above , an l dwarf is possibly a differentially rotating object . features located at different latitudes may respond to different wind speeds , resulting in different rotation periods around the object . the resulting light curve would be a complex composite of several periodic signals with different amplitudes , periods , and phases . photometric `` noise '' from smaller cloud features would degrade the periodic signals further , essentially making them undetectable . since most of the light curves for the objects studied here have data randomly scattered about a mean value , it is likely that the atmospheres for these objects are non - uniformly covered with clouds and do not have any single dominant feature . we have conducted a photometric monitoring program of eighteen l dwarfs in the @xmath6 photometric filter . we find that seven of these eighteen display statistically significant variations above the 95.4% probability level and with rms scatter between 0.010 and 0.083 mag . an additional five objects have probabilities for variability @xmath180% , suggesting that these objects are possibly variable . the remaining targets have probabilities @xmath9235% , indicating that they are non - variable or have amplitudes below our detectability . only three variable l dwarfs have prominent periods in their clean power spectra that have power @xmath393 times the noise . the 24.1-hour period of 2mass 0345 + 25 is not intrinsic to the l dwarf and needs more investigation . 2mass 1300 + 19 has a 9-day section of its light curve that could be evidence for the creation and dissipation of a large storm . the source of the 31.0-hour period of 2mass 0746 + 20ab is not clear from the light curve . however , it does have a 70-day section with a puzzling saw - tooth pattern with properties unlike any that would be expected . all of these timescales are much longer than the rotation periods expected for these objects . the light curves of the other variable objects are quite random and have no dominant features . non - uniform cloud coverage with features evolving on timescales less than a few days is likely the source of the variations in these objects . indeed , we have shown that the low ionization fraction predicted by l - dwarf models and the accompanyingly low magnetic reynolds numbers strongly argue against magnetic spots as a plausible cause for the photometric variations . on the other hand silicate and iron grains condense in l - dwarf atmospheres within the photosphere . these clouds are likely responsible for the photometric variations discovered in the various studies , particularly for the later l dwarfs ( about l2 and later ) . since the thermal emission of t dwarfs is also influenced by clouds @xcite we predict that variability will also be found in the opacity window regions of these objects . further work with models and more observations are required to better understand cloud composition and dynamics . the authors wish to thank d.h . roberts , j. peterson , and b. goldman for various contributions and discussions . we also wish to acknowledge the useful comments of the referee . c.g . and m.m . acknowledge support from nasa grants nag2 - 6007 and nag5 - 8919 and nsf grants ast-9624878 and ast-0086288 . work by k.l . supported by nsf grant ast-0086487 . ackerman , a. s. , & marley , m. s. 2001 , , 556 , 872 allard , f. , hauschildt , p. h. , baraffe , i. , & chabrier , g. 1996 , , 465 , l123 allard , f. , hauschildt , p. h. , alexander , d. r. , tamanai , a. , & schweitzer , a. 2001 , , 556 , 357 bailer - jones , c. a. l. & mundt , r. 1999 , a&a , 348 , 800 bailer - jones , c. a. l. & mundt , r. 2001 , a&a , 367 , 218 bailer - jones , c. a. l. & mundt , r. 2001 , a&a , 374 , 1071 basri , g. , et al . 2000 , , 538 , 363 beebe , r. f. , barnet , c. , sada , p. v. , & murrell , a. s. , icarus , 95 , 163 berger , e. 2002 , in press ( astro - ph/0111317 ) berger , e. , et al . 2001 , , 410 , 338 bondar , n. i. 1995 , a&as , 111 , 259 burgasser , a. j. , kirkpatrick , j. d. , reid , i. n. , liebert , j. , gizis , j. e. , & brown , m. e. 2000 , , 120 , 1100 burgasser et al . 2002 , apjl , in press ( astro - ph/0205051 ) burrows , a. , et al . 1997 , , 491 , 856 burrows , a. , & sharp , c. m. 1999 , , 512 , 843 clarke , f. j. , tinney , c. g. , & covey , k. r. , , 332 , 361 fleming , t. a. , giampapa , m. s. , & schmitt , j. h. m. m. 2000 , , 533 , 372 gelino , c. r. 2002 , ph.d . dissertation , in preparation gelino , c. r. , & marley , m. s. 2000 , in asp conf . 212 , from giant planet to cool stars , ed . c. a. griffith & m. s. marley ( san francisco : asp ) , 322 gelino , c. r. , marley , m. s. , & holtzman , j. a. 2001 , to appear in the proceedings of the 12th cambridge workshop on cool stars , stellar systems , and the sun gizis , j. e. , monet , d. g. , reid , i. n. , kirkpatrick , j. d. , liebert , j. , & williams , r. j. 2000 , , 120 , 1085 haisch , b. , strong , k. t. , & rodono , m. 1991 , , 29 , 275 hammel , h. b. , & lockwood , g. w. 1997 , icarus , 129 , 466 hawley , s. l. , reid , i. n. , & gizis , j. e. 2000 , in asp conf . ser . 212 , from giant planets to cool stars , ed . c. a. griffith & m. s. marley ( san francisco : asp ) , 252 johns - krull , c. m. , & valenti , j. a. 1996 , , 459 , l95 kirkpatrick , j. d. , et al . 1999 , , 519 , 802 kirkpatrick , j. d. , et al . 2000 , , 120 , 447 kirkpatrick , j. d. , dahn , c. c. , monet , d. g. , reid , i. n. , gizis , j. e. , liebert , j. , & burgasser , a. j. 2001 , , 121 , 3235 koerner , d. w. , kirkpatrick , j. d. , mcelwain , m. w. , & bonaventura , n. r. 1999 , , 526 , l25 lodders , k. 1999 , , 519 , 793 marley , m. s. 2000 , in asp conf . 212 , from giant planets to cool stars , ed . c. a. griffith & m. s. marley ( san francisco : asp ) , 152 marley , m. s. , & ackerman , a. s. 2001 , proceedings of iau symposium 202 , `` planetary systems in the universe : observation , formation and evolution '' in press ( astro - ph/0103269 ) marley , m. s. , saumon , d. , guillot , t. , freedman , r. s. , hubbard , w. b. , burrows , a. , & lunine , j. i. 1996 , science , 272 , 1919 marley , m. s. , seager , s. , saumon , d. , lodders , k. , ackerman , a. s. , freedman , r. , & fan , x. 2002 , , 568 , 335 martn , e. l. , delfosse , x. , basri , g. , goldman , b. , forveille , t. , & zapatero osorio , m. r. 1999 , aj , 118 , 2466 martn , e. l. , zapatero osorio , m. r. , & lehto , h. j. 2001 , , 557 , 822 monet , d. et al . 1998 , usno - sa2.0 , ( u.s . naval observatory , washington dc ) parker , e. n. 1955 , , 121 , 491 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 1992 . numerical recipes in fortran 77 : the art of scientific computing ( 2d ed ; cambridge : cambridge university press ) priest , e. r. 1982 , solar magnetohydrodynamics ( dordrecht , holland : d. reidel publishing ) reid , i. n. , et al . 1999 , , 521 , 613 reid , i. n. , gizis , j. e. , kirkpatrick , j. d. , & koerner , d. w. 2001 , , 121 , 489 reid , i. n. , kirkpatrick , j. d. , gizis , j. e. , dahn , c. c. , monet , d. g. , williams , r. j. , libert , j. , & burgasser , a. j. 2000 , , 119 , 369 roberts , d. h. , lehr , j. , & dreher , j. w. 1987 , , 93 , 968 rutledge , r. e. , basri , g. , martn , e. l. , & bildsten , l. 2000 , , 538 , l141 saar , s. h. 1994 , in iau symp . 154 , infrared solar physics , ed . d. m. rabin et al . ( dordrecht : kluwer ) , 493 schubert , g. , & zhang , k. 2000 , in asp conf . ser . 212 , from giant planets to cool stars , ed . c. a. griffith & m. s. marley ( san francisco : asp ) , 210 somov , b. v. 1992 , physical processes in solar flares ( dordrecht , holland : kluwer academic publishers ) stephens , d. , marley , m. , noll , k. , & chanover , n. 2001 , , 556 , l97 tinney , c. g. 1998 , , 296 , l42 tinney , c. g. , & reid , i. n. 1998 , , 301 , 1031 tinney , c. g. , & tolley , a. j. 1999 , , 304 , 119 tsuji , t. 2001 , proceedings of the iau workshop on `` ultracool dwarfs : surveys , properties and spectral classification '' ed . h. r. a. jones & i. steele , in press tsuji , t. , ohnaka , k. , aoki , w. , & nakajima , t. 1996 , a&a , 308 , l29 westphal , j. a. , matthews , k. , & terrile , r. j. 1974 , , 188 , l111 yelle , r. v. 2000 , in asp conf . 212 , from giant planets to cool stars , ed . c. a. griffith & m. s. marley ( san francisco : asp ) , 267 2mass 0015 + 35 & l2 & 00:15:44.7 & + 35:16:03 & @xmath93 & y & 1 + 2mass 0036 + 18 & l3.5 & 00:36:15.9 & + 18:21:10 & 16.10 & & 1,2 + 2mass 0058@xmath206 & l0 & 00:58:42.5 & @xmath206:51:23 & @xmath94 & y & 1 + 2mass 0135 + 12 & l1.5 & 01:35:35.8 & + 12:05:22 & @xmath95 & y & 1 + 2mass 0345 + 25 & l0 & 03:45:43.2 & + 25:40:23 & 16.98 & & 3 + 2mass 0746 + 20ab & l0.5 & 07:46:42.5 & + 20:00:32 & 15.11 & y & 1,2 + 2mass 1029 + 16 & l2.5 & 10:29:21.6 & + 16:26:52 & @xmath96 & y & 1 + 2mass 1108 + 68 & l1 & 11:08:30.7 & + 68:30:17 & @xmath97 & y & 4 + 2mass 1146 + 22ab & l3 & 11:46:34.5 & + 22:30:53 & 17.62 & & 3 + 2mass 1300 + 19 & l1 & 13:00:42.5 & + 19:12:35 & @xmath98 & & 4 + 2mass 1338 + 41 & l2.5 & 13:38:26.1 & + 41:40:34 & @xmath96 & & 1 + 2mass 1411 + 39 & l1.5 & 14:11:17.5 & + 39:36:36 & @xmath99 & & 1 + 2mass 1412 + 16 & l0.5 & 14:12:24.4 & + 16:33:12 & @xmath100 & y & 1 + 2mass 1439 + 19 & l1 & 14:39:28.4 & + 19:29:15 & 16.02 & & 2,3 + 2mass 1506 + 13 & l3 & 15:06:54.4 & + 13:21:06 & @xmath101 & y & 4 + 2mass 1615 + 35 & l3 & 16:15:44.1 & + 35:59:00 & @xmath102 & & 1 + 2mass 1658 + 70 & l1 & 16:58:03.7 & + 70:27:01 & @xmath103 & & 4 + 2mass 2224@xmath201 & l4.5 & 22:24:43.8 & @xmath201:58:52 & @xmath104 & y & 1 2mass 0345 + 25 & 35/37 & 209/318 & 108.73 & 7/7 & @xmath3999.99 & 0.030 & 0.027 + 2mass 0746 + 20ab&38/45 & 260/421 & 150.88 & 10/12 & @xmath3999.99 & 0.010 & 0.007 + 2mass 1300 + 19 & 37/39 & 275/384 & 137.78 & 5/7 & @xmath3999.99 & 0.015 & 0.012 + 2mass 2224@xmath201 & 5/6 & 15/21 & 26.01 & 5/7 & 99.88 & 0.083 & 0.057 + 2mass 1108 + 68 & 46/52 & 394/536 & 182.80 & 5/8 & 99.75 & 0.016 & 0.016 + 2mass 0135 + 12 & 18/22 & 110/187 & 30.00 & 5/6 & 99.33 & 0.041 & 0.035 + 2mass 1658 + 70 & 10/13 & 27/38 & 31.99 & 9/14 & 97.31 & 0.024 & 0.019 + 2mass 1615 + 35 & 12/15 & 43/61 & 40.92 & 6/6 & 82.69 & 0.073 & 0.067 + 2mass 0015 + 35 & 5/9 & 21/34 & 230.31 & 6/6 & 80.82 & 0.026 & 0.024 + 2mass 0058@xmath206 & 3/3 & 11/12 & 26.95 & 5/5 & 80.23 & 0.034 & 0.028 + 2mass 1439 + 19 & 11/11 & 28/30 & 38.90 & 4/4 & 79.81 & 0.014 & 0.013 + 2mass 1338 + 41 & 27/29 & 112/140 & 123.86 & 8/9 & 77.10 & 0.039 & 0.035 + 2mass 1029 + 16 & 5/5 & 14/18 & 32.96 & 6/6 & 30.39 & 0.057 & 0.063 + 2mass 0036 + 18 & 10/18 & 36/85 & 53.00 & 6/8 & 26.66 & 0.009 & 0.010 + 2mass 1506 + 13 & 3/4 & 9/12 & 35.91 & 5/6 & 22.46 & 0.024 & 0.030 + 2mass 1412 + 16 & 8/11 & 27/39 & 84.92 & 4/4 & 11.40 & 0.018 & 0.025 + 2mass 1146 + 22ab & 10/16 & 28/48 & 36.90 & 5/5 & 0.54 & 0.034 & 0.057 + 2mass 1411 + 39 & 24/30 & 78/132 & 131.86 & 4/5 & 0.12 & 0.047 & 0.056 2mass 0746 + 20ab & 31.0@xmath460.1 & 4.85 + 2mass 1300 + 19 & 238.@xmath469 . & 3.38 + 2mass 1108 + 68 & 23.8@xmath460.1 & 2.55 + 2mass 1658 + 70 & 1.89@xmath460.01 & 1.64 + 2mass 0135 + 12 & 18.6@xmath460.2 & 1.48 + 2mass 2224@xmath201 & 21.8@xmath460.4 & 0.71 + 2mass 1439 + 19 & 2.60@xmath460.01 & 1.85 + 2mass 1615 + 35 & 1.01@xmath460.01 & 1.24 + 2mass 1338 + 41 & 6.68@xmath460.01 & 1.06 + 2mass 0015 + 35 & 9.12@xmath460.01 & 0.85 + 2mass 0058@xmath206 & 2.34@xmath460.10 & 0.74

a numerical method for particle - laden fluids interacting with a deformable solid domain and mobile rigid parts is proposed and implemented in a full engineering system . the fluid domain is modeled with a lattice boltzmann representation , the particles and rigid parts are modeled with a discrete element representation , and the deformable solid domain is modeled using a lagrangian mesh . the main issue of this work , since separately each of these methods is a mature tool , is to develop coupling and model - reduction approaches in order to efficiently simulate coupled problems of this nature , as occur in various geological and engineering applications . the lattice boltzmann method incorporates a large - eddy simulation technique using the smagorinsky turbulence model . the discrete element method incorporates spherical and polyhedral particles for stiff contact interactions . a neo - hookean hyperelastic model is used for the deformable solid . we provide a detailed description of how to couple the three solvers within a unified algorithm . the technique we propose for rubber modeling / coupling exploits a simplification that prevents having to solve a finite - element problem each time step . we also develop a technique to reduce the domain size of the full system by replacing certain zones with quasi - analytic solutions , which act as effective boundary conditions for the lattice boltzmann method . the major ingredients of the routine are are separately validated . to demonstrate the coupled method in full , we simulate slurry flows in two kinds of piston - valve geometries . the dynamics of the valve and slurry are studied and reported over a large range of input parameters . example.eps gsave newpath 20 20 moveto 20 220 lineto 220 220 lineto 220 20 lineto closepath 2 setlinewidth gsave .4 setgray fill grestore stroke grestore

1703.05158

for systems that involve grains , fluids , and deformable solids , a key challenge is to determine reasonable methodologies to couple very distinct numerical techniques . on their own , systems of dry grains are commonly simulated using the discrete element method ( dem ) , wherein each grain s position is evolved by newton s laws applied by way of contact interactions with other grains . for fluids , a variety of approaches exist including finite volume methods , finite difference methods , and the lattice boltzman method ( lbm ) , which are based on updating fluid data on an eulerian background set . while the former two methods directly simulate navier - stokes , the latter utilizes a lattice discretization of the boltzmann equation , which approaches navier - stokes under the proper refinement . as for solids , in the large deformation limit , finite - element methods are commonly used , typically based on a moving lagrangian node set . systems that mix particles , fluids , and deformable solids require development of methods that allow proper momentum exchange between these disparate representations , which can be computationally quite complex if not reduced . however , because the particles can enter a dense - packed state , we do not wish to reduce the particle - fluid mixture to a simplified dilute - suspension continuum model . the purpose of this paper is three - fold : 1 . we introduce a reduced - order method that permits continuum deformable solid models , represented with finite - elements , to interact with both grains and fluid in a dynamic environment . the fluid - particle implementation is based on a joint lbm - dem method similar to those used in @xcite . lbm is well - suited to this problem because of its ease dealing with many moving boundaries . the solid interaction method introduced uses data interpolation to map deformed solid configurations from separate , individual solid deformation tests to the in - situ solid as it interacts with particles and fluid . because of the inherent complexity in multi - material modeling , the ability to remove the need to simulate large zones of the computational domain can be advantageous , as long as the macro - physics in those zones can be properly represented otherwise . herein , we introduce an lbm sub - routine that allows us to remove a large zone of the computational fluid domain and replace it with a global analytical form , which handshakes back to the lbm simulation domain appropriately . + 3 . as a key example of where these methods may come to use , we demonstrate their usage in two different piston - valve geometries . in both piston - valve geometries , a large piston pushes a particle - laden fluid through a passive valve . the valve is spring - loaded , and opens when the slurry pressure beneath is large enough . the deformable solid aspect comes into play because the valve has a rubber component along its bottom , which is intended to make a seal with the valve seat . we conduct a systematic parameter study of valve behavior under variations in particle size , input packing fraction , and polydispersity as well as variations in fluid viscosity and piston speed . we consider two types of valve setups : ( 1 ) a ` pressure valve ' , in which the valve separates a zone of pressurized slurry above it from a zone of low pressure below it . slurry pushed through the valve is hence pressurized as it passes through . ( 2 ) a ` safety valve ' , whose goal is to ensure the pressure in a flowing conduit does not exceed a critical limit . here , the valve is placed adjacent to a flowing conduit and remains closed unless the pressure is high enough . figure [ fig : opned_closed ] shows mid - simulation snapshots of both valve setups , showing particles , fluid stream lines , rubber deformation , and the geometry of the valve and frame . note that we exploit symmetry about the @xmath0-plane and simulate only half the domain . ) , deformable solid ( colored according to equivalent shear strain @xmath1 ) , valve ( dark gray ) , and frame ( light gray ) . a spring ( not shown ) applies a downward force on the valve . ] in testing our method , we provide numerical validations of the new techniques introduced . we also perform validations of the lbm approach in the simulated valve geometry . in analyzing valve simulation results , we provide physical commentary where possible to justify various observations . the discrete - element method ( dem ) is already a mature tool that is applied in conjunction with experiments both for a better understanding of the micromechanics of granular materials and as a means of @xmath2 experimentation when laboratory experiments are unavailable . in a similar vein , the inclusion of a fluid at the subgranular scale in dem simulations provides a powerful tool in the broad field of fluid - grain mixtures . obviously , the available computational power and research time restrict considerably the number of particles or the size of a physical system . in the case of dry granular materials , statistically representative samples are obtained and simulated with @xmath3 of particles in 2d @xcite . despite enhanced kinematic constraints , 2d simulations often lead to novel physical insights and realistic behaviors that can be easily generalized to 3d configurations . however , with fluid in the pore space , 2d simulations are much less reliable in the dense regime since the pore space is discontinuous with zero permeability . this two - dimensional flaw can be partially repaired by adding artificially a permeable layer on the particles . but only 3d simulations may account for a realistic behavior of particle - fluid mixtures with their natural permeability . moreover , complex geometries / boundaries relating to realistic engineering problems can not be fully captured in 2d simulations or symmetric 2d extensions ( e.g. axis - symmetry ) ; only 3d approaches can handle such problems in full generality . we developed a 3d fluid dynamics algorithm based on the lattice boltzmann method ( lbm ) . this algorithm was interfaced with a dem algorithm with a standard linear spring - dashpot - friction model of contact between particles . the combined lbm - dem method for particle - laden fluid is then further coupled to a deformable solid domain using finite elements to model a rubber - like behavior . the rubber coupling is intentionally simplified . within actual computer power , it is still a significant challenge to model the entirety of most engineering systems and problems . certain sub - scale details and complex interactions are unnecessary to capture the macroscale system response for a given loading . we utilize symmetric boundaries ( where possible ) and a variety of techniques to shrink the system size and average - up sub - scale phenomena . specifically in this work : to handle sub - scale behavior in the fluid we use a large - eddy - simulation ( les ) technique ( see sect . [ subsec : lbm ] ) , to mimic a large fluid domain outside the focus region we have created a technique we denote _ zoom - in with effective boundaries _ ( zieb ) ( see sect . [ subsec : zoominlinktech ] ) , and to reduce simulation time we developed a weak coupling to the rubber domain based on a _ neo - hookean _ model developed in abaqus . the last part is computed separately and only the result is imported into lbm - dem simulation ; the coupling and description of this part is expounded in sec . [ subsec : lbmdemrubbercoupling ] . the dem is based on the assumption of elastic solids with damping and frictional contact behavior @xcite . newton s equations of motion are integrated for all degrees of freedom with simple force laws expressing the normal and friction forces as explicit functions of the elastic deflection defined from the relative positions and displacements at contact points . we treat all quasi - rigid solids in the domain using this dem description , including grains , the valve , and solid system boundaries . correspondingly , all solid - on - solid contact forces ( e.g. grain on grain , grains on valve , grain on solid wall ) are obtained using dem contact laws . the valve and system walls are discretized as a kinematically constrained connected mesh of polyhedral solid ` particles ' . to simplify contact interactions , we assume linear elastic normal and tangential contact forces characterized by a normal stiffness @xmath4 and tangential stiffness @xmath5 . this is applied to all contact interactions , e.g. between spheres , polyhedra , or sphere - polyhedra , though the stiffnesses can vary depending on the two objects in contact . in additional to the elastic part , a dissipative part of the contact force is necessary @xcite . in our model , we use a linear visco - elastic law for normal damping and a linear visco - elasto - plastic law for tangential damping and friction forces where the plastic part uses a coulomb law . the visco - elastic law is modeled by a parallel spring - dashpot model . the contact normal force is defined as : @xmath6 where @xmath7 is the contact normal vector and @xmath8 is the relative velocity along the contact normal . @xmath9 represents a viscosity parameter with a value that depends on the normal restitution coefficient between grains . according to coulomb s law the friction force is given by : @xmath10 and @xmath11 where @xmath12 is the friction coefficient , @xmath13 is the tangential relative velocity , and @xmath14 is a viscosity parameter , which depends on the tangential restitution coefficient . the equations of motion ( both linear and angular momentum balance ) are integrated according to a velocity verlet scheme @xcite . the lbm is based on a material representation of fluids as consisting of particle distributions moving and colliding on a lattice @xcite . the partial distribution functions @xmath15 are introduced to represent the probability density of a fluid particle at the position @xmath16 with a velocity @xmath17 at time @xmath18 along discrete direction @xmath19 . the three components of @xmath20 are given in tab . [ tab : civalue ] . rrr rrr rrr rrr rrr rrr rr @xmath19 & 0 & 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 & 9 & 10 & 11 & 12 & 13 & 14 & 15 & 16 & 17 & 18 + @xmath21 & 0 & -1 & 0 & 0 & -1 & -1 & -1 & -1 & 0 & 0 & 1 & 0 & 0 & 1 & 1 & 1 & 1 & 0 & 0 + @xmath22 & 0 & 0 & -1 & 0 & -1 & 1 & 0 & 0 & -1 & -1 & 0 & 1 & 0 & 1 & -1 & 0 & 0 & 1 & 1 + @xmath23 & 0 & 0 & 0 & -1 & 0 & 0 & -1 & 1 & -1 & 1 & 0 & 0 & 1 & 0 & 0 & 1 & -1 & 1 & -1 + the lattice boltzmann method is often non - dimensionalized when applied to physical problems . the governing macroscopic equations are given in terms of lattice units : @xmath24 where @xmath25 is the lattice spacing , @xmath26 is the lattice speed with @xmath27 the time step , and @xmath28 is the fluid density at zero pressure . for the following , we will describe the method in lattice units . figure [ fig : schemalattice ] shows a cartesian grid where the meshing scheme d3q19 , corresponding to 18 space directions in 3d used in our simulations , is represented . in lbm , the scheme d3q19 is defined for each node where the distribution functions evolve according to a set of rules , which are constructed so as to ensure the conservation equations of mass , momentum and energy ( with dissipation ) , so as to recover the navier - stokes equations @xcite . this holds only when the wave lengths are small compared to the lattice spacing @xcite . at each lattice node , the fluid density @xmath29 and momentum density @xmath30 are defined as @xmath31 @xmath32 and the temperature is given by @xmath33 where @xmath34 is the number of space dimensions , @xmath35 is particle mass , and @xmath36 is the boltzmann constant . the equilibrium state is assumed to be governed by the maxwell distribution : @xmath37 \label{eqn : maxw}\ ] ] where @xmath38 is the mean velocity . by expanding ( eq . [ eqn : maxw ] ) to order 2 as a function of @xmath39 , which is the local mach number with @xmath40 being the lbm sound velocity , a discretized form of the maxwell distribution is obtained and used in the lbm : @xmath41 \label{eq : equilibre_maxwell_boltzmann}\ ] ] where the factor @xmath42 , @xmath43 and the rest of @xmath44 . @xmath45 depend on the scheme with the requirement of rotational invariance @xcite . the lbm sound speed is then given by @xmath46 . the velocities evolve according to the boltzmann equation . in its discretized form , it requires an explicit expression of the collision term . we used the bhatnagar - gross - krook ( bgk ) model in which the collision term for each direction @xmath19 is simply proportional to the distance from the maxwell distribution @xcite : @xmath47 where @xmath48 is a characteristic time . hence , for the d3q19 scheme , we have a system of @xmath49 discrete equations governing the distribution functions : @xmath50 these equations are solved in two steps . in the collision step , the variations of the distribution functions are calculated from the collisions : @xmath51 where the functions @xmath52 designate the post - collision functions . in the streaming step , the new distributions are advected in the directions of their propagation velocities : @xmath53 the above equations imply an equation of state @xcite : @xmath54 the kinematic viscosity is then given by @xcite @xmath55 \label{eq : viscositelbm}\ ] ] with the requirement @xmath56 . as discussed in @xcite , the lattice boltzmann method holds only when the pressure wave lengths are small compared to the lattice spacing unit . this imposes a limitation on mach number @xmath57 and therefore fluid speeds higher than the sound speed can not be simulated . in nature , for a given fluid we have : sound speed @xmath58 , density @xmath28 and viscosity @xmath59 . from eq . [ eq : collision_step ] , we need the relaxation time @xmath48 . this is related to @xmath60 , @xmath59 and @xmath25 by : @xmath61 equation . [ eq : tauphyslbm ] shows that since @xmath60 and @xmath59 are fixed from fluid properties , only @xmath25 can be used to ensure the stability of lbm , which becomes unstable when @xmath62 . numerically , there is a limitation in computer capability regarding the smallest value of @xmath25 . to handle this , a _ sub - grid turbulent model _ based on les with a smagorinsky turbulence model is used @xcite . the viscosity formulation is : @xmath63 = c_s^2\left[\tau + \tau_{_t } - \frac12\right]\ ] ] where @xmath64 is the sub - grid les viscosity and @xmath65 is the sub - grid les lbm relaxation time . the les viscosity is calculated from the filtered strain rate tensor @xmath66 and a filter length scale @xmath67 through the relation @xmath68 where @xmath69 is the smagorinsky constant . in lbm , @xmath70 is obtained from the second momentum @xcite @xmath71 as @xmath72 , where @xmath73 . from @xmath70 and @xmath74 , @xmath65 is expressed as : @xmath75\ ] ] where the filter length scale @xmath76 is the spatial lattice discretization . there exist different techniques to model fluid - structure interaction . the most used in cfd is stress integration , however , in lbm the preferred approach is based on momentum exchange @xcite . this approach is simple in lbm , since in lbm each node already contains information about the derivatives of the hydrodynamic variables in each distribution function @xmath77 @xcite . due to the presence of a solid boundary , after the collision step but before the streaming step , we know the population of @xmath77 except those which reflect off the solid wall as shown in fig . [ fig : fluideparticule_cb ] ( @xmath78 ) . in our simulations , we use the method developed by bouzidi @xcite , which mimics a macroscopic no - slip condition . for clarity , we describe the interaction rules using lattice units . this means that the time step is @xmath79 , the space discretization is @xmath79 , etc . out of the characteristic scales , we will denote , around the fluid - solid boundary , @xmath21 as position and the subscripts @xmath80 , @xmath81 , and @xmath82 respectively will indicate either the _ fluid _ or _ solid _ domain and the _ fluid - solid boundary_. we present the method in simplified 1d form . for more clarity in the way momenta are exchanged between fluid and solid domains , let us introduce @xmath83 . according to the lbm scheme ( collide and stream sect . [ subsec : lbm ] ) in the presence of a solid wall , we have two scenarios depending on the wall position @xmath84 ( see fig . [ fig : fluideparticule_cb ] ) : 1 . @xmath85 where fluid leaves node @xmath86 , reflects off the wall , and reaches @xmath86 again in time less than @xmath27 . 2 . @xmath87 where fluid leaves node @xmath86 , reflects off the wall , and reaches @xmath86 again in time greater than @xmath27 . to handle these scenarios , we introduce a fictitious node @xmath88(see fig . [ fig : fluideparticule_cb ] ) such that after streaming , a fluid particle leaving @xmath88 arrives at @xmath86 exactly in time @xmath27 . as shown in fig . [ fig : fluideparticule_cb ] , if @xmath86 is the last fluid node before we reach the solid boundary , @xmath89 should be a solid node . let @xmath90 be the distribution function such that @xmath90 is the opposite direction of @xmath19 where @xmath19 is the direction oriented from fluid node to solid boundary . by using a linear interpolation , @xmath90 is expressed as follow : @xmath91 where @xmath92 corresponds to the distribution function of @xmath77 after the collision step but before streaming and @xmath93 . the term @xmath94 is calculated from the boundary velocity and is zero if the boundary is stationary . @xmath94 is calculated by considering that the fluid velocity @xmath95 evolves linearly between @xmath96 and @xmath97 . if @xmath98 is the boundary velocity , @xmath95 is then defined by @xmath99 at first order in @xmath95 . the equilibrium value of @xmath77 is given by @xmath100 where @xmath101 is constant and depends on the lattice discretization scheme @xcite . using a linear interpolation , @xmath94 is given by : @xmath102 hydrodynamic forces acting on the structure are calculated by using momentum exchange @xcite . let @xmath103 and @xmath104 be fluid and solid momentum calculated near the boundary . the exchanged momentum is given by : @xmath105 @xmath103 and @xmath104 are calculated as follows : @xmath106 @xmath107 where @xmath34 is the space dimension , and @xmath108 and @xmath109 are respectively the fluid and the solid distribution functions . to be clear , @xmath109 is constructed at a lattice point occupied by solid by taking the solid velocity @xmath110 and density @xmath111 and assigning a maxwell equilibrium distribution , per eq . [ eq : equilibre_maxwell_boltzmann ] . the hydrodynamic force @xmath112 and torque @xmath113 are then given by : @xmath114 @xmath115 where @xmath116 is the distance between the center - of - mass of the solid domain and the boundary node @xmath84 . unlike the coupling between lbm and dem , the lbm - rubber and dem - rubber coupling is indirect . we focus our explanation below on the case of a rubber ring component of a valve , but the idea can be generalized to other cases . a first simulation is performed using abaqus from which the deformed rubber shape and reaction force of the valve seat on the rubber are saved for many states of rubber compression . this simulation uses no fluid or particles . the abaqus simulation consists of compressing the rubber ring geometry against the bare valve seat ( see inset of fig . [ fig : force_vs_displacement ] ) . the rubber is simulated as a nearly incompressible neo - hookean elastic solid with a strain energy @xmath117 ( per unit reference volume ) given by @xmath118 where @xmath119 is the first deviatoric strain invariant defined as @xmath120 , the deviatoric stretches are given by @xmath121 , @xmath122 is the total volume ratio , and @xmath123 are the principal stretches . the ( small - strain ) shear modulus is @xmath124 and bulk modulus is @xmath125 . a frictionless contact between rubber and valve seat is used for simplicity . figure [ fig : abq_deform_snapshot ] shows several snapshots during the abaqus simulation and fig . [ fig : force_vs_displacement ] gives the seat net - force on rubber as a function of ( downward ) displacement @xmath126 of the rubber ring , @xmath127 . we index each deformed rubber configuration by the value of @xmath126 it corresponds to . the abaqus tests are performed under quasi - static conditions but we also assume damping can exist such that upward the force on the rubber satisfies the relation @xmath128 . ] then , the data from the abaqus simulation is used in a weak coupling routine to describe rubber configurations when the ring takes part in a slurry simulation . in short , the method determines which of the deformed rubber configurations from the stand - alone abaqus tests is the best representation of the actual deformed rubber state at that moment in the slurry problem . hence , the utility of this method lies in the fact that the rubber deformation that occurs in the actual slurry problem largely resembles the modes of deformation seen in a purely solid compression experiment . situations where the rubber surface becomes heavily locally deformed could be problematic for this approach . from the lbm - dem point of view , the rubber is composed of tetrahedra ; this allows us to compute contact forces for dem and the exchanged momentum for lbm as if it were a simple collection of polyhedral objects . since the abaqus simulation is performed without fluid or particles , to use its solutions we need to deduce an effective upward force from lbm - dem acting on the bottom rubber surface , which can then be referenced against the abaqus data to infer a deformed rubber shape . the effective force is needed because the rubber in the abaqus simulations has contact only with the valve seat , whereas in the slurry case , there can be additional forces from fluid and particles extending to the lateral surfaces of the rubber . key to our routine is to identify two subsets of the exposed rubber surface , denoted surface a and surface b. surface a is the part that makes contact with the valve seat and surface b remains free in the abaqus simulations ( see left fig .[fig : abq_coupling_config ] ) . in particular , surface @xmath129 and @xmath130 are geometrically defined using the the last frame of the abaqus simulation where the rubber is fully compressed . in the slurry case , we add uniform hydrostatic stress to the observed rubber loading distribution until the mean normal stress acting on surface b vanishes . this leaves us with a loading state that resembles one from abaqus . because the rubber is essentially incompressible , changing the hydrostatic stress uniformly along the surface does not affect the deformed rubber configuration . to be specific , we compute the normal stress on surface @xmath129 using @xmath131 and on surface @xmath130 using @xmath132 where @xmath133 is the stress from hydrodynamic , particle , and valve - seat forces and @xmath134 is the normal vector on section @xmath135 . since the rubber deformation is caused by the shear part of the stress , we uniformly subtract the traction @xmath136 from all normal stresses acting on the rubber . this modified loading now resembles the abaqus loading ( inset fig . [ fig : force_vs_displacement ] ) in which only an upward force on surface @xmath129 exists . therefore , we define the effective upward force on surface @xmath129 as @xmath137 . the rubber shape is updated using a four step loop , which is performed after particle positions and fluid data are updated . the goal of the iteration routine is to implicitly solve eq . [ constit ] for @xmath126 so that the effective force from particles , fluid , and the valve seat on the final rubber state matches the force from eq . [ constit ] . * step 1 : compute effective upward force @xmath138 from fluid , particle , and valve - seat interactions on rubber given the current guess for @xmath126 * step 2 : use this force and eq . [ constit ] to update to a new guess for @xmath126 . * step 3 : check if new and old @xmath126 differ by less than a tolerance . if so break , if not update the rubber shape based on the new @xmath126 . * step 4 : update applied force on surfaces @xmath129 and @xmath130 according to the rubber shape given by the new guess for @xmath126 . we assume that fluid forces do not change throughout the iteration procedure . this is true by assuming a small incremental rubber shape change between iterates , so only particle and valve - seat forces on the rubber are updated during step 4 . in step 2 we utilize the following update rule @xmath139 \frac{\delta t}{\eta } + \delta}{1 + \frac{\nu}{\eta } } \label{eqn : rubber_eqn_2}\ ] ] where @xmath140 is the actual displacement of the rubber at the beginning of the time - step . the coefficient @xmath141 is numerically selected to aid convergence . note that if the updated value of @xmath126 matches the value inputted , then eq . [ constit ] is solved . the update rule above attempts to move @xmath126 toward such a solution with each pass through the iteration loop . we check convergence of @xmath126 ( step 3 ) using a tolerance that is scaled by the error of the first iterate , @xmath142 , where @xmath143 is the value obtained for @xmath126 after the first pass of the loop . we use simple linear interpolation to compute values of @xmath127 when @xmath126 is between two neighboring values from the abaqus output . and @xmath130 , and the initial and final shape of the rubber as it proceeds through the iteration loop to solve eq . [ constit ] , with a representation of particles in contact shown . the right figure gives a 1d analog of the model to compute the rubber deformation . ] the third step consists of updating the rubber shape using the new guess for @xmath126 obtained in step 2 . again , a linear interpolation is applied as is done for @xmath127 . for the shape , we export the abaqus mesh , which allows the interpolation . for example , if the new @xmath126 lies directly between two frames of the abaqus data , the rubber shape is updated by moving the nodes of the rubber to positions halfway between those of the neighboring frames . the fourth step consists of recomputing the contact force between the tetrahedra representing the rubber , and the particles / valve - seat . the contact force is computed using eq . [ eq : normal_force ] for the normal part and eq . [ eq : friction_force ] for the tangential part . one specificity here is that we do not update @xmath144 during the coupling loop routine unless when @xmath145 where @xmath146 . a schematic model of one update through the iteration loop is presented in fig . [ fig : abq_coupling_config ] ( left ) and a 1d mechanical model of the treatment of the rubber interaction is visualized in fig . [ fig : abq_coupling_config ] ( right ) . the numerical method described in previous sections is implemented using the algorithm displayed in fig . [ fig : algo ] . note that we compute the valve acceleration by enjoining the applied force and mass of the rubber and valve components together in the verlet update ; rubber deformations are not included in the calculation of net acceleration of the valve / rubber composite as they are small compared to the movement of the valve overall . as shown in fig . [ fig : algo ] , the lbm step is computed in parallel with dem force calculation and rubber force calculation . the _ zoom - in with effective boundaries _ ( zieb ) technique replaces a fluid reservoir / domain with an analytical solution that interacts dynamically with the remainder of the domain . the challenge is to determine the correct effective dynamics at the fictitious interface , and to transfer the analytical result to lbm distribution functions . in this study we model valves , which can be positioned far from the pump that leads the slurry to the valve . the goal is to avoid having to calculate flow in the expansive pump region . from a computational point of view , one might assume a simple input velocity boundary condition should solve the problem , however , for a compressible fluid , the imposed flow and pressure may depend on the total removed volume and feedback with the dynamics within the valve system . in this section , we first detail how to obtain the analytical solution then explain how to implement this solution as an lbm boundary condition . ) and the closest virtual lattice point ( at @xmath147 ) . , width=336 ] per fig . [ virtual ] , we assume the virtual ( i.e. removed ) domain is a cylinder and piston . the cylinder is initially full of fluid and has total volume @xmath148 . as the piston moves , fluid is pushed into the simulated domain . let the movement of the piston be given by some prescribed @xmath149 , where @xmath150 measures the piston displacement . the cross - sectional area of the piston ( and cylindrical domain ) is @xmath151 . the piston velocity , @xmath152 , is simply defined from the time - derivative of @xmath150 . define as @xmath153 the mean in - flowing fluid velocity component on the interface between the domains . let @xmath29 be the average density of fluid in the virtual cylinder between the piston head and the interface . further , we make the simplifying assumption that in the cylinder region the fluid density is in fact uniform , such that it is equal to @xmath29 throughout . conservation of fluid mass in the cylinder domain can be expressed by balancing the mass rate within the cylinder against the mass flux into the simulated domain : @xmath154 \label{eq : mass_conv}\ ] ] in a fully continuum framework , the above equation would need to be augmented with momentum balance in order to provide the in - flowing velocity , @xmath153 , at the fictitious interface . however , using the lbm description , we can update @xmath153 in another way , which is consistent with momentum balance on the small scale . at a given time @xmath155 , we assume @xmath156 is given in the cylinder domain and is equal to the density at @xmath157 , where , per fig . [ virtual ] , @xmath157 is the lattice point in the simulated domain that is adjacent to the interface with the virtual domain . we suppose the velocity at @xmath157 is the interfacial velocity @xmath158 . both density and velocity at time @xmath155 at @xmath157 are defined by eq . [ dens ] and [ momentum ] through distribution functions @xmath159 . the distribution functions are updated to @xmath160 under the following procedure , which is applied after the collision step but before streaming . first we update and store the density @xmath161 at @xmath157 using explicit integration of eq . [ eq : mass_conv ] : @xmath162 \label{eq : ziebdesity8}\ ] ] next , a partial lbm streaming step is performed at @xmath157 using the distributions at time @xmath155 . during this step @xmath157 streams to and from its neighboring ` real ' lattice points within the simulated domain . however , it only streams out of the interface with the virtual domain and does not receive any distributions from the virtual domain . define @xmath163 as the density at @xmath157 after this partial streaming step . the next step is to back - solve the needed distributions to be streamed in from the virtual domain in order to guarantee the final density at @xmath157 equals @xmath161 . for example , consider a setup as shown in fig . [ virtual ] and suppose the fictitious interface is normal to the @xmath164 direction ( see fig . [ fig : schemalattice ] ) . after the partial streaming step , updated ( though not finalized ) distribution values exist for all the @xmath77 except for the values associated to @xmath165 and @xmath166 . these five distribution values are all unknown after the partial streaming step . to compute them , first we modify only the value of the @xmath167 distribution , which is the distribution that streams into the simulated domain normal to the fictitious boundary : @xmath168 this advects all the missing density at @xmath169 from a fictitious node @xmath147 ( see bottom of fig . [ virtual ] ) . with these distributions , the velocity at time @xmath160 is computed at @xmath157 according to @xmath170 . because the distributions in the @xmath171 and @xmath166 directions are still unknown at this point , the maxwell equilibrium function eq . [ eq : equilibre_maxwell_boltzmann ] is then used to redistribute all the distributions at @xmath157 to a more natural , equilibrium state . this updates all the distributions to their final values , at @xmath160 . we notice here that the initial value of @xmath29 at the beginning of the simulation should be a normalized value ( in lattice units ) otherwise an additional step of normalizing by the physical fluid reference density will be necessary before using it in eq . [ eq : mass_conv ] and eq . [ eq : ziebdesity9 ] . we first test some of the individual components of the routine . in this section , we provide separate numerical validations of the zieb technique , the rubber deformation model , and the lbm . to validate the zieb method , we performed an analysis of fluid flow in a geometry comprised of a piston driving fluid passing through a narrow restriction . this flow field is then compared to that obtained using a `` virtual piston '' in which the domain containing the moving piston is removed and in its place an effective boundary condition from zieb is used , see fig . [ fig : zoomnlink_config ] . the real piston begins positioned such that the fluid volume between the piston and input section is @xmath172 ; the same @xmath148 is used in eq . [ eq : ziebdesity8 ] for the virtual piston . we use @xmath173 and @xmath174 . as input parameters , we use a ( pressure - free ) fluid density @xmath175 , dynamic viscosity @xmath176 and a smagorinsky constant of @xmath177 for the sub - grid turbulence model @xcite . figure [ fig : velocity_density_opened ] shows the comparison between the two simulations regarding fluid velocity and the normalized input fluid density computed in the same domain ( see fig . [ fig : zoomnlink_config ] ) . the agreement is strong , even the time - dependent fluctuations , confirming the correctness of the zieb method . the test of the rubber coupling and rubber deformation is performed running a loading / unloading test without fluid . a force ( loading / unloading ) @xmath178 is directly applied on the valve which presses the rubber into contact with twelve frozen spheres ( see fig . [ fig : rubberloadunloadtest ] ) . two phases are considered : a loading phase with @xmath179 then an unloading phase with @xmath180 . we use a frictionless contact type between the rubber and spheres where normal stiffness is set to @xmath181 and no normal damping is used to insure that all dissipation comes from internal rubber damping . the rubber coupling parameters ( eq . [ eqn : rubber_eqn_2 ] ) are set to @xmath182 and @xmath183 . the valve density is set to 7850 @xmath184 and the rubber density to @xmath185 . the time step is set to @xmath186 . [ fig : rubberloadunloadtest ] shows the loading / unloading force , the reaction force @xmath138 of spheres on the rubber , the displacement @xmath126 ( right inset ) and the corresponding force @xmath127 . the agreement of @xmath178 , @xmath187 , and @xmath138 after a relaxation time verifies the coupling . the last test is focused on verifying the fluid lbm simulation by comparing flow of fluid in the rubber channel of a pressure - valve assembly against an analytical flow solution ( recall fig . [ fig : opned_closed ] ) . we can run a simulation where the fluid viscosity is large , such that the flow in the channel will be in the stokes limit . to aid in calculating an analytical solution , we treat the flow as radially directed and assert the lubrication approximation . in view of fig . [ fig : opned_closed ] for the definition of the @xmath22 direction , we obtain the following system of equations , which includes momentum and mass balance under the lubrication limit : @xmath188 this is solved by @xmath189 where @xmath129 is an undetermined constant and @xmath190 where @xmath69 is a constant and @xmath129 is the same as in the velocity field equation . from the @xmath191 equation , the pressure difference @xmath192 between @xmath193 and outer at @xmath194 ( see fig . [ fig : channel_velocity_pressure ] ) is given by : @xmath195 . using @xmath196 , @xmath197 and @xmath198 , we find @xmath199 ( see fig . [ fig : channel_velocity_pressure ] left ) from our numerical data , giving @xmath200 . hence , the predicted pressure difference is @xmath201 which is quite close to the obtained pressure difference from the simulation ( @xmath202 , see fig . [ fig : channel_velocity_pressure ] right ) . [ fig : channel_velocity_pressure_vs_r ] shows the analytical solution as a function of @xmath203 in comparison with the numerical data and agreement is found . in this test , the fluid density is @xmath175 and dynamic viscosity is @xmath204 . the valve density is set to @xmath205 and the rubber density to @xmath185 . we use zieb on the input section ( see fig . [ fig : system_cofig ] ) with a virtual piston velocity of @xmath206 , and we apply a constant pressure of @xmath207 at the output section ( see fig . [ fig : system_cofig ] ) . we present numerical examples utilizing the valve geometries presented in fig . [ fig : system_cofig ] . the two systems are used to mimic a safety and pressure valve . the different parts and their corresponding names are presented in fig . [ fig : system_cofig ] . their dimensions are given in appendix [ app : dimension ] . the valve is spring - loaded from above . initially , the valve is closed due to force from the spring ( and prescribed overpressure above in the pressure - valve geometry ) . the spring s force - displacement relation is chosen to be non - linear and is expressed as follows : @xmath208 ; @xmath209 is the stiffness , @xmath210 is the preload spring displacement , @xmath126 is the load displacement and @xmath211 is the spring maximum compression . the fluid region above the valve begins unpressurized in the safety - valve case and pressurized by the selected @xmath212 in the pressure - valve case . for each simulation , both for safety and pressure valve , we start by introducing fluid through the input section beneath valve domain using the zieb technique at constant virtual piston velocity @xmath213 . when the beneath valve domain reaches a large enough pressure , it will overcome the spring - load ( and possible top pressure ) on the valve to open it . we continue displacing the virtual piston and then turn off @xmath213 when we assume that the flow has reached a steady state . we then wait until the valve is closed . we check the behavior of the valve systems with and without particles but the closure phenomena is only investigated for the case where we have particles . in the presence of particles , we start with a constant packing fraction in the domain beneath the valve corresponding to the imposed packing fraction @xmath214 at the input section . during the simulation , each particle that exits the simulated domain is removed and each new particle introduced at the input section starts with the fluid input velocity as its initial velocity . to control the input particle flow , we insert grains to ensure that the input particle packing fraction is constant within a width of @xmath215 from the interface with the virtual domain . we introduce / remove particles in the simulated system every 50 steps . physical parameters involved in the valve problem are displayed in tab . [ tab : summarizedparamvalvebh ] ) , which include the geometry of the valve system ( e.g. safety - valve or pressure - valve ) , each of which has fixed size dimensions ( see appendix ) . for all tests , we fix the solid density of particles @xmath216 , ( pressure - free ) fluid density @xmath217 ( small - strain ) rubber shear modulus @xmath218 and bulk modulus @xmath219 , rubber damping @xmath220 , rubber@xmath221valve mass @xmath222 , and valve spring stiffness @xmath223 . since mono - disperse particles may induce a crystallisation phenomena at higher packing , a random size distribution is used uniformly between @xmath224 and @xmath225 . the distribution may be described by a mean particle size @xmath226 and polydispersity @xmath227 . llllllll particles & & valve + rubber & & fluid & & system + mean diameter & @xmath228 $ ] & mass & @xmath229 $ ] & dynamic viscosity & @xmath230 $ ] & system geometry & @xmath231 $ ] + solid density & @xmath232 $ ] & rubber shear modulus & @xmath233 $ ] & pressure - free density & @xmath234 $ ] & system size & @xmath235 $ ] + polydispersity & @xmath236 $ ] & rubber bulk modulus & @xmath237 $ ] & output pressure & @xmath238 $ ] & piston speed & @xmath239 $ ] + input packing fraction & @xmath240 $ ] & valve spring stiffness & @xmath241 $ ] & & + & & rubber damping & @xmath242 $ ] & & & & + to generalize the valve dynamics and flow behavior , we choose the natural units of our system to be the input section radius @xmath243 = r$ ] for length , time @xmath244 = \sqrt{\rho_{_f } r^3/k_{_v}}$ ] and mass @xmath245 = \rho_{_f } r^3 $ ] . from these units , a dimensionless parametric space is represented by : + @xmath246 + where @xmath247 is the system geometry . taking into account the fixed parameters , the dimensionless parametric space we will explore is described by the following groups : + @xmath248 + the second group is only relevant to pressure valves and the latter five can be independently controlled through the selection of @xmath249 , and @xmath250 . the parameters for all tests are summarized in tab . [ tab : spacerangeopenclosednp ] and tab . [ tab : spacerangeopenclosedyp ] . as indicated in the tables , the tests are conducted in order to observe the dependence of the valve behavior on each of @xmath251 and @xmath214 independently ; for each variable , a sequence of tests is performed where it is varied over a range while the others are held fixed . c|c geo@xmath252safety valve & geo@xmath252pressure valve + lc|c & range & fixed + @xmath59 & 1e-03 to 3.16e+01 & ll name & value + @xmath212 & 0 + @xmath213 & 12.5 + @xmath253 & 0 + @xmath254 & 0 + @xmath214 & 0 + + @xmath213 & 1 to 12.5 & .range of parameters investigated for safety and pressure valve simulations without particles . all units are in @xmath255 , [ \text{m } ] , [ \text{s}]$ ] . [ cols= " < , < " , ] + + the contact model ( dem solver and dem - rubber coupling ) , fluid turbulence model ( les ) and numerical parameters are displayed in tab . [ tab : contactcouplingnumericalparameter ] 2 lll stiffness & normal & tangantial + seat - valve & 1e+11 & 0 + seat - rubber & 1e+06 & 0 + seat - particle & 1e+07 & 0.8e+07 + valve - particle & 1e+07 & 0.8e+07 + rubber - particle & 1e+05 & 0.8e+05 + particle - particle & 1e+07 & 0.8e+07 + ll friction coefficient & + seat - valve & 0 + seat - rubber & 0 + seat - particle & 0.4 + valve - particle & 0.4 + rubber - particle & 0.4 + particle - particle & 0.4 + lll damping & normal & tangantial + seat - valve & 1e+03 & 0 + seat - rubber & 0 & 0 + seat - particle & 3 & 0 + valve - particle & 3 & 0 + rubber - particle & 0 & 0 + particle - particle & 3 & 0 + ll rubber and numerical parameters + smagorinsky constant & 0.4000 + lattice speed & 2.5e+03 + fluid space discretization & 3.0e-04 + dem time step & 5.0e-08 + numerical rubber convergence ( @xmath141 ) & 1.0e+01 + & + in this section , we discuss the effect of fluid viscosity , piston velocity , and input packing fraction on the opening , steady flow , and closure behavior of the valve for a pressure valve configuration . as shown in tab . [ tab : spacerangeopenclosednp ] and tab . [ tab : spacerangeopenclosedyp ] , when varying a parameter of interest , we fix the others to a control set taken from @xmath206 , @xmath256 , @xmath257 and @xmath258 . we will focus our analysis on the pressure valve and will give a brief analysis of the results for the safety valve in sec . [ sec : safetyvalvelift ] . during the opening phase , we observe a delay between the initiation of the piston and the initiation of valve opening . the effect is not due to packing fraction fig . [ fig : closed_yp_lift_v_xxoxxx_sf_0o040_080_0o125_0o150_0o175 ] ( left ) , polydispersity fig . [ fig : closed_np_lift_d_xxoxxx ] ( left ) or mean particle diameter fig . [ fig : closed_np_lift_d_xxoxxx ] ( right ) , rather , the delay increases with fluid viscosity fig . [ fig : np_closed_lift_eta_v_xxoxxx ] ( left ) , and decreases when piston velocity increases fig . [ fig : np_closed_lift_eta_v_xxoxxx ] ( right ) ( simulation without particles ) and fig . [ fig : closed_yp_lift_v_xxoxxx_sf_0o040_080_0o125_0o150_0o175 ] ( right ) ( simulation with particles ) . the lack of dependence of the delay time on particle inputs is because the mean particle diameter is bigger than the initial valve lift so it does not modify fluid behavior in the valve - rubber channel ( see schematic in fig . [ fig : np_closed_prb_conf ] ) . the more dominant effect is negative suction pressure , which develops in the valve - rubber channel as the valve initially displaces upward , as shown in fig . [ fig : np_closed_pressure_bc_br_vs_t_zoom ] . the delay increases with increasing viscosity because this increases the suction force due to lubrication effects in the narrow valve - rubber channel . at the same time , in the beneath valve region where the fluid domain is not thin , the pressure is mostly independent of viscosity as we observe in fig . [ fig : np_closed_pressure_eta_v_xxoxxx ] ( left ) where before the first peak of the valve lift ( fig . [ fig : np_closed_lift_eta_v_xxoxxx ] ( left ) ) the pressure evolution is the same . and @xmath206 . ( without particles ) . ] increasing the piston velocity reduces the delay because the suction pressure in the valve - rubber channel is balanced by faster growth of the beneath valve pressure . figure [ fig : np_closed_pressure_eta_v_xxoxxx ] ( right ) shows that during the pressurization phase , the pressure slope increases with piston velocity , as expected . the valve displacement begins to approach a steady value ( with or without oscillations ) after the first peak in valve lift ( @xmath259 ) and ends when the piston motion is stopped at @xmath260 . as shown in fig . [ fig : np_closed_lift_eta_v_xxoxxx ] , the steady lift position , for the simulations without particles , increases with fluid viscosity and piston velocity . for lower viscosity , the valve has an ` underdamped ' response characterized by decaying oscillations , whereas for larger viscosities , an ` overdamped ' response can be seen ( see fig . [ fig : np_closed_lift_eta_v_xxoxxx ] ( left ) ) . the presence of particles can modify the valve lift behavior . figure [ fig : closed_np_lift_d_xxoxxx ] ( left ) shows that virtually no effect on valve lift is observed for different polydispersity in the range we tested . increasing the mean diameter of particles , fig . [ fig : closed_np_lift_d_xxoxxx ] ( right ) , increases the steady lift position but this appears to be primarily a particle size effect ; after the initial upward motion of the valve , the valve descends downward and is stopped by a monolayer of mobile particles in the rubber channel , which holds the valve position at roughly @xmath261 high . further tests would be needed at higher fixed piston speeds to determine if the valve positioning depends more robustly on @xmath253 . tests involving variation of the packing fraction or the piston velocity , fig . [ fig : closed_yp_lift_v_xxoxxx_sf_0o040_080_0o125_0o150_0o175 ] , show a non - trivial valve lift behavior in which three approximate regimes can be identified : 1 . a lower input particle flux behavior : @xmath262 . 2 . a transition input particle flux behavior : @xmath263 . 3 . a higher input particle flux behavior : @xmath264 . where the input flux is defined by @xmath265 . in the first regime , we observe a simple particle suspension flow through the valve and rubber channel with a quasi - constant lift as shown in fig . [ fig : closed_yp_lift_v_xxoxxx_sf_0o040_080_0o125_0o150_0o175 ] . the beneath valve packing fraction shows also a quasi - constant value fig . [ fig : closed_yp_bvpackingfraction_sf_xxoxxx_v_xxoxxx ] . this regime is observed for @xmath266 ( from fig . [ fig : closed_yp_lift_v_xxoxxx_sf_0o040_080_0o125_0o150_0o175 ] ) . the second regime is characterized by unsteady motion of the valve that oscillates between notably disparate high and low positions . this regime , for the range of parameters we tested , appears to be limited by @xmath267 and @xmath268 . to better understand the valve lift behavior in this regime let us analyze the lift for @xmath269 . figure [ fig : closed_yp_lift_packing_0o1250 ] and fig . [ fig : closed_yp_pressure_0o1250 ] show the time dependence of the lift , beneath valve packing fraction , and beneath valve pressure . notice that the peaks and valleys of the beneath valve pressure and packing fraction are relatively in sync . the peaks in the lift plot are delayed with respect to those of the pressure and packing fraction . this can be understood as follows : when the valve position is low , particles aggregate under the valve and as they do so , they form something of a ` plug ' that causes the pressure beneath the valve to build up . when the pressure is sufficiently high , the valve will open up to release the pressure , which causes the backed - up grains beneath the valve to escape through the open valve . when this happens it causes the beneath valve packing fraction and the pressure to decrease , which immediately allows the valve to recover to its initial lift ( fig . [ fig : closed_yp_lift_packing_0o1250 ] , fig . [ fig : closed_yp_pressure_0o1250 ] ) . this phenomena can be distinguished from the lower input particle flux regime , in that in the lower flux case , the flux of grains into the system is not enough to back - up sufficiently under the valve and induce a pressure build - up . . ] . according to fig . [ fig : closed_yp_lift_packing_0o1250 ] ( right ) , the first two packing fraction peaks correspond to @xmath270 and @xmath271 , and the first two valley correspond to @xmath272 and @xmath273 . ] figure [ fig : closed_yp_packingsmapshot_0o1250 ] shows several snapshots of the beneath valve region for @xmath269 ( from fig . [ fig : closed_yp_lift_v_xxoxxx_sf_0o040_080_0o125_0o150_0o175 ] ( left ) ) . according to fig . [ fig : closed_yp_lift_packing_0o1250 ] ( right ) , the first two packing fraction peaks correspond to @xmath270 and @xmath271 , and the first two valleys to @xmath272 and @xmath273 . the rest of the snapshots correspond to the time between the peaks and the valley . comparing the first packing fraction peak and the first valve lift peak in fig . [ fig : closed_yp_packingsmapshot_0o1250 ] , we observe a delay ; the first peak in packing fraction occurs at @xmath270 , and the peak for valve lift occurs between @xmath274 and @xmath272 . using fig . [ fig : closed_yp_lift_packing_0o1250 ] , we find that the delay is @xmath275 . the same delay is observed for all peaks and valleys . contrary to the valve lift , between the pressure and packing fraction peak / valleys , no delay is observed . this is in agreement with the lift behavior fig . [ fig : closed_yp_lift_v_xxoxxx_sf_0o040_080_0o125_0o150_0o175 ] where the valve lift is a consequence of the packing fraction / pressure evolution . the third regime of valve behavior corresponds to a high particle flux such that the beneath valve slurry develops a sustainably high pressure able to push and hold the valve at a maximal lift . this is observed for @xmath276 on fig . [ fig : closed_yp_bvpackingfraction_sf_xxoxxx_v_xxoxxx ] ( left ) from @xmath277 . out of the three phases , one outlier phenomenon is observed for @xmath278 . in fact , fig . [ fig : closed_yp_lift_v_xxoxxx_sf_0o040_080_0o125_0o150_0o175 ] ( right ) shows that when the valve is opened , from @xmath279 to @xmath280 , we have a constant but small lift . during this phase , it turns out that a portion of the particles are stuck at the entrance of the valve channel but without entering fully into the channel . the force from these stuck particles and fluid pressure is enough to hold open the valve at a constant lift . in this section , we focus on the closure phase of the valve simulations with particles in order to investigate the effect particles have on the final lift of the valve and to study the degree to which particles become stuck in the valve - rubber channel , which could have detrimental effects on valve performance in practice . since the rubber plays the role of a seal between valve and seat , preventing grain trapping during closure could be a key design goal in such systems . the closure phase starts when the piston velocity @xmath213 is turned off . in the range of the parametric study we investigated , the final lift is mostly affected by the mean particle diameter @xmath253 as shown in fig . [ fig : closed_yp_stuck_lift_dmean_poly ] ( for @xmath253 and @xmath254 ) and fig . [ fig : closed_yp_stuck_lift_packing_vp ] ( for @xmath214 and @xmath213 ) . this behavior is simply explained by the fact that the closing valve squeezes out all but a monolayer of particles , which become stuck in the valve channel as illustrated in fig . [ fig : closing_stuck_particles ] where a zoom - in on the valve channel shows how geometrically the lift depends on the stuck particle diameter @xmath253 . using the size of particles and the geometry of the valve channel fig . [ fig : closing_stuck_particles ] , we calculated the envelope giving the maximum lift ( upper bound ) and minimum lift ( lower bound ) which should be obtained if only big particles ( with @xmath281 ) or small particles ( with @xmath282 ) were stuck . the two bounds are given by : @xmath283 where @xmath284 is either @xmath281 or either @xmath282 and @xmath285 is the valve channel inclination as shown on fig . [ fig : closing_stuck_particles ] . @xmath286 is obtained from fig . [ fig : seat ] ( detail b ) . as expected , the final lift is always between these two limits . figure [ fig : closed_yp_stuck_lift_packing_vp ] ( right ) for @xmath278 shows a lift of @xmath287 which is less than the minimum lift ( @xmath288 ) for the smallest particle diameter in the polydispersity to travel through the valve . in fact here , as discussed previously in section _ quasi - steady open valve phase _ for the effect of @xmath213 on the lift , no particle flow is observed in the valve / rubber channel since the lift is less than one particle diameter . therefore , when @xmath213 is turned off , the rubber descent is unimpeded by grains . once the rubber makes contact with the valve seat , the fluid beneath the valve can not escape and therefore a pressure and residual lift of @xmath287 remains , which is the lift when the rubber is in contact with the seat with zero deformation . we give here a first quantitative view on which variables among @xmath289 and @xmath213 matter most in affecting the quantity of particles that get stuck beneath the valve during closure . figure [ fig : closed_yp_stuck_norm_dmean_poly ] and fig . [ fig : closed_yp_stuck_norm_packing_vp ] show the projected area of particles on the seat ( beneath valve - rubber area ) normalized by the beneath valve - rubber area , i.e. the ` normalized area of stuck particles ' ( nasp ) . on these four figures , we find that @xmath214 matters most whereas there is little variation due to changes in @xmath281 , @xmath213 and @xmath254 . we can assume the packing fraction in the open valve channel , during the quasi - steady valve phase , is bounded below by @xmath214 . as the valve descends during closure , particles are squeezed out and , as a further lower bound , we can approximate that a single monolayer at the same packing fraction @xmath214 remains . this implies the total number of stuck particles in the rubber - valve channel is approximated by : @xmath290 where the final lift is given by @xmath291 , and @xmath129 is the projected rubber - valve area . the projected total particle area is : @xmath292 normalizing the @xmath293 by @xmath129 , we obtain : @xmath294 the above lower bound formula assumes that the final packing fraction of grains stuck in the valve is greater than the input value , @xmath214 . in our tests we have observed that this is always true except for the one outlier case mentioned previously ( @xmath278 , fig . [ fig : closed_yp_lift_v_xxoxxx_sf_0o040_080_0o125_0o150_0o175 ] ( right ) ) where no particles travel through the channel because the beneath valve fluid pressure is less than the necessary pressure to open the valve to a lift greater than @xmath295 . this case is observed in fig . [ fig : closed_yp_stuck_norm_packing_vp ] ( left ) ( @xmath278 ) where the normalized stuck area is zero . if particles get stuck , they can also potentially break depending on the force of contact with the valve and seat , and grain properties such as the particle size and strength . a loose approximation of the possible volume of debris created can be made by assuming stuck particles all break . this may be expressed in terms of the final lift and the nasp , and then normalized by @xmath296 , giving @xmath297 where @xmath129 is the valve - rubber projected area . supposing the lift obeys eq . [ eq : lift ] where @xmath298 and nasp obeys eq . [ eq : nasp ] , we suggest the debris variable is approximated by : @xmath299 figure [ fig : closed_yp_stuck_debris_packing_dmean ] ( left ) shows the debris as a function of @xmath253 and fig . [ fig : closed_yp_stuck_debris_packing_dmean ] ( right ) as a function of @xmath214 . we show comparisons to our approximate formula ( eq . [ eq : debris2 ] ) . many of the behaviors in the safety valve configuration mimic those of the pressure valve . here we summarize the safety valve data . the used input parameters are resumed in tab . [ tab : spacerangeopenclosednp ] for the simulations without particles and tab . [ tab : spacerangeopenclosedyp ] for the simulations with particles . figure [ fig : np_opened_lift_eta_v_xxoxxx ] shows the time evolution of the valve lift for different fluid viscosity ( left ) and for different piston velocity @xmath213 ; both simulations are without particles . the delay at the beginning of the simulation is less marked because of the absence of the above valve pressure . the steady lift shows many of the same behaviors as the pressure valve , except there can be non - zero @xmath213 ( 1 m / s and 2.5 m / s ) and the valve may not ever open since the open end beneath the valve can prevent the beneath - valve pressure from building up enough to overcome the valve s spring force to open it . the three observed regimes for the pressure valve during the steady lift are also evidenced here as shown in fig . [ fig : opened_yp_lift_v_xxoxxx_sf_xxoxxx ] for different @xmath214 and @xmath213 . the @xmath300 and @xmath301 need to be calculated here taking into account the outgoing flux through the modular closure of the safety valve system . this then becomes : @xmath302 where @xmath303 is calculated from the input region and @xmath304 is calculated from the flux of outgoing particles through both outlet sections . figure [ fig : opened_np_lift_d_xxoxxx ] ( left ) shows the time evolution of the valve lift for different @xmath254 where a small deviation in different values of @xmath254 is observed . as shown in fig . [ fig : closing_stuck_particles ] , the envelope of the final lift can be predicted by @xmath305 , @xmath306 however , if the stuck particles are not entirely lodged in the valve channel , this prediction is wrong . this is the case in fig . [ fig : opened_np_lift_d_xxoxxx ] ( right ) for @xmath307 where the final lift is less than the observed lift for @xmath308 . figure [ fig:_force_dmax_01o6e03_1 _ ] shows that in fact no particles are fully stuck in the valve channel but there are some stuck in the rubber channel , as indicated by a non - zero normal force @xmath309 . we see in fig . [ fig:_force_dmax_01o6e03_1 _ ] that even though contact exists between a partially stuck particle and the valve channel region , the total force coming from the valve channel is close to @xmath310 whereas the total force observed in the rubber channel is close to @xmath311 ; this means that the main force balancing the valve spring force comes from the rubber channel and therefore , the final lift is overpredicted by the previous formula , @xmath305 . showing ( partially ) stuck particle in the valve entry and normal forces , @xmath312 , supported by all particles . ] the obtained results for both valves suggest that we may need more investigation with a broader range of particle input parameters to better make a comparison between the safety and the pressure valve geometries . in this paper , we have presented a detailed implementation of a 3d dem - lbm - rubber coupling in two complex valve geometries . with different focused tests , we have validated the implemented methods . the coupling of the three types of materials shows a good agreement with our physical predictions . we also have demonstrated the validity of the zieb technique , which allowed us to run simulations without having to simulate the entire domain simulations performed without particles give realistic behaviors . we observe a lubrication effect causing suction that delays the opening of the valve after piston motion commences . we find that increasing fluid viscosity increasingly overdamps the valve lift , reducing or removing the valve oscillations in the quasi - steady regime . we have validated a stokesian pressure drop across the valve channel when the fluid being driven through is sufficiently viscous . in the simulations performed with particles in the pressure valve geometry , we found , for the steady lift portion , three qualitative lift behaviors . the different regimes appear to be governed by the total particle flux , which combines the imposed piston velocity @xmath213 and packing fraction @xmath214 through a flux @xmath313 . the flux variable appears to indicate when the open - valve dynamics transition from a steady value of small lift , to an oscillating value that traverses between high and low positions , to a steady high lift value . further investigation may be needed to calibrate the robustness of the @xmath314 variable in determining qualitative valve dynamics . the valve closure was also investigated , which occurs when the ( virtual ) piston motion is stopped . the pressure valve shows a dependency of the final lift on particle size and we give a prediction of the lift envelope based on the minimum and maximum particle sizes in the polydispersity . we show that if the maximum lift during the open phase does not exceed @xmath315 , the final lift at closure can be less than @xmath316 because the particles are not entirely stuck in the valve channel . lastly we demonstrate the robustness of the approach by switching to a safety valve configuration , in which the above - valve region is not pressurized and the below - valve region has another exit . similar qualitative behaviors are observed as compared to the pressure valve , both with and without particles , albeit at different specific values of the piston speed and input particle packing fraction . this work was supported by aro grant w911 nf-15 - 1 - 0598 and schlumberger technology corporation . pm and kk would like to thank j .- y . delenne ( inra , umr iate montpellier ) for his helpful and useful discussions on dem - lbm coupling and sachith dunatunga for his help in streamlining the numerics . conflict of interest : the authors declare that they have no conflict of interest . in this appendix , we give the safety and pressure valve configuration and the dimensions . units are millimeters @xmath317 and degrees @xmath318 .

the relation between two - dimensional conformal quantum field theories with and without a timelike boundary is explored . dipartimento di matematica , + universit di roma `` tor vergata '' , via della ricerca scientifica 1 , 00133 roma , italy + institut fr theoretische physik , universitt gttingen , + friedrich - hund - platz 1 , d-37077 gttingen , germany _ dedicated to klaus fredenhagen on the occasion of his 60th birthday _

0712.2140

in @xcite , the authors have formulated boundary conformal field theory in real time ( lorentzian signature ) in the algebraic framework of quantum field theory . bcft is a local mbius covariant qft @xmath2 on the two - dimensional minkowski halfspace @xmath3 ( given by @xmath4 ) , which contains a ( given ) local chiral subtheory @xmath5 , e.g. , the stress - energy tensor . the reward of this approach was the surprisingly simple formula ( ( [ relcomm ] ) below ) , expressing the von neumann algebras of local observables @xmath6 in a double cone @xmath7 in terms of an ( in general nonlocal ) chiral conformal net @xmath8 of localized algebras associated with intervals along the boundary ( the time axis @xmath9 ) . the net @xmath8 is mbius covariant and contains the local chiral observables @xmath5 : a(i)b(i ) for each interval @xmath10 . the reduction to a single chiral net is responsible for a kinematical simplification , explaining , e.g. , cardy s observation @xcite that in bcft , bulk @xmath11-point correlation functions are linear combinations of chiral @xmath12-point conformal blocks . the algebra @xmath6 is a relative commutant of @xmath13 within @xmath14 , [ relcomm ] b_+(o ) = b(k)b(l ) , where @xmath15 are a pair of open intervals on the boundary @xmath16 such that the disconnected complement @xmath17 is the set of advanced and retarded times @xmath18 associated with points in @xmath19 ( see fig . 1 ) . although the chiral net @xmath8 is not necessarily local , the intersections ( [ relcomm ] ) do commute with each other when two double cones are spacelike separated . the main result in @xcite is that every bcft is contained in a maximal ( haag dual ) bcft of the form ( [ relcomm ] ) . * figure 1 : * intervals on the boundary and double cones in the halfspace . this leads to a somewhat paradoxical conclusion : on the one hand , each local bulk observable is defined as a ( special ) observable from a chiral cft . thus , superficially , the `` degrees of freedom '' of a bcft are not more than those of a chiral cft , containing only a single chiral component of the stress - energy tensor ( virasoro algebra ) . one might argue that such a `` reduction of degrees of freedom '' is a characteristic feature of qft with a boundary . but this point of view can not be maintained , because on the other hand , the resulting bcft @xmath2 is locally equivalent to another cft @xmath20 on the full two - dimensional ( 2d ) minkowski spacetime , which has all the degrees of freedom of a 2d qft , and in particular contains a full 2d stress - energy tensor ( two commuting copies of the virasoro algebra ) . even in the simplest case , when the chiral net @xmath8 on the boundary coincides with @xmath5 ( sometimes known as `` the cardy case '' ) , the associated bulk qft contains apart from the full 2d stress - energy tensor more ( `` non - chiral '' ) local fields that factorize into chiral fields with braid group statistics . locally , also the bcft contains the same fields . this paradoxical situation is not a contradiction ; it rather shows that `` counting degrees of freedom '' of a qft is an elusive task . trivially , there is no obstruction against a proper inclusion of the form @xmath21 if @xmath22 is an infinite - dimensional hilbert space . but `` counting degrees of freedoms '' , e.g. by entropy arguments , requires the specification of the hamiltonian . the bcft shares the hamiltonian and ground state ( vacuum ) of the chiral cft , while the associated 2d cft has a different hamiltonian and a different ground state . thus , with respect to different hamiltonians , the spacetime dimension ( measured through some power law behaviour of the entropy ) may assume different values ( 1 or 2 , in the present case ) . looking at the issue from a different perspective , we may start from a vacuum representation of the virasoro algebra . the latter integrates to a unitary projective representation of the diffeomorphism group of the circle @xmath23 , which contains the diffeomorphism group of an interval @xmath24 as a subgroup . for two open intervals with disjoint closures , there is a canonical identification between @xmath25 and @xmath26 . in terms of the stress - energy tensor @xmath27 , this amounts to an isomorphism between @xmath28 and @xmath29 , when @xmath30 and @xmath31 have disjoint support . it would be hard to see this local isomorphism directly in terms of the virasoro algebra . the mathematical theorem underlying these facts is the well - known split property @xcite , which can be derived in local qft in any dimension under a suitable phase space assumption . in chiral local cft , a sufficient assumption is the existence of the conformal character @xmath32 . in the algebraic framework , the _ chiral _ observables of a bcft ( e.g. , the stress - energy tensor ) localized in a double cone @xmath33 are operators belonging to the von neumann algebra @xmath34 where @xmath35 and @xmath36 are two open intervals of the time axis ( `` advanced and retarded times '' ) such that @xmath37 , @xmath38 for @xmath19 ( this justifies the notation @xmath39 ) , and @xmath40 are the von neumann algebras generated by the unitary exponentials of chiral fields smeared within @xmath35 . in contrast , the _ chiral _ observables in a 2d cft are operators in the algebra @xmath41 where @xmath35 and @xmath36 are regarded as two open intervals of the lightcone axes , and @xmath42 and @xmath43 are generated by left and right chiral fields . our present association between bcft and 2d cft applies to the case when @xmath44 , i.e. , the left chiral observables @xmath45 are isomorphic with the right chiral observables @xmath46 , and both are isomorphic with the chiral observables @xmath40 of the bcft . let @xmath47 denote the vacuum hilbert space for the chiral cft described by the algebras @xmath40 . the split property states that if @xmath35 and @xmath36 are two intervals with disjoint closures , there is a canonical unitary @xmath48 implementing an isomorphism [ splitiso ] ( a(i)a(j))^ * = a(i)a(j ) . the split isomorphism does not preserve the vacuum vector , i.e. , the canonical `` split vector '' @xmath49 is an excited state in @xmath47 . by construction , the split state @xmath50 on @xmath51 has the property that its expectation values for either subalgebra @xmath40 or @xmath52 coincide with those in the vacuum state , but the correlations between observables @xmath53 and @xmath54 are suppressed : [ splitstate ] ( , a_1a_2 ) = ( , a_1)(,a_2 ) = ( , a_1)(,a_2 ) . the split isomorphism depends on the pair of intervals @xmath35 and @xmath36 . it trivially restricts to algebras associated with subintervals , but it does not , in general , extend to larger intervals . when the intervals touch or overlap , a split state and the split isomorphism cease to exist . while the split isomorphism is well known , we discuss in this paper its extension to `` non - chiral '' local observables , which do _ not _ belong to @xmath51 in the bcft , and to @xmath55 in the 2d cft . as a concrete demonstration for the resolution of the above `` paradox '' , we present two simple but nontrivial examples where the algebraic relations outlined can be easily translated into the field - theoretic setting , i.e. , we characterize the local algebras of the various qfts in terms of generating local wightman fields . let us translate ( [ relcomm ] ) into the field - theoretic language . the intervals @xmath35 and @xmath36 shrink to the points @xmath18 when @xmath39 shrinks to a point @xmath56 . thus , we have to approximate a field @xmath57 of the bcft by observables in @xmath58 ( where the interval @xmath59 approximates @xmath60 from the outside ) , that commute with all fields localized in the interval @xmath61 ( which approximates @xmath60 from the inside ) . this will be done in sect . 2 . a crucial point here is that generating the local algebra @xmath58 involves `` non - pointwise '' operations , e.g. , typical observables may be exponentials of smeared field operators , so that an element of the relative commutant is not necessarily localized in the disconnected set @xmath62 . a second , somewhat puzzling feature of the algebraic treatment of bcft is the fact that the description of the local algebras @xmath6 in terms of the chiral boundary net ( eq . ( [ relcomm ] ) ) is much simpler than that of the local algebras @xmath63 of the associated @xmath64 conformal qft without a boundary . the latter are ( rather clumsily ) defined as jones extensions of the tensor products @xmath55 in terms of a q - system constructed from the chiral extension @xmath65 with the help of @xmath66-induction @xcite . one purpose of this work is to present a more direct construction of the 2d cft without boundary from the bcft . the obvious idea is to take a limit as the boundary is `` shifted to infinity '' . but we shall do more , and establish the _ covariant _ local isomorphism between the subnets @xmath67 and @xmath68 as @xmath69 , i.e. , the restriction of the aqfts to any double cone @xmath70 within the halfspace @xmath4 , at finite distance from the boundary . the main problem here is , of course , the enhancement of the conformal symmetry , i.e. , the reconstruction of the unitary positive - energy representation of the two - dimensional conformal group @xmath71 from that of the chiral conformal group @xmath72 . this is done by a `` lift '' of the chiral mbius covariance of the local chiral net @xmath5 , using the split property which allows to `` embed '' the 2d chiral algebra @xmath55 into a local bcft algebra @xmath6 . this will be done in sect . the point is that only a single local algebra of the bcft is needed for this reconstruction of the 2d conformal group and the full 2d cft . in sect . 4 , we show that the 2d cft can also be obtained through a limit where the boundary is `` shifted to the left '' , or equivalently , the bcft observables are `` shifted to the right '' . the translations in the spatial direction `` away from the boundary '' do not belong to the chiral mbius group of the bcft . but they are at our disposal by the previous lifting of the 2d mbius group into the bcft . therefore , we can study the behavior of correlation functions in the limit of `` removing the boundary '' . as we shift the boundary , the retarded and advanced times are shifted apart from each other . the convergence of the vacuum correlations of the bcft to the vacuum correlations of the 2d cft is therefore a consequence of the cluster behavior of vacuum correlations of the chiral cft @xmath5 . we add three appendices containing some related observations . the purpose of this section is to illustrate the construction ( [ relcomm ] ) in a field - theoretic setting . it is convenient to assume the trivial chiral extension @xmath73 since even in this case the construction ( [ relcomm ] ) is nontrivial , i.e. , non - chiral local bcft fields that factorize into nonlocal chiral fields can be constructed from local chiral fields only . we exhibit local bcft fields in a region @xmath74 as `` neutral '' chiral operators , that behave like products of `` charged '' chiral operators localized in @xmath35 and @xmath36 in the limit of large distance from the boundary . the limit of pointlike localization is also discussed , and reproduces familiar vertex operators . consider the free @xmath75 current @xmath76 with commutator @xmath77=2\pi i\delta'(x - y)$ ] and charge operator @xmath78 . the unitary weyl operators @xmath79 for real test functions @xmath30 satisfy the weyl relation [ weyl ] w(f)w(g ) = e^-i(f , g)w(f+g ) = e^-2i(f , g ) w(g)w(f ) and have the vacuum expectation value [ ground ] ( w(f ) ) = e^-i(f_-,f_+ ) = e^- 12__+kdk f(k)^2 where the symplectic form is [ symp ] ( f , g)= 12_dx(f(x)g(x ) - f(x)g(x ) ) = 12i_kdkf(-k)g(k ) , and @xmath80 ( @xmath81 ) correspond to the restrictions to positive ( negative ) values of @xmath82 of the fourier transform @xmath83 . with these conventions , @xmath84 is a state with charge density @xmath85 . the vacuum correlations of weyl operators are [ corr ] ( w(f_1)w(f_n ) ) = e^-i(_i(f_i-,f_i+ ) + 2 _ i < j(f_i-,f_j+ ) ) the weyl operators @xmath86 with @xmath87 define the local von neumann algebras of the chiral net @xmath88 . we fix a double cone @xmath89 . if @xmath30 is a test function that vanishes outside @xmath59 and is constant in @xmath61 , then @xmath86 belongs to @xmath58 and commutes with @xmath90 by ( [ weyl ] ) and ( [ symp ] ) , hence [ neut ] w(f)b_+(o ) = a(k)a(l ) . these are examples of operators that belong to @xmath6 but ( if @xmath91 ) not to @xmath92 . weyl operators can also be defined for smooth functions @xmath30 such that @xmath93 has compact support , and the relation ( [ weyl ] ) holds . then @xmath94 is called the charge . however , @xmath95 diverges , and the vacuum expectation value ( [ ground ] ) vanishes unless @xmath96 ( see below ) . this implies that correlation functions ( [ corr ] ) of charged weyl operators vanish whenever the total charge is non - zero ( charge conservation ) , while the ir divergences in each term in the exponent of ( [ corr ] ) cancel for neutral correlations . the neutral weyl operators ( [ neut ] ) in @xmath6 are ( up to a phase factor ) products of charged weyl operators with charge densities localized in @xmath36 and in @xmath35 . in the limit of sharp step functions @xmath97 ( requiring a regularization @xcite ) , the regularized weyl operators @xmath98 become the well - known vertex operators of charge @xmath99 and scaling dimension @xmath100 @xcite , which are formally written as v_-q(u ) = : ( iq_u^ j(y)dy ) : . thus , as @xmath33 shrinks to a point @xmath101 , and @xmath35 and @xmath36 shrink to the points @xmath102 and @xmath103 , the ( regularised ) weyl operators @xmath104 behave as _ q(t , x ) = v_q(t+x ) v_-q(t - x ) . the correlation functions of vertex operators are given by = _ 0 _ i < j ( ) ^-q_iq_j if @xmath105 , and @xmath106 otherwise , from which the well - known anyonic commutation relations can be read off . it is then easily seen that @xmath107 commutes with @xmath108 when either @xmath109 or when @xmath110 , because in these cases the anyonic phase factors cancel . it also commutes with @xmath111 if @xmath112 . these are precisely the requirements for locality of the fields @xmath113 among each other , and relative to the conserved current j_0(t , x ) = j(t+x ) + j(t - x),j_1(t , x ) = j(t+x ) - j(t - x ) defined for @xmath4 , i.e. , @xmath114 and @xmath115 are local fields on the halfspace @xmath3 . the correlation functions of @xmath11 fields @xmath116 are correlations of @xmath12 vertex operators ( @xmath12-point conformal blocks ) . after this digression to pointlike fields , let us resume the study of the correlation functions ( [ corr ] ) of the smooth weyl operators @xmath117 , and their behavior as @xmath33 is shifted away from the boundary . we choose @xmath11 test functions of the form [ g - h ] f_i = g_i - h_i where @xmath118 , @xmath119 are smooth step functions with values @xmath120 at @xmath121 and @xmath122 at @xmath123 , such that @xmath124 is supported in @xmath36 and @xmath125 is supported in @xmath35 ( see fig . 2 ) . * figure 2 : * a test function @xmath30 such that @xmath86 belongs to @xmath6 , but not to @xmath126 . @xmath127 and @xmath128 are smooth step functions , @xmath129 , @xmath130 . the neutral states @xmath131 carry the charge @xmath122 in @xmath35 and the charge @xmath132 in @xmath36 . the neutrality condition for each weyl operator @xmath133 can be written [ charge ] _ dxg_i(x ) - _ dxh_i(x)= 0 g_i(0 ) - h_i(0 ) = 0 . the exponent in ( [ corr ] ) is a linear combination of terms of the form ( using @xmath134 ) 2i(f_i-,f_j_+ ) = _ _ + k dx(g_i(x)-h_i(x))dy ( g_j(y)-h_j(y ) ) e^-ik(x - y ) which are ir finite because of ( [ charge ] ) . the separate contributions from @xmath135 and @xmath136 , however , are ir divergent . therefore , we first regularize at @xmath137 by the subtraction @xmath138 ( @xmath139 arbitrary ) , which does not change the result because of ( [ charge ] ) , and then compute the contributions from @xmath31 and @xmath140 separately . we are interested in the behavior of the correlation function ( [ corr ] ) as @xmath33 is shifted away from the boundary . this means that the functions @xmath135 are shifted by a distance @xmath141 to the left , and @xmath136 are shifted by the same distance to the right . the @xmath31-@xmath31 contributions and the @xmath140-@xmath140-contributions to @xmath142 are obviously invariant under this shift , while in the mixed @xmath140-@xmath31 contributions @xmath143 is replaced by @xmath144 : 2i_h_i , g_j(a ) : = - _ i dx h_i(x ) _ j dy g_j(y ) _ _ + k and similar for the @xmath31-@xmath140 contributions . the last integrand can be split into two parts : ( e^-ik(x - y+2a)-1)e^-k/ + e^-ik(x - y+2a)(1-e^-k/ ) so that the first contribution to the momentum integral equals -(1+i(x - y+2a ) ) while the second ( distributional ) contribution is of order @xmath145 in the limit of large @xmath141 . because the remaining integrals have compact support , we obtain _ a _ h_i , g_j(a ) = q_iq_j(2ia ) + o(a ) . together with the @xmath31-@xmath140 contributions @xmath146 , these terms in the exponent of ( [ corr ] ) cumulate up to the factor _ i ( 2a)^-q_i^2 _ i < j ( 2a)^-2q_iq_j = ( 2a)^-q^2 , where @xmath147 is the total charge within @xmath35 . thus ( [ corr ] ) vanishes in the limit @xmath148 if @xmath149 , enforcing `` chiral charge conservation '' in the limit . if @xmath96 , the remaining @xmath31-@xmath31 and @xmath140-@xmath140 contributions yield _ a ( w(f_1)w(f_n ) ) = ( w(g_1)w(g_n))(w(-h_1)w(-h_n ) ) involving charged weyl operators . these expressions are well - defined ( and independent of @xmath150 ) because @xmath151 and @xmath152 are neutral precisely due to @xmath96 . the factorization of the vacuum correlations in the limit @xmath148 is the desired feature we wanted to illustrate by this example . in the limit , @xmath133 have the same correlations as @xmath153 , which are charged observables of the associated 2d cft . notice that in the limit of sharp test functions ( see above ) , one obtains v_q(t+x ) v_-q(t - x ) which are local fields in the entire two - dimensional minkowski spacetime @xmath154 . * remark : * the above construction can be generalized to the @xmath155 current algebra . the frenkel - kac representation of @xmath155 currents at level 1 is given by @xmath156 and @xmath157 . then @xmath158 commutes with @xmath159 at @xmath160 provided @xmath161 . hence the field _ 122(t , x ) = v_122(t+x ) v_-122(t - x ) is local ( as before ) and relatively local w.r.t . the conserved currents @xmath162 ( @xmath163 ) j^a_0(t , x ) = j^a(t+x ) + j^a(t - x),j^a_1(t , x ) = j^a(t+x ) - j^a(t - x).@xmath164 is a neutral combination of charged primary fields of dimension @xmath165 , transforming in the spin-@xmath166 representation of @xmath155 , localized at @xmath102 and @xmath103 . the description of this model in terms of smooth weyl operators is rather straightforward , see e.g. , @xcite : weyl operators with integer multiples of the charge @xmath167 belong to @xmath40 , while operators with half - integer multiples of the charge @xmath167 in @xmath35 and in @xmath36 belong to @xmath168 . the mechanism of `` charge separation '' described here for obtaining elements of @xmath6 that do not belong to @xmath126 is very general @xcite , although in general it can not be formulated in terms of weyl operators . in sect . 4 we shall show that also the factorization behavior far away from the boundary is a general feature , which allows to recover the 2d cft from the bcft . we work in this section with a fixed `` chiral extension '' @xmath65 . here , @xmath5 is a haag dual mbius covariant local net @xmath169 of von neumann algebras on its vacuum hilbert space @xmath47 , satisfying the split property and having finitely many irreducible dhr sectors of finite dimension ( these properties together are called `` complete rationality '' @xcite ; in the case of diffeomorphism covariant nets , haag duality = strong additivity is a consequence of the other properties @xcite . the fact that the @xmath75 weyl algebra in sect . 2 is _ not _ completely rational , indicates that the results to be reported in this section hold also in more general situations ) . @xmath8 is a mbius covariant net @xmath170 on its vacuum hilbert space @xmath171 such that for each @xmath35 the inclusion @xmath172 holds and is an irreducible subfactor , which has automatically finite jones index @xcite equal to the statistical dimension of the ( reducible ) representation of @xmath5 on @xmath171 @xcite . the net @xmath8 may be non - local , but is required to be relatively local w.r.t . @xmath5 . if only @xmath5 is specified , the irreducible chiral extensions @xmath8 of @xmath5 can be classified in terms of q - systems of @xmath5 @xcite . the complete classification has been computed for @xmath5 the virasoro nets with central charge @xmath173 ( and implicitly also for the @xmath155 current algebras ) in @xcite . with @xmath65 one can associate a boundary cft @xmath2 on the halfspace @xmath3 and a two - dimensional cft @xmath20 on minkowski spacetime @xmath154 . to describe the former , we introduce a convenient notation ( see fig . 1 ) . for any quadruples of four real numbers such @xmath174 we define @xmath175 , @xmath176 , @xmath177 , @xmath178 , and @xmath179 . every double cone @xmath7 is of this form and determines @xmath180 , and similarly every pair of open intervals @xmath181 ( `` @xmath35 is to the right = future of @xmath36 '' ) determines @xmath182 , and @xmath39 . then the bcft associated with @xmath65 is the net ( [ relcomm ] ) , i.e. , @xmath183 . we have shown in @xcite that @xmath6 contains @xmath184 as a subfactor with finite index , @xmath2 is local and haag dual on @xmath3 , every haag dual bcft with chiral observables @xmath5 arises in this way ( namely the chiral extension @xmath8 can be recovered from the bcft ) , and every non - haag - dual local bcft net is intermediate between @xmath185 and @xmath2 . if @xmath73 , @xmath6 equals the four - interval subfactor @xmath186 on the circle @xcite ( @xmath187 ) . the 2d cft @xmath20 associated with @xmath65 has been constructed in @xcite . its local algebras are extensions ( with finite jones index ) of the tensor products @xmath55 , specified in terms of a q - system constructed from the chiral extension @xmath65 with the help of @xmath66-induction . we know from @xcite that @xmath2 and @xmath20 are locally isomorphic , i.e. , for each @xmath7 there is an isomorphism @xmath188 such that [ lociso ] ^o(b_+(o_1 ) ) = b_2d(o_1 ) o_1o . however , the hilbert space and the vacuum state for the two theories are very different . in this section , we wish to understand the relation between these two nets , by giving an alternative construction of the 2d cft directly from the bcft . the crucial point is the construction of the enhanced mbius symmetry of the 2d cft , and its ground state ( the 2d vacuum ) which is different from the bcft vacuum . we first construct the hilbert space @xmath189 for the 2d cft . we choose a fixed reference double cone @xmath190 . the subfactor @xmath191 is irreducible with finite index @xcite , and hence has a unique conditional expectation @xmath192 , which is automatically normal and faithful . let @xmath193 be the canonical split vector for @xmath194 as in ( [ splitstate ] ) . the split state @xmath195 on @xmath196 extends to the state @xmath197 on @xmath198 . let @xmath199 , @xmath200 and @xmath201 denote the gns hilbert space , gns vector and gns representation for @xmath202 . we also write @xmath203 for @xmath204 . let us analyze the structure of @xmath199 . the structure of @xmath198 has been described in @xcite . by complete rationality , @xmath5 has finitely many irreducible superselection sectors @xcite . choose for each irreducible sector of @xmath5 a representative dhr endomorphism @xcite @xmath205 localized in @xmath206 , and a representative @xmath207 localized in @xmath208 . ( for the vacuum sector , @xmath209 . @xmath210 and @xmath211 are the representatives of the conjugate sector . ) then the elements of @xmath198 are ( weak limits of ) sums of operators of the form @xmath212 where @xmath213 is the injection @xmath214 , @xmath215 , @xmath216 , and @xmath217 generalize the weyl operators @xmath86 ( [ g - h ] ) of sect . 2 : they are ( for each pair @xmath218 ) `` charged '' intertwiners in @xmath219 . we may express these intersections in a different way : let @xmath220 denote the endomorphisms of @xmath8 extending the dhr endomorphisms @xmath221 of @xmath5 by `` @xmath66-induction '' @xcite , where @xmath222 ( @xmath223 ) acts trivially on @xmath224 localized to the right = future ( left = past ) of the interval where @xmath221 is localized . thus @xmath225 acts trivially on @xmath226 , because @xmath227 . hence [ chint ] ( , |)b(k_0) = ( _ b,_^-_|^+ ) . ( for an alternative characterization of the charged intertwiners by means of an eigenvalue condition , see app . [ appeig ] . ) if @xmath228 is another double cone in the halfspace , the algebra @xmath229 is generated by @xmath230 and charged intertwiners [ chint1 ] _ 1 = ( u|u)(_b,__1 ^ -_|_1^+ ) with unitary charge transporters @xmath231 and @xmath232 , where @xmath233 is localized in @xmath234 and @xmath235 is localized in @xmath236 . e.g. , if @xmath73 ( the `` cardy case '' ) , the charged intertwiners ( generalizing the weyl operators @xmath86 in ( [ g - h ] ) of sect . 2 ) are of the form @xmath237 . this implies that @xmath207 and @xmath205 are representatives of the same sector . thus , the charges of bcft fields are in 1:1 correspondence with the dhr sectors of @xmath5 . in the general case , when @xmath238 and @xmath239 are two charged intertwiners , @xmath240 is an intertwiner @xmath241 . this space is zero unless @xmath242 and @xmath243 , and @xmath244 @xcite . therefore , we may choose ( for each pair @xmath218 ) a basis of charged intertwiners @xmath238 which is orthonormal w.r.t . the inner product @xmath240 . * lemma 1 : * _ the subspaces @xmath245 of @xmath199 spanned by @xmath246 are mutually orthogonal . each subspace @xmath245 factorizes as a representation of @xmath196 according to [ hpsi ] _ _ _ | where @xmath247 and @xmath248 carry the representations @xmath205 and @xmath211 of @xmath249 and @xmath250 , respectively . _ _ proof : _ the computation of matrix elements in a dense set of vectors ^*(a_1a_2)((a_1a_2 ) ) ^*(a_1a_2)= ( , a_1^*a_2^*((a_1a_2)^*)a_1a_2 ) = + = ( , a_1^*a_2^*|(a_1a_2)a_1a_2 ) = ( a_1,(a_1)a_1)(a_2,|(a_2)a_2 ) proves the claim . we may therefore identify the vectors @xmath251 with @xmath252 in the representation @xmath253 under the split isomorphism , such that in particular , the gns vector @xmath254 corresponds to the 2d vacuum vector @xmath255 we write the extended hilbert space @xmath199 in the form [ h2d ] _ 2d _ , z _ , _ _ | ( the `` 2d hilbert space '' ) . the nonnegative integer multiplicities are [ dim ] z _ , = ( _ ^+,_^- ) by the above characterization ( [ chint ] ) of the spaces of charged intertwiners . the chiral factorization ( [ h2d ] ) of the gns construction from the extended state @xmath256 may be viewed as the remnant of the original `` splitting behavior '' of the split vector @xmath257 . as shown in @xcite by comparison of the q - system , the local subfactor @xmath258 on @xmath199 is isomorphic to @xmath259 constructed in @xcite . we may therefore consistently denote also the former by @xmath260 . next , we construct the action of the 2d mbius group on @xmath189 , by a `` lift '' of the mbius transformations of the chiral net @xmath5 , using the split isomorphism and the conditional expectation @xmath150 . the action of @xmath71 on @xmath189 will then be used to define @xmath63 as the images of the reference algebra @xmath261 under a 2d mbius transformation @xmath262 taking @xmath70 to @xmath33 . the 2d mbius group @xmath71 is unitarily represented in the vacuum hilbert space @xmath47 of the chiral net @xmath5 by @xmath263 , the preimage of @xmath264 on @xmath265 under the split isomorphism . ( see app . [ appmod ] , how @xmath266 and @xmath267 can be obtained by modular theory directly on the boundary hilbert space . ) we need to lift @xmath263 to @xmath189 . let @xmath268 denote the connected semigroup taking the interval @xmath35 into itself , generated by the one - parameter subgroup preserving @xmath35 and two one - parameter semigroups fixing either of its endpoints . then @xmath269 is the connected semigroup taking the reference double cone @xmath70 into itself . for @xmath270 , the adjoint action of @xmath271 on @xmath272 , @xmath216 is given by the _ independent _ (= product ) action of the chiral mbius transformations given by geometric automorphisms @xmath273 of the chiral net @xmath5 : [ alpha ] ^+_g_1 ^ -_g_2(a_1a_2 ) = _ g_1(a_1)_g_2(a_2 ) , we extend these endomorphisms of @xmath196 to endomorphisms of @xmath198 by [ beta ] ^+_g_1 ^ -_g_2((a_1a_2 ) ) : = ( _ g_1(a_1)_g_2(a_2 ) ) ( z^(g_1)z^|(g_2 ) ) . here @xmath274 are the unitary cocycles @xcite @xmath275 where @xmath276 and @xmath277 are the representations of the mbius group in the vacuum representation and in the dhr representation @xmath221 . * proposition 1 : * _ ( i ) the maps @xmath278 defined by ( [ beta ] ) for @xmath279 are homomorphisms from @xmath198 onto @xmath280 . + ( ii ) for @xmath281 we have @xmath282 , i.e. , @xmath278 `` act geometrically inside @xmath198 '' . + ( iii ) @xmath278 respect the group composition law within the semigroup @xmath283 . + ( iv ) the conditional expectation @xmath150 intertwines @xmath278 with @xmath284 . _ _ proof : _ ( i ) the homomorphism property follows from the composition and conjugation laws of charged intertwiners @xcite and the intertwining and localization properties of the operators and endomorphisms involved . the statement about the range is just a special case of ( ii ) . + ( ii ) it is sufficient to show that a charged intertwiner @xmath285 is mapped to a charged intertwiner in @xmath286 . by virtue of ( [ chint1 ] ) , we compute ^+_g_1 ^ -_g_2(_1 ) = ( _ g_1(u)z^(g_1)_g_2(|u)z^|(g_2 ) ) . then the claim follows , because @xmath287 , and @xmath288 is localized in @xmath289 , and similarly @xmath290 is localized in @xmath291 . + ( iii ) the group composition law follows from the cocycle properties @xcite of @xmath292 . + ( iv ) the intertwining property of @xmath150 is due to the fact that @xmath150 annihilates all charged intertwiners except the neutral one ( @xmath293 ) . next , we adapt a well - known lemma about the implementation of ( groups of ) automorphisms to the case of ( semigroups of ) endomorphisms . * lemma 2 : * _ let @xmath294 be a von neumann algebra on a hilbert space @xmath22 with a cyclic and separating vector @xmath295 . let @xmath296 be an endomorphism of @xmath294 , preserving the state @xmath297 . then the closure of the map @xmath298 is an isometry @xmath299 . if @xmath295 is cyclic also for @xmath300 , then @xmath299 is unitary . for two endomorphisms @xmath296 , @xmath301 with the same properties , such that @xmath295 is cyclic for @xmath300 , one has @xmath302 . _ _ proof : _ that @xmath299 is an isometry is an obvious consequence of the invariance of the state . since @xmath303 is a dense subset , the @xmath299 is surjective , hence unitary . for the last statement it is sufficient to notice that @xmath304 is densely defined on @xmath303 . we apply the lemma to the endomorphisms @xmath278 of @xmath198 . using ( iv ) of prop . 1 , we see that @xmath278 leave the gns state @xmath305 invariant because the split state @xmath50 on @xmath196 is invariant under @xmath284 . the vector @xmath200 is cyclic and separating for each @xmath306 ( @xmath281 ) because @xmath150 is faithful and @xmath257 is cyclic and separating for each @xmath126 , which in turn follows by the split isomorphism because @xmath307 is cyclic and separating for @xmath308 and for @xmath309 . thus , lemma 2 applies : * corollary 1 : * _ the homomorphisms @xmath278 induce unitary operators on @xmath310 , which satisfy the group composition law within the semigroup @xmath283 . together with the inverse unitary operators , they generate a covering representation @xmath311 of @xmath71 on @xmath189 . _ the last statement is due to the fact that @xmath283 and its inverse generate @xmath312 , and the group law within @xmath283 secures the commutation relations of the lie algebra . by construction , for @xmath270 , @xmath313 on the subspace @xmath314 is equivalent to @xmath315 on @xmath316 under the isomorphism ( [ hpsi ] ) . by ( ii ) of prop . 1 , the adjoint action of @xmath313 takes @xmath229 to @xmath286 for @xmath281 . by constructing @xmath263 , we have thus furnished the local subnet @xmath317 of the bcft with a covariant `` two - dimensional re - interpretation '' . in the representation @xmath201 on @xmath310 , this is precisely the local isomorphism @xmath318 referred to in ( [ lociso ] ) . the present discussion shows that @xmath318 intertwines the global 2d mbius covariance with a `` hidden '' symmetry of the bcft , which is induced by the extended split state @xmath319 and acts locally geometric . we now define for arbitrary double cones @xmath320 the associated local algebras of the 2d conformal net on @xmath189 by varying @xmath321 in the connected neighborhood of unity for which @xmath322 , and putting [ b2d ] b_2d(o ) : = u(g_1,g_2 ) b_2d(o_0 ) u(g_1,g_2)^ * o = g o_0m^2 . for @xmath69 , this coincides with @xmath323 by virtue of ( ii ) of prop . 1 . notice that @xmath324 is uniquely defined as long as @xmath325 because in this case any two @xmath31 with the same image @xmath326 differ by an element of @xmath283 , while it requires the passage to a covering space when @xmath154 is conformally completed . * theorem 1 : * _ the net of von neumann algebras @xmath327 defined by ( [ b2d ] ) is covariant , isotonous , and local . _ _ proof : _ the covariance is by construction . isotony and locality of the 2d net follow from the geometric action inside @xmath70 , ( ii ) of prop . 1 , and the fact that every pair of double cones in @xmath154 such that either @xmath328 are @xmath329 can be moved inside @xmath70 by a mbius transformation , where we know ( from the boundary cft ) that isotony and locality hold . * corollary 2 : * _ the extension @xmath330 is isomorphic to the extension constructed in @xcite . _ _ proof : _ since the local subfactor @xmath260 constructed in @xcite is isomorphic to @xmath331 , and the isomorphism intertwines the representations of the 2d mbius group , the global isomorphism follows . we have associated with the bcft a 2d local cft , that is locally isomorphic . the association is intrinsic in the sense that it requires only the subnet @xmath317 together with the covariance of the dhr sectors of the underlying chiral cft @xmath5 . it should be noticed that the construction is up to unitary equivalence independent of the choice of the reference double cone @xmath332 . the reason is essentially that the charge structure of @xmath333 exhibited by the multiplicities @xmath334 in ( [ h2d ] ) is independent of the pair @xmath15 . we conclude this section with an observation concerning diffeomorphism covariance : * proposition 2 : * _ if @xmath65 is a chiral extension of a diffeomorphism covariant chiral net @xmath5 , then the ( possibly non - local ) chiral net @xmath8 , the bcft net @xmath2 defined by ( [ relcomm ] ) , and the 2d net @xmath20 associated with @xmath2 by thm . 1 are also diffeomorphism covariant . _ _ proof : _ the chiral net @xmath5 is diffeomorphism covariant if for a diffeomorphism @xmath335 of @xmath336 there is a unitary operator @xmath337 on @xmath47 such that @xmath338 . haag duality of @xmath5 implies that if @xmath335 is localized in an interval @xmath35 ( i.e. , acts trivially on the complement ) , then @xmath337 is an observable in @xmath40 . for a chiral extension @xmath65 we claim that if @xmath335 is localized in @xmath206 , then for @xmath339 one has @xmath340 , i.e. , @xmath341 implement the local diffeomorphisms . namely , @xmath342 is generated by @xmath343 and @xmath344 where @xmath345 is the canonical charged intertwiner @xmath346 for the canonical dhr endomorphism @xmath347 localized in @xmath206 @xcite ( see also app . [ appeig ] ) , and @xmath348 is an equivalent dhr endomorphism localized in @xmath234 . we find ( w_)v_1(w_^ * ) = ( w_u ( w_^*))v . now , @xmath349 , and @xmath350 is localized in @xmath351 . this proves the claim . the diffeomorphism covariance of the chiral net @xmath8 follows because the diffeomorphisms localized in @xmath206 together with the mbius group generate the diffeomorphism group of @xmath336 . the argument for the boundary cft and for the 2d cft are very similar : we first show that for diffeomorphisms @xmath352 where @xmath353 is localized in @xmath206 and @xmath354 localized in @xmath208 , the adjoint action with @xmath355 takes @xmath229 to @xmath356 if @xmath281 . again , it is sufficient to verify the action on the charged intertwiners ( [ chint1 ] ) of @xmath229 : [ diff ] ( w__1w__2 ) _ 1 ( w__1w__2)^ * = ( ( w__1u(w__1^*))(w__2|u|(w__2^*)))where @xmath357 and @xmath358 , and @xmath359 is localized in @xmath360 and @xmath361 is localized in @xmath362 . hence ( [ diff ] ) is a charged intertwiner of @xmath363 . this proves the claim . then the diffeomorphism covariance of @xmath2 and @xmath20 follow because the diffeomorphisms localized in @xmath70 together with the mbius group generate all diffeomorphisms . let @xmath364 be bcft observables localized within any fixed double cone @xmath74 . we wish to consider the behavior of a vacuum correlation [ b1bn ] ( , _ x(b_1b_n)),where @xmath365 is the one - parameter semigroup of `` right shifts '' ( @xmath4 , away from the boundary ) , that take @xmath35 to @xmath366 and @xmath36 to @xmath367 , represented as homomorphisms from @xmath6 to @xmath368 , see ( [ beta ] ) . in sect . 3 ( with @xmath33 as the fixed reference double cone ) we have given the re - interpretation of @xmath369 in the gns representation @xmath201 of the state @xmath256 as observables of the associated 2d cft , with the 2d vacuum @xmath370 given by the gns vector . we shall show * theorem 2 : * _ let each @xmath371 ( @xmath372 be of the form @xmath373 with charged intertwiners @xmath374 and @xmath375 and @xmath376 . as @xmath377 goes to @xmath123 , the bcft vacuum correlations ( [ b1bn ] ) converge to the 2d vacuum correlations _ [ corr2d ] ( _ 2d,(b_1b_n)_2d ) = ( b_1b_n ) . _ proof : _ we compute the limit and the 2d vacuum expectation value separately . using the decomposition of products @xmath378 into finite sums of operators of the form @xmath379 @xcite , where @xmath380 are intertwiners between dhr endomorphisms of @xmath5 , we see that the product @xmath381 is a finite sum of operators of the same form @xmath382 . for the present purpose , it is more convenient to write the charged intertwiners as @xmath383 where @xmath384 and @xmath385 ( frobenius reciprocity ) . then , because @xmath386 , we get @xmath387 . hence , the product @xmath388 is a finite sum of operators of the form [ ata ] ( a_1)t(a_2 ) . thus , the above vacuum correlation function is a finite sum of expectation values f(x ) = ( , _ x((a_1)t(a_2 ) ) ) = + = ( , ( _ x(a_1 ) z^(x))t ( z^(-x)^ * _ -x(a_2 ) ) ) = + = ( , _ x(a_1 ) z^(x)(t ) z^(-x)^ * _ -x(a_2 ) ) . here , @xmath389 is the global conditional expectation @xmath390 , which preserves the vacuum state @xcite . in particular , therefore , the expression vanishes identically unless @xmath205 and @xmath207 belong to the same sector . in the latter case , we express the cocycles as @xmath392 , and @xmath393 , giving f(x ) = ( , a_1 u_(x)^ * ( t ) u_(x)^ * a_2 ) = ( , a_1 u_(-2x ) ( t ) a_2 ) , because the intertwiners between dhr endomorphisms also intertwine the representations of the mbius group @xcite . by the spectrum condition , @xmath394 has a bounded analytic continuation to the lower complex halfplane . @xmath395 weakly converges in every direction @xmath396 ( @xmath397 , @xmath398 ) to the projection onto the zero eigenspace of the generator , and the latter projection is nonzero only if @xmath399 is the vacuum representation ; in this case @xmath400 . thus , @xmath401 converges in these directions to the vacuum expectation value [ limit ] _ , 0_,0 ( , a_1 ) ( , a_2 ) . next , we consider = ( , _ x((a_2^*)t^*(a_1^ * ) ) ) . let @xmath402 and @xmath403 . then we can write @xmath404 , where @xmath405 . using the locality properties of @xmath53 , @xmath54 , we can rewrite = ( , _ x((r_^*|(a_1^*))|t(|(a_2^*)r _ ) ) ) . this expression can be computed in the same way as @xmath394 before , giving = ( , r_^*|(a_1^*))u_|(-2x ) ( |t ) |(a_2^*)r _ ) . thus @xmath394 also has a bounded analytic continuation to the upper complex halfplane , and converges to the same limit ( [ limit ] ) also in the directions @xmath396 ( @xmath406 , @xmath398 ) . from this , we may conclude the cluster limit [ cluster ] _ x(,_x((a_1)t(a_2 ) ) ) = _ , 0_,0 ( , a_1 ) ( , a_2 ) . on the other hand , we now compute ( [ corr2d ] ) and show that it coincides with the factorizing cluster limit of ( [ b1bn ] ) . for each contribution of the form ( [ ata ] ) , we have ( _ 2d,((a_1)t(a_2)),_2d ) = ( ( a_1)t(a_2 ) ) = ( a_1(t)a_2 ) . but @xmath407 is an intertwiner in @xmath408 which vanishes unless @xmath399 and @xmath409 both belong to the vacuum sector . in the latter case , @xmath410 thus , [ cblock ] ( ( a_1)t(a_2))= _ , 0_,0 ( a_1a_2 ) = _ , 0_,0 ( , a_1 ) ( , a_2 ) . this coincides with the cluster limit ( [ cluster ] ) `` far away from the boundary '' . recall that @xmath411 and @xmath412 in ( [ ata ] ) were obtained by multiplying @xmath381 and successively decomposing the products of the charged intertwiners . thus , the vacuum expectation values @xmath413 in ( [ cblock ] ) are precisely the chiral conformal blocks of the corresponding 2d correlation functions . a variant of the conformal cluster theorem @xcite should also give a quantitative estimate for the rate of the convergence , depending on the charges of the operators involved through the corresponding spectrum of @xmath414 . we have studied the passage from a local conformal quantum field theory defined on the halfspace @xmath4 of two - dimensional minkowski spacetime ( boundary cft , bcft ) to an associated local conformal quantum field defined on the full minkowski spacetime ( 2d cft ) . there are essentially two ways : the first is to consider bcft vacuum correlations of observables localized far away from the boundary . in the limit of infinite distance , these correlation factorize into chiral correlations ( conformal blocks ) of charged fields . we have traced this effect back to the cluster property of the underlying local chiral subtheory . the second method exploits the split property , i.e. , the existence of states of the underlying local chiral cft in which correlations between observables in two fixed intervals at a finite distance are suppressed . with the help of the split property one can algebraically identify a fixed local algebra of the bcft with a fixed local algebra of the 2d cft , and one can generate a unitary representation of the 2d mbius group in the gns hilbert space of a suitable `` extended split state '' of this algebra . its ground state is different from the bcft vacuum . then , by acting with the 2d mbius group , one can obtain _ all _ local algebras of the 2d cft in the same hilbert space . the converse question : can one consistently `` add '' a boundary in any 2d cft ( without affecting the algebraic structure away from the boundary ) , is not addressed here . however , there arises a necessary condition from the discussion in app . [ apphmi ] : the 2d partition function should be either modular invariant , or at least it should be intermediate between the vacuum partition function and some modular invariant partition function . we hope to return to this problem , and find also a sufficient condition . * acknowledgements : * khr thanks the dipartimento di matematica of the universit di roma `` tor vergata '' for hospitality and financial support , and m. weiner and i. runkel for discussions related to the subject . in @xcite it was shown that a unitary representation of the mbius group @xmath72 is generated by the modular groups of a `` halfsided modular triple '' , i.e. , three von neumann algebras @xmath415 ( @xmath416 ) with a joint cyclic and separating vector @xmath295 such that if @xmath417 is the modular group for @xmath418 , then @xmath419 for @xmath420 . ( here , @xmath421 is understood mod 3 . ) specifically , when @xmath35 is an open interval and @xmath422 are the subintervals obtained by removing an interior point from @xmath35 , the three algebras @xmath423 , @xmath424 , @xmath425 in a local chiral cft together with the vacuum vector @xmath307 define a halfsided modular triple . this means that the entire local net can be recovered from these data . we want to show here , how this construction can be applied to construct a unitary representation of the 2d mbius group @xmath312 from six suitable algebras in the split state @xmath257 associated with a pair of intervals @xmath35 and @xmath36 , see ( [ splitstate ] ) . let @xmath422 arise from @xmath35 be removing a point , and similarly @xmath426 . tensoring by 1 , the two halfsided modular triples ( a(i)1,a(i_1)1 , a(i_2)1 ) + ( 1 a(j) , 1 a(j_1),1 a(j_2 ) ) in the state @xmath427 generate @xmath264 . under the split isomorphism , these triples turn into ( a(i)n , a(i_1),a(i_2)),(a(j)n,a(j_1),a(j_2 ) ) in the split state @xmath257 , where @xmath428 is the canonical intermediate type @xmath35 factor between @xmath40 and @xmath429 . @xmath257 is cyclic and separating for these algebras in the subspaces @xmath430 and @xmath431 , respectively . the latter halfsided modular triples thus generate the two commuting representations @xmath432 of @xmath72 directly in @xmath47 . the charged intertwiners @xmath238 for a given chiral extension @xmath433 , that together with @xmath126 generate @xmath6 , are elements of the finite - dimensional spaces @xmath434 . in ( * eq . ( 5.12 ) ) a linear condition on @xmath435 was given which guarantees that @xmath436 commutes with @xmath437 . here @xmath438 is a homomorphism conjugate to the injection @xmath439 , such that @xmath440 on @xmath13 is a canonical endomorphism for @xmath441 and @xmath442 is the dual canonical endomorphism , which is a dhr endomorphism of @xmath5 localized in @xmath61 @xcite . unfortunately , the condition displayed in @xcite does not take into account that @xmath436 belongs to @xmath443 ( i.e. , is in the range of @xmath444 ) . we want to reformulate this condition so that it is equivalent to @xmath238 belonging to @xmath445 . we first notice that every element of @xmath13 is of the form @xmath446 where @xmath447 is the canonical isometry intertwining @xmath335 . then @xmath448 if and only if @xmath449 . this already secures that @xmath450 , and since @xmath347 is localized in @xmath61 , @xmath238 commutes with @xmath451 . hence it commutes with @xmath13 iff it also commutes with @xmath452 . this is equivalent to the relation [ comm ] yx ! = ( y)x ( _ , |)_,^ * ( y)x where @xmath453 . the statistics operators @xmath389 are trivial @xcite due to the localizations of @xmath205 in @xmath35 , @xmath211 in @xmath36 , and @xmath347 in @xmath61 , but we have displayed them in order to make the condition covariant under unitary deformations of @xmath444 and @xmath454 , possibly changing the localization of @xmath347 and leading to nontrivial statistics operators . the condition ( [ comm ] ) can be equivalently written as the eigenvalue equation [ eigen ] ( y):= ^12 ( 1_r^ * 1_|)(_,^*_,|^ * ) ( 1_y1_)x_2 ! = y. here @xmath455 where @xmath456 is the dual canonical isometry ( such that @xmath457 form a q - system ) ; @xmath458 , and @xmath459 is the index @xmath460 $ ] . @xmath461 and @xmath462 are the concatenation and the monoidal product in the tensor category of dhr endomorphisms of @xmath5 . the map @xmath463 defined by ( [ eigen ] ) is a linear map @xmath464 . ( [ eigen ] ) obviously follows from ( [ comm ] ) by left multiplication with @xmath465 and right multiplication with @xmath377 . to see that ( [ eigen ] ) implies ( [ comm ] ) , one may insert ( [ eigen ] ) into both sides of ( [ comm ] ) and repeatedly use the relations of the dual q - system @xmath466 to get equality . we thank i. runkel who has pointed out to us that @xmath463 is in fact a projection . hence the charged intertwiners @xmath238 are precisely given by @xmath467 where @xmath468 is in the range of @xmath463 . the multiplicities @xmath334 in ( [ dim ] ) equal the dimension of the range of these projections ( for each pair @xmath218 ) . if @xmath5 is completely rational , the c * tensor category defined by its dhr superselection sectors is modular @xcite , i.e. , the unitary @xmath469 and @xmath27 matrices defined by the statistics @xcite generate a representation of the group @xmath470 . by the verlinde formula @xcite , these matrices also describe the modular transformation behavior of chiral partition functions ( `` characters '' ) . by @xcite , the matrix @xmath471 given by ( [ dim ] ) is a modular invariant ( it commutes with @xmath469 and @xmath27 ) , hence the partition function of the 2d cft @xmath20 on @xmath189 is invariant under modular transformations . we want to point out an interesting relation of this fact to haag duality of the associated bcft . as mentioned before , every bcft defined by ( [ relcomm ] ) is automatically haag dual , and any non haag dual bcft @xmath472 with the same chiral observables is intermediate between @xmath185 and @xmath2 @xcite . therefore , the charged intertwiners @xmath473 constitute linear subspaces of the spaces of charged intertwiners in @xmath2 . let the dimensions of these spaces be @xmath474 , and at least one of them @xmath475 ( i.e. , @xmath472 is strictly contained in @xmath2 ) . then the matrix @xmath476 can not be a modular invariant by the following simple argument : consider the @xmath477 component of @xmath478 . because each @xmath479 is positive , ( s^*z s)_00 = _ ij s_0is_0j z_ij is strictly smaller than @xmath480 . if @xmath481 were modular invariant , we would conclude @xmath482 , which is impossible . we notice that the construction of a 2d cft associated to a bcft described in sect . 3 takes an intermediate bcft @xmath483 to an intermediate 2d cft @xmath484 . its hilbert space is of the form ( [ h2d ] ) with @xmath471 replaced by @xmath476 . hence , we conclude that the partition function of the associated 2d cft is modular invariant if and only if the bcft is haag dual . commun . math . phys . * 208 * ( 1999 ) 429487 . b ( proc . suppl . ) * 5b * ( 1988 ) 2056 . * b 240 * ( 1984 ) 514532 . * 99 * ( 1985 ) 347364 . * 110 * ( 1987 ) 325348 , [ erratum - ibid . * 116 * ( 1988 ) 175176 ] . * 75 * ( 1984 ) 493536 . * 23 * ( 1971 ) 199230 , commun . * 35 * ( 1974 ) 4985 . * 176 * ( 1996 ) 541554 . commun . math . * 125 * ( 1989 ) 201226 . rev . math . phys . * si1 * ( special issue , 1992 ) 113157 . invent . * 62 * ( 1980 ) 2366 . * 192 * ( 1998 ) 217244 . * 160 * ( 2004 ) 493522 . * 219 * ( 2001 ) 631669 . * 271 * ( 2007 ) 375385 . commun . math . * 237 * ( 2003 ) 730 * 7 * ( 1995 ) 567597 . * 16 * ( 2004 ) 909960 . * 251 * ( 2004 ) 321364 . * 211 * ( 2000 ) 395406 . * d 11 * ( 1975 ) 15091520 . . phys . * b 300 * ( 1988 ) 360376 .

we have extended the large - scale atomic / molecular massively parallel simulator ( lammps ) to support directional bonds and dynamic bonding . the framework supports stochastic formation of new bonds , breakage of existing bonds , and conversion between bond types . bond formation can be controlled to limit the maximal functionality of a bead with respect to various bond types . concomitant with the bond dynamics , angular and dihedral interactions are dynamically introduced between newly connected triplets and quartets of beads , where the interaction type is determined from the local pattern of bead and bond types . when breaking bonds , all angular and dihedral interactions involving broken bonds are removed . the framework allows chemical reactions to be modeled , and use it to simulate a simplistic , coarse - grained dna model . the resulting dna dynamics illustrate the power of the present framework . dynamic directional bonds , coarse - grain dna models , chemical reactions , molecular and dissipative particle dynamics

1108.4214

when performing molecular dynamics simulations , we distinguish between bonded and non - bonded interactions.@xcite effectively , this means that the interactions have been coarse - grained on the energy scale of the simulation . certain degrees of freedom are frozen , and we describe them as being permanent bonded . other degrees of freedom remain dynamic , and we describe them with relatively weak non - bonded interactions . however , this situation is less clear when simulating systems undergoing chemical reactions where bonds are created or broken . another example is dna molecules where hybridization bonds are broken at high temperatures and reformed when cooling the system . for such systems , it can be computationally more efficient to model these degrees of freedom as being dynamically bonded . the problem of bond dynamics is closely related to the question of how to represent chemical reactions in a molecular dynamics simulation . reactive force fields such as reaxff and empirical valence bond ( evb ) can be used to model chemical reactions.@xcite bond order potentials are interesting since they allow three body interactions in the neighborhood of a bond to modify the strength of the bond.@xcite when coarse - graining systems capable of chemical reactions , it is important to note that the reaction radius and probability also has to be appropriately coarse - grained.@xcite when the bonds become dynamic , this also induces a dynamic for the angular and dihedral interactions . when breaking a bond , all angular and dihedral interactions involving that bond become invalid , and should be removed . similarly , when creating a bond , we have to identify which angular and dihedral interactions to create in the bond neighborhood . this ensures that after melting and renaturing of a system , it is again governed by the same set of interactions and return to the same equilibrium structure . dna molecules are comprised of the four bases adenine ( a ) , cytosine ( c ) , guanine ( g ) , and thymine ( t ) . the bases are attached to a 2-deoxyribose sugar ring . for naturally occurring dna , sugar rings are linked to each other through phosphodiester bonds , that connect the 3 to 5 carbons in consecutive sugar rings . this builds a molecular directionality into the back bone of a dna strand , which will have a 3 and a 5 end . the strand is also characterized by a specific sequence of bases . together the phosphate backbone , the sugar and the base is denoted a nucleotide , which is the repeat unit of a single dna strand . a - t and c - g are watson - crick pairs and can form hydrogen bonds with each other . the energetically favorable stacking interactions allow two complementary single strands to form 3-5/ 5-3 anti - parallel aligned double strands . double stranded dna can be melted and renatured by repeated cycling the temperature around the melting point or by varying solvent conditions . dna is a very complex molecule and numerous models exists to describe behavior from atomistic properties to mesoscopic mechanical properties . the molecular structural details of short dna oligomers can be studied with atomistic molecular dynamics simulations such as amber@xcite and charmm@xcite . however , when we want to understand the large scale properties of dna molecules or materials in which dna molecules are a component , coarse - grained dna models are essential . coarse - graining is the statistical mechanical process by which uninteresting microscopic details are systematically removed , leaving a coarse - grained , effective model that is described by an effective free energy functional.@xcite a major advantage of coarse - grain models are that we can use them to simulate the interesting large scale dynamics of a system directly without wasting time on uninteresting details . this allows larger systems to be studied for longer times which paves the way for studying e.g. the properties materials rather than single molecules . a number of coarse - grain dna molecular dynamics models exists . in the `` three site per nucleotide '' model of de pablo and co - workers , a single nucleotide is represented by a phosphate backbone site , a sugar group site , and a base site , respectively@xcite . the model uses an implicit representation of counter ions at the level of debye - hckel theory , but has recently been generalized to explicit counter ions.@xcite a version of this model has also been generalized to include non - watson - crick base pairing such as hoogsteen pairing.@xcite there is also a number of `` two site per nucleotide '' models where one site represents the back bone and the sugar ring . the other site represents the base.@xcite one challenge to `` one site per nucleotide '' models are to represent the dna double helix . savelyev and papoian@xcite does this by special `` fan '' shaped pair - interactions between a bead and a large number of beads on the opposite strand . this model does not allow for dna melting . trovato and tozzini@xcite produce a helical structure using angular and dihedral interactions along the double strand . in the case where the large scale dna mechanical properties are of interest , it can be advantageous to coarse - grain a whole base - pair to a single rigid ellipsoidal or plate - shaped object and regard dna as a latter - like chain of such objects.@xcite here the coarse - graining has eliminated the melting and renaturation dynamics all together . other types of coarse - grain dna models are applied to study behavior of dna functionalized nano - particles . the dna molecules can e.g. be modeled as rigid rods with a single sticky site on one end and tethered to the surface of the nano - structure by the other end@xcite , as semi - flexible polymers with attractive sites on the monomers@xcite , or the whole dna molecule can be modeled as a single sticky site that can be hybridized to complementary free sticky sites.@xcite here the coarse - graining has completely eliminated the chemical structure , while the melting , renaturing , and sequence specificity has been retained in the dynamics . the two most prevalent statistical mechanical models of rna and dna melting are the poland - scheraga@xcite ( ps ) and the dauxois - peyrard - bishop@xcite ( dpb ) models . the poland - scheraga model describes dna as a 1d lattice model where a base - pair can either be hybridized or open . the free energy expression for the ps model contains empirical stacking free energies each stack of hybridized base - pairs as well as contributions from the strand configuration entropy due to internal bubbles , frayed ends and empirical initiation terms . the dpb model also describes dna as a 1d lattice model , but where each base - pair is characterized by a continuous base - pair distance . contrary to the ps model , the dpb model has a hamiltonian where the base - base potential is described by an anharmonic potential representing hydrogen bonding , and deviations between nearest neighbor base - pair extensions are penalized by a harmonic term . a generalization of the ps model exists , where the strand conformations are represented explicitly as lattice polymers . this provides a conceptual simplification since the conformational entropy of bubbles and frayed ends emerges naturally from the polymer model . this real space lattice ps model has been studied using exact enumeration techniques@xcite , a version of the model has also been applied to study rna folding using monte carlo simulations.@xcite the dynamic bonding framework allows us to study classes of dna models where hybridization bonds , angular bonds , and dihedral bonds are created and broken dynamically . these dynamic bonding dna models are intermediates between the real space lattice ps models , the coarse - grained molecular dynamics models , and the sticky dna models described above . in the ps model , base pairs can either be hybridized or open and are characterized by a corresponding free energy . in a dynamic bonding model , base pairs will be either hybridized or open and a free energy will also characterize this transition . in the coarse - grained molecular dynamics models and the dpb model , base pairs are represented by a continuous non - bonded pair - potential . in the dynamic bonding dna models , base pairs are characterized by a continuous bond potential . the dynamic bond dna models can also be regarded as being off - lattice generalizations of the real space lattice ps model , where a single strand is described as a semi - flexible bead - spring polymer where complementary monomers will form hybridization bonds when they are close . the dynamic bonded dna models are `` one site per nucleotide '' models , but we can also lump sequence of nucleotides into a single coarse - grained bead . in this case , we can as a first approximation assume that only beads representing complementary sequences can hybridize , and that the breaking of a hybridization bond corresponds to the creation of a dna bubble . this would be a `` many nucleotides per site '' dynamic bonding dna model more akin to the sticky site dna models used to study dna functionalized nano - particles . the dynamic bonded dna models ensure anti - parallel strand alignment in the double strand state , through the interplay between the dihedral interactions and the directional bonds . such degrees of freedom are absent from both the ps and dpb 1d lattice models . the coarse - grained models use angular and dihedral interactions to ensure a structure resembling the real chemical structure of dna molecules . in dynamic bonded dna models , the angular and dihedral interactions are dynamically introduced when hybridization bonds are formed to promote a zipper - like closing dynamic . similarly angular and dihedral interactions are dynamically removed as hybridization bonds are broken to promote zipper - like opening dynamic . hence in dynamic bond dna model , we utilize the interplay between dynamic bonded , angular , and dihedral interactions to model cooperative effects in the dna bubble and zippering dynamics , rather than to model chemical structure . the simplicity and success of the ps model in predicting sequence specific dna melting temperatures suggests that the essential physics of dna hybridization , melting and renaturing can , in fact , be accurately captured in a model without chemical details , and where the key property is the dynamics of hybridization . this is our motivation for developing the dynamic bonding framework . we will use it to develop and apply models to study the properties of hybrid materials containing both dna molecules and soft - condensed matter . we have implemented directional bonds and dynamic bonding in the large - scale atomic / molecular massively parallel simulator@xcite ( lammps ) . lammps is a versatile , parallel , highly optimized , open source code for performing molecular dynamics ( md ) and dissipative particle dynamics ( dpd ) simulations of coarse - grained models . due to the modular design , lammps is easy to extend with new interactions and functionality . the dynamic bonding implementation is also modular and easy to extend with new functionality . our extension is by no means limited to modeling dna , but could equally well be used for simulations of chemical reactions such as living polymerization , cross - linking of stiff polymers , coarse - grained dynamics of worm - like micelles and active driven materials . a snapshot of the lammps code with the directional bonds and dynamic bonding implementation can be obtained from the cpc program library . included with the code is also the documentation necessary for porting the directional and dynamic bonding framework to future lammps versions . [ sec : implementation1 ] is a summary of the implementation of directional bonds and the dynamic bonding framework . we present a simplified dna model based on the dynamic bonding framework in sect . [ sec : dna - model ] , which is provides the examples of dna dynamics shown in sect . [ sec : illustrations ] . we conclude with our conclusions in sect . [ sec : conclusions ] , and present the details of the directional bonds and dynamic bonding implementation in an appendix . double stranded dna only exists in a state where the two strands are aligned anti - parallel 3-5/5-3. in order distinguish between parallel and anti - parallel strand alignment , we regard the 3-5 back bone structure as a property of the back bone bonds , which become directional . this is necessary since the chemical structure of the nucleotides have been coarse - grained to a single structureless site . the directional bonds will also play a crucial role when introducing angular and dihedral interactions in a double stranded dna molecule , since this affects the stability , zippering dynamics , and mechanical properties . to implement directional bonds in lammps , we make use of the fact that newtons 3rd law is optional when calculating bond forces . when newtons 3rd law is enabled , each bond force is only calculated once , but subsequently has to be communicated to the bond partner . when it is disabled , lammps calculates the bond force twice , once for each of the two bond partners . in this case , each of the two bond partners store information about the bond type and the identity of the other bond partner . we can denote this situation by @xmath0 and @xmath1 , which shows that the @xmath2 bead stores @xmath3 as the type of the bond to @xmath4 , and the @xmath4 bead stores @xmath3 as the type of the bond to @xmath2 . with a few modifications , lammps will load and store different bond types in the two bond partners . hence , we can have @xmath0 and @xmath5 , where the bond type @xmath6 from @xmath4 to @xmath2 and the bond type @xmath3 from @xmath2 to @xmath4 differs . when the two bond types refer to the same bond potential , newtons 3rd law still applies , and the dynamics is unaffected . however , we can interpret the pattern of bond types as the directionality of the strand . note that if we instead use different bond potentials in the two directions or only a `` half '' bond , the result would be a net force along the bond , which can be used to model driven active matter . we shall not pursue this situation further in the present paper . define a dynamic bond fix : + fix fixid beadgroup bonddynamics _ everystep _ [ paircheck13 ] [ paircheck14 ] < list of rules > + + each dynamic bonding rule is one of : + createbond _ bondtype _ _ beadtype1 _ _ beadtype2 _ maxdistance probability + createdirbond _ bondtype1 _ _ bondtype2 _ _ beadtype1 _ _ beadtype2 _ maxdistance probability + breakbond bondtype mindistance probability + convertbond bondtype1 bondtype 2 probability + killbond _ bondtype _ mindistance + createangle _ angletype _ _ beadtype1 _ _ beadtype2 _ _ beadtype3 _ _ bondtype1 _ _ bondtype2 _ + createdihedral _ dihedraltype _ _ beadtype1 _ _ beadtype2 _ _ beadtype3 _ _ beadtype4 _ _ bondtype1 _ _ bondtype2 _ _ bondtype3 _ + maxbond _ bondtype _ _ beadtype _ _ maxnumber _ the dynamic bonding framework allows a number of rules to be specified , that completely define the bond dynamics . these rules are applied to a specified group of reactive beads with a specified frequency . the application of the rules is conditional on the types of beads , types bonds , distance between beads and length of bonds involved . in particular , we have implemented rules for stochastic creation of symmetric and directional bonds within a certain reaction distance , stochastic removal of symmetric bonds larger than a breaking distance , removal of all symmetric bonds exceeding a certain length , and stochastic conversion of a symmetric bond from one type to another . furthermore , all bond creation rules ensure that a bead can never have more than a specified number of bonds of a given type . the implementation is structured such that it is easy to implement new types of rules . besides the bond dynamics , the consistency of the angular and dihedral interactions should be ensured at all times . after bonds have been broken , all invalid potential angular and dihedral interactions involving broken bonds should also be removed . after bonds have been formed , all triplets or quartets of beads that could be connected by at least one new bond are checked to see if they require the creation of an angular or dihedral interaction . we discard cyclic triplets and quartets where the same bead appears more than once . an angular creation rule specifies which angular interaction can be introduced between a triplet of connected beads @xmath2 , @xmath4 , and @xmath7 . since the triplets are not ordered , the rule should match either @xmath8 or @xmath9 . to test if the @xmath8 bead order matches , we first compare the types of the @xmath8 beads with the bead types the rule specifies . we then compare the two bond types @xmath3 and @xmath6 with the bond types the rule specifies , where the bond types are defined directionally as @xmath1 and @xmath10 . if the @xmath8 bead order did not match , it is repeated @xmath9 bead order , where the bonds types are defined directionally as @xmath11 and @xmath12 . if a rule match , then the specified angular interaction is introduced between the three beads . a creation rule for a dihedral interaction specifies four bead types and three bond types . again we test both @xmath13 and @xmath14 ordered bead quartets . first the bead types of the quartet are compared to the bead types specified by the rule , subsequently the bond types are compared , the bond types are defined directionally as @xmath15 , @xmath16 , @xmath17 , and @xmath18 . the bond types match if @xmath19 , @xmath6 or @xmath20 , and @xmath3 matches the three bond types specified by the rule . if the @xmath13 bead order did not match , it is repeated @xmath14 order . if a dihedral rule match a quartet of beads , the specified dihedral interaction is introduced between the four beads . these rules allows us to selectively and dynamically introduce angular and dihedral interactions taking both bead types and directional bond types into account . note that the same directionality applies to matching the bead type and bond type patterns . to have an efficient parallel implementation , we implement the bond creation and breaking by an pair matching algorithm inspired from the bond / break and bond / create fixes already implemented in lammps . in the dynamic bonding fix , preferred bond creation / breakage partners are identified in each simulation domain . this information is communicated between and aggregated across neighbor simulation domains . afterwards , the bonds selected for breakage are removed . the local neighborhood of all reactive beads are checked for angular and dihedral interactions , that should be removed because they cross broken bonds . then bonds are created between partners selected for bonding . again , we check the local neighborhood of all reactive beads to introduce angular and dihedral interactions . after this final step , we broadcast bond statistics to all simulation domains . note that due to the pair matching algorithm , each bead can maximally have one bond created and broken at each call to the dynamic bonding fix . all rules are applied to a bead pair ( in the specified order ) when identifying if they are eligible for matching . if multiple rules apply to the same bead pair , the last matching rule will be always be chosen . hence , if this last rule has a very low reaction probability , it will completely shadow more probable rules specified earlier . these shadowing issues does not apply to the dna model below , and will not play a role at low concentrations of reacting beads . the details of the implementation and shadowing issues are discussed in appendix [ sec : appendix - a ] . the lammps syntax of the dynamic bond fix is shown in fig . [ fig : syntax ] . the first line defines the name of the particular instance of the fix , the group of reactive beads ( _ beadgroup _ ) , and how often the bond dynamics fix is applied ( _ everystep _ ) . by default creation rules only apply to potential bonding bead pairs , that are further than 4 bonds apart or not bonded . the optional _ paircheck13 _ and _ paircheck14 _ switches includes 1 - 3 and 1 - 4 chemically distant beads in the search of potential bonding partners . the line is followed by a number of dynamic bonding rules . _ createbond _ rules specify pairs of bead types , that can be bonded , if they are within a certain maximum reaction distance from each other . if a bead has more than one potential bond partners , the closest partner is chosen , and a bond with the specified type is then created with the given probability . _ createdirbond _ rules does the same as _ createbond _ , but creates a directional bond with the two specified bond types between the two bead types . _ breakbond _ rules identifies bonded bead pairs with bonds longer than the specified minimum distance and breaks the bond with the specified probability . if a bead has more than one potential bond break partner , then the most distant partner is chosen . since only a single bond can be removed per bead per call to the dynamic bonding fix , a breakbond rule with unit probability does not ensure that all bonds longer than the minimum distance are broken . hence , we have also implemented _ killbond _ rules . these rules operate directly on the bond structures , and are not limited by the pair matching algorithm . _ convertbond _ rules stochastically convert symmetric bonds of one type into another type . this is implemented as nominating the bond pair for removal of the old bond , followed by creation of the new bond . the dynamic bonding framework ensures that angular and dihedral interactions across the bond are also converted accordingly . _ createangle _ and _ createdihedral _ rules defines which angular and dihedral interaction types should be created between triplets and quartets of beads with the specified types of bead , and types of bonds between the beads as discussed above . _ createangle _ and _ createdihedral _ rules do not specify a probability , since they are created as required by the local neighborhood around new bonds . note that angular and dihedral interactions are only introduced as a consequence bond creation events , they are not introduced between already bonded beads even though the bead types and bond types match the rule . when checking potential beads for bond creation , all _ maxbond _ rules are checked to discard beads that already have the maximal number of the specified bond types . we have chosen the present dna model has been chosen because it produces a simple ladder like equilibrium structure , which allows us to illustrate the power of the dynamic bonding framework , and to visualize all the interactions that are dynamically introduced and removed . real dna molecules performs a whole twist every 10.45 base pairs , and to model twist we need a somewhat more complex force field , but exactly the same dynamic bonding rules . because we are interested in studying dna programmed self - assembly , we choose to use dissipative particle dynamics ( dpd)@xcite . dpd is given by a force field comprising a conservative soft pair - force @xmath21 , a dissipative friction force @xmath22 , and a stochastic driving force @xmath23 given by @xmath24 where the forces contributions are given by @xmath25 here @xmath26 and @xmath27 denotes the separation and relative velocity between two interacting beads @xmath28 and @xmath29 , respectively . @xmath30 denotes a gaussian random number with zero mean and unit variance , and the thermostat coupling strength is @xmath31 . the weighting function is @xmath32 . we integrate the dpd dynamics with a velocity verlet algorithm with a time step @xmath33 . the unit of energy is @xmath34 , where we chose to set boltzmann s constant to unity , such that temperature is measured in energy units . we use @xmath35 in all of the simulations except the dna bubble simulation where @xmath36 . the unit of length @xmath37 is defined by the pair force cut - off @xmath38 . the mass is @xmath39 for all beads , this allows us to define the unit of time as @xmath40 . the dpd pair - force parameter is @xmath41 between all species of beads . the viscosity is @xmath42 . non - bonded pair interactions are switched off between beads in molecules that are less than 3 bonds apart . the dna molecule is simulated in an explicit solvent at a density @xmath43 . we represent a nucleotide by a single dpd bead , and let the four atcg nucleotides correspond to bead types 1 - 4 , respectively . they are colored red , green , blue , and magenta , respectively , in figures below . red and green beads ( a - t ) are complementary as are blue and magenta ( c - g ) beads . a single strand of dna is represented as a string of beads joined by permanent directional back bone bonds . the two 3 to 5 and 5 to 3 backbone bond potentials ( bond type 2 and 3 , respectively , colored green and blue in the bond visualizations ) are given by the same potential @xmath44 with @xmath45 , @xmath46 , and @xmath47 . the hybridization bond potential ( bond type 1 , colored red in the bond visualizations ) is given by @xmath48 with @xmath49 , @xmath50 , and @xmath51 . besides the dna interactions , we need to define the bonding dynamics of the dna beads . the corresponding dynamic bonding fix command is shown in fig [ fig : fix - dna - dynamics ] . hybridization bonds are created with probability one when two complementary beads are within a distance of @xmath52 . bead type @xmath53 and @xmath54 are able to form a 5 3 backbone bond when they are within a distance of @xmath46 from each other . the probability of creation of a back bone bond is @xmath55 . this is a simplification for the oligomer - template simulation below . only hybridization bonds can be broken , and they are removed if they are longer than @xmath38 . to control hybridization , we only allow all bead types ( @xmath56 ) to have maximally one hybridization bond ( type 1 ) , one 3 end ( type 2 ) and one 5 end ( type 2 ) of a back bone bond . in the model all nucleotides has the same interactions , hence use @xmath56 for all the bead types rule specifications . 1 : fix dnadyn dna bonddynamics 1 paircheck14 2 : createbond * * * * 0.6 1.0 3 : createbond * * * * 0.6 1.0 4 : createdirbond * * * * 0.3 0.1 5 : killbond 1.0 6 : maxbond 1 * 7 : maxbond 1 * 8 : maxbond 1 * 9 : createangle _ _ * * * 10 : createangle _ _ * * * , 11 : createdihedral _ _ * * * * , 12 : createdihedral _ _ * * * * 13 : createdihedral _ _ * * * * 14 : createdihedral _ _ * * * * the model has two angular interactions , which are described by the potential @xmath57 , where @xmath58 defines the angular spring constant and @xmath59 the equilibrium angle . the first angle interaction ( type 1 ) promotes a straight angle between back bone bonds . this interaction is shown as red angles in the angle visualizations , and it has parameters @xmath60 and @xmath61 . type 1 angles are dynamically introduced for bonding patterns @xmath62 , @xmath63 and @xmath64 , @xmath65 ( i.e. for model bonds types @xmath53 @xmath54 , since @xmath9 order matches @xmath54 @xmath53 ) . the second angle interaction ( type 2 ) promotes a right angle between back bone and hybridization bonds . this interaction is shown as green angles in the angle visualizations , and it has @xmath66 and @xmath67 . type 2 angles are dynamically introduced for bonding patterns @xmath68 , @xmath69 and @xmath70 , @xmath71 ( i.e. model bond types @xmath72 and @xmath73 , since @xmath9 order matches the reverse pattern ) . the dna model has three dihedral interactions , which are described by the potential @xmath74 . we use dihedral spring constant @xmath75 , and @xmath76 ( @xmath77 ) for promoting trans ( cis ) conformations . the first dihedral interaction ( type 1 , shown red in dihedral visualizations ) promotes a cis conformation when a back bone bond connects two hybridized nucleotide pairs . this corresponds to the bonding patterns @xmath68 , @xmath78 , @xmath63 , @xmath79 and @xmath68 , @xmath80 , @xmath65 , @xmath79 , where @xmath81 denotes a hybridization bond ( i.e. model bond numbers @xmath72 @xmath73 @xmath72 ) . the second dihedral interaction ( type 2 , shown green in the dihedral visualizations ) promotes a cis conformation of the two beads that are connected by back bone bonds to a hybridized bead pair and is located on the same side of the bead pair . the bonding pattern is @xmath62 , @xmath82 , @xmath71 , @xmath83 ( i.e. model bond numbers @xmath53 @xmath72 @xmath54 ) . the third interaction ( type 3 , shown blue in the dihedral visualizations ) promotes a trans conformation of the two bead that are connected by back bone bonds to a hybridized bead pair but are localized on opposite sides of the bead pair . the bonding patterns are @xmath62 , @xmath82 , @xmath71 , @xmath84 and @xmath64 , @xmath82 , @xmath71 , @xmath83 ( i.e. model bond numbers @xmath53 @xmath72 @xmath53 and @xmath54 @xmath72 @xmath54 ) . note that without the directional bond , we would be unable to distinguish between these two last types of dihedrals . the examples belows are included as test cases with the dynamic bonding code submitted to the cpc program library , and require less than a cpu hour of computational effort . oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig:"]oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig:"]oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig : " ] oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig:"]oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig:"]oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig : " ] oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig:"]oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig:"]oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig : " ] oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig:"]oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig:"]oligomer - dna template hybridization ( rows 1 - 4 ) showing the dynamics of bond , angular , and dihedral interactions ( columns a - c ) for times @xmath85 , @xmath86 , @xmath87 , and @xmath88 into the simulation . bead and interaction colors match those in fig . [ fig : fix - dna - dynamics ] . note that back bone bond directionality is only shown in the first row for simplicity.,title="fig : " ] to illustrate the dynamic bonding framework with the dna model , we simulate a @xmath89 template in the presence of two @xmath90 oligomers . the first oligomer is already hybridized with the template , while the second is placed in the vicinity of the template . fig . [ fig : example - hybridization ] shows snapshots along the trajectory where the remaining oligomer hybridizes with the template . the top left visualization shows the initial designed configuration . the blue - green pattern of the hybridized oligomer backbone shows it has 3 5 direction , while the green - blue pattern of the template backbone shows the 5 3 direction . the top center visualization shows the angular interactions of the initial configuration . the back bone stiffness is controlled by the red angular interactions between back bone bond pairs , which promote a straight back bone configuration . the green angular interactions promote hybridization bonds that are perpendicular to the strand axis . the top right visualization shows the dihedral interactions of the initial configuration . the hybridized template shows red and green dihedral interactions which are promote cis arrangement of stacked bead pairs , while the blue dihedral interaction promotes trans arrangement . together they stabilize the ladder - like structure of the double strand . without the bond directionality , we would have no way to distinguish between green and blue dihedral interactions , and hence control over the stiffness of the double strand relative to that of the single strands . as we let the simulation run ( left column top to bottom ) initially two hybridization bonds are introduced between the two beads at right most end of the template . later a third and a fourth hybridization bond are also introduced . along with the hybridization bonding dynamics , angular and dihedral interactions ( center and right columns ) are also created . the angular interactions cause the free oligomer to align with the template , while the dihedral interactions creates a torque that ensures that the alignment is anti - parallel . back bone ligation reaction by addition of directional back bone bond ( rows 1 - 3 ) showing dynamic of bond , angular and dihedral interactions ( columns a - c ) for the simulation in fig . [ fig : example - hybridization ] continued to times @xmath91 , @xmath92 , and @xmath93 , respectively.,title="fig:"]back bone ligation reaction by addition of directional back bone bond ( rows 1 - 3 ) showing dynamic of bond , angular and dihedral interactions ( columns a - c ) for the simulation in fig . [ fig : example - hybridization ] continued to times @xmath91 , @xmath92 , and @xmath93 , respectively.,title="fig:"]back bone ligation reaction by addition of directional back bone bond ( rows 1 - 3 ) showing dynamic of bond , angular and dihedral interactions ( columns a - c ) for the simulation in fig . [ fig : example - hybridization ] continued to times @xmath91 , @xmath92 , and @xmath93 , respectively.,title="fig : " ] back bone ligation reaction by addition of directional back bone bond ( rows 1 - 3 ) showing dynamic of bond , angular and dihedral interactions ( columns a - c ) for the simulation in fig . [ fig : example - hybridization ] continued to times @xmath91 , @xmath92 , and @xmath93 , respectively.,title="fig:"]back bone ligation reaction by addition of directional back bone bond ( rows 1 - 3 ) showing dynamic of bond , angular and dihedral interactions ( columns a - c ) for the simulation in fig . [ fig : example - hybridization ] continued to times @xmath91 , @xmath92 , and @xmath93 , respectively.,title="fig:"]back bone ligation reaction by addition of directional back bone bond ( rows 1 - 3 ) showing dynamic of bond , angular and dihedral interactions ( columns a - c ) for the simulation in fig . [ fig : example - hybridization ] continued to times @xmath91 , @xmath92 , and @xmath93 , respectively.,title="fig : " ] back bone ligation reaction by addition of directional back bone bond ( rows 1 - 3 ) showing dynamic of bond , angular and dihedral interactions ( columns a - c ) for the simulation in fig . [ fig : example - hybridization ] continued to times @xmath91 , @xmath92 , and @xmath93 , respectively.,title="fig:"]back bone ligation reaction by addition of directional back bone bond ( rows 1 - 3 ) showing dynamic of bond , angular and dihedral interactions ( columns a - c ) for the simulation in fig . [ fig : example - hybridization ] continued to times @xmath91 , @xmath92 , and @xmath93 , respectively.,title="fig:"]back bone ligation reaction by addition of directional back bone bond ( rows 1 - 3 ) showing dynamic of bond , angular and dihedral interactions ( columns a - c ) for the simulation in fig . [ fig : example - hybridization ] continued to times @xmath91 , @xmath92 , and @xmath93 , respectively.,title="fig : " ] fig . [ fig : example - backbone ] shows how the nick in the dna molecule is closed by forming a back bone bond . the interactions between the two oligomers and the template ensures that they are both aligned anti - parallel to the template backbone axis . the single red dihedral interaction across the nick promotes a cis configuration , and twists the two oligomers towards the same side of the template . finally the missing back bone bond is created following the 3-5 directionality of the strand , along with all the angular and dihedral interactions to produce a double stranded configuration . together figs . [ fig : example - hybridization ] and [ fig : example - backbone ] simulates a chemical reaction where a dna template and two complementary oligomers first hybridize due to their complementary sequences , and then ligate to produce the complementary template sequence . dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig:"]dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig:"]dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig : " ] dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig:"]dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig:"]dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig : " ] dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig:"]dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig:"]dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig : " ] dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig:"]dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig:"]dna unzipping by a weak vertical force @xmath94 applied to the left most bead pair ( rows 1 - 4 ) for bond , angular and dihedral interactions ( columns a - c ) . the rows corresponds to times @xmath95 , @xmath96 , @xmath97 , @xmath98 , respectively , starting from a straight double strand conformation at @xmath99.,title="fig : " ] to melt the double strand , we can e.g. apply an external force to tear the two strands apart@xcite or increase the temperature to let thermal fluctuations do the work . fig . [ fig : example - pulling ] shows the result of applying an external opposing force to left most nucleotide pair . progressively the left most hybridization bond snaps . along with the breakage of hybridization bonds , we also see the gradual removal of green angular interactions and all the dihedral interactions . the external force is opposed by a single left most hybridization bond along with the angular and dihedral interactions across the gab . during the unzipping process , often the hybridization bonds are transiently reformed just after breakage if thermal fluctuations pull them within the hybridization reaction distance . time series of dna unzipping by a strong horizontal force @xmath100 applied to the left and right most beads of the two strands ( a - h ) . the snapshots corresponds to times @xmath101 , @xmath102 , @xmath103 ( top row ) , @xmath104 , @xmath105 , @xmath106 ( middle row ) , and @xmath107 , @xmath108 , @xmath109 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series of dna unzipping by a strong horizontal force @xmath100 applied to the left and right most beads of the two strands ( a - h ) . the snapshots corresponds to times @xmath101 , @xmath102 , @xmath103 ( top row ) , @xmath104 , @xmath105 , @xmath106 ( middle row ) , and @xmath107 , @xmath108 , @xmath109 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series of dna unzipping by a strong horizontal force @xmath100 applied to the left and right most beads of the two strands ( a - h ) . the snapshots corresponds to times @xmath101 , @xmath102 , @xmath103 ( top row ) , @xmath104 , @xmath105 , @xmath106 ( middle row ) , and @xmath107 , @xmath108 , @xmath109 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig : " ] time series of dna unzipping by a strong horizontal force @xmath100 applied to the left and right most beads of the two strands ( a - h ) . the snapshots corresponds to times @xmath101 , @xmath102 , @xmath103 ( top row ) , @xmath104 , @xmath105 , @xmath106 ( middle row ) , and @xmath107 , @xmath108 , @xmath109 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series of dna unzipping by a strong horizontal force @xmath100 applied to the left and right most beads of the two strands ( a - h ) . the snapshots corresponds to times @xmath101 , @xmath102 , @xmath103 ( top row ) , @xmath104 , @xmath105 , @xmath106 ( middle row ) , and @xmath107 , @xmath108 , @xmath109 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series of dna unzipping by a strong horizontal force @xmath100 applied to the left and right most beads of the two strands ( a - h ) . the snapshots corresponds to times @xmath101 , @xmath102 , @xmath103 ( top row ) , @xmath104 , @xmath105 , @xmath106 ( middle row ) , and @xmath107 , @xmath108 , @xmath109 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig : " ] time series of dna unzipping by a strong horizontal force @xmath100 applied to the left and right most beads of the two strands ( a - h ) . the snapshots corresponds to times @xmath101 , @xmath102 , @xmath103 ( top row ) , @xmath104 , @xmath105 , @xmath106 ( middle row ) , and @xmath107 , @xmath108 , @xmath109 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series of dna unzipping by a strong horizontal force @xmath100 applied to the left and right most beads of the two strands ( a - h ) . the snapshots corresponds to times @xmath101 , @xmath102 , @xmath103 ( top row ) , @xmath104 , @xmath105 , @xmath106 ( middle row ) , and @xmath107 , @xmath108 , @xmath109 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series of dna unzipping by a strong horizontal force @xmath100 applied to the left and right most beads of the two strands ( a - h ) . the snapshots corresponds to times @xmath101 , @xmath102 , @xmath103 ( top row ) , @xmath104 , @xmath105 , @xmath106 ( middle row ) , and @xmath107 , @xmath108 , @xmath109 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig : " ] in fig . [ fig : example - force - horizontal ] we perform another pulling experiment , where a much stronger horizontal force is applied to the left most bottom strand and right most top strand beads of the double strand . initially the whole molecule is sheared , as all the green angular interactions cooperate in opposing the deformation . gradually bonds snaps from either end towards the center . interestingly , since the two molecules have a 4-nucleotide long repeating sequence , when the hybridization bonds are broken , they very rapidly reform with the complementary beads one repeat sequence further down the molecule . the shear process repeats for the second hybridization sequence until it too is broken , and two single strands are formed . time series showing bubble opening and closing dynamics for dna at an elevated temperature @xmath36 ( a - h ) . the snapshots are from times @xmath110 , @xmath111 , @xmath112 ( top row ) , @xmath113 , @xmath114 , @xmath115 ( middle row ) , and @xmath116 , @xmath117 , @xmath118 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series showing bubble opening and closing dynamics for dna at an elevated temperature @xmath36 ( a - h ) . the snapshots are from times @xmath110 , @xmath111 , @xmath112 ( top row ) , @xmath113 , @xmath114 , @xmath115 ( middle row ) , and @xmath116 , @xmath117 , @xmath118 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series showing bubble opening and closing dynamics for dna at an elevated temperature @xmath36 ( a - h ) . the snapshots are from times @xmath110 , @xmath111 , @xmath112 ( top row ) , @xmath113 , @xmath114 , @xmath115 ( middle row ) , and @xmath116 , @xmath117 , @xmath118 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig : " ] time series showing bubble opening and closing dynamics for dna at an elevated temperature @xmath36 ( a - h ) . the snapshots are from times @xmath110 , @xmath111 , @xmath112 ( top row ) , @xmath113 , @xmath114 , @xmath115 ( middle row ) , and @xmath116 , @xmath117 , @xmath118 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series showing bubble opening and closing dynamics for dna at an elevated temperature @xmath36 ( a - h ) . the snapshots are from times @xmath110 , @xmath111 , @xmath112 ( top row ) , @xmath113 , @xmath114 , @xmath115 ( middle row ) , and @xmath116 , @xmath117 , @xmath118 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series showing bubble opening and closing dynamics for dna at an elevated temperature @xmath36 ( a - h ) . the snapshots are from times @xmath110 , @xmath111 , @xmath112 ( top row ) , @xmath113 , @xmath114 , @xmath115 ( middle row ) , and @xmath116 , @xmath117 , @xmath118 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig : " ] time series showing bubble opening and closing dynamics for dna at an elevated temperature @xmath36 ( a - h ) . the snapshots are from times @xmath110 , @xmath111 , @xmath112 ( top row ) , @xmath113 , @xmath114 , @xmath115 ( middle row ) , and @xmath116 , @xmath117 , @xmath118 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series showing bubble opening and closing dynamics for dna at an elevated temperature @xmath36 ( a - h ) . the snapshots are from times @xmath110 , @xmath111 , @xmath112 ( top row ) , @xmath113 , @xmath114 , @xmath115 ( middle row ) , and @xmath116 , @xmath117 , @xmath118 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig:"]time series showing bubble opening and closing dynamics for dna at an elevated temperature @xmath36 ( a - h ) . the snapshots are from times @xmath110 , @xmath111 , @xmath112 ( top row ) , @xmath113 , @xmath114 , @xmath115 ( middle row ) , and @xmath116 , @xmath117 , @xmath118 ( bottom row ) starting from a straight double stranded conformation at @xmath99.,title="fig : " ] dna can be molten by raising the temperature . the melting temperature depends on the sequence , the length of the strands as well as the strand concentration.@xcite prior to melting , bubbles of open nucleotide sequences appear since they contribute configuration entropy and hence lower the free energy similar to vacancies in crystals . at increased temperatures , the number and the size of these bubbles grow and cause the two strands to melt.@xcite in fig . [ fig : example - bubble ] we show a time series of a bubble , that is created by breaking a single hybridization bond , the bubble grows until it breaks the last hybridization bond . however , the two frayed strands form a hybridization bond at the end , and progressively the bubble closes again . simulating the chain for sufficiently long time at an elevated temperature will cause the double stands to melt with a transition very much like the one shown in fig . [ fig : example - bubble ] . we have implemented a versatile framework for studying the effects of dynamic bonding of ordinary and directional bonds in coarse - grained models within the context of the large - scale atomic / molecular massively parallel simulator ( lammps)@xcite . the dynamic bonding framework ensures that angular and dihedral interactions are kept consistent during bond breakage and creation . the code has been parallelized and optimized to the case where the bond formation or breakage probability for each bead is relatively low . since the dynamic bonding code is very modular it will be easy to extend with other types of bonding rules . the dynamic bonding framework was written with the aim of developing a new type of coarse - grained models of dna dynamics . we have illustrated a dynamic bonding dna model using dna hybridization and ligation , as well as two geometries of force induced unzipping and bubble dynamics . clearly the present dna model is very simple , nonetheless it qualitatively captures some of the fundamental phenomena of dna molecules . the dynamic bonding framework will allow us to build dna models , that we expect will provide quantitative predictions as good as the poland - scheraga model@xcite , while we can use these dna models as components in molecular dynamics and dissipative particle dynamics simulations of hybrid materials containing both soft - condensed matter and dna molecules . c.s . gratefully acknowledges discussions with h. fellermann , r. everaers , p .- a . monnard , m. hanczyc , and s. rasmussen . when newtons 3rd law is not applied to bonded interactions , lammps has an bond interaction table for each bead listing the other beads it is bonded to and the type of the bond . similar angular and dihedral interaction tables exists for each bead . lammps also has a neighbor structure where bonded neighbors , next nearest neighbors , and third nearest neighbors are stored . this information is derived from the bonding structure , and used to enable or disable non - bonded interactions between beads connected by up to three bonds . initially when lammps reads the control file to set up a simulation , the dynamic bonding fix is called to parse the entire set of rules such as those in fig . [ fig : fix - dna - dynamics ] . the rules and their parameters are sanity checked and stored internally in the fix . when the simulation is initialized , the dynamic bonding framework starts by having each simulation domain count how many bonds of each type each reactive bead has . then at a specified frequency the code does : 1 . forward communication of ghost particle positions to neighboring nodes and the table of bond counts . this is required for testing distances and for applying maximum rules . 2 . creation nomination . each reactive bead can nominate a single preferred bonding partner . the search for partners is performed over all beads in the reactive group and each creation rule is tested in succession . the test of rules is done in the order they are specified , and if more than one rule match the same bead pair , the last matching rule will apply . the search is over all non - bonded beads and optionally over beads 2 or 3 bonds away from the current bead . for each bead pair and creation rule , their types are tested and if they within the maximum reaction distance . beads that already have the maximal number of bonds of the type , that would be produced by the current rule are discarded . of all the potential bonding partners , the closest partner in the same simulation domain ( if any ) is nominated for bonding . 3 . bond breakage nomination . each reactive bead can nominate a single preferred partner to break an existing a bond to . the search for partners is performed over all beads in the reactive group and each bond break rule is tested in succession . the test of rules is done in the order they are specified , and if more than one rule match the same bead pair , the last matching rule applies . for each bead pair and bond breakage rule , it is tested if the bond between them has the specified type , and if they are further apart than the minimum bond breakage distance . of all the potential bond breakage partners , the partner most distant in the same simulation domain ( if any ) is nominated for bond breakage . bond conversion is internally represented as a bond pair that nominates each other for a bond breakage and creation of the new bond . hence bond conversion over rules both bond breakage and creation in case they occur simultaneously . 4 . communication . the nominated partners are distributed to and aggregated across neighboring simulation domains and the closest partner is chosen for creation and the most distant partner is chosen for bond breakage . information about which rule lead the nomination of each partner are also distributed along with a random number for stochastic bond breakage and a random number for stochastic bond creation . 5 . bond breakage . if any killbond rules are defined , all beads check , if they are part of a bond longer than the cut - off distance , and if that is the case then the bond is marked for removal . if two bonds nominate each other as bond breakage partners , then bond breakage is attempted . each bead contributes a uniform random number for bond breakage , these are averaged and compared to the specified bond breakage probability . in case the random number is smaller than the probability , the bond is marked for removal . this ensures that beads on different simulation domains makes the same random choice . when bonds are marked for removal the bond type in the corresponding entry in the bond interaction tables is set to -1 . if a maximum rule applies to that particular bond and bead type , the table of bead functionalities is also updated . the outdated neighbor structure is retained . removing angular and dihedral interactions . to ensure parallelism , each reactive bead is alone responsible for all its angular and dihedral interactions . if a bond has been broken in its local neighborhood , the bead has to remove any angular and dihedral interactions involving that bond . this is done by generating all non - cyclic paths of length three and four either starting at or crossing the present bead using the outdated neighbor structure ( which still contains the broken bonds ) . the beads checks each path for bond breakage events ( using the bond interaction tables , which shows if a bond has been marked for breakage ) . if a path involves a broken bond , then the bead removes the corresponding entry in its angular and dihedral interaction tables , if they exist . the lammps neighbor structure is updated , and the broken bond entries are removed from the bond interaction tables . if no bonds are to be created , we can jump directly to 10 . . bond creation . if two bonds nominate each other as bond creation partners , then an attempt is made at creating the bond . each bead contributes a uniform random number for bond creation , these are averaged and compared to the specified bond creation probability . again this ensures the same random choice for beads residing in different simulation domains . the new bond is added to the bond interaction table for the bead . the neighbor structure is also updated . if a maximum rule applies to the bond and bead type , the table of bead functionalities is also updated . creating angular and dihedral interactions . again each reactive bead is responsible for determining if a bond was created in their local neighborhood . this is done the same way as angular and dihedral interactions are removed . since the neighbor structure now contains the new bonds , we can generate non - cyclic paths of length three and four starting at or crossing the present bead using the updated neighbor structure ( which now contains the new bonds ) . each path is checked for bond creation events using the bond interaction tables . if the bead determines that it is part of a new triplet or quartet of beads , then it compares the bead types and directional bond types with all the angular and dihedral creation rules . if a match is found , then the bead adds the corresponding interaction to its interaction table . 10 . distribution of statistics of the total number of bonds , angles , dihedrals introduced and removed in the current time step . since bond creation requires a distance check , the lammps pair communication distance should be at least the longest reaction distance , otherwise bonds will only be created between bead pairs within the communication distance from each other . since the implementation also depends on all beads knowing about all their bonded , angular , and dihedral interactions , it will not work without newtons 3rd law being disabled for bonded interactions . this is also required for the implementation of directional bonds . the dynamic bonding framework transparently handles symmetric bonds , hence they are just special cases of directional bonds . the dynamic bonding code is optimized to the situation , where the density of reacting beads is so low that at most one bond breakage and bond creation event is likely to occur per bead per time step . for instance , the match making algorithm does not attempt to make matches between rejected partners , that could still be eligible for bond breakage or bond creation rules . nor does the match making algorithm attempt to pick the most likely of multiple possible reaction path ways . for instance , if multiple bond creation rules applies to a single bead , then only the last nominated bond creation partner is stored . hence a creation rule with a low reaction probability can overwrite the bonding partner nominated by a prior creation rule with much higher reaction probability . in this case , the high probability reaction will never happen . similar issues apply when multiple bond break rules involve the same bead . since the bond conversion rules are implemented as bond deletion followed by bond creation , these can interfere with both bond creation and bond breakage rules . killbond rules are completely safe , since they are not implemented using the match making algorithm . for the dna model , none of these caveats apply . * program summary * _ manuscript title : lammps framework for dynamic bonding and an application modeling dna . _ + _ authors : carsten svaneborg _ + _ program title : lammps framework for directional dynamic bonding _ + _ journal reference : _ + _ catalogue identifier : _ + _ licensing provisions : gpl _ + _ programming language : c++ _ + _ computer : single and multiple core servers _ + _ operating system : linux / unix / windows _ + _ ram : 1 gb _ + _ number of processors used : single or parallel _ + _ keywords : dynamic bonding , directional bonds , molecular dynamics . _ + _ classification : 16.11 polymers , 16.13 condensed - phase simulations _ + _ nature of problem : simulating coarse - grain models capable of chemistry e.g. dna hybridization dynamics . _ + _ solution method : extending lammps to handle dynamic bonding and directional bonds . _ + _ unusual features : allows bonds to created and broken while angular and dihedral interactions are kept consistent . _ + _ running time : hours to days _ 10 url # 1`#1`urlprefixhref # 1#2#2 # 1#1 b. r. brooks , c. l. brooks , iii , a. d. mackerell , jr . , l. nilsson , r. j. petrella , b. roux , y. won , g. archontis , c. bartels , s. boresch , a. caflisch , l. caves , q. cui , a. r. dinner , m. feig , s. fischer , j. gao , m. hodoscek , w. i m , k. kuczera , t. lazaridis , j. ma , v. ovchinnikov , e. paci , r. w. pastor , c. b. post , j. z. pu , m. schaefer , b. tidor , r. m. venable , h. l. woodcock , x. wu , w. yang , d. m. york , m. karplus , charmm : the biomolecular simulation program , j. comput . phys . 30 ( 2009 ) 1545 .

quantum information processing devices need to be robust and stable against external noise and internal imperfections to ensure correct operation . in a setting of measurement - based quantum computation , we explore how an intelligent agent endowed with a projective simulator can act as controller to adapt measurement directions to an external stray field of unknown magnitude in a fixed direction . we assess the agent s learning behavior in static and time - varying fields and explore composition strategies in the projective simulator to improve the agent s performance . we demonstrate the applicability by correcting for stray fields in a measurement - based algorithm for grover s search . thereby , we lay out a path for adaptive controllers based on intelligent agents for quantum information tasks .

1407.1535

when building devices for quantum information processing one has to take changing environment conditions and device imperfections into account . it is therefore necessary to include adaptive mechanisms that characterize and calibrate the device from within . furthermore , it is desirable for these devices to obtain a certain degree of autonomy in maintaining their functional state despite detrimental environment influences , in particular , when they are assembled to a larger quantum information processing infrastructure . in the attempt to miniaturize current implementations of quantum devices , we will reach the point where these devices will be of microscopic scale and require short reaction times . for such microscopic systems we can no longer assume that their internal controllers are full - fledged universal computers that can carry out arbitrary programs . instead , controllers will be small physical systems that are specialized for their respective purpose with a program that emerges from the controller s analog dynamics . in this paper we explore the applicability of a controller in form of an intelligent learning agent that has access to a _ projective simulator _ @xcite . within this agent framework , the aim is to demonstrate adaptive calibration and compensation strategies against stray external fields when carrying out quantum information tasks . the agent shall thereby implement a simple form of adaptive error avoidance and implicit parameter estimation . algorithms from machine learning have been used to find strategies for parameter estimation , and optimal strategies for parameter estimation are known for specific cases , see e.g. @xcite . here , however , we focus on strategies that arise naturally from the adaptive dynamics of the underlying physical system , for which we choose a projective simulator . the projective simulator is a platform that has been proposed as a physical model for reinforcement learning @xcite , and it effectively reproduces input output reward correlations from an internal adaptive stochastic process . with the restriction to this particular system , one can not hope for the best possible strategy to emerge while keeping the rules governing the dynamics reasonably simple and computational overhead low . both requirements are necessary to allow for an actual physical realization . as an additional feature , the projective simulator offers a natural route to quantization as indicated in @xcite and thereby a way to intelligent agents that benefit from internal quantum dynamics , as demonstrated in the reflective quantum projective simulator @xcite . agent quantization is not explored further in the present work as we focus on the application of a classical agent to quantum information processing first . for recent comprehensive reviews in the domain of quantum physics and artificial intelligence or machine learning see @xcite and @xcite . as illustration of our method of adaptive quantum information processing we study grover s quantum search algorithm @xcite in the paradigm of measurement - based quantum computation . grover s algorithm provides a fast way to find a marked item in an unsorted database with @xmath0 elements . in particular , it provides a quadratic speed - up with @xmath1 database look - ups over a search by means of a classical computer with @xmath2 look - ups . first proof - of - principle implementations of grover s algorithm with nuclear magnetic resonance techniques @xcite and entangled photons @xcite employed the circuit model of quantum computation , where individual unitary quantum logic gates are applied to a register of qubits to process information . measurement - based quantum computation ( mbqc ) @xcite is a different paradigm of quantum computation , where the computation is carried out by measuring single qubits of an initially highly entangled resource state @xcite . the first experimental demonstration of mbqc in a system of entangled photons @xcite ( and with trapped ions @xcite ) also demonstrated the grover algorithm in its smallest realization with a database of 4 entries ( 2-qubits ) by using a 4-qubit cluster state as computation resource . as preparation for the full measurement - based algorithm we first study a basic setting . we situate a quantum system , a single qubit , in an unknown external magnetic field . an artificial agent , the controller , is endowed with a projective simulator and the ability to measure the quantum system and thereby prepare quantum states . we hardwire the learning process , i.e. , the update rule in the reinforcement learning process of the projective simulator , such that the agent effectively carries out the following tasks : ( i ) adapt measurement directions to changes of the external magnetic field , and dynamically improve the sensing resolution . ( ii ) learn to adapt simultaneously for multiple measurement directions needed for general mbqc - algorithms in a feedback scheme . ( iii ) carry out a quantum information task , the grover algorithm @xcite in the setting of measurement - based quantum computation , with unknown stray magnetic fields . this provides a completely worked - out example , starting from the physical system that generates the actions of an adaptive `` intelligent '' agent , here a projective simulator , to a controller tailored to a specific quantum information task , e.g. measurement - based grover s search algorithm . first , we describe an approach that allows the projective simulator to effectively obtain a notion of the strength of an external magnetic field and hence carry out a primitive form of parameter estimation . however , there is a conceptual difference between our approach and parameter estimation . after the agent has learned , the information on the strength of the magnetic field will _ not _ be available as a number that the agent gives as an output . instead , this information is only indirectly incorporated into the dynamics and decision patterns of the agent , and it can be exploited to _ do _ certain things that are adapted to the external field . therefore , we will analyze the learning process of the agent from two different perspectives : from an _ operational _ perspective we characterize how well the agent adapts its actions to the external field , and from an _ informational _ perspective we quantify how much of the information about the external field is really contained in the parameters that define the dynamics of the agent . we start with a detailed description of the setting , that is , of the agent and its interaction with the measurement apparatus , the dynamics of the projective simulator , and an analysis of the learning process . in the present setting the magnetic field direction is promised to be fixed along the @xmath3-axis , and the agent needs to estimate its strength @xmath4 . the following steps are visualized in figure [ fig : agentsetup ] . the agent starts by preparing a single qubit in the state @xmath5 , which in the presence of the field evolves according to the hamiltonian @xmath6 , where the frequency @xmath7 is proportional to the magnetic field strength @xmath4 . after some fixed time interval @xmath8 , the initial state has evolved into @xmath9 up to a global phase , with @xmath10 . estimating the field strength @xmath4 amounts to estimating @xmath11 and obtaining information about the angle @xmath12 between this state and the initial state in the equatorial plane of the bloch sphere . in a linear optics setup @xcite , the unknown angle @xmath12 would correspond to an unknown phase shifter in the beam line . the agent measures the qubit in the unknown state @xmath11 in various directions and incorporates the measurement outcomes to change its choice of measurement directions . the measurements applied by the agent are in general described by povms @xcite . for simplicity , we will restrict our analysis to projective measurements . we shall comment on the general case at the end of the paper . the challenge is to effectively realize a probability distribution for the unknown angle @xmath12 without explicitly performing computations and analyzing the measurement data . rather it should emerge dynamically as the result of a feedback loop by reinforcing certain actions on the quantum system . therefore , we choose an approach where the internals of the agent are wired such that it tries to optimize the direction of a measurement . in the optimal case @xmath11 is the @xmath13 eigenstate of this measurement . qubit observables whose eigenstates with eigenvalues @xmath14 lie in the equator of the bloch sphere are given by @xmath15 where @xmath16 is of the form with angle @xmath17 . both eigenstates lie on opposite sides of the equator . the probability to obtain the measurement outcome @xmath14 is @xmath18 that is , the closer the angles @xmath17 and @xmath12 the higher is the probability to obtain the @xmath13 measurement outcome . to simplify notation we often consider the projector onto the @xmath13 eigenstate @xmath19 instead of the observable @xmath20 , and measurements of the projector with outcomes 1 and 0 . for qubits , measuring @xmath21 gives the same statistics of measurement outcomes and resulting states as measuring the observable @xmath20 because there is a unique state orthogonal to @xmath16 . the projective simulator inside the agent employs an adaptive stochastic process that is modeled by a random walk of an excitation in a network of so - called `` clips '' @xcite . for now the clip network takes the form of a directed weighted graph depicted in figure [ fig : network_1to4 ] . the random walk starts at the only `` percept clip '' , which is excited by an internal trigger of the agent with a time interval @xmath8 after the qubit has been prepared ( cf . figure [ fig : agentsetup ] ) . the excitation propagates in the network according to the weights of the links that connect the percept clip to the action clips . once the excitation reaches an action clip , the corresponding action is performed and the process inside the projective simulator is finished . a single action is a measurement of a certain @xmath21 at the qubit . if the measurement outcome is @xmath13 it is fed back as reward to the agent to re - enforce and strengthen the link between the percept clip and the last action clip . the process is repeated for the next measurement . the probabilities to select certain measurements , however , change as a result of previous measurement outcomes . this makes measurements with angles @xmath17 closer to @xmath12 more likely . these probabilities in effect represent a coarse - grained , discrete probability distribution over angles @xmath12 . -clip , which then undergoes random walk dynamics according to the weights of the links . the action clip where the excitation arrives determines the measurement direction . ] in detail , each link in the clip network carries a weight @xmath22 . the probability to jump from the percept clip `` @xmath23 '' to the action clip corresponding to @xmath21 is given by the normalized weight of all edges from the percept clip , that is , @xmath24 at the beginning of the learning process all weights are initialized with @xmath25 . after the measurement of @xmath21 in the @xmath26-th round , the measurement outcome ( 0 or 1 ) is rescaled by a factor @xmath27 and fed back into the projective simulator as a reward @xmath28 to the transition with weight @xmath22 . regardless of whether or not a transition has been taken , all weights are damped by a small amount with rate @xmath29 . after the @xmath26-th round , in which @xmath17 was the measurement angle , all weights are changed according to the following update rule : @xmath30 as a result , the projective simulator converges to a state ( set of @xmath31-values ) that increases the chances of obtaining @xmath13 measurement outcomes and thereby increases the probability to measure in directions close to @xmath12 . from the perspective of the projective simulator only an outcome @xmath13 denotes success because the action that led to this outcome will be reinforced . this `` subjective '' success probability is @xmath32 an action that leads to a reward ( measurement result @xmath13 ) is also the correct action from an operational point of view . the transition probabilities @xmath33 provide an internal representation of a discretized probability distribution for the angle @xmath12 . the change of @xmath34 as a function of the number of rounds ( measurements on the quibt ) is depicted in figure [ fig : learning_single_phi ] for several examples of @xmath12 . the results in figure [ fig : learning_single_phi ] show that the agents learns to obtain rewards more often and thus obtains information about the state @xmath11 and thereby about @xmath4 . in our example we start with 4 projectors at angles every @xmath35 , which corresponds to the projectors onto the eigenstates of the observables that are given by the pauli matrices @xmath36 and @xmath37 . if @xmath38 , measurements of @xmath39 will always give outcome @xmath13 and hence be rewarded . the two adjacent projectors at @xmath40 and @xmath41 are rewarded in half of the measurements , and measurements in the direction @xmath42 are never rewarded . in this situation the projective simulator builds a strong link to @xmath39 , somewhat less strong links to @xmath43 and @xmath44 and leaves the link for @xmath45 at its initial value . the coarse - grained discrete probability distribution for @xmath12 is consequently peaked at @xmath38 and within statistical fluctuations symmetric around this direction ( figure [ fig : learning_single_phi ] top left inset ) . if @xmath12 is between two of the projectors , say @xmath46 , measurements of @xmath39 and @xmath43 will only be rewarded with only @xmath47 probability , and measurements of the opposite projectors with @xmath48 probability . the distribution of measurement probabilities will also be symmetric around the direction @xmath49 but less pronounced as shown by a broader distribution in figure [ fig : learning_single_phi ] ( bottom left inset ) . a broad distribution for measuring in the direction @xmath17 results in a lower success probability for angles that have a large distance to all projectors , e.g. , @xmath50 . at this point a smaller damping rate @xmath29 and a larger multiplier of the rewards @xmath27 both lead to a larger value of rewarded transitions in the steady - state and hence to a larger success probability and a probability distribution that is more peaked . at the same time increasing both @xmath27 and @xmath29 speeds - up the learning process leading to learning curves with a steeper initial rise . note , however , that extremal cases with too large rewards or too weak damping favor situations in which the agent prefers actions that just by luck led to a reward in the past although they are not highly rewarded on average . un - learning such an initial `` misunderstanding '' and building a probability distribution that reflects the actual probabilities of being rewarded may take a long time . this aspect leads to larger fluctuations in the success probability of an ensemble of agents and a slower final convergence . _ asymptotic success probability._for the asymptotes of the success probability we can find a first - order approximation by assuming a steady state of the transition probabilities @xmath51 and the respective @xmath31-values . the resulting steady - state success probability is @xmath52 . when coarse - graining over many measurements the time average of the reward for each action is given by @xmath53 , and the steady - state probability to measure in direction @xmath17 is @xmath54 . with these assumptions the update rule turns into a set of coupled equations for the steady - state values @xmath55 , @xmath56 in which the loss terms given by the damping @xmath29 and the gain terms given by the time - averaged reward are in equilibrium . this set of nonlinear equations can be solved numerically and yields a very good approximation for the ensemble average as seen in figure [ fig : learning_single_phi ] . the asymptotic value obtained in this approximation only depends on the ratio @xmath57 . : states are summarized as the position of the resulting vectors @xmath58 for each agent @xmath59 in the complex plane ( equatorial plane ) . the degeneracy with respect to the expected reward for @xmath60 provides a manifold of equally successful states , which is populated by the ensemble . blue dot on the unit circle gives the angle of the ensemble average , the black dot is the angle @xmath12 . ] ( left ) the action pair @xmath39 and @xmath43 is degenerate with respect to the expected reward . for @xmath61 ( middle ) , i.e. , not exactly between two projectors , the agent measures more often into the direction @xmath62 . fluctuations in the measurement probabilities do not necessarily show in the success probability . for comparison , the ensemble averages of 1000 agents after 1000 measurements are given as dashed lines . larger rewards @xmath27 and damping @xmath29 ( both rescaled by a factor 10 ) decrease the timescale of the fluctuations while maintaining approximately the same time average ( right ) . the agent jumps between different preferred action and stays for extended times . ] _ time average vs. ensemble average._the fluctuations of the probabilities to choose certain actions ( see insets in figure [ fig : learning_single_phi ] ) show that even after 1000 iterations of the update rule not all agents have converged to a single state ( figure [ fig : ensembleaverage ] ) . many steady states occur if there is a whole manifold that is rewarded equally , that is , when two or more actions have the same expected reward . for example , for @xmath60 both actions @xmath39 and @xmath43 have equal chances of being rewarded and thus the there is no preference of either action as long as one of them is carried out . actions that have the same expected reward span a subspace for which the sum of the probabilities for doing these actions is approximately constant in the steady state , however , their relative ratio is not . the states of the whole ensemble of agents fills this degenerate subspace of action probabilities . the ensemble average yields an approximation of @xmath12 . although the ensemble has learned , i.e. , the success probability has converged , the dynamics of each individual is not necessarily converged to a single state where it remains . in the course of time the state of a single agent explores the whole degenerate reward manifold while keeping the success probability constant as we numerically illustrate in figure [ fig : timeaverage ] . we find that the time average of a single agent for long times equals the ensemble average because the state of the single agent assumes all the different steady states that an ensemble produced after short time as in figure [ fig : ensembleaverage ] . hence , to obtain an ensemble average a snapshot after a relatively short time is sufficient . however , if there is a degenerate space in the reward scheme , the state of a single agent at a fixed time gives only an imprecise estimate of @xmath12 , and even its time average does when considered only for a short time . a larger damping parameter @xmath29 and higher rewards @xmath27 facilitate a faster exploration of the degenerate reward manifold and thus provide a better time average for a single agent for shorter times . for @xmath60 in figure [ fig : timeaverage](right ) , the agent selects either @xmath62 or @xmath35 for an extended time and then suddenly switches between these equally rewarded choices . this jumping behavior occurs for large reward and damping , whereas for smaller values ( left ) also equal probabilities occur for longer durations . the way that the agent uses the rewards to change its actions to do measurements more often along angles that are close to @xmath12 , is a way of representing information about @xmath12 . we regard this probability distribution of actions along the discrete set of angles as a probability distribution of @xmath12 @xcite , and compare it to standard computational analysis procedures employed in state and parameter estimation . by its actions and the returned rewards the agent effectively samples the reward distribution @xmath63 . the same data , namely the measurement direction and outcome , however , can also be used in a bayesian update rule to explicitly build a probability distribution @xmath64 , or the data can be used to reconstruct the state @xmath11 via state tomography . we compare the angular distribution of actions that the agent maintains to the angular distribution that a bayesian update would produce , and also to a simple state tomography by estimating expectation values from the same measurement data . a simple form of state tomography can be done by calculating the expectation values @xmath65 and @xmath66 from the measurement results of the four projective measurements . together with the initial assumption @xmath67 , these expectation values give an approximation of the state s bloch vector . our four measurement directions @xmath68 give the same measurements as the pauli matrices with expectation values @xmath69 where expectation values of the observables can be related to those of the projectors by @xmath70 for a total of @xmath71 measurements , of which @xmath72 are done in direction @xmath17 , with individual measurement outcomes @xmath73 for observable @xmath20 , the expectation values can be approximated with the mean @xmath74 the resulting bloch vector with coordinates @xmath75 provides an angle with the @xmath76-axis and thereby an estimate of @xmath12 . in a bayesian analysis , we update an initially flat prior distribution @xmath77 with the information obtained from each measurement . after each measurement , the distribution is updated with result @xmath78 for measurement in direction @xmath79 , e.g. , for the first update @xmath80 where we include the knowledge of quantum mechanics and the statistics of measurement outcomes for the underlying system with @xmath81 given by . after @xmath82 measurements the resulting probability distribution is @xmath83 with normalization @xmath84 . for an efficient update and a compact representation of the conditioned probability distribution we expand it in a fourier series , which has at most @xmath82 higher harmonics , and construct an recursive update rule for the expansion coefficients following the approach in @xcite for parameter estimation with a single fixed observable but variable time delays . for our choice of measurement directions , with @xmath17 being a multiple of @xmath35 , the fourier expansion generally contains @xmath85 and @xmath86 terms . the recursive update rules for the expansion coefficients are given in the appendix . . the data points represent the measurement data of 10 agents from 1500 measurements each , and they give an estimate of the angle @xmath12 as the mean of the 10 distributions . the 10 data points are supplemented by a black error bar , which indicates their circular mean and circular standard deviation . for the projective simulator , the blue error bar indicates the circular mean and circular standard deviation of the discrete probability distribution over the 4 actions , after averaging over all 10 agents . for most examples of @xmath12 the projective simulators generate a distribution of the mean angles that coincide with @xmath12 except for @xmath61 , where similar as in figure [ fig : ensembleaverage ] a bias towards the nearest available projector ( @xmath62 ) occurs . ] to compare the estimates of @xmath12 by these three approaches , we fix the angle @xmath12 , and let 10 agents with a projective simulator do 1500 measurements each and according to the dynamics arising from using the projective simulator . after these 1500 measurement each agent has built a probability distribution of actions @xmath33 , we take the mean of each distribution as the estimate of @xmath12 . figure [ fig : distributioncomparison ] shows these estimates as the blue data points . because the probability distributions for @xmath12 that we obtain from the @xmath33 have support only on 4 angles , which are uniformly and discretely spaced on the circle , each distribution has a large variance . we average the distributions of all 10 agents and give the mean and circular standard deviation of the resulting distribution as the blue error bar in figure [ fig : distributioncomparison ] for comparison . clearly , when @xmath12 is close to one of the possible choices of @xmath17 , the projective simulator captures @xmath12 accurately , but for values of @xmath61 or @xmath49 the estimates are biased towards one of the @xmath17 as in figure [ fig : ensembleaverage ] . for @xmath60 the angular means are widely spread , and their distribution has a large variance , which is reminiscent of the distributions given in the insets in figure [ fig : learning_single_phi ] and the distribution of means of a large ensemble in figure [ fig : ensembleaverage ] . the estimates for @xmath12 obtained from the expectation values of the pauli matrices , i.e. , the simple state tomography , are calculated from the same measurement record for each agent and are given by the orange data points in figure [ fig : distributioncomparison ] . for the bayesian update scheme , we construct the conditional probability distributions for @xmath12 , again from the same measurement record that the projective simulator generated . all of the resulting distributions assume an approximate gaussian shape with a narrow peak ( @xmath87 ) . the means are given as red data points in figure [ fig : distributioncomparison ] . both approaches can estimate @xmath12 correctly within the error bars . surprisingly , for @xmath12 along one of the @xmath17 , the estimates from the expectation values spread more than in the other two approaches . the reason is that in these cases the projective simulator samples most of measurements along a single direction and only few for the other observable , which causes a rather large uncertainty in one of the coordinates . although state tomography and bayesian estimation perform generally equally good or better than the projective simulator , the big conceptual difference between these approaches is that very little knowledge of quantum physics and measurement statistics is build into the projective simulator . the projective simulator does not assume that the rewards originate from measurement probabilities of a quantum state and , therefore , it is `` model free '' . the update rule causes a learning dynamics that drive the agent to measure more often into directions that give a + 1 measurement outcome and thereby implicitly align measurement directions with @xmath12 . even when no optimal measurement direction is available the agent learns how to deal with a system such that reward is most likely to occur . in principle , it could even adapt to artificial situations , where measurements along the @xmath76-axis always give a @xmath13 outcome and measurements along @xmath88 always give the outcome @xmath89 , something which can not be explained by measuring a qubit in a defined fixed state . therefore , it is not surprising that methods that make use of additional information , namely measurement probabilities predicted by quantum physics , can extract more information about @xmath12 from the measurement results . given that an agent with a projective simulator lacks this additional information it does comparably well , and , conceivably , it can be improved by changing the update rule to incorporate more knowledge about the underlying quantum physics . for example , a positive measurement result and reward in one direction can be combined with a negative reward into the opposite direction , or , for each measurement result the reward is distributed according to how close all potential actions are to the rewarded one . an important feature of the projective simulator is its ability to forget and thus to adapt to a changed situation . this ability distinguishes the present setting from schemes of parameter estimation , where the unknown parameter is assumed to be constant . for example , for a changing parameter standard bayesian updating can not be applied because past information needs to be disregarded and only recent information should be considered for estimating the current parameter . the projective simulator , in contrast , keeps track of an integrated average of past rewards for each action and is endowed with an element , the damping quantified by @xmath29 , to forget these rewards . the agent has the ability to completely change its behavior regardless of what has been rewarded earlier and irrespective of its earlier state . we shall consider two of such relearning scenarios in the following . we analyze the relearning by means of two quantities , the asymptotic success probability and the learning time it takes the agent to adapt . leading to a shift from @xmath12 to @xmath90 . an ensemble of 1000 agents first learns with @xmath12 for @xmath91 measurements , after which the angle changes to @xmath90 for another @xmath91 measurements , @xmath92 , @xmath93 . * top : * examples of learning curves for 4 different switches . @xmath12 and @xmath90 are given in the same color as the corresponding learning curve ( @xmath94 for blue and @xmath95 for the other examples ) . * middle : * the asymptotic success probability ( analytically obtained ) shifts due to the change in angles . * bottom : * the relearning time @xmath96 to reach 90% of the asymptotic success probability after the field has changed also gives rise to a periodically repeating pattern and shows a structure commensurate with the choice of projectors in intervals of @xmath35 . ] _ relearning after a switched field._ changes of @xmath4 result in a different @xmath12 and require the agent to adapt and relearn . for a single sudden change in @xmath4 the angle @xmath12 changes only once at a certain time to a new angle @xmath90 . depending on the values of @xmath12 and @xmath90 the agents shows a rich landscape of relearning patterns as illustrated in figure [ fig : relearning ] . after the switch the asymptotic success probability is always that of the new @xmath90 and may lie above or below the success probability of the old @xmath12 ( figure [ fig : relearning ] top ) . the change in @xmath97 is illustrated in figure [ fig : relearning ] ( middle ) . the sudden drop or increase in success right after the change of the angle and the time to reach a success probability depends strongly on the relation of the two angles and how much of the internal state ( @xmath31-values ) needs to be changed to reach the new state . these effects in the relearning time appear in addition to the known effects of changing the reward scaling @xmath27 and damping rate @xmath29 @xcite . a summary of the relearning times and change in asymptotic efficiencies is given in figure [ fig : relearning ] ( middle and bottom ) . _ time - dependent fields._an important feature for applications is the agent s ability to adapt its actions to slowly changing external fields . an agent s state is the result of a dynamical equilibrium between rewarded actions in the past and forgetting this information on a time scale given by @xmath29 . therefore , the speed at which an agent can adapt is limited by the speed with which it can modify its internal state . the agent can adapt to a change in the reward landscape caused by a changing field as long as it has enough time to sample the modified reward landscape and modify its internal state accordingly , which depends on @xmath27 and the timescale given by @xmath29 . figure [ fig : timedependentfields ] shows two examples . the first example ( left ) is a setting with a fast oscillating field , i.e. , one with @xmath98 as a function of the measurement round @xmath26 , where only the time average is learned because the agent effectively takes samples from the entire reward landscape . the state vector converges to angle @xmath38 and reaches almost unit length . the second example ( right ) shows a setting with a linearly increasing magnetic field , giving rise to @xmath99 . as @xmath12 moves anticlockwise on the unit circle as a function of the measurement round , the agent can keep up as quantified by @xmath100 with a state trajectory that also moves counterclockwise albeit with a slight delay and the length of the state vector is longer , i.e. , the field is learned better , for a slower rate of change . as averaged over an ensemble of 1000 agents . the state vector of the ensemble starts in the origin . * left : * for fast oscillating fields @xmath101 with @xmath102 ( red ) and @xmath103 ( blue ) the agents adapt to the average angle @xmath38 over 5000 measurements . data points are explicitly indicated for the first 20 measurements and joined by a line . * right : * linearly drifting fields @xmath104 can be learned by the agent the better the slower they change on the timescale of the learning time : @xmath105 ( blue ) shown for 10000 measurements , and @xmath106 ( red ) , @xmath107 ( orange ) shown for 4000 measurements each . the ensemble follows the field and the state trajectory converges to a limiting cycle . ] the choice of projectors that the agent can measure affects the agent s success in two ways : on one hand it fixes the available angles and thereby ability to measure the correct angle . a finer grained sample of measurement angles is beneficial because it will contain an angle that is closer to the actual angle and allow for almost perfect measurements . it also avoids efficiency minima due to the coarse - graining as they appear in figure [ fig : learning_single_phi ] ( top , right ) . a finer resolution of measurement angles , however , will introduce many angles that are almost equally successful and are hard to distinguish by their average reward . on the other hand , the choice of measurement angles fixes the discrete support on which the probability distribution for @xmath12 can be built , which contains the information on the angle @xmath12 . a drawback of a coarse - grained support is the arising large variance in the distribution . a fine - grained support , however , needs lots of sampling to evaluate each individual point in the distribution . measurement directions initially available to an agent ( blue ) , uniformly spaced on the circle , and with additional bisection compositions ( red ) . data points are angular averages , and the vertical region denotes the maximum and minimum of the success probability . these data summarize the angular dependence of @xmath97 as depicted for several examples in the inset . the reason for a decrease of @xmath97 with an increase in measurement directions is due to damping . ] as there are advantages and disadvantages to the number of measurement directions , we ask if there is an optimal number of fixed projectors . in order to distinguish directions on a circle , at least three directions are needed . ( for just two directions that are equally successful , it is not possible to decide which of the two angles between those two directions is the correct one . ) we have calculated the asymptotic success probability for an agent that has access to @xmath108 measurement angles that are equidistantly spaced on the unit circle , i.e. , it can measure @xmath109 observables of the form with two eigenstates each that are opposite on the equator of the bloch sphere . the angular dependence of the success probability in figure [ fig : initialprojectorscomposition ] ( inset ) shows maxima at angles that are measurement directions and minima between two neighboring angles . as more observables are added , the worst cases in between two neighboring projectors improve , but at the same time also the optimal cases decrease because the optimal angle is not chosen as frequently due to slightly better neighboring angles that are also rewarded more often . the success probability averaged over all angles first increases and then decreases as summarized in figure [ fig : initialprojectorscomposition ] . in the limit of large @xmath109 the success probability converges to 50% because of the constant damping @xmath29 . in the example case with @xmath92 and @xmath93 we can give these recommendations : for optimizing the best case , the number of 2 projectors is optimal , for the best worst case success probability , 8 projectors are best , and for the best average success probability 6 projectors are the best initial choice . a strategy to mitigate the decrease in overall efficiency for a more refined angular resolution is _ composition _ , which is one of the original features of projective simulation @xcite . with composition the projective simulator is endowed with the ability to generate new clips based on the composition of already existing ones . for parameter estimation the projective simulator can insert new clips with new measurement directions only where additional resolution is needed . the composition mechanism is an additional dynamical element in the projective simulator . based on the state of the projective simulator it is triggered and inserts a new clip based on existing ones . these new elements , i.e. , the trigger mechanism , constructing the new clip , and how the new clip is inserted into the network must be specified and leave room for arbitrarily complicated rules . we will restrict to the simplest mechanisms , which will also draw some intuition from actual conceivable physical dynamics . _ bisecting composition._the first composition mechanism simply operates by bisection and refining the resolution in the relevant regions . after the agent has learned with its initial set of projectors , the two actions clips with the largest @xmath31-values are selected and used to compose a new clip between the two . in situations with angles @xmath60 or @xmath110 the action clips with @xmath62 and @xmath35 will have the largest @xmath31-values and give rise to the creation of a new clip with @xmath50 , which improves the resolution of the discretization in the first quadrant . the success probability before and after one such composition is depicted in the inset in figure [ fig : initialprojectorscomposition ] as light blue and red curve , respectively . for the angle @xmath60 the success probability is increased from a minimum of 83.4% to a maximum with 96.2% without adding unnecessary projectors in the remaining quadrants , which would lower @xmath97 to 93.2% of the curve with 8 projectors . when always adding a single additional angle in the middle of the quadrant in which @xmath12 lies , worst case scenarios for @xmath97 appear only for angles like @xmath110 and @xmath111 with 93.32% , which still is a slight improvement over the coarse graining with only 4 projectors ( 93.26% ) . for @xmath61 the composition at @xmath50 is helpful but suboptimal . for angles at the projectors , e.g. @xmath38 , an additional composition is harmful and decreases @xmath97 from 97.1% to 96.1% . a single composition that doubles the angular resolution in one quadrant is qualitatively similar to 8 initial measurement angles , but with a higher success probability . a second composition step that adds another projector with an angle of odd multiples of @xmath110 , effectively reproduces the resolution of 16 initial angles but only in one octant of the unit circle . it improves the worst cases at the cost of a slightly reduced overall success probability . even more bisections will further increase the angular resolution but reduce the overall success to the point that they are counterproductive . although a bisecting composition is very simple approach , it provides the advantage of a larger number in initial projectors while avoiding a large penalty in overall efficiency due to a large action space with the same parameters . _ composition with the glow mechanism._the second mechanism departs from the strict bisection strategy of the first mechanism . the agent reaches an optimal success probability if it can measure along the direction @xmath112 . the bisection strategy only approximates @xmath12 and sometimes introduces unnecessarily many angles , e.g. , for @xmath61 the additional angle @xmath50 has to be built first . we overcome this disadvantage by a better use of the information provided by the measurement results to estimate which new projector angle should be inserted as addition action . we employ a variant of the `` edge glow mechanism '' @xcite to compose a single new action clip in the following way . we assign a second degree of freedom to each edge called `` glow '' and denote it by @xmath113 . instead of updating the @xmath31-values with the reward according to , we first accumulate rewards in the @xmath113 according to the following update rule : @xmath114 with initial values @xmath115 . the change in the update rule for @xmath31 effectively amounts to setting @xmath116 and @xmath117 . the behavior of the agent remains unchanged as the @xmath31-values remain at their initial values @xmath25 , i.e. , the agent measures equally often in all available directions . however , since there is no bias in the frequency of available measurement direction , the accumulated rewards in the respective @xmath113 provide a measure of the average reward for each direction . once the agent sampled enough measurement results , e.g. , when the first @xmath113 surpasses the threshold @xmath118 , a new action clip is composed and inserted into the projective simulator . the new measurement direction @xmath119 is composed from all @xmath17 and weighted by the @xmath113 : @xmath120 and we set the new @xmath121 . in order to prevent that a direction is inserted that is already present , the agent first checks that @xmath119 is sufficiently different from all already existing @xmath17 , e.g. , by inserting @xmath119 only if it differs from @xmath17 by more than @xmath122 circular standard deviations of the circular distribution given by the @xmath113 . if @xmath119 is too close to one @xmath17 , the @xmath22 of this @xmath17 is instead strengthened and set equal to the sum of all @xmath113 , and no new clip is inserted . after the composition , we continue with the usual update rule for the @xmath31-values . . the @xmath31-values are only updated with @xmath27 and @xmath29 after the composition . for @xmath12 coinciding with an existing @xmath17 only 4 agents compose a new angle @xmath119 , which is more than @xmath123 away from an existing @xmath17 , whereas all agents compose angles for the other examples of @xmath12 . the position of the step depends on the choice of the threshold for composition , here @xmath124 , which is chosen for large statistics but can be decreased without much penalty in the asymptotic efficiency albeit at the cost of slightly less accurate composed angles . ] the learning curves for the this form of glow composition are shown in figure [ fig : glow ] . starting with 4 angles and a @xmath118 , at least 2000 measurements have to be done on average before the first composition can occur . this threshold can be decreased leading to a faster composition , albeit at worse statistics , which result in inaccuracies of the composed angles . inaccurate compositions , however , impact the success probability only to a small extent because it decreases with the cosine of the angular difference between @xmath12 and the composed angle . in the direction of an existing angle , e.g. , @xmath38 or @xmath35 , the first amplifications of the respective @xmath17 occurs starting with 2000 measurements . for the direction @xmath60 , more measurement need to be done on average to reach @xmath125 or @xmath126 because these direction are not rewarded with certainty , and composition occurs on average later , with @xmath60 being one of the four latest instances . after the composition the success probability jumps from 50% to about 99% . since the newly set @xmath31-value for the best measurement direction is larger than the steady - state value for our choice of @xmath92 and @xmath93 , the success probability decreases slightly to approach @xmath97 from above . in an ensemble of 1000 agents only 4 compose an angle when @xmath38 or @xmath35 , whereas all do a composition for @xmath61 or @xmath49 . in our numerical experiment , the distribution of composed angles @xmath119 is sharply peaked around @xmath12 with a @xmath87 . by using the glow mechanism to obtain an effective average reward for each measurement direction , and then composing a mean angle from the reward distribution , the agent effectively creates a weighted sum of directions . it thereby embodies a method similar to the estimation of expectation values done in state tomography . so far we have demonstrated how an agent equipped with a suitable projective simulator can align a single measurement direction , e.g. , @xmath36 for the state @xmath127 , with an initially unknown state @xmath11 , which emerged from @xmath127 due to a magnetic field . since one of the aims is to employ the agent as a means to carry out measurement - based quantum computation ( mbqc ) @xcite in an unknown external field , all measurement directions that are required to run a specific algorithm in mbqc need to be adapted to this unknown stray field . we therefore need to extend the projective simulator to learn several measurement directions , which shall be given as the respective input . ultimately , the agent would translate the measurement directions necessary for the algorithm to the reference frame that rotates due to the magnetic field . we modify the inital agent setup depicted in figure [ fig : agentsetup ] in the following way . the step that prepares the defined initial state @xmath127 is removed and the qubit is simply left in the state that is prepared by the previous measurement . the projective simulator now receives as an input not just a trigger event , which activated the @xmath23-clip , but now it receives the previous measurement direction and the obtained measurement result as a percept . the initial state and percept can be chosen arbitrarily , e.g. , at random , as they do not matter in the subsequent feedback loop . and obtainable measurement outcome ( 0 or 1 ) . percepts that correspond to the same state prepared by the previous measurement are colored equally . ] for the new scheme , we also extend the clip network of the projective simulator to 8 percept clips , which represent all combinations of previous measurement direction and obtained reward , as depicted in figure [ fig : network_8to4 ] . effectively , the extended clip network consists of 8 copies of the previous simple clip network , which are activated according to the actions and results of the previous time step . the agent enters a feedback cycle , where measured directions and outcomes are fed back to the agent . the information about which state preparation method was used is available to the agent as percept , and thereby it indirectly receives a hint about which state has been prepared . given each prepared state , which then evolves to acquire an additional shift in the angle by @xmath12 , the agent learns which measurement direction most likely matches this rotated initial state . to give an example , consider the test qubit in the initial state @xmath128 , which evolves into @xmath11 . the agent measures this state , say along @xmath40 , and obtains result @xmath129 . it thereby prepares the test qubit in state @xmath130 , which again evolves for time @xmath8 into @xmath131 for the next measurement . this next measurement is chosen according to the @xmath31-values of edges originating from the percept clip `` @xmath132 '' to each of the four actions . . plotted are conditioned success probability , i.e. , given a percept , what is the probability of obtaining a + 1 measurement outcome , where each curve corresponds to one percept , solid lines represent outcome-@xmath129 preparation methods , dashed lines those for outcome @xmath133 . color codings are the same as for the clip network . curves are averages over 100 agents , with @xmath92 and @xmath93 . for the last time step the clip network with @xmath31-values encoded in the thickness of the edges are given in the inset . colors and clips match those of figure [ fig : network_8to4 ] . ] the clip network is now much larger than before and the agent needs more measurements to update all the connections until the @xmath31-values converge into those of the steady state . naively , we can expect an 8-fold increase , however , since the agent converges to a state in which measurements that give outcomes @xmath129 are preferred , learning the right measurements for a outcome-@xmath133 preparation is delayed . this learning behavior is shown in figure [ fig : completepslearning ] , where outcome-@xmath129 preparations converge early and outcome-@xmath133 preparations later , which in turn also delays the overall convergence . naturally , the training of the whole network is faster in situations where the @xmath133 outcomes occur more often , e.g. , for @xmath60 , or in situations that lead to different measurement directions , e.g. , for @xmath134 . as the agent encounters situations with different percepts , the number of time steps in between two successive activations of the same percept is now increased on average . this leads to a qualitative and quantitative change in the learning curves as compared to the previous simple agent with only one percept . the number of times that the damping reduces the @xmath31-value of each edge would increase and lead to a reduced efficiency , because the agent forgets too quickly in between rewards . to maintain high @xmath31-values for rewarded transitions we could adjust @xmath29 to a lower value , but we choose to simply restrict the application of the update rule , and the application of the damping in particular , to a subgraph of the clip network , namely , only those edges that are connected to the activated percept clip . thereby we maintain the quantitative behavior of the simple clip network used in the previous sections . percepts give the preparation procedure of the test qubit and thereby effectively encode information about which state has been prepared . a closer inspection reveals that each state is represented twice because is can be prepared in two ways , e.g. , @xmath127 can be prepared by a measurement of @xmath39 with outcome 1 or by @xmath45 with outcome 0 . preparation procedures that result in the same prepared states are highlighted with the same color of the percept clip in figures [ fig : network_8to4 ] and [ fig : completepslearning ] . this redundancy increases the learning times because the same behavior has to be learned twice . the clip network could be optimized with an additional intermediate layer that first maps preparation methods to states , which may be learned first without a stray field , and then the prepared states to best measurement directions in a stray field . once the agent has adapted its measurement directions to the unknown external field with a test qubit , it can be used as a translator between intended measurement directions and their corresponding directions in the rotated reference frame . this application of a trained agent works as follows . after a training period , we fix all the @xmath31-values . instead on the test qubit , the agent now acts on the qubit that needs to be measured along a certain direction according to a mbqc scheme , for example . we then excite a percept of the agent that corresponds to the direction of the intended measurement direction in zero field . the agent then chooses most likely the measurement direction that corresponds to this measurement in the rotated frame , i.e. , the measurement that takes the field into account . we first briefly repeat the mbqc variant of the grover search algorithm for a database with 4 elements @xcite and adapt it to our notation and use of projective measurements @xmath21 . the initial resource state is a cluster state of 4 qubits in ring form , i.e. , starting from the state @xmath135 we apply a controlled phase gate between the qubit pairs 12 , 23 , 34 , 41 , and obtain @xmath136 a database with 4 entries ( i.e. , with elements @xmath137 , @xmath138 , @xmath139 , and @xmath140 ) only requires a single grover step to find the marked element . the algorithm starts by doing this one necessary query to the database and thereby marks the database entry that is to be found . a measurement of the projectors @xmath39 or @xmath45 on qubits 1 and 4 realizes the specific database , where each pair of measurement directions @xmath137 , @xmath141 , @xmath142 , and @xmath143 corresponds to marking the database element @xmath137 , @xmath138 , @xmath139 , and @xmath140 , respectively . for each of the two measurements of @xmath39 or @xmath45 both measurement results @xmath144 or @xmath129 appear with probability @xmath145 . therefore , the results alone do not allow us to infer the measurement directions and thereby the marked element . in the problem setting of the algorithm the choice of measurement directions is hidden . only from the measurement results of qubits 1 and 4 , and from the measurements done on the remaining two qubits , we should infer the marked element . on the remaining qubits we therefore measure the observable @xmath39 , whose measurement outcome depends on the measurement directions on qubits 1 and 4 , and is correlated to the previous two outcomes . finally , the calculation of @xmath146 , i.e. , addition of the measurement outcomes modulo 2 , reveals the two bits of the marked element with certainty . although , at the present point the mbqc version of grover s algorithm appears to merely uncover ( anti-)correlations between measurement directions , there is an explicit mapping between the quantum circuit of gover s algorithm on one hand , and the circuit for creating and measuring the cluster state on the other @xcite . along the equator of the bloch sphere . data is obtained in independent runs with marked element @xmath137 for all fields giving rise to angles @xmath12 between 0 and @xmath147 in steps of @xmath148 . noisy data is the fraction of an ensemble of agents that identifies the marked element correctly when performing all four measurements in grover s algorithm . red ( analytical ) and orange ( numerical , 3000 agents ) curves give the success of the grover search ( all four measurements ) without taking into account the field in the measurement direction . the light blue curve gives the success for an ensemble of 1000 agents that each have a perfectly trained projective simulator with 4 measurement directions , which has learned the external magnetic field before doing the measurements for the grover search algorithm . the dark blue curve is an ensemble of 1000 agents that employs the glow mechanism to build a measurement direction that is adapted to the external magnetic field before using it to perform the grover search . ] if the initial state is placed in an unknown external field pointing along the @xmath3-direction , the state @xmath149 is transformed into @xmath150 with the local unitary rotations @xmath151 . if we recall that @xmath152 it is straightforward to see that the measurement protocol of the grover algorithm will no longer give the correct marked element because the external field effectively shifts the measurement directions by the angle @xmath153 with respect to the original measurement directions . as a result the probability to identify the correct marked element , i.e. , the success probability of the algorithm , is periodically modulated by @xmath12 and a straightforward calculation gives @xmath154 for @xmath12 being a multiple of @xmath155 the algorithm works perfectly because the local rotations align the qubits reference frames again with the @xmath76-axis . the grover algorithm is invariant under the inversion of all measurement directions ( i.e. , changing all directions @xmath133 to @xmath155 and vice versa ) . in the worst case , for @xmath12 being an odd multiple of @xmath35 , the chance of identifying the right element is 1/4 , as the measurements of the intended grover search actually reveal no useful information because they are unbiased with respect to the required measurement direction . in figure [ fig : grover ] , the analytic results for identifying the correct marked element @xmath137 in the rotated state match the trials with 1000 agents that simply measure all 4 qubits the direction @xmath62 and then try to identify the marked element from the obtained measurement results . for testing the agent with a projective simulator we restrict to a realization with a single marked element , namely @xmath137 , which can be implemented with measurements along the @xmath76-axis , all in the direction @xmath62 . the agent first learns with a test qubit exposed to the external field and adapts to the field strength . we then fix these obtained @xmath31-values and use the agent without update rule to carry out the four measurements on the cluster state , one after the other , according to its available measurement directions and internal probabilities . the first example is an agent that has only 4 fixed measurement directions available ( @xmath17 being a multiple of @xmath35 ) , which we first train to achieve optimal success probability with the test qubit . the optimal performance is reached in the limit @xmath156 , which amounts to @xmath157 for the single @xmath17 that is closest to @xmath12 and all others zero . the light blue curve in figure [ fig : grover ] shows the fraction of the agent ensemble that identifies the marked element with this projective simulator correctly . the grover search is recovered perfectly for fields with @xmath12 being a multiple of @xmath35 , which can be matched exactly by the available measurement directions . the second example is an agent that first learns with a test qubit in the external field with composition according to the glow mechanism . that is , after 2000 measurements on average , the agent composes a new measurement direction or strengthens an existing one that matches @xmath12 . the @xmath31-values after the composition remain fixed and the agent measures the cluster state according to the available measurement directions and probabilities . the dark blue curve in figure [ fig : grover ] illustrates that an ensemble of this kind of agent is highly successful in doing the grover search for all angles @xmath12 . the shortfall from a perfect performance ( the average success probability is 99.0% with a standard deviation of 0.3% ) originates in the slight deviations of the composed angle from @xmath12 and the non - zero probability to chose the remaining non - optimal measurement directions . we presented an autonomous adaptive system that is able to perform quantum information processing in changing environments . the controller is a learning agent endowed with a projective simulator that adapts measurement directions in a setup of measurement - based quantum computation by reinforcement learning . our approach thus combines elements from embodied artificial intelligence with the purpose of carrying out robust quantum information processing . in an exemplary setup of adapting measurement directions to an unknown stray magnetic field in a fixed direction , we have characterized the learning process of the projective simulator and its adaption to time - varying fields using numerical studies . we found that an agent using projective simulation is able to adapt to such unknown stray fields . we provide analytical estimates of its success probability in limiting cases of the non - linear learning process . in our scenario the agent may adapt the measurement direction by drawing from a initially provided set of fixed measurement directions . we have characterized the performance of the agent for different sets of available measurement directions and we explored composition mechanisms to create new and better measurement directions on the fly , together with the corresponding internal structure in the projective simulator . strategies with composed measurement directions surpass strategies with fixed sets of directions in both learning speed and resulting efficiency . as a demonstration of adaptive quantum information processing , the agent successfully carries out a measurement - based version of grover s search algorithm in the presence of a detrimental unknown external magnetic field . the present approach can be readily extended and improved in several directions as indicated in the respective sections in the paper . first and foremost , the agent effectively develops and embodies rules to cope and operate with quantum mechanical systems , which are seeded by the specific form the update rule together with the reward scheme , and the composition mechanisms . both of these elements start from simple primitives , e.g. `` prefer a specific measurement if it more likely results in a + 1 measurement outcome '' for the update rule , and give rise to a sensible and sufficient behavior in our problem setting . both can be improved by effectively incorporating more information about the quantum mechanical nature of the underlying problem domain , however , at the expense of more complicated update and composition rules . errors or imperfect measurements can be straightforwardly incorporated into the present scheme by using povms instead of projective measurement , or by adding a classical noise , e.g. bit flips , to the measurement outcomes . such errors lead to a diluted information about which measurement directions are correct and give @xmath13 measurement outcomes . in the presence of errors , spurious rewards appear for wrong measurement directions and the average reward for correct measurement directions is reduced . both effectively diminish the contrast in the reward landscape , which is equivalent to a lower reward scaling factor @xmath27 . we expect that the agent will still be able to learn in such situations , but it will take longer to do so and reach a lower asymptotic success probability . the latter can partly be recovered by adjusting @xmath27 and @xmath29 , however , an increase of the learning time over a noiseless scenario will remain . the long - term goal of this investigation is to develop integrated and autonomous schemes for measurement - based quantum information processing that can adapt to changing environments . in our scheme , learning is not realized by feedback from some external macroscopic sensor , e.g. a magnetometer , but it uses only information drawn from measurements on qubits , which are also the operations that drive the processing of the quantum information . in this sense our approach is related to recent work on intelligent quantum error correction @xcite . the approach that we have presented in the present paper can be generalized and integrated into a scheme of universal measurement - based quantum computation , where measurements of stabilizer operators of a cluster state are used both for the correction of errors on the resource state and , at the same time , for the adaption of measurement directions that drive the quantum computation . this will be reported elsewhere . we note that the projective simulator does not assume that rewards originate from measurement probabilities of a quantum state and , therefore , it is `` model free '' . this also opens the path to study foundational questions such as , to what extent can a machine effectively learn the rules of quantum mechanics through simple reinforcement processes . we thank wolfgang dr for initial discussions on this topic and vedran dunjko for comments on the manuscript . we acknowledge support from the austrian science fund ( fwf ) through the sfb foqus : f4012 , and the templeton world charity foundation grant twcf0078/ab46 . 499 briegel , h. j. & de las cuevas , g. projective simulation for artificial intelligence . * 2 * , 400 ( 2012 ) . mautner , j. , makmal , a. , manzano , d. , tiersch , m. & briegel , h. j. projective simulation for classical learning agents : a comprehensive investigation . _ new gen . comp . _ * 33 * , 69114 ( 2015 ) . melnikov , a. a. , makmal , a. & briegel , h. j. projective simulation applied to the grid - world and the mountain - car problem . _ preprint _ arxiv:1405.5459 [ cs.ai ] ( 2014 ) . melnikov , a. a. , makmal , a. , dunjko , v. & briegel , h. j. projective simulation with generalization , _ preprint _ arxiv:1504.02247 [ cs.ai ] ( 2015 ) . hentschel , a. & sanders , b. c. machine learning for precise quantum measurement . lett . _ * 104 * , 063603 ( 2010 ) . hentschel , a. & sanders , b. c. efficient algorithm for optimizing adaptive quantum metrology processes . lett . _ * 107 * , 233601 ( 2011 ) . lovett , n. b. , crosnier , c. , perarnau - llobet , m. , & sanders , b. c. differential evolution for many - particle adaptive quantum metrology . _ phys . lett . _ * 110 * , 220501 ( 2013 ) . sergeevich , a. , chandran , a. , combes , j. , bartlett , s. d. & wiseman , h. m. characterization of a qubit hamiltonian using adaptive measurements in a fixed basis . a _ * 84 * , 052315 ( 2011 ) . granade , c. e. , ferrie , c. , wiebe , n. & cory , d. g. robust online hamiltonian learning . _ new j. phys . _ * 14 * , 103013 ( 2012 ) . hayes , a. j. f. & berry , d. w. swarm optimization for adaptive phase measurements with low visibility . rev . a _ * 89 * , 013838 ( 2014 ) . sutton , r. s. & barto , a. g. _ reinforcement learning : an introduction 1st edn _ ( mit press , 1998 ) . russel , s. j. & norvig , p. _ artificial intelligence a modern approach 2nd edn _ ( prentice hall , 2003 ) . paparo , g. d. , dunjko , v. , makmal , a. , martin - delgado , m. a. & briegel , h. j. quantum speedup for active learning agents . x _ * 4 * , 031002 ( 2014 ) . dunjko , v. , friis , n. & briegel , h. j. quantum - enhanced deliberation of learning agents using trapped ions . _ new j. phys . _ * 17 * , 023006 ( 2015 ) . wittek , p. _ quantum machine learning : what quantum computing means to data mining_. ( academic press , 2014 ) . schuld , m. , sinayskiy , i. & petruccione , f. an introduction to quantum machine learning . _ preprint _ arxiv:1409.3097 [ quant - ph ] ( 2014 ) . grover , l. k. a fast quantum mechanical algorithm for database search in _ proceedings of the 28th annual symposium on the theory of computing _ , 212219 , ( acm press , 1996 ) . grover , l. k. quantum mechanics helps in search for a needle in a haystack . lett . _ * 79 * , 325328 ( 1997 ) . chuang , i. l. , gershenfeld , n. & kubinec , m. experimental implementation of fast quantum searching . lett . _ * 80 * , 34083411 ( 1998 ) . jones , j. a. , mosca , m. & hansen , r. h. implementation of a quantum search algorithm on a quantum computer . _ nature _ * 393 * , 344346 ( 1998 ) . kwiat , p. g. , mitchell , j. r. , schwindt , p. d. d. & white , a. g. grover s search algorithm : an optical approach . opt . _ * 47 * , 257266 ( 2000 ) . raussendorf , r. & briegel , h. j. a one - way quantum computer . lett . _ * 86 * , 51885191 ( 2001 ) . raussendorf , r. , browne , d. e. & briegel , h. j. measurement - based quantum computation on cluster states . rev . a _ * 68 * , 022312 ( 2003 ) . briegel , h. j. & raussendorf , r. persistent entanglement in arrays of interacting particles , _ phys . * 86 * , 910913 ( 2001 ) . walther , p. et al . experimental one - way quantum computing . _ nature _ * 434 * , 169176 ( 2005 ) . prevedel r. et al . high - speed linear optics quantum computing using active feed - forward . _ nature _ * 445 * , 6569 ( 2007 ) . lanyon , b. p. et al . measurement - based quantum computation with trapped ions . lett . _ * 111 * , 210501 ( 2013 ) . knill , e. , laflamme , r. & milburn , g. j. a scheme for efficient quantum computation with linear optics . _ nature _ * 409 * , 4652 ( 2001 ) . helstrom , c. w. _ quantum detection and estimation theory_. ( academic press , 1976 ) . fisher , n. i. , lewis , t. & embleton , b. j. j. _ statistical analysis of spherical data_. ( cambridge university press , 1987 ) . mardia , k. v. & jupp , p. e. _ directional statistics_. wiley series in probability and statistics ( john wiley & sons ltd , 2000 ) . combes , c. et al . in - situ characterization of quantum devices with error correction . _ preprint _ arxiv:1405.5656 [ quant - ph ] ( 2014 ) . the angular probability distribution for @xmath12 given the @xmath82 measurement outcomes @xmath158 in directions @xmath79 , which are multiples of @xmath35 , is given by @xmath159 with normalization @xmath84 , and can be expanded into a fourier sum @xmath160 where the normalization is solely contained in the coefficient @xmath161 . updating this probability distribution with the next measurement result @xmath162 amounts to multiplication with the factor @xmath163 , which we again expand into a fourier sum . comparing the coefficients we obtain the following recursion relations for the @xmath164 and @xmath165 : @xmath166 \\ s_{m+1}(q ) & = \frac{s_m(q)}{2 } + \frac{r_{m+1}}{4 } \big [ \sin(\alpha_{m+1 } ) \big ( c_m(q-1 ) - c_m(q+1 ) \big ) + \cos(\alpha_{m+1 } ) \big ( s_m(q-1 ) + s_m(q+1 ) \big ) \big],\end{aligned}\ ] ] the advantage of the fourier representation is that circular moments of the probability distribution can be straightforwardly calculated : @xmath170 the first circular moment @xmath171 gives rise to the mean angle @xmath172 and the circular standard deviation @xmath173 @xcite .

recent measurements of the kerr parameters for two black - hole binaries in our galaxy , gro j1655 and 4u 1543 of and , respectively , fitted well the predictions of , of . in this report we also note that predicted for 80% of the soft x - ray transient sources . the maximum available energy in the blandford - znajek formalism for givesergs , orders of magnitude larger than the energy needed for the grb and hypernova explosion . we interpret the soft x - ray transients to be relics of grbs and hypernovae , but most of them were subluminous ones which could use only a small part of the available rotational energy .

astro-ph0612461

recent estimates of the kerr parameters @xmath0 for two soft x - ray transients ( sxts ) @xcite , gro j1655@xmath140 ( nova sco ) and 4u 1543@xmath147 ( il lupi ) , facilitate a test of stellar evolution , in that the spins of the black holes in these binaries should be produced in common envelope evolution which begins with the evolving massive giant and companion donor , and ends up in helium - star donor binary , the hydrogen envelope of the massive star having been stripped off and the helium having been burned . @xcite ( hereafter denoted as lbw ) assumed common envelope evolution to begin only after he core burning has been completed ; i.e. , case c mass transfer @xcite . otherwise the he envelope , if laid bare , would blow away to such an extent that the remaining core would not be sufficiently massive to evolve into a black hole @xcite . the black - hole - progenitor star , in which the helium core burning has been completed , is tidally locked with the donor ( secondary star ) so the spin period of the helium star is equal to the orbital period of the binary . in this tidal locking , lbw assumed uniform rotation of he star by assuming that the inner and outer parts of he are strongly connected due to the presence of a strong magnetic field . the c - o core of the helium star drops into a rapidly spinning black hole due to angular momentum conservation . in this process , the spin of the black hole depends chiefly on the mass of the donor because the orbital period chiefly depends on the donor mass as we explain later ( in section [ sec - lmd2ham ] ) . lbw calculated this kerr parameter ( @xmath0 ) as a function of binary orbital period . the results are given in their fig . 12 which we reproduce as fig . [ fig12 ] . the agreement of the natal kerr parameters with the @xcite measured ones means that only a small amount of angular momentum energy could have been lost after the formation of the black hole . the good agreement in this comparison supports the assumption of case c mass transfer and the tidal locking at the donor - he star stage assumed in the lbw calculations . in the lbw calculation , the @xmath0 for nova sco was slightly greater than for il lupi . combining this with observation , @xmath8 would be our best estimate for both binaries . in this report we also note that lbw predicted @xmath9 for 7 soft x - ray transient sources with main sequence companions and @xmath10 for xte j1550@xmath1564 and gs 2023@xmath11338 ( v404cygni ) with evolved companions . the maximum available energy in the blandford - znajek formalism for @xmath6 gives @xmath7ergs , orders of magnitude larger than observed in the grb and hypernova explosion . based on this observation , we interpret the soft x - ray transients to be relics of grbs and hypernovae . it should be noted that the way in which the hypernova explodes can be similar to the woosley collapsar model . the main advantage in our scenario is that the h envelope in our binary is removed by the donor and the rotational energy is naturally produced in the common envelope evolution . the necessity for case c mass transfer , given galactic metallicity , and the measured system velocity lock us into the kerr - parameter values we find . in section [ casec ] we elaborate on the determination of the kerr parameters of soft x - ray transient black - hole binaries discussed above , which could be read off the figures of lbw by a discussion of tidal locking , and the connection of tidal locking to case c mass transfer . we also discuss the energetics for grbs and hypernovae based on the black - hole spin . in section [ sec - lmd2ham ] we show that the angular - momentum energy of the black - hole binary is determined mainly by the mass of the donor . we discuss 12 galactic transient sources with angular - momentum energies @xmath12ergs , so that all of these are relics of grbs and hypernovae . the energies of the grb and hypernova explosion powered by these , as we shall develop , should be subtracted from the natal rotational energies , to give the explosion energy . case c mass transfer implies that the mass transfer takes place late , after the he in the giant progenitor of the black hole has been burned . the proof of our scenario was given in the measured kerr parameters for nova sco and il lupi @xcite which agreed with the predictions of lbw . again , aside from the fact that we use the tidal coupling of the donor to spin of the black hole progenitor , the rest of our scenario , especially the collapse , is the same as in the woosley collapsar model . the grb and hypernova explosions are all of type @xmath13 , so that he lines do not appear . in the woosley collapsar model , he burning is not necessarily finished before explosion , but a ) the interacting he may fall into the black hole or , b ) the he may not mix with the @xmath14ni , so that in either case he lines would not be seen . we remark here that the explosion on which a record number ( 119 ) of astronomers concentrated their attention @xcite , sn2006aj was a @xmath15 explosion ; i.e. ( see additional information on @xcite ) convective carbon comes to an end just at the zams mass at which black holes begin to form , as explained in @xcite ( see fig.1 of that paper ) because the entropy loss by neutrino emission during convective carbon burning is shut off by the absence of carbon and the entropy that increases with zams mass must go into adding nucleons to the iron core of the star , before it collapses . the @xmath15 nature of the explosion identifies the central engine , therefore , as a black hole . in the 4 nature papers focussed on grb060218/sn2006aj the authors speculated on the central engine being a magnetar , but , in fact , it must have been a black hole and the grb and hypernova run by the blandford - znajek mechanism . otherwise there would have been carbon lines . in @xcite the black hole formation was described by a blaauw - boersma explosion , which should be sufficient for calculating the system velocity of the binary because conservation laws are respected . however , we believe the woosley collapsar model to give a more detailed description of the black hole formation and the hypernova explosion . the @xcite description includes magnetohydrodynamics in the form of the blandford - znajek mechanism . however , mass loss in the explosion and conservation laws are those of blaauw - boersma . lbw found that there are two classes of soft x - ray transients , those with main sequence companions ( denoted as aml ) , and others with evolved companions ( denoted as nu ) . due to the angular momentum loss , via gravitational wave radiation and magnetic braking , after black - hole formation the orbits of amls are shortened . based on this argument and the current observation , lbw traced back the orbital period at the time of black - hole formation in their fig . 10 . by the tidal locking , the estimated kerr parameters for amls are @xmath16 ( about half of them are @xmath17 ) . the maximum available energies for these system via the blandford - znajek formalism are @xmath18 ergs . so we believe that those black - hole binaries with main sequence companions are the relics of grbs and hypernovae . the evolution of nu s after black - hole formation is mainly controlled by the donor which is evolving beyond the main - sequence stage , and the orbit is widened due to the conservative mass transfer from the less massive donor to the black hole . we will discuss the possibilities of nu s as relics of grbs and hypernovae . we will be brief in reconstructing nova sco as a relic of grb and hypernova because this was constructed in considerable detail with assumed kerr parameter of @xmath19 in @xcite . with the smithsonian - harvard - mit measurement of @xmath20 we do nt need to change the brown et al . discussion by much . indeed , the smithsonian - harvard - mit @xmath21 gives its value after powering the grb and hypernova explosion , whereas the @xcite assumed value was before the explosion , so the two are not significantly different , since the explosion can be powered by the energy from an @xmath22 change in @xmath0 when @xmath0 is large . the hypernova aspect of the explosion in nova sco was clear by the accretion of @xmath23-particle nuclei onto the donor as a result of the hypernova explosion . @xcite found that o , mg , si and s have abundances on the f - star donor @xmath24 times solar . these nuclei were presumably absorbed by the donor , which acts as witness to the explosion . due to the similarity in the orbital period between nova sco and il lupi as summarized in fig . [ fig11 ] ( lbw ) , we argue that il lupi is also a relic of grb and hypernova . although the black hole masses are not known as well in xte j1550@xmath1564 and gs 2023@xmath11338 ( v404 cygni ) , it can be seen from fig . [ fig12 ] and fig . [ fig11 ] that using their reconstructed preexplosion periods they have a kerr parameter of @xmath25 , possibly somewhat less definite than the prediction of the @xmath21 for nova sco . the latter two binaries have black holes with nearly double the mass as the first two , and , therefore , larger accretion disks . from our arguments , in appendix [ app2 ] , we believe that they may be able to accept more rotational energy , which could be checked by subtracting the measured kerr parameters from the natal ones . @xcite remarked that for nova sco _ after the first second the newly evolved black hole has @xmath26erg of rotational energy available to power these . the time scale for delivery of this energy depends ( inversely quadratically ) on the magnitude of the magnetic field in the neighborhood of the black hole , essentially that on the inner accretion disk . the developing supernova explosion disrupts the accretion disk ; this removes the magnetic fields anchored in the disk , and self - limits the energy the blandford - znajek mechanism can deliver . " _ this , together with the total rate of creations of binaries of our type of @xmath27galaxy@xmath28yr@xmath28 estimated by @xcite will be shown in section [ sec - pop ] to reproduce the population of subluminous bursts in nearby galaxies . it is clear that they are subluminous , at least in the cases of nova sco and il lupi because their kerr parameters measured by the smithsonian - harvard coalition are indistinguishable from the natal predicted @xmath29 within observational errors . the disruption of the black hole disk will be discussed in detail in appendix [ app2 ] . our evolution of black hole binaries in our galaxy might appear to be irrelevant for the long ( high luminosity ) @xmath30-ray bursts because @xcite show that these come chiefly from low metallicity , very massive stars in galaxies of limited chemical evolution , quite unlike our milky way . however , we can construct a quantitative theory of the rotational energies of the black holes which power the central engine for the grbs and hypernovae in our galaxy because we can calculate the black hole kerr parameters . having this quantitative theory it is straightforward to apply it to explosions in low metallicity galaxies , the high luminosities resulting because the donors are more massive than those in our galaxy and can accept more rotational energy from the binary than the high metallicity binaries with low mass donors , although there may be some overlap of the high metallicity stars with the same mass donor as the low metallicity stars , as we shall discuss . we predict that this will be so with xte j1550 - 564 ( v383 normae ) , in which the kerr parameter will be measured ( j. mcclintock , private communication ) so that with our calculation presented here we will be able to obtain the energy used up in the explosion which should be nearly that of cosmological ( high luminosity ) grbs . recently the eclipsing massive black hole binary x@xmath17 has been discovered in the nearby spiral galaxy messier 33 @xcite . since the metallicity is @xmath31 solar , we believe it to mimic low metallicity stars which are more massive than the galactic ones . the donor has mass @xmath32 now , possibly @xmath33 earlier . we expect that this system may have gone through a dark explosion due to the high donor mass , which implies a low rotational energy as we discuss in next section . a highly relevant discussion for the v4641 sgr evolution , which will be a template of an earlier grs 1915@xmath11105 development was given by @xcite , for cyg x@xmath11 . they discussed the latter as if the donor and black hole were very nearly equal in mass which we shall show will happen in the future for cyg x@xmath11 , although its donor is now @xmath34 and black hole @xmath35 . the donor could have been substantially more massive when the black hole was born after common envelope evolution and lost mass by wind . the donor in @xcite had a stellar wind of @xmath36 yr@xmath28 @xcite throughout the evolution . once the mass of the donor became reduced to a mass comparable to the mass of the black hole , the donor established thermal equilibrium and filled its roche lobe transferring mass at the rate of @xmath37yr@xmath28 . because of the continuing wind loss the donor shrank significantly below its roche lobe and the system widened . the donor started to expand again after it had exhausted all of the hydrogen in the core and filled its roche lobe a second time . in this phase the mass transfer reached a second peak of @xmath38 yr@xmath28 , where mass transfer was driven by the evolution of the h - burning shell . the most interesting feature of this calculation was that the system became detached after the first initial time scale because of the stellar wind from the donor . since the donor is close to filling its roche lobe , such a wind may be focussed towards the accreting black hole , as is inferred from the tomographic analysis of the mass flow in cyg x-1 by @xcite . @xcite decompose the stellar wind of the supergiant into two moments , one representing the approximately spherically symmetrical part of the wind and the second representing the focussed enhancement of wind density in the direction of the black hole . the latter component of the wind transferred mass in an essentially conservative way ( although the former would bring about mass loss ) . we shall use the wind in both , v4641 sgr and grs 1915@xmath11105 , to accrete matter from the donor to the black hole later on . note that the wind , transferring matter at hypercritical ( much greater than eddington ) rate basically shuts off the initial roche lobe overflow , because it can transfer mass sufficiently by itself . the second period of roche lobe overflow transfer is driven by the evolution of the h - burning shell ; i.e. , by the secondary star going red giant . @xcite say that irrespective of whether this particular model is applicable to cyg x@xmath11 , the calculation ... illustrates that it is generally more likely to observe a high - mass black - hole x - ray binary in the relatively long - lived wind mass - transfer phase following the initial thermal timescale phase which only lasts a few @xmath39 yr . in this example , the wind phase lasts a few @xmath40 yr , but it could last as long as a few @xmath41 yr if the secondary were initially less evolved . " we believe the above scenario to apply not only to cyg x@xmath11 , v4641 sgr and grs 1915@xmath11105 , but also to lmc x@xmath13 and m33 x@xmath17 , binaries with donors more massive than the black - hole companion . we find that these all had dark explosions , as we develop below , basically because the donor had too high a mass at the time of the explosion to give an energetic grb . we suggest that the wind in these cases may resemble the tidal stream of @xcite . in the two - dimension system studied by these authors , they show that when @xmath42 becomes less than @xmath43 , where @xmath44 is the distance between o - star and black hole and @xmath45 is the radius of the primary ( or accreting ) star , the tidally - enhanced - wind accretion exceeds bondi - hoyle accretion , a factor that increases to several as @xmath42 decreases . we are thus now able , given the evolution of @xcite , to describe in detail the crude mass transfer used in the evolution of grs 1915@xmath11105 by @xcite , as conservative mass transfer in wind . we disagree , however , with the procedure of @xcite to limit the accretion to eddington . obviously there is no surface in the black hole onto which the binding energy of the infalling matter can be accreted and then furnish photons etc . which will be emitted so as to impede the other infalling matter . in fact , once the matter goes across the event horizon it can no longer influence the matter that has not . in the case of 1915@xmath11105 , @xcite found the average mass transfer rate to be @xmath46yr@xmath28 , about 200 times eddington ( see @xcite p.355 ) . none of the three binaries we consider had appreciable natal @xmath21 s because of the high masses of the donors ( secondary stars ) . thus , we can see that the high kerr parameter must come from mass accretion ( see fig . [ fig - a ] ) and that the accreted mass must be slightly greater than the natal mass in order to have such a high kerr parameter for grs 1915@xmath11105 . taking the final black hole mass to be @xmath47 we need a natal mass of @xmath48 . if v4641 sgr is to be a template for grs 1915@xmath11105 , then its black hole should have had a mass of @xmath48 when born , which would have evolved to its present value of @xmath49 by accretion by wind from the donor . thus , the present @xmath50 donor and @xmath49 black hole will have essentially interchanged masses from the time of the birth of the black hole until present , through mass exchange from the donor to the black hole . as outlined in lbw , v4641 sgr will transfer by wind another @xmath51 from donor to black hole to reach the present grs 1915@xmath11105 with black hole mass @xmath47 . thus , the total mass accreted by grs 1915@xmath11105 is estimated to be @xmath52 . the companion mass evolved in this way is @xmath53 , somewhat larger than the @xcite mass of @xmath54 . some mass is , however , lost in jets , which is not taken into account in our approximation of conservative mass transfer . our evolution of the black hole birth is similar to the second evolution version of @xcite which would give the right chemical abundances to the secondary star in v4641 sgr . these investigators suggested that the black hole mass at birth was @xmath55 , and to obtain the right surface abundances they proposed that the explosion was a dark one ; i.e. , one of low energy . @xcite suggest that the explosion in cyg x@xmath11 was a dark explosion , much less energetic than the one in nova sco . we shall see that the explosions are dark because the donors ( secondary stars ) have high masses . we will construct spin energies for the transient sources in the next section , but we want to make some general comments here . the question with grbs is whether there is enough angular momentum to power the grb and hypernova explosion . in the case of the widely accepted theory , woosley s collapsars , this question is unanswered , although one may take the point of view that we observe grbs and hypernova explosions , so there must be enough angular momentum , which is an integral part of the mechanism . none the less , @xcite say when recent estimates of magnetic torques @xcite are added , however , the evolved cores spin an order of magnitude slower . this is still more angular momentum than observed in young pulsars , but too slow for the collapsar model for gamma - ray bursts . " furthermore , the usual scenarios for the interactions of wolf - rayets with other stars is that they slow down the rotation . the above arguments in support of the binary model which was used in @xcite , were made in @xcite , and applied to galactic transient sources . the hypernova formed in 1998bw had @xmath56ergs in energy @xcite . in addition , the jet formation in the grb requires lifting all of the matter out of the way of the jet . @xcite estimates that this costs @xmath57ergs in kinetic energy . at early times the thermal and kinetic energies in supernova explosions are roughly equal , satisfying equipartition . we believe this to be at least roughly true in our explosions here , so that @xmath58ergs would be needed for 980425/sn1998bw and possibly morec@xmath59 would be @xmath60ergs . note that this energy is invisible " , and is not normally included in estimates of grb energies . ] , because @xcite describes 980425 as a smothered " explosion . as noted in appendix of brown , lee and moreno mndez , the blandford - znajek efficiency drops substantially as the kerr parameter decreases below @xmath61 . thus , the available rotational energy will decrease rapidly with increasing donor mass . in appendix [ app2 ] we will estimate that the highest rotational energy of @xmath58ergs can be accepted by a binary with a @xmath62 donor . using m_he=0.08(m_giant/)^1.45[eq : mhe ] lbw found that following common envelope evolution a_f= ( ) ( ) ^-0.55a_i[eq : af ] . here @xmath63 is the final separation of the he star which remains from the giant following the strip off of its h envelope , and @xmath64 is its initial separation . it has inherited the angular momentum of the he star and is tidally locked with the donor . the giant masses found by lbw were all about @xmath65 . from kepler we have the preexplosion period = ( ) ^3/2 ( ) ^1/2 . [ eq : kepler ] & @xmath21 & @xmath66 + & [ @xmath67 & [ @xmath67 & & [ @xmath68 ergs ] + + j1118@xmath11480 & @xmath69 & @xmath70 & @xmath71 & @xmath72 + vel 93 & @xmath73 & @xmath70 & @xmath71 & @xmath72 + j0422@xmath1132 & 6@xmath17 & @xmath70 & 0.8 & @xmath74 + 1859@xmath11226 & 6@xmath17 & @xmath70 & 0.8 & @xmath74 + gs1124 & @xmath75 & @xmath70 & 0.8 & @xmath74 + h1705 & @xmath75 & @xmath70 & 0.8 & @xmath74 + a0620@xmath1003 & @xmath76 & @xmath70 & 0.6 & @xmath77 + gs2000@xmath11251 & @xmath76 & @xmath70 & 0.6 & @xmath77 + + gro j1655@xmath140 & @xmath78 & 1@xmath12 & @xmath71 & @xmath72 + 4u 1543@xmath147 & @xmath78 & 1@xmath12 & @xmath71 & @xmath72 + xte j1550@xmath1564 & @xmath79 & 1@xmath12 & @xmath80 & @xmath81 + gs 2023@xmath11338 & @xmath79 & 1@xmath12 & @xmath80 & @xmath81 + xte j1819@xmath1254 & 6@xmath17 & @xmath79 & 0.2 & @xmath82 + grs 1915@xmath11105 & 6@xmath17 & @xmath79 & 0.2 ( @xmath83 ) & @xmath82 + cyg x@xmath11 & 6@xmath17 & @xmath84 & 0.15 & @xmath85 + from fig . [ fig12 ] ( fig.12 of lbw ) one sees that the kerr parameter increases sharply as the period of the binary @xmath86 decreases . from eq . ( [ eq : af ] ) we see that @xmath63 is proportional to @xmath87 , the donor mass , and from eq . ( [ eq : kepler ] ) that @xmath86 is proportional to @xmath88 . estimated kerr parameters for galactic sources are summarized in table [ tab - a ] , which makes the dependence on the donor clear for the galactic sources . in fact , it would seem priori more natural to have an @xmath35 donor in the binary with a @xmath65 giant than a @xmath89 donor and @xmath65 giant as in nova scorpii . we shall see further on in this section that case c mass transfer requires that the initial @xmath64 is very large , @xmath90 for a @xmath65 giant . the increased gravitational binding between donor and the he - star left from the giant after being stripped of hydrogen must furnish the energy , modulo the efficiency @xmath91 , to remove the hydrogen envelope . the latter decreases inversely with the radius of the giant , so that when the radius is large , the envelope can be removed by a low - mass companion . combining eqs . ( [ eq : af ] ) and ( [ eq : kepler ] ) we have = ( ) ^3/2 ( ) ^1/2 ( ) ^0.83 [ eq : afkepler ] so that @xmath92 for @xmath93 and @xmath94 for higher donor masses . as we develop later , the distances @xmath64 at which mass transfer begins depend very little on donor mass so we can use eq . ( [ eq : afkepler ] ) in order to scale from one donor mass to another . lbw found all giants they needed for the transient sources had mass @xmath95 , and , therfore , @xmath96 , substantially larger than the donor masses for the most energetic grbs . for higher mass donors mass loss in the explosion can be neglected , giving pre and post - explosion periods which are nearly the same . in supernova explosions at short times the kinetic and thermal energies are equal , following equipartition . we are close to finding this to be true for grbs and hypernovae . the kinetic energy , which is an order of magnitude or more greater than the grb energy @xcite , results from the ram pressure which is needed to clear the way for the jet which initiates the grb , as just described . we shall assume the kinetic energy to be the same as the thermal energy , which is more easily measured from the observations . thus , we assume that twice the hypernova energy is needed to power the explosion . on the other hand , the efficiency @xmath97 in depositing the energy in the perturbative region is usually taken to be @xmath98 , the optimum value . this optimum value is obtained by impedance matching as in ordinary electric circuits . the blandford - znajek energy is deposited in a fireball in the perturbative region . @xcite and @xcite have shown that in order to power a grb all that is necessary is for the fireball to have enough energy so that the temperature is well above the pair production temperature ; i.e. , @xmath99mev . then the grb and the afterglow will follow , just from having as source the localized hot fireball . the estimated explosion energies of galactic sources are summarized in table [ tab - a ] . we claim that ours is the first quantitative calculation of the explosions giving rise to grbs and hypernovae , in the sense that we calculate the energy that is supplied in the form of angular momentum energy . what the central engine does with this energy is another matter . we can not calculate how much of the energy is accepted by the accretion disk in detail , because we can not calculate analytically the rayleigh - taylor instability in magnetic field coupling to the accretion disk of the black hole ( but see appendix [ app2 ] ) . we do know from observations that accretion disks have been formed ; the kerr parameters have been measured by relativistic doppler effects as the matter goes into the black hole . we also know that jets are launched sporadically from the soft x - ray transient sources in our galaxy . the rotational energy that is not accepted by the black hole does , however , remain in the binary and appears later in the kerr parameter of the black hole . thus far , in nova sco and il lupi a tiny part of the available rotation energy was accepted by the central engine , so little that the final rotational energy could not be discriminated from our calculated initial energy , because the uncertainty in the measurement of kerr parameters was the same order of magnitude as the explosion energy . we can understand why amls and nova sco , il lupi , xte j1550@xmath1564 and gs 2023@xmath11338 among nus have such rotational energies from case c mass transfer . the lower the companion mass , the greater the radius @xmath100 that the supergiant must reach before its roche lobe meets the companion . given giants such as shown in fig . [ fig2 ] , the typical separation distance between giant and companion is @xmath101 , higher ( because of the roche lobe of the companion star ) than the @xmath102 the giant radius reaches . the binding energy of the supergiant envelope goes as @xmath103 and at such a large distance it can be removed by the change in binding energy of an @xmath104 donor , spiralling in from @xmath105 to @xmath106 . a higher mass donor would end up further out , since it would not have to spiral in so close in order to release enough binding energy . thus low mass @xmath107 companions can naturally deposit their increase in gravitational binding energy in removing the high @xmath100 envelopes . a detailed discussion of these matters is given in @xcite . thus case c mass transfer naturally gives the ultra - high rotational energies of the binaries with low - mass donors discussed as relics of grb and hypernova explosions in our galaxy . * cyg x-1 : * one can see that there was essentially no mass loss in the cyg x@xmath11 explosion because the space velocity of cyg x@xmath11 relative to cyg ob3 in the cluster of o - stars of @xmath108 km s@xmath109 is typical of the random velocities of stars in expanding associations @xcite . in comparison , nova sco had a very strong explosion from the fact that its space velocity after explosion is @xmath110 km s@xmath28 , although only a small part of the available energy is used up in the system velocity . our reconstruction of the cyg x@xmath11 evolution had the explosion take place when the black hole mass was about @xmath111 , the @xmath112 less than the black hole mass is today having accreted from the donor ( secondary star ) . the mass transfer from donor to black hole is nonconservative because of the higher mass of the donor . the donor would have been substantially more massive than it is today at the time of the explosion , at least @xmath95 . for a @xmath65 donor we obtain the maximum available energy @xmath113ergs . @xcite believe that cyg x-1 had a dark explosion . as discussed in appendix [ app1 ] , the efficiency @xmath97 can be taken to be @xmath80 for the higher @xmath0 , say @xmath5 , but it decreases for small @xmath0 , so that for @xmath114 we calculate it to be @xmath115 and for @xmath116 we calculate @xmath117 . the @xmath118ergs is clearly not enough for an explosion in cyg x@xmath11 . * binaries in low metallicity galaxies : * as discussed in a previous section , @xcite have recently measured the extra galactic m33 x@xmath17 in a neighborhood where the metallicity is @xmath31 solar . following our prediction that the donors of low - metallicity galaxies are generally more massive than those of our galaxy , the now @xmath119 ( @xmath33 at the time of common envelope evolution ) is much more massive . in fact , it is too massive in that with the small kerr parameter we estimate to be @xmath120 ; it probably went through a dark explosion . far from being irrelevant , as @xcite imply ( because of their lack of dynamics ) , the measurements of the harvard - smithsonian group teach us how to calculate the energies of grb and hypernova explosions . having a dynamical theory , we can easily extend it to low metallicity galaxies , by increasing the donor masses . all of the grbs in our binary model come from the same mechanism , but their angular momentum energy will be decided by the mass of their donor . our mechanism suggests , however , that the binaries are usually left spinning with the measured kerr parameter . there must be a goldilocks " scenario for the energy needed to power a high - luminosity grb , neither too big nor too small . galactically , with the @xmath121 low mass donors , the energy is clearly too large . we know this because the calculated initial kerr parameters were essentially the same as those found by @xcite ; thus , very little of the energy had been used up in the explosion . on the other hand we have m33 x@xmath17 , which we will discuss in more detail later , which had a donor of @xmath33 , with kerr parameter @xmath120 which probably went into a dark explosion , like cyg x@xmath11 . so we have bracketed ( but rather widely ) the luminous explosions . initially in supernova explosions , the kinetic energy , the main part of which results from clearing out the matter in the way of the jet that accompanies the grb , is about equal to the thermal energy of the hypernova . @xcite finds this to be at least approximately true and it would follow from equipartition of energy . we should be able to connect grs 980425 with our galactic grbs because of its high metallicity , nearly solar @xcite . ( the galaxy of grb 980425 is incorrectly put in the class of low - metallicity by @xcite and by @xcite . ) note that the high luminosity grbs turn out to be only a small fraction of the total number , even if we use a beaming factor of 100 for them . thus , they must be formed in very special circumstances ( see appendix [ app2 ] ) . the question is whether there are enough transient sources to supply subluminous grbs in nearby galaxies . @xcite estimated that in our galaxy the total rate of creation of the transient source binaries was @xmath122galaxy@xmath28yr@xmath28 . given @xmath40 galaxies within @xmath123mpc this number translates into @xmath124gpc@xmath125yr@xmath28 . @xcite find a beaming factor typically less than 14 ; such a beaming factor would reduce our number to @xmath126gpc@xmath125yr@xmath28 , essentially that of @xcite of @xmath127gpc@xmath125yr@xmath28 . this is much higher than their rate of high - luminosity grbs of @xmath128gpc@xmath28yr@xmath28 . the usual beaming factor for the high luminosity bursts is @xmath129 . even with such a large factor , the high luminosity grbs are estimated to be much less in number , by a factor of @xmath130 , than the subluminous ones . although one should add the woosley collapsar rate to our binary rate , we have enough binaries to account for all of the bursts . @xcite estimate that @xmath131 of the stars above @xmath132 can , under certain circumstances , retain enough angular momentum to make grbs . the effect of cutting down the wind losses galactically gave a hint about how the rotational energy in the binaries could be decreased so as to be in the ballpark needed for high lumiosity grbs . the winds are particularly high because of galactic metallicity . low metallicity stars have much less wind . in general the stars are more massive than galactic ones , which we believe has the effect of scaling up all of the galactic masses . we pursue the question of cosmological grbs and their abundances in appendix [ app2 ] . if one accepts the @xcite numbers literally , then case c mass transfer is actually limited to a narrow interval of zams masses about @xmath133 , @xmath134 as found by @xcite . this is because the binary orbit widens with mass loss of the supergiant so that in order to initiate mass transfer only after helium burning the supergiant has to expand sufficiently that this widening of the orbit is compensated for . a graphic display of this is shown in fig . 1 of @xcite . lbw realized that in order to reproduce black holes from the interval of zams masses @xmath135 , necessary for their evolution of transient sources , they had to cut down the wind losses in the red giant stage ( by hand -see lbw fig . this was clearly necessary because @xcite had shown that high mass x - ray black hole binaries could be evolved with the black holes coming from zams masses @xmath135 provided case c mass transfer was used . it may be that the donor mass for high luminosity grbs has to be higher than @xmath62 . we do not yet know how rapidly the binaries are left rotating after the explosion . measurements of black hole binaries with donor masses @xmath136 would be very helpful . in lbw it seemed strange that the giant progenitors of the black - hole binaries nova sco , il lupi and grs 1915@xmath11105 all came from @xmath137 giants , whereas black holes were formed from zams mass @xmath135 in case c mass transfer in our galaxy , and the lower mass black holes are certainly more copius than @xmath137 ones . in the case of nova sco and il lupi the explosion was so energetic that the black hole of @xmath138 was only about half of the progenitor he star mass , i.e. the explosion was so violent that nearly half of the mass of the system was lost in the explosion ; a loss of half or more resulting in system breakup . grs 1915@xmath11105 and cyg x@xmath11 did have black holes of @xmath139 , with little mass loss in the explosion , which came from @xmath140 progenitors ( our evolution of grs 1915@xmath11105 in the present paper is an improvement over that in lbw ) . thus , the black holes in galactic soft x - ray transient sources do really come from a wide range of zams mass progenitors . our theory of grbs and hypernova explosions was developed in @xcite and is essentially unchanged . in the meantime we learned in @xcite how to calculate the kerr parameters of the black holes , essentially through an understanding of the tidal locking . our kerr parameters have been checked in our galaxy by the measurement of the kerr parameters of nova sco and il lupi by the smithsonian - harvard coalition . both @xcite and @xcite have shown that when sufficient energy has been delivered to the fireball ( so that the temperature is above the pair - production threshold ) the grb and hypernova explosions follow and the afterglow is that as observed . in this sense the production of energy and the explosion decouple , but the latter follows from the former once sufficient energy is furnished . in this sense we have a complete calculable theory of grbs and hypernovae . the grb and hypernova explosions are just those of the collapsar model of woosley , but with an important improvement ; namely , any required amount of rotational energy is obtained from the tidal spin up of the black - hole progenitor by the donor . the donor then , after furnishing the angular momentum , acts as a passive witness to the explosion , but can show some detail of the latter in the chiefly alpha - particle nuclei which it accretes . the magnetic field lines threading the disk of the black hole are well placed to power the central engine in the blandford - znajek mechanism . the jet formation and hypernova explosion are powered just as in the @xcite collapsar . we check by population synthesis that our binaries are sufficient in number to reproduce all grbs . the great advantage that the soft x - ray transient sources in our galaxy have is that their properties can be studied in detail . they are , however , a special class because of the high metallicity in our galaxy . none the less , it is easy to extend our galactic description to one of low - metallicity galaxies , because the angular momentum energy is determined by the mass of the donor . donors in low - metallicity galaxies tend to be more massive than in high - metallicity ones , furnishing a lower rotational energy . we find that the subluminous grbs come from two sources : 1 ) the galactic metallicity with low - mass donors , where the magnetic field coupling to the black hole disk is so high that it dismantles the central engine before much angular momentum energy can be delivered . nova sco and il lupi are excellent examples of these , in that we have shown that only a tiny part of the angular momentum energy was used up in the explosion of these . 2 ) binaries with low metallicity donors which are massive , going up to the @xmath141 donor in m33 x@xmath17 . somewhere in between these extremes the binaries will have the rotational energies of the cosmological grbs . we estimate the high luminosity grbs to come from binaries with donor masses @xmath142 , but this is uncertain until the kerr parameters of binaries such as xte j1550@xmath1564 are measured . with such a kerr parameter in hand , we can subtract the rotational energy left in the binary from the preexplosion energy which we calculate ( see table [ tab - a ] ) . this will tell us the energy of the explosion . in the cases of nova sco and il lupi , the energy used up in the explosion was tiny compared with the initial rotational energy , but this must change as the initial rotational energy decreases , and black hole mass increases . we have shown that galactically there are 12 relics of grbs and hypernova explosions , and that 3 ( xte j1819@xmath1254 , grs 1915@xmath11105 , cyg x@xmath11 ) might have gone through a low - energy dark explosion , although the first two of these may have gone through grbs and hypernova explosions . so we believe that the soft x - ray black hole binaries are the major sources for the subluminous grbs . we suggest that the luminous cosmological grbs result from a goldilocks " phenomenon , and that only the binaries with donor masses @xmath143 have enough energy for cosmological grbs . we would like to thank jeff mcclintock for many useful discussions . we believe that the smithsonian - harvard group and collaborators , who did not know of our predictions for the @xmath0 s of nova scorpii and il lupi have opened up an exciting new field of activity . g.e.b . was supported in part by the us department of energy under grant no . de - fg02 - 88er40388 . chl was supported by national nuclear r&d program(m20808740002 ) of mest / kosef . in fig . [ bzcirc ] we show that a rotating black hole operates like a generator of electricity ; this is the blandford - znajek @xcite mechanism , which is summarized in the caption . we rely on the relatively complete review by @xcite . the rotational energy of a black hole with angular momentum @xmath144 is a fraction of the black hole mass energy e_rot = f(a _ ) m_bh c^2 , [ eqa1 ] where f(a _ ) = 1 - . for a maximally rotating black hole ( @xmath145 ) f=0.29 . in blandford - znajek mechanism the efficiency of extracting the rotational energy is determined by the ratio between the angular velocities of the black hole @xmath146 and the magnetic field velocity @xmath147 , _ = _ f/_h . for optimal energy extraction @xmath148 . this just corresponds to impedance matching between that in the generator ( fig . [ bzcirc ] ) and that in the perturbative region where the energy is delivered . one can have the analytical expression for the energy extracted e_bz = 1.810 ^ 54_f(a _ ) . [ ebz ] the rest of the rotational energy is dissipated into the black hole , increasing the entropy or equivalently irreducible mass . although the use of @xmath149 is close to the actual efficiency for high @xmath0 , it decreases with @xmath0 from @xmath150 at @xmath151 to @xmath152 at @xmath153 ( @xcite , see @xmath154 , @xmath155 where @xmath156 is the marginally stable radius ) . there would be a further decrease of @xmath157 to the @xmath0 of @xmath115 of cyg x@xmath11 and m33 x@xmath17 . this low efficiency virtually ensures that these two binaries will have gone through dark explosions . the hypernova results from the magnetic field lines anchored in the black hole and extending through the accretion disk , which is highly ionized so the lines are frozen in it . when the he star falls into a black hole , the latter is so much smaller in radius that it has to rotate much faster than the progenitor he star so as to conserve angular momentum . initially large amount of energy up to @xmath158 ergs were attributed to gamma ray bursts . however , when correction was made for beaming the actual gamma ray burst energy is distributed about a mere " @xmath159 ergs @xcite . however , @xmath57ergs are required to clear the way for the jet . hypernova explosions are usually modeled after the nearby supernova 1998bw . the hypernova by @xcite that @xcite compared with nova scorpii had an energy of @xmath160ergs . the amount of energy poured into the accretion disk of the black hole , and , therefore , also pressure is almost unfathomable . , the @xmath161ergs being @xmath162 times the energy of a strong supernova explosion , the latter being spread over a much larger volume than that of the accretion disk . also , @xmath161ergs@xmath163 . near the horizon of the black hole , the physical situation might become quite complicated @xcite . field - line reconstruction might be common and lead to serious breakdowns in the freezing of the field to the plasma ; and the field on the black hole sometimes might become so strong as to push it back off the black hole and into the disk ( rayleigh - taylor instability ) concentrating the energy even more . during the instability the magnetic field lines will be distributed randomly in globs " , the large ones having eaten the small ones . it seems reasonable that the blandford - znajek mechanism is dismantled . later , however , conservation laws demand that the angular momentum not used up in the grb and hypernova explosion be reconstituted in the kerr parameter of the black hole . the radius of the ( kerr ) black hole is @xmath164 given the above scenario of the very high magnetic couplings dismantling the disk by rayleigh - taylor instability , in nova sco , we wish to make a guestimate " of the same effect for xte j1550@xmath1564 , since its kerr parameter is being measured by the smithsonian - harvard collaboration . the black hole radius is proportional to @xmath165 . the ratio of @xmath166 , actually more like @xmath167 because grs 1655@xmath140 has a kerr black hole and in xte j1550@xmath1564 the black hole is about halfway between kerr and schwarzschild . thus , the area of the last stable circular orbit is @xmath168 times larger for xte j1550@xmath1564 . taking the field strength for rayleigh - taylor instability , to go as the inverse of the area , this means that its effect would be cut down by a factor of 6 in going from grs 1655@xmath140 to xte j1550@xmath1564 . if our scenario that the magnetic field coupling is correct for nova sco , not much of the effect would be left in xte j1550@xmath1564 which should accept most of the energy for a cosmological grb . the question then is , what is the latter ? we know that sn1998bw had a hypernova energy of @xmath169bethe . from the equipartition of energy , we would estimate the kinetic energy to clear out a path for the jet in the grb to be about equal to the thermal energy , which is roughly true in @xcite . so the total energy would be @xmath170bethes , which we know can be accepted by the binary . @xcite suggested that the accompanying grb 980425 was smothered " , so that may be a lower limit , although it is larger than estimates we have seen to date , so we choose it as the energy of a cosmological grb . now , 60bethes is @xmath171 of our estimated angular momentum energy for xte j1550@xmath1564 . however , the decrease in kerr parameter necessary to go from 300 to 240bethes is only @xmath172 from our natal @xmath61 , or @xmath173 , which requires an accurate measurement in kerr parameter , but is none the less much larger than the @xmath174 decrease estimated for nova sco . ( note that for schwarzschild black holes , the rotational energy goes approximately as @xmath175 . ) on the other hand , this may be the minimal difference between natal and present angular momentum energies because no other model leaves the system spinning so rapidly . thus , if no difference is conclusively seen because observational errors are @xmath173 then this is also very interesting . suppose a decrease of more than @xmath176 is seen . then this means that we have underestimated the energy of the explosion , but our above estimates are as large as any proposed ones . in any case , the possibility that the system , following the explosion , is left rotating is a new and interesting one . we thus see that we can fit the @xcite condition for no high luminosity grbs in our high metallicity stars in it , because all of the donor masses of the grbs that received the highest rotational energies had companion masses of @xmath107 and the rotational energy furnished to them was so great that the accretion disks were dismantled . the result is that the grbs were subluminous , like the vast majority of grbs . the @xmath168 times larger surface area should be helpful in allowing xte 1550@xmath1564 to accept the rotational energy , but the amount is still tremendous . we do not have any donors of @xmath142 in our galaxy , but such a donor would bring the energy down by @xmath177 , since it goes inversely with donor mass , to @xmath170bethes , that we estimate for grb 980425 , which as noted has nearly galactic metallicity . although grb 980425 is essentially smothered " @xcite , @xmath178bethes is the highest energy anyone has attributed to the grb and hypernova explosion . thus , our estimates would say that donors of @xmath142 would give the most luminous grbs , and that xte 1550@xmath1564 may or may not have been able to accept @xmath170bethes , but since we view this as an upper limit , this binary should still be spinning furiously . we should enter a proviso here . our main thesis is that the black hole binary must have a donor sufficiently massive to slow it down enough so that the black hole can accept the strong magnetic coupling through its accretion disk without the rayleigh - taylor instability entering . by increasing the area of the accretion disk by a factor @xmath168 in going from the @xmath142 black hole in nova sco , to the @xmath35 black hole in xte 1550@xmath1564 the density of magnetic coupling is decreased by a factor @xmath168 . however , the donor in xte 1550@xmath1564 is only @xmath179 , about the same size as in nova sco . therefore , the grb in xte 1550@xmath1564 may still have been subluminous , but less so than nova sco because of the larger disk area . the donor masses at the time of common envelope evolution of xte 1819@xmath1254 and grs 1915@xmath11105 were @xmath180 . however they have only @xmath181 of the energy of the binary with @xmath62 donor , @xmath182bethes . these black holes accrete a lot of matter , more than doubling the black hole mass in the case of grs 1915@xmath11105 and , ultimately will double that mass in xte 1819@xmath1254 . such binaries with substantial mass exchange do not obey our simple scaling which is designed for natal angular momentum . they must be evolved in detail . we believe , that similar evolutionary arguments will apply to binaries with higher mass donors , and , anyway , the angular momentum energy will be cut down by the higher donor masses . thus we expect that only binaries with donor masses @xmath142 will give highly luminous grbs with rotational energy @xmath170bethes . in a very rough estimate , using a flat distribution of donors with mass up to @xmath141 we can estimate that the number of cosmological grbs , those with high luminosity will be @xmath183 of the total , not far from the ratio of grbs with high luminosity to subluminous ones found by @xcite . in summary , the commonly accepted estimates of the explosion energies in grbs are orders of magnitude less than the natal angular momentum energy in nova sco . the measured kerr parameter of @xmath151 @xcite has a present rotational energy indistinguishable from the natal one , within error bars . however , the dismantling of the accretion disk from the very high magnetic couplings should be less in xte j1550@xmath1564 and the rotational energy is somewhat less , the donor being about the mass of the he star in the woosley collapsar model , so we propose that xte j1550@xmath1564 can accept substantial rotation energy but probably less than the energy of a cosmological grb . measurement of this energy is being carried out by the smithsonian - harvard collaboration . if our suggested scenario is confirmed , then this should be strong support for the @xcite binary scenario for grbs . @xcite calculated analytically hypercritical spherical accretion onto compact objects and in particular for the fall back in sn 1987a and obtained the same result @xcite obtained numerically . along this report we have used the evolution of cyg x@xmath11 by @xcite where they find mass transfers from the secondary to the black hole as large as @xmath184 per year for a period on the order of @xmath39 years only to limit the accretion to eddington s limit in the end , ejecting most of the transferred mass out of the binary . we have used the @xcite path in order to evolve not only cyg x@xmath11 , but also v4641 sgr and grs 1915@xmath11105 , except we believe a large fraction of this mass gets accreted hypercritically into the black hole . similarly , @xcite show that m33 x@xmath17 can only be evolved into its current state @xcite , if hypercritical accretion takes place in this system . next we outline the calculation by @xcite : @xcite obtain , for sn 1987a , an accretion rate of = 1.1510 ^ 22 g s^-1=1.8110 ^ -4yr^-1 , [ eq : dotm ] which is the same order of magnitude as the one in @xcite , so we simply follow their results . the eddington accretion rate is _ edd==5.9210 ^ -8yr^-1 , [ eq : edd ] where @xmath185 cm is the radius of the compact object and @xmath186 is the opacity , which we take to be @xmath187 cm@xmath59 g@xmath28 . this is an estimate that applies over a range of temperature and density similar to that present here @xcite . the eddington luminosity @xmath189 , the luminosity for which the pressure of outward traveling photons balances the inward gravity force on the material , is obtained from @xmath190 . if @xmath191 exceeds @xmath192 then some of the accretion energy must be removed by means other than photons . in the present case , = 0.3110 ^ 4 . when @xmath191 exceeds @xmath192 by so large a factor , the accretion rate is called hypercritical , and was considered by @xcite . @xcite finds a trapping radius @xmath193 such that photons within @xmath194 are advected inward with accreting matter faster than they can diffuse outward . we follow his derivation in slightly modified form . we start with the same type of equation as is used in deriving the bondi @xmath191 : = 4r^2v . [ eq : blondi ] here , however , we consider any radius @xmath195 and take @xmath196 to be the free fall velocity , that is @xmath197 . we divide equation [ eq : blondi ] by equation [ eq : edd ] to find _ es = rr_s^-1/2r^-3/2 , where @xmath198 is the schwarzchild radius ; @xmath199 is about @xmath200 km for a @xmath201 compact object . we now find the optical depth for electron scattering to be _ es^_r _ esdr=2r(rr_s)^-1/2 , [ eq : optdep ] where @xmath195 is the distance from the compact object where the photon begins its journey . from random walk , the time for the photon to diffuse over a distance @xmath202 is _ diff=_es , [ eq : diffti ] where @xmath203 is the photon mean free path . the approximation in the last step follows if we assume @xmath202 is large enough that most of the scatterings that the photon undergoes on its trip to infinity have already occurred by the time it has traveled distance @xmath202 . in order to obtain the photon trapping radius , we set this equal to the dynamical time , which is the time it would take for a piece of matter at position @xmath195 to travel the distance @xmath202 at its instantaneous speed at @xmath195 , and approximates the time it takes for a photon to be advected inwards distance @xmath202 : t_dyn==d()^1/2 , [ eq : dynti ] where @xmath204 is the free - fall velocity at @xmath195 . using equation [ eq : optdep ] for @xmath205 and substituting @xmath194 , the trapping radius , for @xmath195 we find r_tr=2r . [ eq : trapr ] since solution of the diffusion equation for radial diffusion in three dimensions decreases the diffusion time by a factor of @xmath206 , we must likewise decrease @xmath194 by this factor . we finally obtain r_tr=0.6r=1.8610 ^ 9 cm . [ eq : trapr3d ] any photon flux emitted much below @xmath194 is unable to diffuse upstream and thus can have no effect on the luminosity reaching the observer at infinity . chevalier and collaborators tool into account that neutrinos can carry away accretion energy and developed self consistent solutions for hypercritical accretion ( chevalier ( 1989 ) , ( 1990 ) ; @xcite ) . in particular they find an expression for the radius of the accretion shock in terms of @xmath191 , for a neutron star . we follow the derivation in chevalier ( 1989 , p.854 ) with small modification . we first derive an expression for @xmath207 , the pressure at the surface of the neutron star , in terms of @xmath191 and @xmath208 , the shock radius . we then examine neutrino cooling near the surface of the neutron star , producing an equation in terms of @xmath207 . insertion of our expression for @xmath207 gives @xmath208 in terms of @xmath191 . since the pressure is radiation dominated , the accretion envelope forms an @xmath209 ( @xmath210 ) polytrope . thus , inside the shock = _ sh()^-3 ; p = p_sh()^-4 ; v = v_sh ( ) , [ eqs : rhopv ] where the subscript sh refers to the value at the shock front . because of the adiabatic compression by factor @xmath211 , _ sh=7_0,[eq : comp ] where @xmath212 is the density just outside the shock front . we neglected the ( small ) decrease in @xmath213 because of increased ionization of the material going through the shock . as in @xcite , @xmath212 is calculated as follows : _ 0= , [ eq : rho0 ] where @xmath214 is the free - fall velocity at the shock radius . from conservation of mass flow across the shock front , v_sh =- v_in . thus , the kinetic energy of the accreting matter is diminished by a factor of @xmath215 ; that is , it is almost entirely converted into thermal energy , so we can estimate the thermal energy density as _ sh_0v_in^2 , the factor 7 entering because of compression ( equation [ eq : comp ] . in this derivation we neglected @xmath216 as compared with @xmath214 . the small correction that would be generated had we taken account of the decrease in @xmath213 due to ionization across the shock in obtaining eq . [ eq : comp ] . since the pressure is radiation dominated , p_sh=_sh_0v_in^2 . inserting equation [ eq : rho0 ] and @xmath217 for @xmath214 , and using the second equation of [ eqs : rhopv ] , we find the pressure at the surface of the neutron star p_ns=1.8610 ^ -12r_sh^3/2dyn cm^-2 , [ eq : pns ] where @xmath191 is expressed in g s@xmath28 and @xmath208 in cm . we took the radius of the neutron star to be @xmath218 km . the energy loss by neutrino pair production per unit volume is @xcite _ n=1.0610 ^ 25t^9c()ergs cm^-3s^-1 , [ eq : doteps ] where @xmath219 is the electron chemical potential , @xmath220 is in mev , and @xmath221 is a slowly varying function of @xmath222 which can be computed from the paper of @xcite . for @xmath223 , @xmath224 ; we shall use this value , because the electrons are not very degenerate . in the region where @xmath225 is operative , @xmath226 mev so that @xmath227 , @xmath228 pairs as well as photons , contribute to the radiation pressure . with @xmath220 in mev , the photon blackbody energy density is w=1.3710 ^ 26t^4ergs cm^-3 . inclusion of the @xmath227 , @xmath228 pairs multiplies this by a factor of @xmath229 , and we divide by three to obtain the pressure p=1.2610 ^ 26t^4ergs cm^-3 . [ eq : p ] combining equations [ eq : doteps ] and [ eq : p ] yields _ n=1.8310 ^ -34p^2.25 , where @xmath230 is in ergs cm@xmath125 s@xmath28 when @xmath232 is in ergs cm@xmath125 . the neutrino cooling is taken to occur within a pressure scale height @xmath233 of the neutron star , or in a volume of @xmath234 . energy conservation gives r_ns^31.8310 ^ -34p_ns^2.25= , [ eq : encons ] with , again , everything in cgs units . inserting equation [ eq : pns ] into equation [ eq : encons ] , we solve for @xmath208 : r_sh6.410 ^ 8()^-10/27 cm . the power @xmath235 is the same as that obtained by @xcite using the accurate neutrino cooling function , and not that of chevalier ( 1989 ) . the detailed calculation of @xcite finds the only substantial correction to our schematic calculation to arise from general relativity , which can be taken into account by multiplying the expression for @xmath208 by @xmath236 . thus , r_sh2.610 ^ 8()^-0.370cm=6.310 ^ 9 cm , where we have used equation [ eq : dotm ] for @xmath191 . @xcite go on to calculate the critical time after which the neutron star left behind by sn 1987a should be visible and find it is less than a year after the explosion . up to this moment , no neutron star has been observed . nevertheless the hypercritical accretion onto a black hole will be just as efficient , if not more than onto a neutron star , since the black hole has no surface against which matter would hit and produce radiation pressure , but only an event horizon which it would ram through without resistance . bethe , h.a . , brown , g.e . , and lee , c .- h . 2003 , _ formation and evolution of black holes in the galaxy : selected papers with commentary _ , world scientific series in 20th century physics vol.33 ( world scientific , 2003 ) . nomoto , k. , mazzali , p.a . , nakamura , t. , iwamoto , k. , maeda , k. , suzuki , t. , turatto , m. , danziger , i.j . , and patat , f. 2000 , _ the greatest explosions since the big bang : supernovae and gamma - ray bursts _ , edited by m. livio , n. panagia , and k. sahu , cup : cambridge , 144 , astro - ph/0003077 . van den heuvel , e.p.j . 1994 , interacting binaries : topics in close binary evolution . in _ lecture notes of the 22nd advanced course of the swiss society for astronomy and astrophysics ( ssaa ) _ ( eds nussbaumer h. and orr a. ) 263 - 474 ( berlin - springer , 1994 ) .

in this article , we study the , , form - factors with the light - cone qcd sum rules , where the-meson light - cone distribution amplitudes are used . in calculations , we observe that the line - shapes of the-meson light - cone distribution amplitude have significant impacts on the values of the form - factors , and expect to obtain severe constraints on the parameters of the-meson light - cone distribution amplitudes from the experimental data in the future . + zhi - gang wang + department of physics , north china electric power university , baoding 071003 , p. r. china pacs numbers : 12.38.lg ; 13.20.he * key words : * meson , light - cone qcd sum rules

1011.3200

the semi - leptonic and radiative @xmath3-decays to the light tensor mesons play an important role in testing the standard model and searching for new physics . at the quark level , the transitions @xmath5 and @xmath6 occur through the flavor changing neutral currents , which are forbidden at the tree - level in the standard model , provide fertile ground for testing the standard model at the loop - level as well as obtaining useful information about new physics effects . the @xmath7 form - factors enter the semi - leptonic decays and radiative decays , and serve as the basic parameters , their values with improved precision can reduce the theoretical uncertainties . experimentally , the radiative decays @xmath8 and @xmath9 have been observed in the past years @xcite . other radiative decays and semi - leptonic decays are expected to be observed at the kek - b and the lhcb in the future . the tensor mesons can not be produced from the local @xmath10 currents of the standard model , the @xmath7 form - factors also play an important role in studying the two - body hadronic decays @xmath11 ( with @xmath12 ) , the calculations based on the naive factorization approach and qcd - improved factorization approach can not account for the experimental data satisfactorily @xcite , the non - factorizable contributions maybe large enough . comparing with the two - body hadronic @xmath3-decays , the semi - leptonic and radiative @xmath3-decays suffer from much less theoretical uncertainties involving the hadronic matrix elements . the @xmath7 form - factors concern the nonperturbative sector of qcd and have been calculated in the isgw model @xcite and its improved version ( the isgw2 model ) @xcite , the covariant light - front quark model @xcite , the light - cone sum rules approach ( where the light - cone distribution amplitudes of the tensor mesons are used ) @xcite , the large energy effective theory @xcite and the perturbative qcd approach @xcite , etc . another calculation based on the light - cone qcd sum rules is useful . in refs.@xcite , khodjamirian et al obtain new sum rules ( the so - called @xmath3-meson light - cone qcd sum rules ) from the correlation functions expanded near the light - cone in terms of the @xmath3-meson distribution amplitudes for the @xmath13 form - factors . furthermore , they suggest qcd sum rules motivated models for the three - particle @xmath3-meson light - cone distribution amplitudes , which satisfy the relations between the two - particle and three - particle @xmath3-meson light - cone distribution amplitudes derived from the qcd equations of motion and heavy - quark symmetry @xcite . the new qcd sum rules have been applied successfully to calculate the @xmath14 form - factors , etc @xcite . in this article , we use the @xmath3-meson light - cone qcd sum rules to study the @xmath15 , @xmath1 , @xmath2 form - factors . the @xmath3-meson light - cone distribution amplitudes play an important role in the exclusive @xmath3-decays , the inverse moment of the two - particle light - cone distribution amplitude @xmath4 enters many factorization formulas @xcite . the two - particle @xmath3-meson light - cone distribution amplitudes have been studied with the qcd sum rules and renormalization group equation @xcite . although the qcd sum rules can not be used for a direct calculation of the distribution amplitudes , it can provide constraints which have to be implemented within the qcd motivated models @xcite . comparing with the light pseudoscalar mesons and vector mesons , the @xmath3-meson light - cone distribution amplitudes have received relatively little attention . our knowledge about the nonperturbative parameters which determine those light - cone distribution amplitudes is limited and an additional application ( or estimation ) based on the light - cone qcd sum rules is useful . on the other hand , if those form - factors are extracted from the experimental data on the semi - leptonic decays and radiative decays at the kek - b and lhcb in the future , we get feedback , and obtain severe constraints on the input parameters of the @xmath3-meson light - cone distribution amplitudes . at the lhcb , the @xmath16 pairs will be copiously produced with the cross section about @xmath17 . the article is arranged as : in sect.2 , we derive the @xmath15 , @xmath1 , @xmath2 form - factors with the light - cone qcd sum rules ; in sect.3 , the numerical result and discussion ; and sect.4 is reserved for our conclusion . in the following , we write down the definitions for the @xmath19 form - factors @xmath20 , @xmath21 , @xmath22 , @xmath23 , @xmath24 , @xmath25 , @xmath26 , and @xmath27 @xcite , @xmath28 \right\ } \ , , \nonumber\\ \langle k^*_2(p)|\bar{s}(0)\sigma^{\mu\nu}q_\nu b(0)|b(p)\rangle&=&2i \epsilon^{\mu\lambda\rho\tau } p_\lambda p_\rho \epsilon_\tau^*t_1(q^2 ) \ , , \nonumber\\ \langle k^*_2(p)|\bar{s}(0)\sigma_{\mu\nu}\gamma_5q^\nu b(0)|b(p)\rangle&=&t_2(q^2)\left [ \left(m^2_b - m_{k^*_2}^2\right)\epsilon_\mu^*-\epsilon^ * \cdot p(p+p)_\mu \right ] \nonumber\\ & & + t_3(q^2)\epsilon^ * \cdot p\left[q_\mu-\frac{q^2}{m^2_b - m_{k^*_2}^2}(p+p)_\mu \right ] \ , , \end{aligned}\ ] ] where @xmath29 @xmath30 , @xmath31 , and the @xmath32 is the polarization vector of the tensor meson @xmath33 . in this article , we write down the calculations for the @xmath34 form - factors explicitly , and obtain others using the flavor @xmath35 symmetry . we study the form - factors with the two - point correlation functions @xmath36 , @xmath37s(x)+\bar{u}(x)\gamma_\beta \left[\overrightarrow{d}_\alpha(x)-\overleftarrow{d}_\alpha(x)\right]s(x)\right\}\ , , \\ j^1_\mu(x)&=&\bar{s}(x)\gamma_\mu b(x)\ , , \nonumber \\ j^2_\mu(x)&=&\bar{s}(x)\gamma_\mu\gamma_5 b(x)\ , , \nonumber \\ j^3_\mu(x)&=&\bar{s}(x)\sigma_{\mu\nu}q^\nu b(x)\ , , \nonumber \\ j^4_\mu(x)&=&\bar{s}(x)\sigma_{\mu\nu}\gamma_5q^\nu b(x)\ , , \end{aligned}\ ] ] where @xmath38 , @xmath39 , @xmath40 , and @xmath41 . according to the quark - hadron duality @xcite , we can insert a complete set of intermediate states with the same quantum numbers as the current operator @xmath42 into the correlation functions @xmath43 to obtain the hadronic representation . after isolating the ground state contributions from the pole term of the tensor meson @xmath33 , we obtain the results , @xmath44 + \cdots \ , , \\ \pi^3_{\alpha\beta\mu}(p , q ) & = & \frac{if_{k^*_2}m_{k^*_2}^2m_bt_1(q^2 ) } { m_{k^*_2}^2-p^2 } \left(-v_\alpha\varepsilon_{\beta\mu\lambda\tau}p^\lambda v^\tau -v_\beta\varepsilon_{\alpha\mu\lambda\tau}p^\lambda v^\tau\right ) + \cdots \ , , \\ z^\alpha z^\beta \pi^4_{\alpha\beta\mu}(p , q ) & = & \frac{f_{k^*_2}m_{k^*_2}^2 } { m_{k^*_2}^2-p^2 } \left[\left(m_b^2-m_{k^*_2}^2\right)t_2(q^2)z_\mu z\cdot v - m_b^2\widetilde{t}_3(q^2)(z\cdot v)^2v_\mu \right ] + \cdots \ , , \end{aligned}\ ] ] where @xmath45 \ , , \nonumber\\ \widetilde{a}_3(q^2)&=&a_2(q^2)-\frac{2m_{k^*_2}\left(m_b+m_{k^*_2}\right)}{q^2}\left [ a_3(q^2)-a_0(q^2)\right]\ , , \nonumber\\ \widetilde{t}_3(q^2)&=&t_2(q^2)-t_3(q^2)\left(1- \frac{q^2}{m_b^2-m_{k^*_2}^2}\right)\ , , \end{aligned}\ ] ] the @xmath46 is a four - vector with @xmath47 , @xmath48 , and the decay constant ( or pole residue ) @xmath49 is defined by @xmath50 the tensor current @xmath42 maybe also have non - vanishing coupling with the vector meson @xmath51 , @xmath52 where the @xmath53 is the polarization vector and the @xmath54 is the decay constant ( or pole residue ) . in this article , we derive the qcd sum rules with the tensor structures @xmath55 , @xmath56 , @xmath57 and @xmath58 to avoid possible contaminations from the @xmath51 meson . in the following , we briefly outline the operator product expansion for the correlation functions @xmath36 in perturbative qcd . the calculations are performed at the large space - like momentum region @xmath59 and @xmath60 , @xmath61 . we write down the propagator of a massive quark in the external gluon field in the fock - schwinger gauge and the @xmath3-meson light - cone distribution amplitudes firstly @xcite , @xmath62\right\}\ , , \end{aligned}\ ] ] @xmath63\gamma_5\right\}_{\beta\alpha } \ , , \nonumber\end{aligned}\ ] ] @xmath64\gamma_5\right\}_{\beta\alpha}\,,\end{aligned}\ ] ] where @xmath65 the @xmath66 and @xmath67 are parameters of the @xmath3-meson light - cone distribution amplitudes . in this article , we take the two - particle and three - particle @xmath3-meson light - cone distribution amplitudes suggested in ref.@xcite and ref.@xcite , respectively . they obey the powerful constraints derived in ref.@xcite and the relations between the matrix elements of the local operators and the moments of the light - cone distribution amplitudes , if the conditions @xmath68 and @xmath69 are satisfied @xcite . we contract the quark fields with wick theorem , substitute the @xmath70 quark propagator and the @xmath3-meson light - cone distribution amplitudes into the correlation functions @xmath36 , then complete the integrals over the variables @xmath71 and @xmath72 , finally obtain the representation at the level of quark - gluon degrees of freedom . in calculations , we take into account the contributions from the two - particle and three - particle @xmath3-meson light - cone distribution amplitudes , the gluons presented in the covariant derivatives @xmath73 , @xmath74 ( or emitted from the vertex ) and emitted from the intermediate quark lines , which correspond to the diagrams ( b ) and ( c ) respectively in fig.1 , both contribute to the three - particle @xmath3-meson light - cone distribution amplitudes . in calculating the diagrams ( b - c ) in fig.1 , we use the fock - schwinger gauge @xmath75 to express the gluon field @xmath76 in terms of the gluon field strength tensor @xmath77 , @xmath78 , and then extract the three - particle @xmath3-meson light - cone distribution amplitudes . in the region of small @xmath79 , the exponential form of ( or the gaussian - like ) distribution amplitude @xmath4 is numerically close to the more elaborated model ( the braun - ivanov - korchemsky model , or the bik model ) suggested in ref.@xcite , @xmath80\ , , \end{aligned}\ ] ] where @xmath81 , and @xmath79 is in unit of gev . the parameters @xmath82 and @xmath83 are determined by the qcd sum rules including the radiative and nonperturbative corrections in the heavy quark effective theory . there are other phenomenological models for the two - particle @xmath3-meson light - cone distribution amplitudes , for example , the @xmath84 factorization formalism @xcite . in this article , we use the qcd sum rules motivated models . after matching with the hadronic representation below the continuum thresholds @xmath85 , we obtain the sum rules for the form - factors @xmath20 , @xmath21 , @xmath86 , @xmath87 , @xmath25 , @xmath26 and @xmath88 , respectively , @xmath89 \right\}e^{-\frac{s}{m^2 } } \right.\nonumber\\ & & + \int_{0}^{\sigma_0}\widetilde{d\sigma}\left\ { \frac{\left[2um_s+(2u-1)\omega\right]\left[\psi_a(\omega,\xi)-\psi_v(\omega,\xi)\right]}{\bar{\sigma}^2m^2 } -\frac{2\left[u\widetilde{\omega}+(1-u)\omega\right]\psi_v(\omega,\xi)}{\bar{\sigma}^2m^2}\right . \nonumber \\ & & + \frac{2u(\widetilde{\omega}+\omega)x_a(\omega,\xi)}{\bar{\sigma}^2m^2 } + \frac{2m_s(\omega-2u\widetilde{\omega})\widetilde{y}(\omega,\xi)}{\bar{\sigma}^3m^4 } + \frac{\widetilde{x}_a(\omega,\xi)}{\bar{\sigma}^2m^2 } \left [ \frac{1 + 8u}{2 } \right.\nonumber \\ & & \left.\left . \left.+\frac{2(2um_s^2-m_s\omega+\omega\widetilde{\omega})}{\bar{\sigma}m^2 } + \frac{2\omega}{m_b\bar{\sigma}}\left ( 1-\frac{\widetilde{m}_b^2}{2m^2}\right)\right ] \right\ } e^{-\frac{s}{m^2}}\right\ } \ , , \end{aligned}\ ] ] @xmath90\frac{m_s}{\bar{\sigma}^2 } \left [ \frac{m_s-\omega'}{m^2}-\frac{1}{2m_b}\left(1-\frac{\widetilde{m}_b^2}{m^2}\right)\right]\right\}e^{-\frac{s}{m^2 } } \nonumber\\ & & + \int_{0}^{\sigma_0}\widetilde{d\sigma}\left\ { \left[\psi_a(\omega,\xi)-\psi_v(\omega,\xi)\right]\left[-\frac{u}{\bar{\sigma } } + \frac{2um_s(m_s-\widetilde{\omega})+\omega(m_s+\widetilde{\omega})-2u\omega\widetilde{\omega}}{\bar{\sigma}^2m^2}\right . \nonumber \\ & & \left.-\frac{2u(m_s+\omega)-\omega}{2m_b\bar{\sigma}^2}\left(1-\frac{\widetilde{m}_b^2}{m^2 } \right ) \right ] + \frac{\psi_v(\omega,\xi)}{\bar{\sigma}^2}\left [ \frac{2\left[u\widetilde{\omega}(\widetilde{\omega}-m_s)+(1-u)\omega\widetilde{\omega}\right]}{m^2}\right . \nonumber \\ & & \left.+\frac{u(\widetilde{\omega}-\omega)+\omega}{m_b}\left(1-\frac{\widetilde{m}_b^2}{m^2 } \right ) \right]+\frac{2ux_a(\omega,\xi)}{\bar{\sigma}^2 } \left [ \frac{\widetilde{\omega}(m_s-\widetilde{\omega}-\omega)}{m^2}-\frac{\omega+\widetilde{\omega}}{2m_b}\left(1-\frac{\widetilde{m}_b^2}{m^2 } \right ) \right ] \nonumber \\ & & + \frac{\widetilde{x}_a(\omega,\xi)}{\bar{\sigma}^2 } \left [ \frac{m_s+6\omega+(1 - 12u)\widetilde{\omega}}{2m^2 } -\frac{1 + 8u}{4m_b}\left(1-\frac{\widetilde{m}_b^2}{m^2}\right ) + \frac{4um_s^2(m_s-\widetilde{\omega}-\omega)}{\bar{\sigma}m^4 } \right.\nonumber \\ & & \left.-\frac{4um_s^2}{\bar{\sigma}m_bm^2}\left(1-\frac{\widetilde{m}_b^2}{2m^2}\right ) -\frac{4\omega}{\bar{\sigma}m^2}\left(1-\frac{s}{2m^2 } \right)-\frac{2\omega}{\bar{\sigma}m_b^2 } \left ( \frac{1}{2}-\frac{\widetilde{m}_b^2}{m^2}+\frac{\widetilde{m}_b^4}{4m^4}\right)\right]\nonumber \\ & & + \frac{2\widetilde{y}_a(\omega,\xi)}{\bar{\sigma}^2m^2 } \left[(2u-1)(\widetilde{\omega}+\omega)+2u\widetilde{\omega } + \frac{m_s\left[2u\widetilde{\omega}(\widetilde{\omega}-m_s)-\omega(\widetilde{\omega}+m_s)+2um_s\omega\right]}{\bar{\sigma}m^2 } \right.\nonumber \\ & & \left.\left.\left . + \frac{m_s(2u\widetilde{\omega}-\omega)}{\bar{\sigma}m_b}\left(1-\frac{\widetilde{m}_b^2}{2m^2 } \right)\right]\right\}e^{-\frac{s}{m^2 } } \right\}\ , , \end{aligned}\ ] ] @xmath91 \right\}e^{-\frac{s}{m^2 } } \right . \nonumber\\ & & + \int_{0}^{\sigma_0}\widetilde{d\sigma}\left\ { \frac{2\left[(1 - 2u)\omega\widetilde{\omega } -2m_s\omega-2u\widetilde{\omega}^2 \right]\left[\psi_a(\omega,\xi)-\psi_v(\omega,\xi)\right]}{\bar{\sigma}^2m^2 } \right.\nonumber \\ & & + \frac{4\widetilde{\omega}\left[(1-u)\omega - u\widetilde{\omega}\right]\psi_v(\omega,\xi)}{\bar{\sigma}^2m^2 } + \frac{4u\widetilde{\omega}(\omega+\widetilde{\omega})x_a(\omega,\xi)}{\bar{\sigma}^2m^2 } \nonumber \\ & & + \frac{\widetilde{x}_a(\omega,\xi)}{\bar{\sigma}^2m^2 } \left [ ( 12u-1)\widetilde{\omega}-6\omega+\frac{8um_s^2(\omega+\widetilde{\omega})}{\bar{\sigma}m^2 } -\frac{4\omega\widetilde{\omega}}{\bar{\sigma}m_b}\left(1-\frac{\widetilde{m}_b^2}{2m^2}\right ) \right.\nonumber \\ & & \left.\left.\left . + \frac{8\omega}{\bar{\sigma}}\left(1-\frac{s}{2m^2 } \right)\right ] + \frac{4\widetilde{y}_a(\omega,\xi)}{\bar{\sigma}^3m^4 } \left[\omega\widetilde{\omega}(2m_s-\widetilde{\omega})+2u(\omega+\widetilde{\omega})\widetilde{\omega}^2 \right ] \right\}e^{-\frac{s}{m^2 } } \right\}\ , , \end{aligned}\ ] ] @xmath92 \right\}e^{-\frac{s}{m^2 } } \right . \nonumber\\ & & + \int_{0}^{\sigma_0}\widetilde{d\sigma}\left\ { \frac{\left[\psi_a(\omega,\xi)-\psi_v(\omega,\xi)\right]\left[2u(m_s-2\widetilde{\omega}-\omega)-\omega \right]}{\bar{\sigma}^2m^2 } -\frac{2\psi_v(\omega,\xi)\left[(1-u)\omega+u\widetilde{\omega}\right]}{\bar{\sigma}^2m^2 } \right.\nonumber \\ & & + \frac{2ux_a(\omega,\xi)(\omega+\widetilde{\omega})}{\bar{\sigma}^2m^2 } + \frac{\widetilde{x}_a(\omega,\xi)}{\bar{\sigma}^2m^2 } \left[\frac{1}{2}+4u+ \frac{2\omega(m_s+\widetilde{\omega})+4um_s^2}{\bar{\sigma}m^2 } + \frac{2\omega}{\bar{\sigma}m_b}\left(1-\frac{\widetilde{m}_b^2}{2m^2}\right ) \right]\nonumber \\ & & \left.\left . + \frac{2\widetilde{y}_a(\omega,\xi)}{\bar{\sigma}^3m^4 } \left[\omega(m_s-2\widetilde{\omega})+2u\widetilde{\omega}(2\widetilde{\omega}-m_s+2\omega ) \right ] \right\}e^{-\frac{s}{m^2 } } \right\}\ , , \end{aligned}\ ] ] @xmath93\right\ } \right.\nonumber\\ & & e^{-\frac{s}{m^2}}+\int_{0}^{\sigma_0}\widetilde{d\sigma}\left\{\left[\frac{u}{\bar{\sigma}}+ \frac{\left[\omega(m_b-\widetilde{\omega}-m_s)-2um_s(m_s+m_b-\widetilde{\omega})+2u\omega(\widetilde{\omega}-m_b)\right ] } { \bar{\sigma}^2m^2}\right ] \right . \nonumber \\ & & \left[\psi_a(\omega,\xi)-\psi_v(\omega,\xi)\right ] + \frac{2\left[(1-u)\omega(m_b-\widetilde{\omega})+u\widetilde{\omega}(m_b+m_s-\widetilde{\omega } ) \right]\psi_v(\omega,\xi)}{\bar{\sigma}^2m^2}\nonumber\\ & & + \frac{2u\left[\widetilde{\omega}(\widetilde{\omega}-m_b - m_s)+\omega(\widetilde{\omega}-m_b)\right]x_a(\omega,\xi)}{\bar{\sigma}^2m^2 } + \frac{\widetilde{x}_a(\omega,\xi)}{\bar{\sigma}^2m^2 } \left[- \frac{6\omega+\widetilde{\omega}+m_b+m_s}{2}\right . \nonumber \\ & & + 4u(\widetilde{\omega}-m_b)+2u\widetilde{\omega}+\frac{2\left[m_b\omega(m_s-\widetilde{\omega})+2um_s^2(\widetilde{\omega}-m_b - m_s+\omega)\right]}{\bar{\sigma}m^2 } \nonumber\\ & & \left.+\frac{4\omega}{\bar{\sigma}}\left(1-\frac{s}{2m^2}\right)-\frac{2\omega(\widetilde{\omega}+m_b - m_s)}{m_b\bar{\sigma}}\left ( 1-\frac{\widetilde{m}_b^2}{2m^2}\right)\right]+\frac{2\widetilde{y}_a(\omega,\xi)}{\bar{\sigma}^2m^2}\left [ ( 1 - 2u)\omega \right.\nonumber\\ & & \left.\left.\left.+(1 - 4u)\widetilde{\omega } + \frac{m_s\left[\omega(m_s+\widetilde{\omega}-m_b)+2u\widetilde{\omega}(m_s-\widetilde{\omega}+m_b)-2um_s\omega\right]}{\bar{\sigma}m^2}\right]\right\}e^{-\frac{s}{m^2}}\right\}\ , , \end{aligned}\ ] ] @xmath94 \right.\right . \nonumber\\ & & \left.+\frac{m_s}{\bar{\sigma}^2 } \left[\frac{m_b(\omega'-m_s)}{m^2}+\frac{m_b+\omega'-m_s}{2m_b}\left(1-\frac{\widetilde{m}_b^2}{m^2}\right ) -\left(1-\frac{s}{m^2}\right)\right ] \left[\widetilde{\phi}_+(\omega')-\widetilde{\phi}_-(\omega')\right ] \right\ } \nonumber\\ & & e^{-\frac{s}{m^2}}+\int_{0}^{\sigma_0}\widetilde{d\sigma}\left\{\left[\psi_a(\omega,\xi)-\psi_v(\omega,\xi)\right ] \left[-\frac{u}{\bar{\sigma}}\left(m_b-\frac{\widetilde{m}_b^2}{2m_b}\right ) \right.\right . \nonumber \\ & & + \frac{m_b\left[2um_s(m_s-\widetilde{\omega})+\omega(m_s+\widetilde{\omega})-2u\omega\widetilde{\omega}\right ] } { \bar{\sigma}^2m^2}+\frac{2u(m_s+\omega)-\omega}{\bar{\sigma}^2}\left(1-\frac{s}{m^2}\right ) \nonumber\\ & & \left . + \frac{\omega(\widetilde{\omega}+m_b+m_s)-2um_s(\widetilde{\omega}+m_b - m_s)-2u\omega(m_b+\widetilde{\omega})}{2m_b\bar{\sigma}^2}\left(1-\frac{\widetilde{m}_b^2}{m^2}\right)\right ] \nonumber\\ & & + \frac{2\psi_v(\omega,\xi)}{\bar{\sigma}^2 } \left [ \frac{m_b\left[(1-u)\omega\widetilde{\omega}+u\widetilde{\omega}(\widetilde{\omega}-m_s)\right]}{m^2 } -\left[(1-u)\omega+u\widetilde{\omega}\right]\left(1-\frac{s}{m^2}\right)\right.\nonumber\\ & & \left.+\frac{(1-u)\omega(m_b+\widetilde{\omega})+u\widetilde{\omega}(\widetilde{\omega}+m_b - m_s)}{2m_b } \left(1-\frac{\widetilde{m}_b^2}{m^2}\right)\right]+\frac{2ux_a(\omega,\xi)}{\bar{\sigma}^2}\nonumber\\ & & \left [ \frac{m_b\widetilde{\omega}(m_s-\widetilde{\omega}-\omega)}{m^2 } -\frac{\widetilde{\omega}(\widetilde{\omega}+m_b - m_s)+\omega(\widetilde{\omega}+m_b)}{2m_b}\left(1-\frac{\widetilde{m}_b^2}{m^2}\right ) + ( \widetilde{\omega}+\omega)\right.\nonumber\\ & & \left.\left(1-\frac{s}{m^2}\right)\right ] + \frac{\widetilde{x}_a(\omega,\xi)}{\bar{\sigma}^2 } \left [ \frac{m_b(m_s+6\omega+\widetilde{\omega})-12um_b\widetilde{\omega}}{2m^2}+ \frac{1 + 8u}{2}\left ( 1-\frac{s}{m^2}\right)\right . \nonumber \\ & & + \frac{4um_s^2m_b(m_s-\widetilde{\omega}-\omega)}{\bar{\sigma}m^4 } + \frac{6\omega+\widetilde{\omega}+m_s - m_b-8u(\widetilde{\omega}+m_b)-4u\widetilde{\omega}}{4m_b}\left(1-\frac{\widetilde{m}_b^2}{m^2}\right ) \nonumber\\ & & \left.+\frac{8um_s^2 - 4\omega(m_b+m_s-\widetilde{\omega})}{\bar{\sigma}m^2}\left(1-\frac{s}{2m^2}\right ) -\frac{4um_s^2(\widetilde{\omega}+m_b - m_s+\omega)}{m_b\bar{\sigma}m^2}\left ( 1-\frac{\widetilde{m}_b^2}{2m^2}\right)\right.\nonumber\\ & & \left.-\frac{2\omega(m_b+m_s-\widetilde{\omega})}{m_b^2\bar{\sigma}}\left(\frac{1}{2}-\frac{\widetilde{m}_b^2}{m^2}+ \frac{\widetilde{m}_b^4}{4m^4}\right)\right]+\frac{2\widetilde{y}_a(\omega,\xi)}{\bar{\sigma}^2}\left [ \frac{m_b\left[2u\widetilde{\omega}+(2u-1)(\omega+\widetilde{\omega})\right]}{m^2 } \right.\nonumber\\ & & + \frac{(2u-1)(\omega+\widetilde{\omega})+2u\widetilde{\omega}}{2m_b}\left(1-\frac{\widetilde{m}_b^2}{m^2}\right ) + \frac{m_sm_b\left[2u\widetilde{\omega}(\widetilde{\omega}-m_s)-\omega(m_s+\widetilde{\omega})+2um_s\omega\right]}{\bar{\sigma}m^4 } \nonumber\\ & & + \frac{m_s\left[2u\widetilde{\omega}(m_b+\widetilde{\omega}-m_s)-\omega(\widetilde{\omega}+m_b+m_s)+2um_s\omega\right]}{m_b\bar{\sigma}m^2}\left(1-\frac{\widetilde{m}_b^2}{2m^2}\right ) \nonumber\\ & & \left.\left.\left.+\frac{2m_s(\omega-2u\widetilde{\omega})}{\bar{\sigma}m^2}\left(1-\frac{s}{2m^2}\right)\right]\right\}e^{-\frac{s}{m^2 } } \right\ } \ , , \end{aligned}\ ] ] @xmath95 \left[m_b \right.\right.\right.\nonumber\\ & & \left.\left.+\frac{m_b\omega'(\omega'-m_s)}{\bar{\sigma}m^2 } + \frac{2\omega'-m_b}{\bar{\sigma}}\left(1-\frac{s}{m^2}\right)\right ] \right\ } e^{-\frac{s}{m^2}}+\int_{0}^{\sigma_0}\widetilde{d\sigma } \left\{\left[\psi_a(\omega,\xi)-\psi_v(\omega,\xi)\right ] \right . \nonumber\\ & & \left[-\frac{um_b}{\bar{\sigma}}+ \frac{m_b\left[2u\widetilde{\omega}(m_s-\widetilde{\omega})-\omega(m_s+\widetilde{\omega})+2u\omega\widetilde{\omega}\right ] } { \bar{\sigma}^2m^2}-\frac{2u\left[(2\widetilde{\omega}-m_b)+2\omega\right]}{\bar{\sigma}^2}\right.\nonumber \\ & & \left.\left(1-\frac{s}{m^2}\right ) + \frac{2u\omega}{\bar{\sigma}^2}\left(1-\frac{\widetilde{m}_b^2}{m^2}\right)\right ] + \frac{2m_b\widetilde{\omega}\left[u(m_s-\widetilde{\omega})+(u-1)\omega\right]\psi_v(\omega,\xi)}{\bar{\sigma}^2m^2 } \nonumber\\ & & + \frac{2u\left [ m_b\widetilde{\omega}(\widetilde{\omega}-m_s+\omega)\right]x_a(\omega,\xi)}{\bar{\sigma}^2m^2 } + \frac{\widetilde{x}_a(\omega,\xi)}{\bar{\sigma}^2m^2}\left [ -\frac{m_b(6\omega+\widetilde{\omega}+m_s-12u\widetilde{\omega})}{2}\right . \nonumber\\ & & \left.+\frac{4um_s^2m_b(\widetilde{\omega}-m_s+\omega)}{\bar{\sigma}m^2}+ \frac{4\omega(m_b+2m_s)}{\bar{\sigma}}\left ( 1-\frac{s}{2m^2}\right)-\frac{2m_b\omega(m_s+\widetilde{\omega})}{\bar{\sigma}m_b}\left(1-\frac{\widetilde{m}_b^2}{2m^2}\right)\right ] \nonumber \\ & & + \frac{2\widetilde{y}_a(\omega,\xi)}{\bar{\sigma}^2m^2}\left [ m_b\left[(1 - 2u)(\omega+\widetilde{\omega})-2u\widetilde{\omega}\right]+\frac{m_bm_s\left[\omega(m_s+\widetilde{\omega } ) + 2u\widetilde{\omega}(m_s-\widetilde{\omega})-2um_s\omega \right]}{\bar{\sigma}m^2 } \right.\nonumber\\ & & \left.\left.\left.-\frac{2\widetilde{\omega}\left[2u\widetilde{\omega}+(2u-1)\omega\right]}{\bar{\sigma}}\left(1-\frac{\widetilde{m}_b^2}{2m^2}\right ) + \frac{4\widetilde{\omega}\left[2u\widetilde{\omega}+(2u-1)\omega\right]}{\bar{\sigma}}\left(1-\frac{s}{2m^2}\right)\right]\right\}e^{-\frac{s}{m^2 } } \right\ } \ , , \nonumber\\\end{aligned}\ ] ] where @xmath96 with a simple replacement , @xmath97 we can obtain the corresponding sum rules for the @xmath98 and @xmath99 form - factors , respectively . in ref.@xcite , lange and neubert observe that the evolution effects drive the light - cone distribution amplitude @xmath4 toward a linear growth at the origin and generate a radiative tail that falls off slower than @xmath100 , even if the initial function has an arbitrarily rapid falloff . the normalization integral of the @xmath101 is ultraviolet divergent . in this article , we derive the sum rules without the radiative @xmath102 corrections , the ultraviolet behavior of the @xmath4 plays no role at the leading order @xmath103 . furthermore , the duality thresholds in the sum rules are well below the region where the effect of the tail becomes noticeable . the nontrivial renormalization of the @xmath3-meson light - cone distribution amplitudes is so far known only for the @xmath101 , we use the light - cone distribution amplitudes of order @xmath103 , which satisfy all qcd constraints . the input parameters for the @xmath3-meson light - cone distribution amplitudes are taken as @xmath104 , @xmath105 @xcite , @xmath106 @xcite , @xmath107 , and @xmath108 @xcite . the masses , decay constants ( or pole residues ) , threshold parameters and borel parameters of the tensor mesons @xmath2 , @xmath1 and @xmath109 are determined by the conventional two - point qcd sum rules . the values are @xmath110 , @xmath111 , @xmath112 , @xmath113 , @xmath114 , @xmath115 , @xmath116 , @xmath117 , @xmath118 , @xmath119 , @xmath120 and @xmath121 @xcite . the tensor meson @xmath109 was originally studied with the qcd sum rules by t. m. aliev et al @xcite . in the conventional two - point qcd sum rules for the tensor mesons , we determine the borel windows by imposing the two criteria ( pole dominance and convergence of the operator product expansion ) to reproduce the experimental values of the masses @xcite . in this article , we take the tensor currents @xmath122 to interpolate the tensor mesons @xmath33 , @xmath1 and @xmath2 as in the conventional two - point qcd sum rules , and expect that the parameters survive . with variation of the borel parameter @xmath123 , other parameters are taken to be the central values . , width=377 ] taking into account all uncertainties of the relevant parameters , we obtain the numerical values of the form - factors @xmath20 , @xmath21 , @xmath86 , @xmath87 , @xmath25 , @xmath26 and @xmath88 . the values of the form - factors @xmath124 , @xmath125 , @xmath126 , @xmath127 , @xmath128 , @xmath129 are very stable with variations of the borel parameters in a large range , the uncertainties originate from the borel parameters are small ( for example , see fig.2 ) ; while the values of the form - factors @xmath130 are not stable enough with the variation of the borel parameter even in the borel window . the form - factors can be parameterized in the double - pole form , @xmath131 where the @xmath132 denote the @xmath20 , @xmath21 , @xmath86 , @xmath87 , @xmath25 , @xmath26 and @xmath88 , the @xmath133 and @xmath134 are the corresponding coefficients . the values of the form - factors at zero momentum transfer and the @xmath133 , @xmath134 are presented in table 1 . in table 2 , we present the central values of the ratios among the form - factors at zero momentum transfer . in this article , we calculate the uncertainties @xmath135 with the formula @xmath136 where the @xmath137 denotes the @xmath7 form - factors , the @xmath138 denotes the input parameters @xmath66 , @xmath139 , @xmath49 , @xmath140 . as the partial derivatives @xmath141 are difficult to carry out analytically , we take the approximation @xmath142 ^ 2 $ ] in the numerical calculations , and take into account the values @xmath143 ^ 2 \neq\left[f(\bar{x}_i- \delta x_i)-f(\bar{x}_i)\right]^2 $ ] . . the values of the form - factors at zero momentum transfer and the parameters of the @xmath7 form - factors , where the bik denotes the @xmath101 is taken as the bik light - cone distribution amplitude . [ cols="^,^,^,^,^,^,^,^",options="header " , ] in calculation , we observe that the dominating contributions in the sum rules for the form - factors come from the two - particle @xmath3-meson light - cone distribution amplitudes , the contributions from the three - particle @xmath3-meson light - cone distribution amplitudes are of minor importance . is defined by @xmath144 . in the sum rules for the @xmath26 and @xmath88 , the dominating contributions come from the two - particle and three - particle @xmath3-meson light - cone distribution amplitudes , respectively . it is obvious that the form - factor @xmath27 acquires its value mainly from the two - particle @xmath3-meson light - cone distribution amplitudes . ] this is the prominent advantage of the @xmath3-meson light - cone qcd sum rules . it is not un - expected that the dominating uncertainty comes from the parameter @xmath66 ( or @xmath82 ) , which determines the line - shapes of the two - particle and three - particle light - cone distribution amplitudes . we can take the value @xmath145 from the qcd sum rules in ref.@xcite , where the @xmath3-meson light - cone distribution amplitude @xmath4 is parameterized by the matrix element of the bilocal operator at imaginary light - cone separation . in tables 3 - 5 , we also present the values ( and the ratios among the central values ) of the @xmath146 form - factors from other theoretical calculations , such as the improved version of the isgw model @xcite , the covariant light - front quark model @xcite , the light - cone sum rules approach @xcite , the large energy effective theory @xcite and the perturbative qcd approach @xcite . from those tables , we can see that the central values of the present predictions are at least ( or almost ) twice as large as the existing estimations . if we take the @xmath101 to be the more elaborated model ( the bik model ) suggested in ref.@xcite , the values of the @xmath7 form - factors at @xmath147 are even larger , see table 1 , where only the central values are presented for simplicity . the two - particle @xmath3-meson distribution amplitudes in the @xmath84 factorization formalism have the same gaussian - like distribution amplitudes @xmath101 and @xmath148 besides additional factors describing the @xmath149-parameter dependence @xcite . in this article , we do not factorize out the @xmath84 explicitly , and can not take the distribution amplitudes in @xmath84 factorization formalism . the form - factors @xmath25 , @xmath26 and @xmath27 are related to the radiative decays @xmath150 . the branching ratio of the radiative decay @xmath151 can be written as @xmath152 where the @xmath153 and @xmath154 are the ckm matrix elements , the @xmath155 is the wilson coefficient for the operator @xmath156 , the @xmath157 is the perturbative @xmath158 corrections @xcite , the @xmath159 is the @xmath3-meson lifetime . in eq.(27 ) , we take into account the factorizable contribution and its perturbative @xmath158 corrections in the standard model . from the experimental data of the radiative decays @xmath160 and @xmath161 @xcite , we can obtain the value @xmath162 , which relates to the parameter @xmath163 in refs.@xcite , the errors originate from the uncertainties of the experimental data and the @xmath149-quark pole mass , respectively . compared with the experimental value , the present prediction @xmath164 seems too large , see table 1 . there are two possible explanations for the apparent discrepancy . one possibility is that the non - factorizable contributions we have neglected in extracting the form - factor @xmath165 from the experimental data ( for example , the contributions of the diagrams where the soft gluon is emitted from the intermediate @xmath166 loops and then absorbed by the @xmath3-meson ) are large enough . if we take into account them consistently , the value extracted from the experimental data maybe compatible with the present calculation . the non - factorizable contributions of the soft - gluon emitted from the intermediate @xmath166 loops have been studied for the @xmath167 and @xmath168 decays @xcite , while the corresponding non - factorizable soft contributions of the @xmath169 decay have not been calculated yet . the other possibility is that the non - factorizable contributions in the decay @xmath169 are small enough to be neglected , the form - factor @xmath165 extracted from the experimental data is precise enough , the apparent discrepancy is due to the shortcoming of the theory . in the following , we perform detailed discussions about the second possibility . in calculation , we observe that if we double the value of the parameter @xmath66 , i.e. taking @xmath170 instead of @xmath171 ( here we introduce a new parameter @xmath172 to avoid confusion ) , we can obtain the value @xmath173 , which is consistent with the theoretical estimations @xmath174 , @xmath175 and @xmath176 from the perturbative qcd @xcite , the light - cone qcd sum rules @xcite and the covariant light - front quark model @xcite , respectively , or the value @xmath177 extracted from the experimental data @xcite . the values ( and the ratios among the central values ) of the @xmath7 form - factors with @xmath178 are presented in tables 1 - 5 , from the tables , we can see that the present predictions of the values of the form - factors are compatible with other theoretical estimations , while the ratios among the central values of the form - factors from different theoretical approaches vary in a large range and no definite conclusion can be made . in the @xmath3-meson light - cone qcd sum rules for the @xmath179 , @xmath180 , @xmath181 , @xmath182 , @xmath183 , @xmath184 , @xmath185 , @xmath186 form - factors @xcite , the main contributions come from the two - particle @xmath3-meson light - cone distributions @xmath187 as in the present case , the form - factors @xmath188 , the model light - cone distribution @xmath189 $ ] with the typical value @xmath190 can give satisfactory results . in the present case , the form - factors @xmath191 due to the derivative in the interpolating currents , if the minor contributions are neglected , the predictions based on the typical value @xmath190 are too large . there are two possible reasons to account for the apparent discrepancy : ( 1 ) the model light - cone distribution amplitudes @xmath187 are universal , but the @xmath3-meson light - cone qcd sum rules are not applicable for the @xmath7 form - factors , although we can prove that the operator product expansion near the light - cone @xmath192 is feasible ; ( 2 ) the @xmath3-meson light - cone qcd sum rules are applicable for the @xmath7 form - factors , but the model light - cone distribution amplitudes @xmath187 are not the ideal ones . a compromise between these two reasons can be suggested , both the model light - cone distribution amplitudes @xmath187 and the @xmath3-meson light - cone qcd sum rules are robust , we can search for the ideal value of the parameter @xmath66 , as the line - shapes of the @xmath3-meson light - cone distribution amplitudes have significant impacts on the values of the form - factors . the analytical expressions of the form - factors are complicated , we can not obtain physical insight on the @xmath66 dependence . we can neglect the tiny contributions from the three - particle @xmath3-meson light - cone distribution amplitudes and other minor contributions , recast the expressions into simple form , and study the @xmath66 dependence . in the following , we will present some typical examples for illustration . the simplified expressions of the form - factors @xmath193 , @xmath194 and @xmath195 in the transitions @xmath196 , @xmath197 and @xmath198 can be written as @xmath199 respectively , where the typical integral kernels are @xmath200 , @xmath201 and @xmath202 , respectively . we can take the input parameters as @xmath203 , @xmath204 , @xmath205 , @xmath206 in the @xmath1 channel , @xmath207 , @xmath208 , @xmath209 , @xmath210 in the @xmath184 channel and @xmath211 , @xmath212 , @xmath213 , @xmath214 in the @xmath215 channel , respectively @xcite , and plot the numerical results in fig.3 . from the figure , we can see that the values of the form - factors decrease monotonously with the increase of the @xmath66 , in the region bellow the typical value @xmath190 @xcite , the form - factors decrease drastically , the curves are very steep , while in the region above the typical value @xmath190 , the form - factors decrease more slowly , and the curves are flatter . if we take the typical value @xmath190 , the values of the form - factors @xmath216 and @xmath195 are consistent with other theoretical estimations @xcite , while the value of the form - factor @xmath193 is too large , on the other hand , if we take the value @xmath217 , the value of the form - factor @xmath193 is consistent with other theoretical estimations while the values of the form - factors @xmath216 and @xmath195 are too small . irrespective of either of the two possible explanations for the apparent discrepancy between the present theoretical calculation and the experimental extraction , we should bear in mind that the values of the parameter @xmath66 from different theoretical approaches differ from each other greatly , @xmath218 from the experiential value in the qcd factorization @xcite , @xmath219 , @xmath220 , @xmath221 from different qcd sum rules @xcite , @xmath222 from the bakamjian - thomas relativistic quark model @xcite , @xmath223 from the operator product expansion @xcite . the two form - factors in the decays @xmath224 are proportional to @xmath225 at the tree - level in the heavy quark limit , those processes are directly related to the parameter @xmath226 and would be the most direct way of measuring it . in searching for the decays @xmath227 , @xmath228 , the babar collaboration has set the upper bounds @xmath229 or @xmath230 @xcite , @xmath231 @xcite . it is difficult to choose the ideal value at the present time . all those theoretical calculations and experimental extractions concern approximations in one or the other ways , and comprehensive theoretical analysis are still needed . at zero momentum transfer , the @xmath232 , @xmath233 and @xmath234 denote the form - factors @xmath235 , @xmath194 and @xmath195 , respectively , the two vertical lines correspond to the typical values @xmath190 and @xmath236 , respectively . in ( ii ) , the form - factors are normalized to 1 at the value @xmath237 . , title="fig:",width=188 ] at zero momentum transfer , the @xmath232 , @xmath233 and @xmath234 denote the form - factors @xmath235 , @xmath194 and @xmath195 , respectively , the two vertical lines correspond to the typical values @xmath190 and @xmath236 , respectively . in ( ii ) , the form - factors are normalized to 1 at the value @xmath237 . , title="fig:",width=188 ] we can extract those form - factors from the precise experimental data on the radiative and semi - leptonic decays @xmath238 at the kek - b and lhcb in the future by including the non - factorizable contributions , new physics effects , etc , in the theoretical analysis , and obtain severe constraints on the input parameter @xmath66 of the @xmath3-meson light - cone distribution amplitudes , although it is a hard work . for example , the semi - leptonic decays @xmath239 and @xmath240 can be described by the effective hamiltonian at the lowest order approximation in the standard model , @xmath241 where the @xmath242 is the ckm matrix element and the @xmath243 is the fermi constant . we can take the form - factors presented in table 1 with @xmath244 as the input parameters to study the partial ( and total ) decay widths @xmath245 ( and @xmath246 ) , and the forward - backward ( fb ) asymmetries @xmath247 of the lepton , @xmath248 where @xmath249 and the @xmath250 is the polar angle of the lepton with respect to the moving direction of the tensor meson in the lepton pair rest frame . taking the other parameters from the review of particle physics @xcite , we can obtain the branching ratios @xmath251 , @xmath251 , @xmath252 , @xmath253 , @xmath253 and @xmath254 for the semi - leptonic decays @xmath255 , @xmath256 , @xmath257 , @xmath258 , @xmath259 and @xmath260 , respectively , which are consistent with the estimations from the perturbative qcd @xcite . the forward - backward asymmetries @xmath261 are shown in fig.4 . in this article , we show the central values explicitly . with the momentum transfer @xmath262 , the @xmath263 , @xmath3 , @xmath264 , @xmath265 , @xmath266 , and @xmath267 correspond to the decays @xmath268 , @xmath269 , @xmath257 , @xmath270 , @xmath271 , and @xmath272 , respectively . , title="fig:",width=188 ] with the momentum transfer @xmath262 , the @xmath263 , @xmath3 , @xmath264 , @xmath265 , @xmath266 , and @xmath267 correspond to the decays @xmath268 , @xmath269 , @xmath257 , @xmath270 , @xmath271 , and @xmath272 , respectively . , title="fig:",width=188 ] in this article , we study the @xmath273 form - factors @xmath20 , @xmath21 , @xmath22 , @xmath274 , @xmath25 , @xmath26 and @xmath27 with the @xmath3-meson light - cone qcd sum rules . in calculations , we observe that the dominating contributions come from the two - particle @xmath3-meson light - cone distribution amplitude @xmath4 , its line - shapes have significant impacts on the values of the form - factors , we can search for the ideal values of the parameter @xmath66 . in the @xmath3-meson light - cone sum rules for the @xmath275 form - factors , the dominating contributions @xmath188 , while in the @xmath3-meson light - cone sum rules for the @xmath7 form - factors , the dominating contributions @xmath276 . if we take the value @xmath277 as in the @xmath3-meson light - cone qcd sum rules for the @xmath275 form - factors , the central values of the present predictions are at least ( or almost ) twice as large as the existing theoretical estimations , and the @xmath128 deviates greatly from the value extracted from the radiative decays @xmath160 and @xmath278 , the non - factorizable contributions are neglected in the extraction . on the other hand , if we take the value @xmath244 , the present predictions are compatible with other theoretical estimations . in calculations , we observe that the main uncertainty comes from the parameter @xmath66 ( or @xmath82 ) , which determines the line - shapes of the two - particle and three - particle @xmath3-meson light - cone distribution amplitudes , it is of great importance to refine this parameter . we can extract the values of those form - factors from the experimental data on the radiative and semi - leptonic decays at the kek - b and the lhcb in the future , and obtain severe constraints on the parameter @xmath66 . this work is supported by national natural science foundation of china , grant number 11075053 , and the fundamental research funds for the central universities .

we study coherent multiple andreev reflections in quantum sns junctions of finite length and arbitrary transparency . the presence of superconducting bound states in these junctions gives rise to great enhancement of the subgap current . the effect is most pronounced in low - transparency junctions , , and in the interval of applied voltage , where the amplitude of the current structures is proportional to the first power of the junction transparency . the resonant current structures consist of steps and oscillations of the two - particle current and also of multiparticle resonance peaks . the positions of the two - particle current structures have pronounced temperature dependence which scales with , while the positions of the multiparticle resonances have weak temperature dependence , being mostly determined by the junction geometry . despite the large resonant two - particle current , the excess current at large voltage is small and proportional to . + pacs : 74.50.+r , 74.80.fp , 74.20.fg , 73.23.ad

cond-mat0101283

transport properties of small conducting structures are strongly influenced by size effects . oscillation of magnetoresistance in thin metallic films , and quantization of conductance in narrow wires and point contacts are examples of such effects . size effects in superconducting tunneling have attracted attention since early experiments by tomasch @xcite . in these experiments , oscillations of the tunnel conductance as a function of applied voltage were found for tunneling from a superconductor to a thin superconducting film of an ns proximity bilayer . the geometric resonance nature of the effect was clearly indicated by the dependence of the period of oscillations on the thickness of the superconducting film . similar conductance oscillations for tunneling into a normal metal film of ns bilayers were reported by rowell and mcmillan @xcite . later on an even more pronounced effect steps on the current - voltage characteristics of sins junctions at applied subgap voltages , @xmath5 was observed by rowell@xcite ( for a review see ref . ) . in addition to the dependence on the thickness of the n - film , the period of the current steps also shows temperature dependence which scales with the temperature dependence of the superconducting gap @xmath3 . the current steps occur at applied subgap voltages , @xmath5 , and they are understood as resonant features due to quasiparticle tunneling through superconducting bound states existing in ins wells at energies lying within the superconducting gap , @xmath6 , de gennes saint - james levels@xcite . recently , properties of superconducting bound states have attracted new attention in connection with observations of conductance anomalies in mesoscopic ns structures . resonant oscillations of the subgap conductance in mesoscopic quasi - ballistic ns junctions have been reported by morpurgo et al@xcite . the zero bias conductance peak , a related resonant phenomenon in diffusive ns junctions , was discovered even earlier@xcite and then intensively studied both theoretically and experimentally ( for a review see ref . ) . in these recent studies of mesoscopic junctions , attention was switched from the single - particle current through superconducting bound states to the two - particle ( andreev ) current . the traditional view of subgap current transport in proximity nins and sins structures considers single - particle tunneling into bound states in the normal region of the ins well@xcite , implicitly assuming that the normal region of the ins well plays the role of equilibrium reservoir . such a model is appropriate for low - transmission tunnel junctions with low tunneling rate compared to the inelastic relaxation rate . however , transparent mesoscopic structures are in a different transport regime where the bound levels are well decoupled from the superconducting reservoirs , and where injected quasiparticles escape from the ins well via andreev reflection.@xcite resonant two - particle current in quantum nins junctions has been theoretically studied in refs .. in sns junctions , the quasiparticles may undergo multiple andreev reflections ( mar ) before they escape into the reservoirs.@xcite in a number of recent experiments with ballistic sns devices fabricated with high mobility 2d electron gas ( 2deg ) @xcite the highly coherent mar transport regime has been realized . the purpose of this paper is to develop a theory of coherent multiple andreev reflections which will be applicable to 2deg - based sns junctions . in 2deg devices the separation of the superconducting electrodes @xmath7 is typically larger than 200 nm , which is of the same order of magnitude as the superconducting coherence length , @xmath8 ( @xmath9 is the fermi velocity of the 2d electrons ) , and superconducting bound states are formed well inside the energy gap . the presence of bound states in the junctions of finite length gives rise to resonances in the mar transport , which dramatically affects the subgap current . furthermore , it is possible in 2deg devices to use electrostatic gates to reach the quantum transport regime with a small number of electron modes and variable transmissivity . a theory of coherent mar has earlier been developed for short superconducting junctions,@xcite @xmath10 , where superconducting bound states do not play any significant role.@xcite such a theory is consistent with the physical situation in atomic - size superconducting point contacts , @xcite where quantization of conduction modes has turned out to be very helpful for detailed comparison between theory and experiment . the purpose of our study is to investigate the interplay between superconducting bound state resonances and coherent mar in long sns junctions , @xmath11 , in the quantum transport regime . in a number of publications , the coherent mar approach has been applied to long sns junctions@xcite . however , these studies were restricted to fully transparent junctions where the bound states are strongly washed out and the resonances are not pronounced ( in fact , as we will show , at zero temperature the current in such junctions does not show any structures ) . we will study junctions with arbitrary transmissivity , @xmath12 and pay special attention to the low - transparency limit , @xmath13 , where the resonance effects are most pronounced . the paper is organized as follows . in sec . [ sninsjunc ] we derive a 1d model for a gated ballistic 2deg device with one transport mode . in sec . [ details ] we construct a scheme for calculating mar amplitudes in terms of wave propagation in energy space . in sec . [ lowtransp ] , single current resonances are studied , and sec . [ interres ] is devoted to a discussion of the interplay between resonances in multi - particle currents . the properties of the total subgap current is discussed in sec . [ discussion ] . we consider an sns junction similar to the one discussed by takayanagi et al.@xcite schematically shown in fig . [ s2degs - dev ] . the junction consists of a normal conducting channel fabricated with a high mobility 2deg , which is confined between superconducting electrodes . the distance between the electrodes is comparable to the superconducting coherence length and small compared to the elastic and inelastic mean free paths and to the normal electron dephasing length . the superconductor-2deg interfaces are highly transmissive , the transmission coefficient typically exceeding a value 0.75 , and the number of conducting modes in the 2deg channel is controlled by a split gate . under these conditions , electrons ballistically move from one electrode to the other while occasionally being scattered by rare impurities or junction interfaces . under a voltage bias applied to the junction , the transport regime corresponds to fully coherent multiple andreev reflections ( mar ) . to calculate the dc current we will apply the scattering theory approach @xcite generalized for superconducting junctions ; see refs . and references therein . the normal electron propagation through the junction is generally described by the @xmath14-channel scattering matrix . by assuming the split gate to select only one transport mode , we will characterize the transport through this mode by energy - dependent transmission amplitude , @xmath15 , and reflection amplitudes , @xmath16 ( the energy @xmath17 is counted from the fermi energy ) . the scattering amplitudes satisfying the unitarity relations , @xmath18 , @xmath19 . the energy dispersion of the scattering amplitudes will introduce the normal electron ( breit - wigner ) and superconducting ( andreev ) resonances in the scattering problem . the effect of narrow breit - wigner resonances on coherent mar was earlier studied by johansson et al.@xcite and levi yeyati et al.@xcite here we will focus on the effect of andreev resonances and only consider breit - wigner resonances which are wide on the scale of the energy gap . this will allow us to neglect the energy dispersion of the junction transparency , @xmath20const . however , the scattering phases may depend on the energy , which yields the andreev resonances . in the simplest case , this dependence is a linear function within the energy interval @xmath21 , and we will write it on the form @xmath22 where @xmath23 are constant . in this case , the scattering properties of the normal channel are similar to those of a 1d nin junction . indeed , the corresponding 1d transfer matrix , @xmath24 can be decomposed into a product of three transfer matrices , @xmath25 where @xmath26 is a pauli matrix . the first and the last matrices describe ballistic propagation of an electron , with wave vector @xmath27 , through the right and left n - regions of an effective junction with lengths @xmath28 and @xmath29 respectively ( from right to left ) , and the matrix @xmath30 describes an effective barrier ( i ) . quasiparticle propagation through the effective 1d snins junction is described by means of the time - dependent bogoliubov - de gennes ( bdg ) equation , @xcite @xmath31 where @xmath32 is the normal electron hamiltonian , @xmath33 is the impurity potential and @xmath34 is the applied voltage . the superconducting order parameter @xmath35 is constant within the superconducting electrodes and zero within the normal region @xmath36(see fig . [ potentialdiag ] ) . in the further calculations , the impurity potential is described by the transfer matrix @xmath37 in eq . ( [ t ] ) . the spatial distribution of the applied potential along the channel is modeled with a step - like function @xmath38 . in fact , the actual spatial distribution of the potential does not play any role in this system : it can be included in the transfer matrix in eq . ( [ t ] ) , leading to an additional energy - independent shift in the scattering phases in the matrix @xmath37 . as we will see later [ comment after eq . ( [ resphase ] ) ] , the energy - independent phases in the @xmath37-matrix do not affect the current , and can therefore be excluded . the phase difference between the two superconductors follows from the josephson relation ( @xmath39 ) and introduces time dependence into the problem . the superconducting electrodes are considered to be equilibrium reservoirs where the quasiparticle wave function is a superposition of electron- and hole - like plane waves , @xmath40 in this equation , the @xmath41 signs in the time - dependent factors refer to the left / right electrode , @xmath42 is the wave vector of electron / hole - like quasiparticles , and @xmath43 , @xmath44 are the bogoliubov amplitudes . the ratio of the bogoliubov amplitudes equals the amplitude of andreev reflection for particles incoming from the neighboring normal region , @xmath45 since the time dependencies of the wave functions in the two reservoirs are different , the quasiparticle scattering by the junction is inelastic , and one has to consider a superposition of plane waves with different energies in order to construct scattering states . we will now proceed with the construction of recurrences for the scattering amplitudes following the method suggested by johansson et al . @xcite . to this end we introduce the wave functions in the left / right normal region ( @xmath46 ) of the junction with respect to the position of the impurity . a particular scattering state , labeled with the energy @xmath17 of the incoming quasiparticle , will consist of a superposition of plane waves with energies @xmath47 , where @xmath48 is an integer , @xmath49 , @xmath50\ ] ] @xmath51.\ ] ] the normal electron / hole wave vector @xmath52 is here defined as @xmath53 , @xmath54 . the meaning of the labels for the scattering ( mar ) amplitudes @xmath55 will be explained below . continuity of the scattering state wave function across the left and right ns interfaces determines the relation between the electron and hole amplitudes in the vicinity of each interface , @xmath56 which describes elastic andreev reflection ( l / r indices are omitted ) . it is convenient to consider scattering amplitudes near the impurity ( at @xmath57 ) rather than at the ns interfaces , and to rewrite eq . ( [ andrreflection ] ) for such amplitudes , combining the amplitudes of the ballistic propagation through the normal regions with the andreev reflection amplitude . then , in vector notation , @xmath58 the modified relation ( [ andrreflection ] ) takes the form @xmath59 where @xmath60 the phase @xmath61 , characterizing @xmath62 , is real inside the energy gap , @xmath63 , where it describes the total energy dependent phase shift due to ballistic propagation and andreev reflection . outside the gap , @xmath64 , the phase @xmath65 has an imaginary part which describes leakage into the superconducting reservoirs due to incomplete andreev reflections . by matching harmonics with the same time dependence in eq . ( [ ansatz ] ) , we derive a relation between scattering amplitudes at the left and the right side of the barrier : @xmath66 where the effective barrier transfer matrix @xmath67 is defined in eqs . ( [ te1 ] ) , ( [ t ] ) . the recursion relations in eqs . ( [ urelation ] ) and ( [ trelation ] ) couple the scattering amplitudes @xmath68 into an infinitely large equation system . this equation system describing coherent mar is illustrated by the mar diagram in fig . [ coeffpic ] . the electron part of the quasiparticle injected at the left ns interface propagates upwards along the energy axis , the amplitudes for this propagation being labeled with @xmath69 . at the injection energy @xmath70 ( amplitude @xmath71 ) , the quasiparticle is accelerated across the barrier ( i ) , where the potential drops . thus , it enters the right normal part of the junction with energy @xmath72 ( @xmath73 ) , undergoes andreev reflection and goes back as a hole ( @xmath74 ) , entering the left normal part of the junction having been accelerated to energy @xmath75 ( @xmath76 ) , and is then again converted into an electron ( @xmath77 ) . the @xmath41 indices label the amplitudes after ( @xmath78 ) and before ( @xmath79 ) the andreev reflection for propagation upwards along the energy axis . there is a similar trajectory of injected holes , which descends in energy , with the mar amplitudes labeled with @xmath80 . due to electron back scattering at the barrier , the upward and downward propagating waves are mixed , e.g. @xmath71 being not only forward scattered into @xmath73 , but also back - scattered into @xmath81 , which opens up the possibility of interference . injection from the left reservoir , shown in fig . [ coeffpic ] , generates a mar path which only connects even side bands at the left side of the junction with odd side bands at the right side . injection from the right reservoir will generate a different mar path , with even side bands at the right side of the junction , i.e. the diagram in fig . [ coeffpic ] will effectively be mirrored around the barrier ( i ) . thus , there are two independent equation systems for the mar amplitudes : injection from the left and from right . the @xmath82 labels in the mar amplitudes can then be omitted since they are uniquely defined by the source term and the side band index . the transport along the energy axis generated by mar , from energy @xmath17 to @xmath83 , is conveniently described by the effective transfer matrix @xmath84 , @xmath85^{-1}\hat { c}_{0- } , \ \ n<0\ ] ] @xmath86 where @xmath87 and @xmath88 for the injection from the left ( for the injection from the right , the even and odd side band indices are interchanged ) . for paths within the superconducting gap , @xmath89 , the matrix @xmath84 satisfies the standard transfer matrix equation , @xmath90 , which provides conservation of probability current along the energy axis , @xmath91 an important consequence of the coherence of mar is the possibility of transmission resonances in energy space . from the form of the @xmath92-matrix , @xmath93 it is evident that when @xmath94 , the two matrices @xmath95 and @xmath67 will cancel each other , and the probability of transmission through this part will be unity , which leads to resonant enhancement of mar . the solutions @xmath96 of the resonance equation , @xmath97 coincide with the spectrum of the de gennes saint - james levels localized in ins quantum wells @xcite . the corresponding bound states are located either on the left or the right side of the junction . without loss of generality , the calculations can be performed with a real matrix @xmath67 . the transformation to such a real matrix is given by @xmath98 , with diagonal unitary matrices @xmath99 whose elements are constructed with the scattering phases , which are energy - independent . it is clear from eq . ( [ mresstruc ] ) that since these energy - independent matrices commute with the matrices @xmath100 , they cancel each other , and the matrix @xmath84 undergoes a similar transformation . this will lead to an overall phase shift of the scattering state which does not affect the current . it is interesting to consider the special case of fully transparent junctions , @xmath101 , which has been studied in the literature @xcite . in this case , all matrices @xmath102 in eq . ( [ mdef ] ) are equal to the unity matrix , and the @xmath92-matrix takes the simple form @xmath103 . the length of the junction then enters only through the phase of the mar amplitudes , which drops out of the side band current . thus the dc current of fully transparent sns junctions is independent of length and equal to the current in quantum constrictions @xcite . in particular , at zero temperature this current does not show any structures in the subgap current . it is also worth mentioning that in this particular case of fully transparent sns junctions , the @xmath92-matrix is diagonal and therefore a closed set of recursive relations can be derived for the mar probabilities ( not just for the mar amplitudes , as in the general case ) , equivalent to the equations for distribution functions derived in the original paper by klapwijk , blonder and tinkham @xcite . equation ( [ mar ] ) describes `` source - free '' propagation along the mar ladder . to complete the set of equations for the mar amplitudes we need to take into account quasi - particle injection , which introduces a source term in eq . ( [ mar ] ) . to this end , let us consider a quasiparticle incoming from the left superconducting electrode with energy @xmath17 , having a wave function of the form @xmath104 .\ ] ] the two terms in this equation refer to electron- ( @xmath105 ) and hole - like ( @xmath106 ) injected quasiparticles . we now include this wave function into the continuity condition at the ns interface at energy @xmath17 , which gives us the following relation between the mar amplitudes @xmath107 and @xmath108 , @xmath109 @xmath110 for quasiparticles injected from the right , a similar equation holds with the substitutions @xmath111 , and @xmath112 . equations ( [ mar ] ) and ( [ source ] ) give a complete set of equations for the mar amplitudes with the boundary conditions @xmath113 at infinity . a formal solution of eqs . ( [ mar ] ) and ( [ source ] ) , which is useful both for numerical calculations and analytical investigations , can be constructed by reducing this infinite set of recursion relations to a finite set by representing the mar process above @xmath114 and below @xmath115 by boundary conditions involving reflection amplitudes @xmath116 and @xmath117 , defined as @xmath118 and @xmath119 . this gives the following representation for the vectors in eq . ( [ coeffdef ] ) , @xmath120 the reflection amplitudes @xmath116 and @xmath117 are independent of the injection , in contrast to the coefficients @xmath121 . furthermore , they are determined by the boundary conditions @xmath113 and can be expressed in terms of the matrix elements of @xmath122 and @xmath123 , where @xmath124 , @xmath125^{-1}\left ( \begin{array}{c } r_{0- } \\ 1 \end{array } \right)=0.\ ] ] in other words , the vectors in eq . ( [ lim ] ) are equal to the asymptotical values of the eigenvectors of @xmath126-matrices corresponding to the eigenvalues which decrease when @xmath14 goes to infinity . the advantage of introducing the reflection amplitudes @xmath116 and @xmath117 is that although they have to be calculated numerically , the recurrences which they obey do not contain resonances and converge rather quickly . this is in contrast to the matrix @xmath84 , which does possess resonances , but which can be calculated analytically in a straight forward way for any given @xmath48 . the solutions of the recursion equations eqs . ( [ urelation ] ) and ( [ trelation ] ) can now be explicitly written down . for any given energy @xmath17 we get four different sets of solutions for four scattering states including electron / hole injection from the left and the right . using the formal expression in eq . ( [ nullvectors ] ) and the matrix elements of @xmath127 , the solutions for injection from the left ( @xmath128 ) have the form , @xmath129 } { m_{22}+m_{21}r_{0-}e^{2i\varphi_{0 } } -m_{12}r_{n+}e^{2i\varphi_{n } } -m_{11}r_{0-}r_{n+}e^{2i\varphi_{0}+2i\varphi_{n } } } \left(\begin{array}{c } 1\\ r_{n+ } \end{array}\right).\ ] ] the solutions for injection from the right can be found through interchanging @xmath130 and calculating all quantities with respect to injection from the right . the solutions for @xmath131 are calculated in a similar manner . now turning our attention to the current , we calculate it in the normal region next to the barrier , using the wave function in this region , @xmath132 , and assuming quasiparticle equilibrium within the electrodes . the current then takes the form , @xmath133 where @xmath134 is the superconducting density of states , and the sum is over the four scattering states at a given energy @xmath17 associated with the electron- and hole - like quasiparticles ( e / h ) injected from the left and right ( l / r ) . the current can be divided into parts with different time dependence and expressed as a sum over harmonics : @xmath135 focusing on the dc ( @xmath136 ) component , and calculating the contribution of each scattering state at the injection side of the junction , we express the current spectral density @xmath137 through the probability currents of electrons and holes at energies @xmath138 ( fig . [ coeffpic ] ) , @xmath139 @xmath140 these currents coinside with the probability currents @xmath141 , eq.([jp ] ) , flowing along the energy axis . it is convenient to introduce a leakage current @xmath142 , defined as the difference of the probability currents before and after andreev reflection , @xmath143 @xmath142 represents the amount of probability current from all the scattering states injected at energy @xmath17 and leaking out of the junction at energy @xmath114 ( fig . [ coeffpic ] ) . the leakage current is zero inside the energy gap due to complete andreev reflection , @xmath144 , @xmath145 [ cf . ( [ jp ] ] . the explicit expression for the leakage current for @xmath146 follows from eq . ( [ leakage ] ) after insertion of eqs . ( [ coeffsol ] ) and ( [ urelation ] ) , @xmath147 it follows from eq . ( [ ipn ] ) that the leakage currents are positive for all @xmath146 , @xmath148 . one can also show that they satisfy the inequality @xmath149 , which is a consequence of the conservation of probability current : the leakage current of all side bands except of the side band @xmath150 does not exceed the probability current injected into four scattering states . furthermore , the leakage current satisfies the important detailed balance equation @xcite , @xmath151 i.e. the leakage at energy @xmath152 due to the injection at energy @xmath17 is the same as the leakage at energy @xmath17 due to injection at energy @xmath152 . using the continuity of current across the barrier , @xmath153 , guaranteed by the transfer matrix @xmath67 , we can express the probability currents in eq . ( [ dccurrent ] ) through the leakage current , eq . ( [ leakage ] ) , @xmath154 @xmath155 by adding and subtracting consecutive terms in the sum . the spectral density of the dc charge current eq . ( [ dccurrent ] ) can then be written on the form @xmath156 since @xmath142 appears in @xmath48 probability currents . this formula has a clear physical meaning : the contribution to the charge current of the @xmath48-th side band is proportional to the leakage current of the side band times the effective transferred charge @xmath157 . the detailed balance of the leakage currents , eq . ( [ cancel ] ) , allows us explicitly to prove that at zero temperature the scattering processes between ( occupied ) states with negative energies , @xmath158 do not contribute to the current , in agreement with the pauli exclusion principle . indeed , by separating the contributions from side bands with @xmath131 and remembering that the leakage current is zero within the gap , we get for zero temperature , @xmath159 where the first and the third terms cancel each other by virtue of eq . ( [ cancel ] ) . at finite temperature , these two terms produce current of thermal excitations while the second term gives the current of real excitations created by the voltage source . keeping only this term which dominates at low temperature , we finally get , @xmath160 we end this section by noting a technically useful symmetry in the current density , namely @xmath161 , seen from the explicit form of the @xmath84-matrix . this allows us to reduce the integration interval in eq . ( [ npartcurrent ] ) to @xmath162 . the approach formulated above provides necessary foundations for numerical calculation of the current for arbitrary transparency and length . however , to get a full understanding of the rich subgap structure in the current - voltage characteristics , which may seem quite random , especially for intermediate transparencies and lengths ( see figs . [ itotd1plots]-[itotl4plots ] ) , we will conduct a detailed analytical study of the limit of low transparency @xmath13 . the separation of currents into @xmath48-particle currents , eq . ( [ npartcurrent ] ) , is our basis for analysis , and we will study each current @xmath163 separately . as explained in the previous section , the de gennes saint - james levels , eqs . ( [ mresstruc ] ) and ( [ resphase ] ) , are important for the current transport through the junction , leading to resonant enhancement of the current . our main attention in this and the next section is on the calculation of the position , height and width of the main current peaks and oscillations which have the magnitude of order @xmath2 . to simplify notations , left / right injection indices are omitted in most cases . the single - particle current , which dominates at large applied voltages , has , according to eq . ( [ npartcurrent ] ) , an onset at @xmath164 . the full numerical solution for the single - particle current is plotted in fig . [ i1d1plots ] . the current shows pronounced oscillations , and the magnitude of the slope at the current onset strongly depends on the junction length . to understand this behavior , we analyze eq . ( [ ipn ] ) in the limit of small transparency @xmath0 , i.e. in the tunnel limit . first we note ( see appendix [ refexp ] ) that the reflection amplitudes @xmath116 and @xmath117 may be expanded as @xmath165 @xmath166 after inserting the explicit form of @xmath167 together with the expansion ( [ reflectionexp ] ) into eq . ( [ ipn ] ) and putting @xmath168 , we can write the single particle current on the form @xmath169 \tanh(|e|/2k_bt),\ ] ] where @xmath170 in analogy with the tunnel formula for the current , @xcite @xmath171 is identified as the tunneling density of states ( dos ) on the left / right side of the junction . in fig . [ densplot ] the energy dependence of the dos is presented . the deviation of this dos from the normal metal density of states is a manifestation of the proximity effect . the expression ( [ densityofstates ] ) for the dos has earlier been derived for proximity ns sandwiches @xcite . note that the dos in our case is constant throughout the n - regions . in junctions with arbitrary length , the dos usually approaches zero at the gap edge @xmath172 ( ref . ) . exceptions are junctions with lengths @xmath173 where a bound state splits off from the gap edge . in this case , the dos diverges at the gap edge . the quantum well structure of the sns junctions also give rise to quasi - bound states in the continuum spectrum , @xmath174 , seen as oscillations in the dos . the single particle current in eq . ( [ i1smalld ] ) is written as the integral over the product of the dos at the entrance energy @xmath17 and the exit energy @xmath175 . the latter depends on the applied voltage , as well as the integration interval , and therefore the dos oscillations produce oscillations of the current @xmath176 as a function of voltage ( rowell - mcmillan oscillations @xcite ) . the oscillations become more pronounced when the junction is sufficiently long , and the differential conductance may even become negative . it is also clear that the dos oscillates as a function of the length of the junction , which give rise to oscillations also in @xmath176 . in short junctions , @xmath177 , the current onset at @xmath164 is very steep , see fig . [ i1d1plots ] . in junctions with finite length , the current onset is smeared and replaced with a smooth oscillating behavior . this can be directly related to the smearing of the singularity in the dos at the gap edge . the length where the crossover between these two behaviors occurs can be taken as a measure of when finite length effects become important . to estimate this length , we write equation ( [ ipn ] ) for small lengths @xmath178 , near the threshold , @xmath179 , @xmath180 , keeping the first order terms in @xmath2 in the denominator . for a symmetric junction , @xmath181 , we get @xmath182 from this formula it is clear that for short junctions ( @xmath183 ) , the current onset has the width @xmath184 . if @xmath7 is of the order of @xmath185 , the size of the onset has substantially diminished , and there is no visible onset at @xmath164 when @xmath186 . this crossover between steep onset and smooth behavior , which happens already for quite short lengths if @xmath2 is small , can be interpreted in terms of a bound state which is situated exactly at the gap edge in short junctions ( @xmath183 ) , and which moves down into the gap when @xmath187 , the effect becoming fully pronounced when the distance from the gap edge , @xmath188 , exceeds the dispersion of the andreev state , @xmath189 , in symmetric junctions@xcite . when @xmath190 approaches @xmath191 , the lowest quasi - bound state in the continuum spectrum approaches the gap edge . this leads to an accumulation of the spectral weight at the gap edge and reappearance of the singularity in the dos , which results in the reappearance of a sharp current onset at @xmath164 , but with smaller magnitude ; see fig . [ i1d1plots ] ( @xmath192 ) . it is of interest to note that in our calculations , based on the scattering theory approach , the bound states are not directly involved in the single - particle transport , which therefore is non - resonant and shows no subgap resonance peaks . within the tunnel model approach the situation is qualitatively different : the dos in eq . ( [ i1smalld ] ) usually includes the contribution of the broadened bound states , and therefore the single - particle current exists and has pronounced resonant features at subgap voltages @xmath5 . this difference results from the fact that , within the tunnel model approach , the superconducting bound states are implicitly assumed to be connected to the reservoirs ( broadening due to inelastic interaction ) , which allows a stationary current to flow through the bound states . in contrast , within the scattering approach , the bound states are disconnected from the reservoirs and have zero intrinsic width . in this case the bound states obtain their width only due to higher order tunneling processes involving andreev reflections , which are manifested by the resonant multi - particle currents . in practice , the relevance of the multi - particle versus single - particle mechanism of the subgap current transport is determined by physics and depends on the ratio of the corresponding dwelling and relaxation times@xcite . in this paper , the inelastic relaxation time @xmath193 which determines the width of the single - particle resonances is assumed to be much larger than the dwelling time of the most important two - particle current , @xmath194 . the two - particle current @xmath195 in quantum point contacts , ( @xmath178 ) is of order @xmath196 when @xmath5 and of order @xmath197 when @xmath198 ( ref . ) . for finite length junctions , the situation is different . for the mar paths where the energy of the andreev reflection coincides with a bound state , the current spectral density @xmath199 is of order unity , due to resonant transmission through this state . for low transparency @xmath0 , this gives a sharp concentration of the current density around the resonant energies . in this limit , the two - particle current is well described by the sum of contributions from these resonances , and to evaluate them we examine the energy dependence of @xmath200 close to the resonant energies , @xmath201 . let us consider the contribution to the leakage current @xmath202^{l}$ ] from quasi - particles injected from the left . as shown in appendix [ twopartres ] , in this case eq . ( [ ipn ] ) reduces to the standard breit - wigner resonance form @xmath203^{l}=\frac{\gamma_{0}^{(m ) } \gamma_{2}^{(m)}}{\displaystyle \left(\frac{\delta e-\delta e^{(m)}}{\delta}\right)^{2}+ \left(\frac{\gamma_{0}^{(m)}+\gamma_{2}^{(m)}}{2 } \right)^{2 } } , \ ] ] where the tunneling rates @xmath204 are given by @xmath205 , @xmath206 , and @xmath207 and the position of the resonance is shifted by @xmath208 an analogous result is valid for quasiparticles injected from the right . after integrating over energy , the two - particle current in the resonance approximation may be written on the form @xmath209 where the summation is over the positive bound level energies , @xmath210 , and the dos @xmath211 should be calculated at the injection side of the junction and @xmath212 $ ] . according to eq . ( [ i2approx ] ) , the two - particle current @xmath213 increases in a step - like manner in the voltage region @xmath214 . the steps occur at every voltage where a new resonant channel through a bound state opens up , at @xmath215 . we note that the step positions depend on temperature and approximately scale with @xmath3 . each current step has the height of order @xmath2 . as seen from eq . ( [ i2approx ] ) , the contribution to the current of a particular bound state @xmath96 is modulated , as a function of voltage , by the oscillations of the density of states at the entrance and exit energies , @xmath216 . in other words , the pronounced oscillations of the two - particle current seen in fig . [ i2d1plots ] reflect how close the entrance and exit energies @xmath217 are to a quasi - bound state in the continuum . for @xmath198 , the two - particle current @xmath195 oscillates around a constant value with an amplitude of oscillation decreasing as @xmath218 for large voltages . it is interesting to compare the resonant structures of the two - particle current with the resonant structures in nins junctions.@xcite in nins junctions , the resonant current steps occur at @xmath219 , and they do not have any modulation because the dos on the normal side of the junction is constant . the distance between the resonances and the resonance widths are proportional to the bound level spacing , and they decrease in long junctions . for sufficiently long junctions , the two - particle current may thus give the appearance to include a series of peaks , as shown on fig . [ i2comp ] . in symmetric junctions , the bound state energies at both sides of the barrier will coincide , reducing the number of steps with a factor of two , and giving current steps of double height . we will conclude this subsection by noting that the difference between the full numerical calculation of the two - particle current and the resonant approximation given in eq . ( [ i2approx ] ) is rather small already when @xmath220 , as can be seen in fig . [ i2comp ] . excess current in sns junctions , i.e. the difference between the current in superconducting junction and in the normal junction at large voltage , @xmath221 is commonly considered as a measure of the intensity of andreev reflection . in tunnel sis junctions and low - transmissive point contacts the excess current is small , @xmath222 , @xmath13 , while in fully transparent contacts the excess current is large , @xmath223 , @xmath101.@xcite accordingly , one would expect large excess current in long sns junctions due to the resonant enhancement of the two - particle current . however , the excess current is small because of a large deficiency , of order @xmath2 , of the single - particle current caused by the broadening of the current onset at the threshold . as we will show , the single - particle and two - particle currents undergo a fine cancellation , yielding small net excess current of order @xmath196 when @xmath13 . the excess current has contributions only from the single- and two - particle currents , since all higher order currents include at least one andreev reflection outside the gap whose probability is of order @xmath224 . in the limit of large voltage , @xmath225 , the relevant part of the current in eq . ( [ npartcurrent ] ) then takes the form @xmath226 @xmath227 these equations are written for symmetric junctions , @xmath228 , and for zero temperature ; small andreev reflection amplitudes @xmath229 have been neglected in eq . ( [ ipn ] ) . the behavior of the current in eq . ( [ ihighvoltage ] ) as a function of voltage is presented in fig . [ excessapproach ] for different lengths . it is clearly seen that the limiting value of the excess current is approached much faster in finite length sns junctions compared to point contacts ( @xmath183 ) . in fig . [ excessplot ] the excess current behavior with respect to the junction length is presented for different transparencies . to analytically examine the excess current in the limit of small transparency , @xmath13 , it is convenient to start with equations ( [ i1smalld ] ) and ( [ i2approx ] ) . to first order of @xmath2 the excess current assumes the form ( @xmath230 ) , @xmath231 @xmath232de,\ ] ] @xmath233 let us consider the contributions to the single - particle current from the left electrode , @xmath234^{l}=-\frac{2ed\delta}{h } + \frac{2ed}{h}\int_{\delta}^{\infty}\left [ n^{l}(e)-1\right]de.\ ] ] inserting @xmath235 from eq . ( [ densityofstates ] ) , this equation can be transformed to the form , @xmath236^{l}= \frac{2ed}{h}\int_{\delta}^{\infty}\left(\frac{e\xi } { \xi^{2}+\delta^{2}\sin^{2}(2el_{l}/\delta\xi_{0})}- \frac{e}{\xi}\right)de= -\frac{ed}{h}\int_{-\infty}^{\infty}d\xi \frac{\sin^{2}(2el_{l}/\delta\xi_{0})}{\xi^{2}+\sin^{2 } ( 2el_{l}/\delta\xi_{0})},\ ] ] where @xmath237 . it is now possible to analytically continue the integral in the upper half plane which will reduce the integral to a sum over the residues of the poles given by the equation @xmath238 . comparing this equation with eq . ( [ resphase ] ) we find that the poles coincide with the energies of the bound states in the gap . the excess current contribution from the left - injected single particle current is thus @xmath236^{l } = -\frac{2d\pi e\delta}{h}\sum_{m\geq0 } \frac{1}{\eta^{(m)}}=-\left[i^{exc}_{2}\right]^{l},\ ] ] where @xmath239^{l}$ ] is the contribution to the two - particle current from the bound state resonances at the left electrode . a similar relation is derived for current from the right electrode . thus , there is exact cancellation of the excess single - particle and two - particle currents to first order in @xmath2 . it is interesting to note that the cancellation effect is related to the conservation of the number of states in a proximity normal metal compared to the conventional normal metal . it follows from eq . ( [ i1excl ] ) that @xmath240 is equal to the difference between the number of continuum states in the proximity metal and the total number of states in a conventional metal , while , on the other hand , the number of the bound states , is equal to @xmath241 according to eq . ( [ iexcsmalld ] ) . for processes with several andreev reflections ( @xmath242 ) , the possibilities for resonances increase . every andreev reflection energy may coincide with a bound state energy and thus be resonant . for some specific voltages , more than one resonance is important , creating a situation of overlapping resonances , which can enhance the current giving peaks in the current - voltage characteristics at these voltages . the three - particle current @xmath243 has a non - resonant value of order @xmath244 . however , @xmath243 is enhanced to order @xmath196 when the energy of one of the two andreev reflections coincides with a bound state energy . for the applied voltage equal to the difference between two bound state energies , @xmath245 , two resonances occur simultaneously , i.e. form a resonance consisting of two overlapping single resonances ; see the inset in fig . [ i3d1plots ] . this will enhance the current to order @xmath2 close to this voltage , giving a peak in the ivc . the number of peaks is equal to the number of bound state pairs . the peaks are located in the voltage interval @xmath246 ; we note that the peak positions are weakly dependent on temperature . to evaluate the height and the width of these peaks , we study the contribution from overlapping resonances at @xmath247 and at @xmath248 . close to these energies , the phases @xmath249 and @xmath250 , defined in eq . ( [ resphase ] ) , are close to zero , and we find the current spectral density for injection of a quasiparticle from the left ( see appendix [ threepartres ] ) , @xmath251^{(km),l}=\frac { d^{3 } n^{l}(e)n^{r}(e_{3 } ) } { \left|d-4\varphi_{1}^{(k)}\varphi_{2}^{(m ) } + id\left(\varphi_{1}^{(k)}n^{r}(e_{3})+\varphi_{2}^{(m)}n^{l}(e ) \right)\right|^{2 } } .\ ] ] we now expand @xmath252 , @xmath253 in the deviation from perfect overlap in energy , @xmath254 , and in voltage , @xmath255 , and find , using @xmath0 , from eq . ( [ ip3res ] ) @xmath256^{(km),l } = \frac{d \gamma_{0}^{(k ) } \gamma_{3}^{(m ) } } { \left(\delta e_{+}\delta e_{-}/\delta^{2 } - d/4\eta^{(k ) } \eta^{(m ) } \right)^{2}+\lambda^{2}},\ ] ] where @xmath257 , @xmath258 . the energy dependence of the current in eq . ( [ doublebreit ] ) has the form of two resonant peaks with width @xmath259 split by the energy interval @xmath260 at @xmath261 , the peak splitting increasing with increasing @xmath262 . after integration over energy , the overlapping resonances give a current contribution in the form of a current peak ( @xmath263 ) , @xmath264 in this equation , the densities of states @xmath171 are taken at the entrance and exit energies , @xmath265 and @xmath266 , and the temperature is taken to be zero . a factor of @xmath267 has been included in eq . ( [ i3res ] ) to take into account the similar resonant process for injection from the right , where @xmath268 and @xmath269 . the curve for the three - particle current versus voltage thus consists of peaks with heights of order @xmath2 and half - width @xmath270 on top of a background of order @xmath196 . the background current increases with voltage in the interval @xmath271 as more single resonances come into the integration region . in the interval @xmath272 , the background current decreases due to broadening of the resonances because of leakage associated with incomplete andreev reflection outside the gap . in long symmetric junctions the current peaks form an interesting triangular pattern . to see this , we first note that if the bound state spectrum were perfectly linear , several of the peaks described by eq . ( [ ip3res])-([i3res ] ) will be situated at the same voltage since @xmath273 ( see also the inset in fig . [ i3long ] ) , and thus the total number of peaks will be reduced while their respective height will be increased . since the bound state spectrum is not linear , the peaks show splitting . however , the deviation from linearity is small and in practice the peaks form clusters , giving combined peaks with height roughly equal to the number of clustering peaks . in the interval @xmath271 , the number of peaks in a cluster increases in steps of three from @xmath274 to @xmath275 , etc . , up to the number of bound states . in the interval @xmath214 the number of peaks in a cluster decreases in steps of one . this gives an appearance of a `` peak triangle '' for very long junctions , shown in fig . [ i3long ] . this `` peak triangle '' is further enhanced by the background current , which has a similar triangular form , as explained above . the four - particle current has a non - resonant value of order @xmath276 , which is enhanced to order @xmath244 when the energy of one of the three andreev reflections coincides with a bound state energy . similar to the three - particle current , overlapping resonances can enhance the magnitude of the current @xmath277 to the order @xmath2 for those voltages where both the first and the third andreev reflections coincide with the bound states , as shown in the upper inset in fig . [ i4d1plots ] . indeed , it is clear from the explicit form of @xmath278 that when @xmath279 and @xmath280 , then @xmath281 , i.e. the transparency of the mar trajectory is enhanced to unity . other combinations of the resonances , e.g. when the first and the second andreev reflection occur at bound state energies , will produce peaks of order @xmath196 or smaller , as described in appendix [ fourpartres ] . focusing on the double resonances which produce large ( @xmath282 ) current peaks , we find that in short junctions with just one pair of bound states , @xmath283 , the double resonance will occur at voltage @xmath284 , provided the energy of the bound state is within the interval @xmath285 . the spectral density of the current has a form similar to the one in eq . ( [ doublebreit ] ) , the major difference being the small peak splitting@xcite , @xmath286 . the height of the resulting current peak ( @xmath263 ) is @xmath287_{max}=\frac{\pi de\delta}{h } \frac{1-\left[a(2e^{(0)})\right]^{4 } } { 1+\left[a(2e^{(0)})\right]^{4 } } , \ ] ] where @xmath288 is the andreev reflection amplitude at energy @xmath289 . for longer junctions , there are many possibilities to have overlapping resonances . two bound states at one side of the junction with energies @xmath290 and @xmath291 can give a peak in @xmath277 if @xmath292 . although the height of all peaks is roughly proportional to @xmath2 , numerically the heights ( and widths ) of the peaks may vary considerably depending on the position of the second andreev reflection . if the second andreev reflection does not occur at the energy of a bound state , the situation is similar to the one described above ; see lower inset in fig . [ i4d1plots ] . however , if a bound state is close to the energy of the second andreev reflection , then the current spectral density @xmath293 consists of the three full - transmission peaks with widths @xmath286 which are split within the interval @xmath294 ( triple resonance ) . the triple resonance has larger spectral weight compared to the double resonance , which results in the larger height and width of the current peak . rigorously speaking , a triple resonance can only occur in asymmetric junctions because it requires equal distance between neighboring resonances , while the bound state spectrum in symmetric junctions is not equidistant . however , in long junctions , the deviation from the equidistant spectrum is small , and quasi - triple resonances may therefore occur also in long symmetric junctions . this effect can be observed in fig . [ i4d1plots ] , where the four - particle current for a symmetric junction with length @xmath295 consists of three peaks with different heights : the central peak corresponding to the quasi - triple resonance while the two side peaks corresponding to the double resonances with the heights given by eq . ( [ i4max ] ) . finally , it is worth noting that , similar to the situation for the three - particle current , the peaks will form clusters , giving a smaller number of current peaks than the number of pairs of bound states in long junctions . the studied properties of multiple resonances in three- and four - particle currents allow us to make some general conclusions about resonant behavior of the high order multi - particle currents which determine the total current at small voltage . the non - resonant magnitude of an @xmath48-particle current is of order @xmath296 at the threshold voltage , @xmath297 , and therefore the total non - resonant current exponentially decreases with the applied voltage ( in transparent junctions , @xmath298 , the current is exponentially small at@xcite @xmath299 ) . however , multiple resonances may enhance the magnitude of the current by several orders of @xmath2 . the major question of interest here concerns the maximum value of the resonant current , in particular whether it can be of order @xmath2 at arbitrary small voltage . to obtain such large current at small voltage , it is necessary to achieve a transmission probability through a high order mar path equal to unity , which implies that the energy of at least every other andreev reflection must coincide with a bound state ( cf . the discussion in the previous subsection ) . for @xmath300 , this means that three or more bound states must be approximately equidistant in energy . since the bound state spectrum is non - equidistant , eq . ( [ resphase ] ) , this is generally not possible if the resonances are narrow ; therefore , in junctions with arbitrary geometry and small transmissivity there are no large current peaks below the voltage @xmath301 . however , the possibility of a large resonant current exists for junctions with sufficiently large transparency . to find the relevant transparency , let us consider a very long symmetric junction and assume for the moment that the bound state spectrum is equidistant , @xmath302 . then , from mapping of the @xmath48-th order mar process on a 1d multi - barrier structure ( see fig.[quasi - resonance ] ) , it is clear that if the applied voltage is commensurate with the level spacing , e.g. , @xmath303 , the multi - barrier structure is periodic , and full transmission is achieved leading to a current peak . this conclusion is valid also for a non - equidistant spectrum if the variation of the interlevel distance does not exceed the width of the full - transmission band . the deviation of the bound state spectrum from the best linear fit does not exceed the value 0.33@xmath304 , fig.[quasi - resonance ] . on the other hand , the width of the full - transmission energy band is @xmath305 for equidistant spectrum and for @xmath306 . thus one should expect large current structures in long symmetric junctions with transparency @xmath307 to occur at voltages @xmath308 . in junctions with smaller transparency , large current structures may appear only at @xmath309 , as explained before ; see fig . [ itotl4plots ] . it is also easy to see that in asymmetric junctions , where the width of the full transmission band for an equidistant spectrum is @xmath286 ( since the relevant resonances at one side of the junction are weakly coupled to each other through the mar process ) , large resonant current at small voltage may exist if @xmath310 . our numerical investigations confirm that in symmetric junctions when @xmath2 is of the order @xmath311 , the multiple resonances are completely blocked and current peaks are exponentially suppressed at @xmath312 . adding up the contributions to the current calculated in this paper , we arrive at a rather complex form of current - voltage characteristics ( ivc ) at subgap voltages , as shown in figs . [ itotd1plots]-[itotl4plots ] . nevertheless , the analysis of the tunnel limit allows us to classify various subgap current structures . here we will summarize the results of this classification . as a reference system we will take a short ( @xmath183 ) junction where the form of the ivc is well studied @xcite . the current structures in short junctions can be interpreted as resonant features due to quasi - bound states situated at the edges of the energy gap @xcite , the resonant conditions selecting voltages equal to the gap subharmonics , @xmath313 . this subharmonic gap structure of the short junction gradually changes with increasing junction length as bound states move down into the gap , giving rise to ivc structures with steps , oscillations and peaks . major points are : \(i ) the current in the subgap region is considerably enhanced , compared to the short junction case . this effect is present as soon as the effective length @xmath314 is comparable to , or larger than , the square root of transparency of the junction , @xmath315 . \(ii ) the main onset of the current in short junctions at @xmath164 shifts downwards in voltage to the value @xmath316 where @xmath317 is the energy of the bound state . this shift is caused by the resonant two - particle current giving a contribution to the total current of the order of the single - particle current . \(iii ) for longer junctions , the current onset transforms into a staircase within the voltage interval @xmath214 with the number of steps corresponding to the number of bound states , the step positions being given by @xmath318 . this is due to the resonances in the two - particle current transported through bound states . resonant channels open up , one by one , as the voltage increases and bound states enter the `` energy window '' available for two - particle processes . the current plateaus are not flat but modulated because of oscillations of the density of continuum states . the period of the modulation is roughly equal to the interlevel distance , and it decreases with the junction length . the amplitude of the modulation , on the other hand , increases with the junction length . thus , in long junctions , the current structures take the form of a series of peaks ( see fig . [ itotd1plots ] ) within the voltage interval @xmath214 . the position of the peaks has pronounced temperature dependence , scaling with the temperature dependence of the order parameter , while the distance between peaks has a weak temperature dependence . \(iv ) there is another series of the current peaks whose positions only weakly depend on temperature and are entirely determined by the bound state spectrum : @xmath319 and @xmath320 . these peaks are caused by the overlap of two resonances in the three- and four - particle currents and they exist in the intervals of applied voltage @xmath246 and @xmath321 respectively . the heights of these peaks are comparable with the heights of the two - particle current structures ( @xmath282 ) . \(v ) at voltage smaller than @xmath301 the resonant current structures generally become smaller in magnitude ( at least by one order in @xmath2 ) if the junction transparency is sufficiently small ( @xmath322 ) , and the current decays exponentially when @xmath323 approaches zero ( although for some particular junction lengths there could be huge ( @xmath282 ) current peaks caused by multiple resonances ) . this qualitative difference of the ivc below and above @xmath301 allows one to expect a cross over from power to exponential dependence of ivc in multichannel junctions . \(vi ) in transparent junctions , all current structures will persist but become smooth ; appreciable current will appear below @xmath301 as soon as @xmath324 . the current structures completely disappear in fully transparent junctions , @xmath101 , where the ivc does not depend on the junction length ; see fig . [ itotl4plots ] . \(vii ) at voltage larger than @xmath325 , the current undergoes oscillations , similar to rowell - mcmillan oscillations @xcite , and the excess current is approached much faster than in short junctions . in low transparency junctions the excess current is small , @xmath326 , despite strong andreev reflection and large pair current @xmath327 . we thank e. n. bratus , j. lantz and t. lfwander for discussions . support from nfr and nutek ( sweden ) and from nedo ( japan ) is gratefully acknowledged . in this appendix , the expansion is derived for the reflection amplitudes in eq . ( [ reflectionexp ] ) for a quasiparticle injected from the left . from the definition of @xmath117 and @xmath328 , eq . ( [ nullvectors ] ) , we know @xmath329 @xmath330 they are related as @xmath331 where @xmath332 and @xmath333 . from this relation , we find @xmath117 in terms of @xmath328 as @xmath334 where @xmath335 . when @xmath336 , we can make an expansion in this parameter to get to the form @xmath337 similarly we also get @xmath338 in this appendix , we derive the resonant form of the two - particle current , eq . ( [ ip2res ] ) , for a quasiparticle injected from the left . the definition of @xmath339 is @xmath340 , which using the pseudo - unitarity of the transfer matrices @xmath341 can be written on the form @xmath342 it simplifies in the limit @xmath0 , @xmath343 to @xmath344 inserting the simplified expansion of @xmath339 and the expansion of @xmath345 and @xmath117 from eq . ( [ reflectionexp ] ) into eq . ( [ ipn ] ) , as well as putting @xmath168 , the leakage current density takes the form @xmath346^{l}={\frac{\displaystyle \frac{1- { \left|1-e^{2i\varphi_{0 } } \right|^{2 } } \frac{1-|a_{2}|^{4}}{\left|1-e^{2i\varphi_{2 } } \right|^{2 } } d^{2 } } { \left|2i\varphi_{1}^{(m)}+\displaystyle\frac{d}{2}\left ( \displaystyle\frac{1+e^{2i\varphi_{0}}}{1-e^{2i\varphi_{0}}}+ \displaystyle\frac{1+e^{2i\varphi_{2}}}{1- e^{2i\varphi_{2}}}\right ) \right|^{2}}.}\ ] ] we make an expansion of the phase @xmath347 , where @xmath348 the two - particle current density now takes a breit - wigner form @xmath346^{l}=\frac{\gamma_{0}^{(m)}\gamma_{2}^{(m ) } } { \displaystyle \left(\frac{\delta e-\delta e^{(m)}}{\delta}\right)^{2}+ \left(\frac{\gamma_{0}^{(m)}+\gamma_{2}^{(m)}}{2 } \right)^{2}}\ ] ] where the tunneling rates are given by @xmath349 where @xmath350 are equal to the dos , eq . ( [ densityofstates ] ) at energy @xmath351 . the resonance is slightly shifted from @xmath96 with @xmath352 in this appendix , the resonant form of the three - particle current , eq . ( [ ip3res ] ) , is derived . the @xmath353-matrix , which by definition is @xmath354 can be transformed using eq . ( [ m20 ] ) to @xmath355 which can be written in the form @xmath356 it simplifies in the limit of @xmath0 , @xmath357 and @xmath358 to @xmath359.\ ] ] inserting this form of the @xmath353-matrix and the expansion ( [ reflectionexp ] ) for @xmath117 and @xmath360 into eq . ( [ ipn ] ) , as well as putting @xmath168 , the probability current density for injection of a quasiparticle from the left takes the form @xmath361^{l}=\frac{(1-|a_{0}|^{4})(1- { \left|1-e^{2i\varphi_{0 } } \right|^{2 } \left|1-e^{2i\varphi_{3 } } \right|^{2}\left|q\right|^{2}}\ ] ] @xmath362 where @xmath0 is once again used . since @xmath363 and @xmath364 and the dos at energies @xmath365 , eq . ( [ densityofstates ] ) , are equal to @xmath366 @xmath367 we arrive at the form @xmath361^{l}=\frac{n^{l}(e)n^{r}(e_{3})d^{3 } } { \left|d-4\varphi_{1}^{(k)}\varphi_{2}^{(m ) } + id\left(\varphi_{1}^{(k)}n^{r}(e_{3})+\varphi_{2}^{(m)}n^{l}(e ) \right)\right|^{2 } } .\ ] ] in this appendix , we discuss the structure of the resonance in the four - particle current . the matrix @xmath368 can be written as @xmath369 @xmath370+\ ] ] @xmath371 @xmath372.\ ] ] from eq . ( [ mmatris ] ) it is clear that in general @xmath373 . when both @xmath374 and @xmath375 are close to a multiple of @xmath376 , @xmath377 , while close to other double resonances , e.g. when @xmath374 and @xmath378 are close to a multiple of @xmath376 , @xmath379 .

we study reversible quantum cellular automata with the restriction that these are also clifford operations . this means that tensor products of pauli operators ( or discrete weyl operators ) are mapped to tensor products of pauli operators . therefore clifford quantum cellular automata are induced by symplectic cellular automata in phase space . we characterize these symplectic cellular automata and find that all possible local rules must be , up to some global shift , reflection invariant with respect to the origin . in the one dimensional case we also find that every uniquely determined and translationally invariant stabilizer state can be prepared from a product state by a single clifford cellular automaton timestep , thereby characterizing these class of stabilizer states , and we show that all 1d clifford quantum cellular automata are generated by a few elementary operations . we also show that the correspondence between translationally invariant stabilizer states and translationally invariant clifford operations holds for periodic boundary conditions . ' '' '' * on the structure of clifford quantum cellular automata * dirk - m . schlingemann , holger vogts and reinhard f. werner ' '' '' 1 . isi foundation , quantum information theory unit + viale s. severo 65 + 10133 torino , italy [ aff-1 ] 2 . institut fr mathematische physik + technische universitt braunschweig + mendelssohnstrae 3 + 38106 braunschweig , germany [ aff-2 ]

0804.4447

classical cellular automata have become a standard modeling tool for complex phenomena . with their discrete time step and their intrinsically high degree of parallelization they are ideally suited for models of diverse phenomena as coffee percolation , highway traffic and oil extraction from porous media . as an abstract computational model cellular automata can simulate turing machines , and even explicit simple automata such as conway s life game have been shown to support universal computation @xcite . on the quantum side , the interest in cellular automata stems from their implementation in optical lattices and arrays of optical microtraps . however , the theory of quantum cellular automata ( qcas ) is still in its early stage . since each cell may influence several others , the dynamics is subject to a `` no - cloning '' constraint , leading to a non - trivial interplay between the conditions of locality and unitarity . it is therefore helpful to have some class of qcas , which can be analyzed in great detail , and which can serve as a testing ground for general ideas about qcas . this paper is concerned with such an analysis , namely of the special class of clifford quantum cellular automata ( cqcas ) , in which the elementary time step is given by a `` clifford gate '' , meaning that it takes tensor products of pauli matrices to tensor products of other pauli matrices . in the theory of gate model computation , and for the one - way quantum computation model , a detailed analysis of what can be done with clifford operations alone turned out to be very useful , even though as the downside of allowing an efficient classical description such gates alone do not allow universal quantum computation . by analogy it is therefore clear that cqcas do not comprise the full complexity of qcas . what one can hope to get , however , is an interesting class of cellular automata , and some tools for understanding this class in great detail . a similar analysis has been done with gaussian quantum cellular automata @xcite , e.g. the qca describes a chain of harmonic oscillators with nearest neighbor couplings . for all these qcas the hilbert space of one elementary cell is infinite dimensional , and the qca maps phase space translations , also referred to as weyl operators , to phase space translations . in our approach we use elementary cells with a finite number of levels , which corresponds to replacing the continuous phase space by a discrete space . by definition , a cellular automaton is a lattice system , which consists of many subsystems ( called `` cells '' ) labeled by a point lattice in space . for simplicity , we will always take the lattice as @xmath2 , the integer cubic lattice in @xmath3 space dimensions , where we discuss periodic boundary conditions , and hence toroidal lattices ] . the cell systems in the classical case may have states like `` occupied '' and `` empty '' . in the quantum case , they will be @xmath4-state quantum systems , for some finite @xmath4 . in either case the group of lattice translations ( `` shifts '' ) is a symmetry of the system . the dynamics will be given by a discrete global time step , or global `` transition rule '' assumed to have the following three properties : * _ translation invariance _ : the time step commutes with the lattice translation symmetries . * _ reversibility _ : there is an inverse rule . for a finite quantum system this would mean unitary dynamics . for the infinite lattice system this will be stated algebraically below . * _ locality _ , or `` finite propagation speed '' : the state of each cell after one step can be computed from the state in a fixed finite region around the cell . these assumption define the class of reversible qcas @xcite . the locality and reversibility conditions are best phrased in the heisenberg picture : if @xmath5 denotes some observable of the system , its expectation after one time step starting from the initial state @xmath6 will be @xmath7 for a suitable observable @xmath8 , where by @xmath9 we denote the expectation of @xmath5 in the state @xmath6 . the transformation @xmath10 is what we will call the _ global rule _ of the automaton . then reversibility ( together with complete positivity , which is required of any dynamical map ) implies that @xmath11 is a homomorphism of the observable algebra of the whole system : @xmath11 is linear , @xmath12 , and @xmath13 . locality means that an observable @xmath14 localized at some lattice point @xmath15 ( i.e. , an observable of the cell at @xmath16 ) will be mapped to an observable localized in the region @xmath17 . that is , @xmath18 will be in the tensor product of the cell algebras belonging to the sites @xmath19 with @xmath20 . by translation invariance this set @xmath21 , called the _ neighborhood scheme _ of the automaton is independent of @xmath16 . the global transition rule @xmath11 is a map on an infinite dimensional space , and hence not readily specified explicitly . however , by using the basic properties of qcas one can see that it suffices to know just a few local data , associated with the region @xmath21 , in order to reconstruct @xmath11 uniquely . suppose we know @xmath18 for every observable @xmath14 in some basic cell @xmath16 . then by translation invariance we know the analogous transformation for _ any _ cell . moreover , since every local observable can expanded in products of one - cell observables , and because @xmath11 is a homomorphism , we can compute @xmath11 for any local observable . so the restriction @xmath22 of @xmath11 to the observables of a single cell @xmath16 can be called the _ local rule _ of the automaton . we can also decide by a finite set of equations , whether a proposed local rule actually belongs to a well - defined global rule : clearly @xmath22 must be a homomorphism . the only further condition one has to check is that the images of observables @xmath14 and @xmath23 localized in the cells indicated also commute , i.e. , @xmath24 , whenever @xmath25 . this is necessary , because @xmath14 and @xmath23 commute , and a moment s reflection shows that this is also sufficient for uniquely reconstructing the images of arbitrary observables under @xmath11 . the commutation conditions on the local rule are trivially satisfied when @xmath16 and @xmath26 are sufficiently far apart , when @xmath17 and @xmath27 are disjoint . hence only finitely many conditions need to be checked . for special classes , the job of specifying a qca via its local rule can be reduced still further , which is where the clifford condition comes in . let us assume now that we have a qubit system , so the local cell dimension is @xmath28 . for each local cell we thus have a basis for the observables , consisting of the identity and the three pauli matrices , which we denote by @xmath29 and @xmath30 . by @xmath31 etc . we denote the corresponding pauli matrix belonging to the cell @xmath16 . finite tensor products of pauli matrices belonging to different sites , perhaps with a sign @xmath32 will be referred to as _ pauli products_. these form a group , called the pauli group . then a _ clifford quantum cellular automaton _ ( cqca for short ) is defined by the condition * _ clifford condition _ : if @xmath5 is any multiple of a pauli product , so is @xmath8 . clearly , this is equivalent to the property that one - cell pauli operators are taken to pauli products , which simplifies the local rule . moreover , it suffices to specify @xmath33 and @xmath34 for some @xmath16 , because we can compute @xmath35 via the homomorphism property . hence a cqca is defined in terms of just two pauli products . [ example-0 ] for the one - dimensional lattice ( @xmath36 ) , consider the relations @xmath37 let us verify that all requirements for a local rule are satisfied . to begin with each of the expressions on the right hand side , as a product of pauli matrices , is hermitian with square one . these are all the required conditions related to just a single line , and are satisfied for any pauli product with a sign @xmath38 . next we have to verify the anti - commutation relation arising from applying a homomorphism @xmath11 to the anti - commutation relation @xmath39 . indeed , @xmath40 . hence the definition @xmath41 again produces a hermitian operator with square @xmath42 , and the local rule is a homomorphism @xmath22 into the algebra on the sites @xmath17 with @xmath43 . finally , we have to check the commutation rules for the images of observables from neighboring sites . for example , we have @xmath44=-[z_x , z_x\otimes x_{x+1}\otimes z_{x+2}]=0 $ ] , and similarly @xmath45=0 $ ] . perhaps the only non - trivial relation to check is @xmath46 = [ z_{x-1}\otimes x_x\otimes z_{x+1},z_{x}\otimes x_{x+1}\otimes z_{x+2 } ] = 0,\ ] ] which holds because the factors on sites @xmath16 and @xmath47 both _ anti-_commute . in principle , we would also have to check the existence of an inverse for the automaton , which is actually given by @xmath48 and @xmath49 , but as was shown in @xcite , this already follows from the homomorphism property . it is clear from this example that the search for cqcas is now a combinatorial problem . we can first look for _ self commuting _ pauli products , i.e. , pauli products , which commute with all translates of itself . only these can appear on the right hand side of local rules . one can then check , for any pair @xmath50 of such products , whether they anti - commute , while all proper translates of @xmath51 commute with @xmath52 . in fact , we began our investigation by running this simple search program . we found , for example , that while there is a rich variety of self - commuting pauli products only reflection symmetric products could appear in a local rule . this will indeed be shown in full generality below . commuting sets of pauli products also play a central role in the problem of determining so - called stabilizer states : these are pure states , which can be characterized by eigenvalue equations for pauli products or , equivalently , by the condition that certain pauli products have expectation @xmath38 . it is easy to check that pauli products which simultaneously have sharp expectations @xmath38 must commute . now for the infinite lattice systems it is natural to ask which pauli products @xmath5 have the property that there is a unique pure state @xmath6 of the infinite system , which has expectation 1 for @xmath5 and all its translates . as the simplest example , let us take @xmath53 , so we ask for states with @xmath54 for all @xmath15 . clearly , this defines the `` all spins up '' state , which is an infinite product state . a slightly more complex example uses the stabilizer operators @xmath55 , which singles out the one - dimensional _ cluster state _ , whose higher dimensional analogs are used as the entanglement resource for universal one - way quantum computing @xcite . showing that these eigenvalue equations define a unique state of the infinite lattice is now very easy , by using the cellular automaton ( [ example - rule ] ) : since this automaton maps @xmath56 to the required stabilizer operator , all existence and uniqueness problems for such a state are mapped to the corresponding trivial questions for the stabilizer operator @xmath56 . in other words , self - commuting pauli products of the form @xmath57 for some cqca @xmath11 characterize a unique translation invariant cluster state . we will show later that ( at least in one dimension ) the converse is also true , so that there is a very close connection between stabilizer states and clifford cellular automata . the definition of cqcas given above applies only to qubit systems . however , all our results are also valid for higher dimensional cells , particular cells of prime dimension @xmath4 . the role of the pauli operators @xmath58 and @xmath30 is then taken by the cyclic shift on @xmath59 , and the multiplication by a phase , i.e. @xmath60 where all ket labels @xmath61 are taken modulo @xmath4 . products of these operators are called _ weyl operators _ , and the appropriate definition of cqcas requires that @xmath33 and @xmath34 are both tensor products of weyl operators . the necessary preliminaries on the pauli group and clifford operations in this extended setting , and the background concerning infinite lattice systems are provided in subsection [ sec-2 - 1 ] . in order to utilize the translation symmetry one would like to use fourier transform techniques . however , in the discrete structures an integral with complex phases makes no sense . it turns out , however , that a `` generating function '' technique does nearly as well . the analogue of the fourier transform is then a laurent - polynomial in an indeterminate variable , i.e. , a polynomial with coefficients in the field @xmath62 with both positive and negative powers . the salient facts about this structure will be provided in subsection [ sec-2 - 3 ] . the description in terms of laurent polynomials can also be adapted to lattices with periodic boundary conditions . this will be described in section [ sec - periodic ] . in order to discuss general clifford quantum cellular automata , that is , for arbitrary lattice and single cell dimension , we introduce in section [ sec-2 ] the necessary mathematical tools . we first review the concept of discrete weyl systems and ( infinite ) tensor products of them , thereby characterizing the underlying `` phase space '' . we show that clifford qcas can be completely characterized in terms of classical symplectic cellular automata . we also introduce our fourier transform techniques and study the structure of isotropic subspaces , because these play an essential role for the characterization of symplectic cellular automata and translationally invariant stabilizer states . in section [ sec-3 ] we will state our main results . we show that symplectic cellular automata can be identified with two - by - two matrices , which have laurent - polynomials as matrix elements . we will find that the determinant of this matrix must be one and that the polynomials must be reflection invariant . in the one - dimensional case we state that every translationally invariant stabilizer state can be prepared out of a product state by a single cqca step . furthermore , we also specify the generators of all 1d qcas . finally , we show in section [ sec - periodic ] that the close connection between translationally invariant stabilizer states and cqcas also holds in the case of periodic boundary conditions even in every lattice dimension . we introduce some mathematical tools , which we will use to study clifford qcas . we start with a short repetition of finite weyl systems , which generalize the pauli operators to systems with prime number dimensions . these weyl operators can be described by phase space vectors and clifford operations are induced by symplectic transformations on the phase space . since we are looking for translationally invariant operations , we also introduce some kind of fourier transform . each single cell in a qca is given by a finite dimensional quantum system , so the observables on a single system can be described by matrices from the algebra @xmath63 . a possible basis for this algebra is given by weyl operators @xmath64 , whereby @xmath58 and @xmath30 are given by the generalized pauli operators from equation ( [ onestep ] ) . these operators fulfill the weyl relations @xmath65 where @xmath66 is the @xmath67 root of unity . > from this equation the commutation relation @xmath68 immediately follows . obviously we get for @xmath28 the standard pauli operators from @xmath69 and the weyl operators are generalizations of the pauli operators to higher dimensional spaces . the indices @xmath70 and @xmath71 are integers modulo @xmath4 , so they are elements of the finite field @xmath72 . in infinite dimensional systems weyl operators describe phase space translations and therefore we call the space @xmath73 a discrete phase space . building a tensor product of weyl operators means that we must assign a phase space vector @xmath74 to each lattice point @xmath75 , so @xmath76 is a mapping from @xmath2 into @xmath73 and we denote for the tensor product @xmath77 this infinite tensor product is well defined , if there are only finitely many of the weyl operators different from @xmath78 . for a mapping @xmath79 we have that only finitely many @xmath16 with @xmath80 are allowed , so the support of @xmath76 is finite . the set of such functions describes the global system and is identified with the global phasespace @xmath81 . we denote the finitely supported functions from @xmath2 to @xmath82 by @xmath83 and we have @xmath84 . the corresponding weyl operators generate an algebra and , by restricting the support of the functions to some finite subset @xmath85 , we get a finite dimensional algebra @xmath86 , also called the local algebra of @xmath87 . by taking the union of these algebras over all finite subsets of @xmath2 and taking the closure ( in operator norm ) we get a quasilocal @xmath88-algebra @xmath89 @xcite , which is used in the general theory of quantum cellular automata @xcite . the local structure is accompanied by the symmetry group of lattice translations . for each lattice translation @xmath75 an automorphism @xmath90 is defined by @xmath91 where @xmath92 is the translation of phase space vectors . given a phase space vector @xmath76 , the translated vector is @xmath93 . so the automorphism @xmath90 shifts the position of each tensor factor by @xmath16 . it follows directly from ( [ translation ] ) that the homomorphism property @xmath94 holds . furthermore , the automorphism @xmath90 maps the local algebra @xmath95 onto @xmath96 . the weyl relations of a single system completely determine the relations of the global system which are given by @xmath97 where we have introduced the bilinear form @xmath98 . the adjoint of a weyl operator is given by @xmath99 which is due to the unitarity of the weyl operators . since commutation relations are essential for validating possible local rules of quantum cellular automata , the commutation relations of weyl operators are most important for us . we get @xmath100 whereby @xmath101 is the canonical symplectic form on @xmath81 . this means that two weyl operators @xmath102 and @xmath103 are commuting if and only if @xmath104 ( and for @xmath28 they anti - commute if @xmath105 ) . in particular , an abelian algebra of weyl operators is given by a subspace of @xmath81 on which the symplectic form vanishes . such a subspace is called isotropic and a maximally abelian algebra corresponds to a maximally isotropic subspace . as already mentioned a clifford quantum cellular automaton is a qca which maps weyl operators to multiples of weyl operators , which are in our case tensor products of single cell weyl operators , so we have the relation ( the `` clifford condition '' ) @xmath106 with a mapping @xmath107 on the phase space @xmath81 and some phase valued function @xmath108 . since @xmath11 is an automorphism we find with equation ( [ com - ccr ] ) that @xmath109 holds , so we have @xmath110 or in other words @xmath107 is a symplectic transformation . for reversible operations the clifford condition is in general equivalent to the weyl covariance ( for general theory on covariant channels we refer to @xcite and for the special case of weyl covariance to @xcite ) of the quantum channel : an automorphism @xmath11 on the weyl algebra @xmath89 fulfills the clifford condition ( [ cqca ] ) if and only if the weyl covariance @xmath111 holds for all operators @xmath112 and some symplectic transformation @xmath107 . because the weyl operators form a basis of @xmath89 we just have to insert @xmath102 for some @xmath113 in the covariance condition , which yields the equation @xmath114 . if @xmath11 is a clifford automorphism we have already seen that @xmath107 is a symplectic transformation and obviously fulfills this equation . in the inverse direction we get that @xmath115 must be a multiple of @xmath116 , because the relation must hold for all @xmath117 and the symplectic form is non degenerate ( note that the support of the phase space vectors is finite and that @xmath107 maps therefore finitely supported vectors to finitely supported vectors , so the commutation relations can be checked in a finite dimensional space ) . since a qca is a translationally invariant automorphism on the quasilocal algebra , it suffices that the clifford condition holds for the local rule , e.g. the qca restricted to operators which are localized in a single cell . furthermore , because of the weyl relations on a single cell , we only need to specify the image of the weyl operators @xmath118 and @xmath119 . to some extend we are free in the choice of the phases @xmath120 , since these phases do not interfere with the commutation relations for the local rule . the only condition is that some power of a weyl operator is always equal to @xmath42 ( we will specify this below ) , and so these phases must be some roots of unity . the two phases @xmath121 and @xmath122 completely determine the function @xmath123 . of course @xmath107 and @xmath123 must be translationally invariant , because @xmath11 is translationally invariant . using the homomorphism property of the qca and equation ( [ ccr ] ) we get @xmath124 , so because the weyl operators form a basis the transformation @xmath107 must be linear and the phase function must fulfill @xmath125 which enables us to calculate the phase @xmath126 for each @xmath113 , if the local rule and therefore @xmath107 and the phases @xmath121 and @xmath122 are given . in total we get the following theorem : [ thm-1 ] if @xmath11 is a clifford quantum cellular automaton ( equation ( [ cqca ] ) ) on the weyl algebra @xmath89 , then @xmath107 is a translationally invariant linear symplectic transformation ( `` symplectic cellular automaton '' ) and the phase function @xmath123 fulfills equation ( [ coboundary ] ) . this means that we are able to study clifford qcas up to some phase function in terms of a classical cellular automaton on the phase space @xmath81 . it is well known that clifford operations allow an efficient classical description , which in the case of qcas turned out to be the group of classical symplectic cellular automata . in the rest of the paper we will study the structure of this kind of cellular automata , thereby characterizing the structure of cqcas . we would like to give a closed expression for the phase function @xmath123 , but this has to be done in dependence of the cell dimension . first we consider the case @xmath127 . then all weyl operators fulfill @xmath128 and because of @xmath129 the phase @xmath126 must be a @xmath67 root of unity . so we can write @xmath130 with a function @xmath131 , which then has to fulfill @xmath132 . this equation determines the function @xmath133 up to some linear functional @xmath134 , which is given by the choice of the phases @xmath135 and @xmath136 . the bilinear form @xmath137 is symmetric , because @xmath107 is a symplectic transformation . if @xmath127 we may divide by @xmath138 and the general solution is @xmath139 . the case of qubits ( @xmath28 ) is slightly more complicated because the weyl operators fulfill @xmath140 . so the phase function must fulfill @xmath141 with @xmath142 . we replace the form @xmath143 by the bilinear form @xmath144 , which is formally given by @xmath145 , so the values of @xmath146 are even elements of @xmath147 and the weyl relation becomes @xmath148 . this means that @xmath149 fulfills @xmath150 with the form @xmath151 . this form is symmetric , so in the decomposition @xmath152 we have @xmath153 and all these elements are even . we can find @xmath154 with @xmath155 , but this choice is not unique in @xmath147 and corresponds exactly to the freedom in the choice of the phases @xmath121 and @xmath122 . the solution for @xmath149 is then given by @xmath156 ( note that @xmath157 and so @xmath158 holds ) . we would like to use fourier transform techniques for the study of the structural properties of symplectic ca , because of translational invariance , and because we know that this is very helpful for symplectic ca with continuous single cell phase space @xcite . so we have to apply a fourier transform to the functions @xmath83 . but the values of these functions are in the finite field @xmath82 and multiplying such a value with a complex number does not really match . it turns out that a slight modification of the usual fourier transform does as well . for a function @xmath159 we define @xmath160 with @xmath161 . now the transformed function @xmath162 is a polynomial or , more precisely , a laurent - polynomial in the variables @xmath163 with coefficients in @xmath82 , which will be denoted by @xmath164 . note that we have indeed polynomials , because the functions in @xmath83 are finitely supported . equation ( [ fourier ] ) identifies functions of @xmath83 with polynomials @xmath165 and this identification is unique , so @xmath83 and @xmath165 are isomorphic . the usual fourier transform would require @xmath166 . we do not further specify the domain of the variables , and this approach can be seen as `` generating function approach '' or `` algebraic fourier transform '' . the convolution @xmath167 is a natural product becomes a `` commutative division ring '' . ] of functions in @xmath83 . the invertible elements with respect to this operation are the functions which are supported on a single lattice point , e.g. @xmath168 ( @xmath169 is the kronecker - delta ) with @xmath170 and @xmath75 , and the unit element is @xmath171 . the nice fact about fourier transform is that the convolution turns into a usual product which is also true for our algebraic version : @xmath172 note that the invertible polynomials are monomials , e.g. they are of the form @xmath173 . of course the unit element is the constant @xmath174 . another important operation is the reflection operation ( or involution ) @xmath175 for @xmath159 . obviously the reflection preserves the convolution , e.g. @xmath176 , and for the transformed function we have @xmath177 . the phase space @xmath81 consists of two - dimensional tuples of functions from @xmath83 and all operations can be defined component - wise - module . ] , so we get that the phase space is isomorphic to @xmath178 . we would like to study the structure of symplectic ca in this polynomial space . the transformation of an operation @xmath179 is defined according to @xmath180 , so @xmath181 is a mapping from @xmath182 to @xmath182 . we introduce the symplectic form @xmath183 by @xmath184 which can be written as @xmath185 , whereby @xmath186 denotes the @xmath187-matrix @xmath188 with polynomial entries . the symplectic form @xmath189 is the best fitting symplectic form for symplectic ca , because it combines both the basic symplectic form @xmath190 as well as the translation invariance : [ sca - fourier ] a linear operation @xmath107 on the phase space @xmath81 is a symplectic cellular automaton , if and only if , the transformed operation @xmath181 leaves the symplectic form @xmath189 invariant . for this proof we introduce the form @xmath191 for @xmath192 . a straightforward computation shows that @xmath193 holds . this means we have @xmath194 , so @xmath189 is the fourier transform of @xmath195 and the invariance of @xmath195 under some operation @xmath107 is equivalent to the invariance of @xmath189 under @xmath181 . now suppose @xmath107 is a symplectic ca . then we have for all @xmath75 that @xmath196 holds , because @xmath107 is translationally invariant and preserves @xmath190 , so @xmath195 is invariant under @xmath107 . if @xmath107 leaves @xmath195 invariant , this holds also for @xmath197 , and because of this @xmath198 holds for all @xmath75 and all @xmath192 and so @xmath107 must commute with the translations @xmath92 . so we can characterize symplectic ca in `` momentum space '' by studying the linear transformations on @xmath182 which leave the symplectic form @xmath189 invariant . in the subsequent we will mainly work in the polynomial space @xmath165 . therefore we will just identify the phase space @xmath81 with @xmath182 and we will omit the symbol @xmath199 for the fourier transform of transformations . as we have already seen in the introduction , commutation relations are important for the verification of local rules of reversible qcas , because a qca is a homomorphism and preserves the algebraic structure . especially the images of @xmath31 and @xmath56 must be `` self - commuting '' , meaning that @xmath200=0=[t(z_x),t(z_y)]$ ] holds for all @xmath201 . so the operators @xmath33 generate a translationally invariant abelian algebra . for weyl operators translationally invariant abelian algebras correspond exactly to isotropic subspaces of @xmath182 with respect to the symplectic form @xmath189 and these subspaces can be easy connected to translationally invariant stabilizer states . therefore it is important for us to study the structure of these subspaces . a @xmath165-subspace @xmath202 is called isotropic , if for all @xmath203 the symplectic form @xmath204 vanishes . an isotropic @xmath165-subspace @xmath205 is called maximally isotropic , if the relation @xmath204 for all @xmath206 implies that @xmath207 holds . for us the form of the generators of isotropic , in particular maximally isotropic , @xmath165-subspaces is important , because this is a substantial step for the characterization of local rules of cqcas and translationally invariant stabilizer states . the following lemma shows that a generator @xmath76 of a singly generated maximally isotropic subspace is reflection invariant and that the components @xmath208 and @xmath209 are coprime . we will call a polynomial @xmath210 ( or a tuple of those ) reflection invariant for some half integer lattice point @xmath211 , if @xmath212 holds . the greatest common divisor of two polynomials @xmath213 will be denoted by @xmath214 . note that the greatest common divisor is defined only up to invertible elements . we will simply write @xmath215 , if @xmath216 and @xmath217 are coprime . + [ lem - iso-1 ] 1 . if the subspace @xmath218 is maximally isotropic , we have @xmath219 . if the subspace @xmath218 is maximally isotropic , @xmath76 is reflection invariant to some point @xmath211 . every reflection invariant polynomial generates an isotropic @xmath165-subspace . suppose @xmath220 is a maximally isotropic @xmath165-subspace and @xmath221 is not invertible . so we can write @xmath222 with @xmath223 , but @xmath224 since @xmath216 is not invertible . but we have that @xmath225 holds , which is a contradiction to @xmath220 being maximally isotropic . suppose that @xmath220 is maximally isotropic . by 1 we have @xmath219 . since @xmath226 , it follows that @xmath227 . so we have @xmath228 with some polynomial @xmath210 . but for the reflected phase space vector @xmath229 we also have that @xmath230 , so @xmath216 must be invertible and therefore a monomial @xmath231 for some @xmath232 . suppose that @xmath233 is reflection invariant . then @xmath234 holds , and @xmath76 generates an isotropic @xmath165-subspace . [ stand - ex]both @xmath235 and @xmath236 are reflection invariant . the corresponding weyl operators @xmath237 and @xmath238 are the same reading from the left and from the right ( `` palindromes '' ) . both phase space vectors generate isotropic subspaces . the subspace generated by @xmath239 is indeed maximally isotropic and the components @xmath240 and @xmath241 are coprime , whereas the subspace generated by @xmath242 is not maximally isotropic because @xmath243 is a nontrivial common divisor . this is also clear in terms of operators , because all operators @xmath56 commute with @xmath244 , but can not be obtained by products of translates of @xmath244 . in particular , the greatest common divisor comes into play . we will be able to state more results in the one - dimensional case ( @xmath36 ) , due to the fact that the ring of polynomials @xmath245 is euclidean is a principal ideal ring , which means that every ideal in @xmath246 is generated by a single element . for this general algebraic theory we refer to @xcite . ] . especially this means that the euclidean algorithm can be applied for finding the greatest common divisor of two polynomials , which is also used for the factorization of wavelet transformations @xcite . [ lem - euclid ] let @xmath247 be a phase space vector . then there exist @xmath248 such that @xmath249 holds . we define the degree of a laurent polynomial @xmath250 by @xmath251 when @xmath252 and @xmath253 are nonzero . suppose @xmath254 and let @xmath255 and @xmath256 . we make a division with remainder and get a polynomial @xmath257 with @xmath258 and a polynomial @xmath259 with @xmath260 such that @xmath261 with this decomposition we get @xmath262 . we repeat this division recursively until the remainder vanishes : @xmath263 then we have @xmath264 . we rewrite the recursion to get the form of equation ( [ euclid ] ) : @xmath265 so we get @xmath266 with @xmath267 and since all entries in the matrices are polynomials we get polynomials @xmath268 and @xmath269 such that @xmath270 holds . we have seen in proposition [ sca - fourier ] that symplectic cellular automata are nothing else but linear functions on the phase space @xmath271 that preserve the @xmath165-symplectic form @xmath189 . such a map on @xmath182 can be represented by a two - by - two matrix with entries in the polynomial ring @xmath165 . the first column is given by @xmath272 ( `` the local rule for @xmath58 '' ) and the second column by @xmath273 ( `` the local rule for @xmath30 '' ) . the commutation relations of the local rule then end up in the following conditions on the column vectors : [ cor - sca ] a two - by - two matrix @xmath107 with entries in @xmath165 is a symplectic cellular automaton , if and only if , the column vectors of @xmath274 fulfill @xmath275 and @xmath276 . the column vectors @xmath277 of a symplectic cellular automaton generate maximally isotropic @xmath165-subspaces @xmath278 , since these are the images of the basis vectors @xmath279 and @xmath280 under the invertible symplectic transformation @xmath107 . because the basis vectors generate by construction maximally isotropic subspaces , this must then also be true for the images @xmath277 . in the next subsection , we shall see that the classification of one - dimensional symplectic cellular automata is easier to handle . a useful observation is that a @xmath3-dimensional symplectic cellular automaton @xmath281 induces for each direction @xmath282 a one - dimensional cellular automaton . to see this , we introduce for each direction @xmath282 a surjective ring homomorphism @xmath283 which maps the polynomial ring @xmath165 of @xmath3-variables @xmath284 onto the ring @xmath246 of one variable @xmath285 . the ring homomorphism @xmath283 assigns to a polynomial @xmath210 the polynomial @xmath286 which only depends on the variables @xmath287 . the ring homomorphism @xmath283 evaluates the polynomial @xmath210 at @xmath288 , for @xmath289 , whereas @xmath290 is the remaining free variable . for a symplectic cellular automaton @xmath281 the conditions @xmath291 and @xmath276 are identities of polynomials . the matrix @xmath292 is build by applying the ring homomorphism @xmath283 to each matrix element individually . obviously , the identities @xmath293 as well as @xmath294 follow . as a consequence we get : [ cor - red - dim ] let @xmath281 be a @xmath3-dimensional symplectic cellular automaton . then for each direction @xmath282 , the two - by - two matrix @xmath292 is a one - dimensional symplectic cellular automaton . now it is easy to show that symplectic cellular automata are reflection invariant and that the determinant is a monomial . it is slightly more involved that we have reflection invariance with respect to a lattice point and not with respect to an half integer lattice point . [ thm - sca-1 ] a @xmath165-linear map @xmath281 is a symplectic cellular automaton , if and only if , the following holds : 1 . the matrix @xmath107 is a reflection invariant with respect to some lattice point @xmath295 . the @xmath165-valued determinant of @xmath107 is @xmath296 . if @xmath107 is a symplectic cellular automaton , then the column vectors @xmath277 generate maximally isotropic subspaces . by lemma [ lem - iso-1 ] it follows that @xmath297 , respectively @xmath298 , is reflection invariant to some half integer lattice point @xmath299 , respectively @xmath300 . since @xmath107 preserves the symplectic form @xmath189 we obtain @xmath301 and therefore @xmath302 for an half - integer lattice point @xmath232 . as a consequence , @xmath107 is reflection invariant for @xmath232 . now , @xmath303 . vice versa , let @xmath107 be a matrix , which is invariant with respect to the reflection at @xmath299 and whose determinant is @xmath296 . then the column vectors @xmath277 are reflection invariant , which implies ( by lemma [ lem - iso-1 ] ) that @xmath291 holds . the determinant of @xmath107 is @xmath304 which implies @xmath276 . thus @xmath107 preserves the symplectic form @xmath189 . by corollary [ cor - red - dim ] , we obtain a one - dimensional symplectic cellular automaton @xmath305 for each lattice direction @xmath282 . we have already shown that the column vectors @xmath277 are reflection invariant for @xmath306 , which implies that for each direction @xmath71 the column vectors @xmath307 are reflection invariant for @xmath308 . we also have that @xmath307 generate maximally isotropic @xmath246-subspaces , since these define valid cellular automaton rules . suppose now , that @xmath309 is reflection invariant for @xmath310 in the half - integer lattice . then we can translate @xmath216 by an even translation @xmath311 , such that @xmath312 is either @xmath313 or @xmath314 . if @xmath216 is of even length , then @xmath315 follows . the polynomial @xmath316 is reflection invariant for @xmath317 and can be expanded as @xmath318 now , for each @xmath319 , the polynomial @xmath320 is a multiple of @xmath321 . thus @xmath216 is also a multiple . ] of @xmath322 . from this we conclude that , if @xmath300 is not an integer , then a reflection invariant @xmath247 is a multiple of @xmath322 and does not generate a maximally isotropic @xmath246-subspace , since @xmath323 is a nontrivial common divisor of @xmath208 and @xmath209 , which is a contradiction . so @xmath308 must be an integer lattice point , that is , @xmath324 . so each symplectic cellular automaton @xmath107 is reflection invariant for the reflection at some lattice point @xmath295 . therefore , the symplectic cellular automaton @xmath325 is reflection invariant with respect to the origin @xmath326 . in the subsequent , we call all symplectic cellular automata , which are reflection invariant with respect to the origin , to be `` centered '' and it is sufficient to classify only those . the polynomials in @xmath165 which are reflection invariant with respect to the origin form a subring @xmath327 and will be simply called reflection invariant ( for @xmath36 we will again omit the index ) . from theorem [ thm - sca-1 ] , we obtain a handy characterization of centered symplectic cellular automata : [ cor - sca-1 ] the group of centered symplectic cellular automata is given by the group @xmath328 of two - by - two matrices @xmath107 with entries in the subring @xmath329 of reflection invariant polynomials and @xmath329-valued determinant @xmath330 . the symplectic transformation corresponding to the `` cluster state qca '' ( eq . [ example-0 ] ) is given by @xmath331 obviously all entries are reflection invariant with respect to the origin and the determinant is equal to one ( modulo 2 ) . a nice aspect of corollary [ cor - sca-1 ] is that the centered symplectic cellular automata can be obtained by the following strategy : choose two arbitrary reflection invariant @xmath332 and find all possible factorizations of the polynomial @xmath333 into a product of two reflection invariant @xmath334 . the corresponding symplectic cellular automaton is then given by @xmath335 even if the task of factorizing the polynomial @xmath336 is quite cumbersome , there is always a `` trivial '' solution , namely , @xmath337 and @xmath338 . the matrix @xmath339 describes the corresponding symplectic cellular automaton . another remarkable fact is that , due to cramer s rule , the inverse of a centered symplectic ca @xmath107 is simply given by @xmath340 similarly we have that for a symplectic ca @xmath107 containing a translation by @xmath299 positions , e.g @xmath341 , the inverse contains a translation by @xmath342 positions . in this subsection we are investigating one - dimensional symplectic cellular automata . as already mentioned , we can achieve more results in this case , because we can apply the euclidean algorithm ( lemma [ lem - euclid ] ) . we will use the euclidean algorithm to show that for every reflection invariant @xmath247 with @xmath219 , there exists at least one corresponding reflection invariant @xmath343 such that @xmath344 holds and is therefore a valid column of a symplectic cellular automaton matrix . we will use this fact to show that every uniquely determined and translationally invariant stabilizer state can be prepared from a product state by applying one timestep of a clifford qca . stabilizer states are studied extensively in the last years ( @xcite and @xcite are just examples , which are useful as introductory texts ) . the basic concept is to fix an abelian group of operators ( usually a subgroup of the pauli group ) , also called stabilizer group , and to define a stabilizer state as common eigenvector of all these operators . in our case we are looking for translationally invariant states , so the stabilizer group is generated by all translates of one single weyl operator @xmath102 for some phase space vector @xmath247 . the state should fulfill @xmath345 for all @xmath346 . the stabilizer formalism is often studied for finitely many qudits . in that case it is known that the stabilizer state is uniquely determined , if the number of generating operators is large enough ( see e.g. @xcite for a quantitative statement ) . in our situation we have infinitely many qudits , so we can not apply this result . but it turns out that the operators @xmath347 must generate a maximal abelian algebra , or equivalently , the subspace @xmath348 must be maximally isotropic . [ structure - thm-1 ] for a phase space vector @xmath247 the following is equivalent : 1 . [ equiv-1 ] there exists a uniquely determined state @xmath349 with @xmath350 for all @xmath346 . [ equiv-2 ] @xmath348 is a maximally isotropic @xmath246-subspace . [ equiv-3 ] there is a clifford qca @xmath11 with @xmath351 . [ equiv-4 ] @xmath76 is a reflection invariant and @xmath219 . [ equiv-2 ] . @xmath352 [ equiv-4 ] . because @xmath348 is a maximally isotropic subspace we conclude from lemma [ lem - iso-1 ] that @xmath76 is reflection invariant with @xmath219 . [ equiv-4 ] . @xmath352 [ equiv-3 ] . we have to find @xmath353 with @xmath344 and @xmath354 . with lemma [ lem - euclid ] we find a solution @xmath355 of the equation @xmath356 and @xmath357 is a solution of @xmath358 . yet we do not know , whether @xmath359 is reflection invariant , or equivalently , whether @xmath360 holds . but if @xmath359 is a solution of @xmath358 then the same is true for @xmath361 . thus we have to solve the condition @xmath362 . the polynomial @xmath363 is anti - symmetric with respect to the reflection @xmath364 and it can be expanded as @xmath365 . by choosing @xmath366 we find that @xmath361 is indeed reflection invariant . the matrix @xmath367 is then a symplectic cellular automaton and induces a clifford qca with the desired property . [ equiv-3 ] . @xmath352 [ equiv-1 ] . consider a state @xmath349 with the desired property . then this state is equal to @xmath368 , whereby @xmath369 is a state with @xmath370 for all @xmath346 , so the stabilizer group of this state is given by all translates of @xmath119 . this means @xmath369 is a translationally invariant product state , which is determined by the equation @xmath371 and corresponds to the one dimensional projector onto the eigenspace of @xmath119 with eigenvalue @xmath314 . therefore this state is uniquely determined and @xmath372 is the unique state with @xmath350 . [ equiv-1 ] . @xmath352 [ equiv-2 ] . suppose @xmath348 is an isotropic @xmath246-subspace but not maximally isotropic . by lemma [ iso - inbedding ] , we know that there exists a phase space vector @xmath373 with @xmath374 . so we have @xmath375 with @xmath216 not invertible and lemma [ lem - iso-1 ] tells us that @xmath373 is reflection invariant . with help of the euclidean algorithm we find a qca @xmath11 and a corresponding symplectic transformation @xmath107 with @xmath376 ( just as step two of this proof ) . now consider a product state @xmath149 with @xmath377 depending on the @xmath378 . we transform this state with @xmath379 and the expectation values of the operators @xmath380 should be all equal to 1 : @xmath381 so we have to solve the equations @xmath382 to get appropriate @xmath378 and therefore states with the desired property . since @xmath216 is not invertible the support of @xmath383 is not a one - elementary set . let @xmath384 be the minimal interval such that @xmath385 . we can choose arbitrary @xmath386 to compute @xmath387 from the equation @xmath388 . recursively all @xmath378 can be calculated from the other equations but the solution will depend from the initial choice of the @xmath386 . this means that there exists more than one state @xmath149 of the above form , such that @xmath389 is fulfilled . so the uniqueness of the state in [ equiv-1 ] . forces @xmath348 to be maximally isotropic so we have shown that every translationally invariant and uniquely determined stabilizer state in a one - dimensional lattice can be prepared out of a product state by a single timestep of a clifford qca . unfortunately we can not generalize this result to higher lattice dimensions , because lemma [ lem - euclid ] is only valid for univariate polynomials . the euclidean algorithm for computing the greatest common divisor can be generalized to multivariate polynomials @xcite , but the extended version ( equation ( [ euclid ] ) ) does not hold . we consider again the phase space vectors @xmath235 and @xmath236 ( see example [ stand - ex ] ) . as already mentioned , the phase space vector @xmath242 is reflection invariant for @xmath317 and generates an isotropic @xmath246-subspace , but none of the statements of theorem [ structure - thm-1 ] holds : the expectation value of @xmath390 is equal to one both in the `` all spins up '' and in the `` all spins down '' state , so there is no uniquely determined stabilizer state . as we have seen in [ stand - ex ] the subspace @xmath391 is not maximally isotropic . the reflection invariance does not hold for an integer lattice point , so @xmath242 is not a valid column of a symplectic ca , and @xmath243 is a common divisor of @xmath392 and @xmath393 , which is not invertible . in contrast @xmath239 fulfills all four conditions . the uniquely determined stabilizer state is given by the one - dimensional cluster state and a possible cqca is given by example [ example-0 ] . we have seen that the set of centered cqcas form a group and that this group is given by @xmath187-matrices with determinant one and reflection invariant polynomials as matrix elements . in the one - dimensional case the group structure can be more clarified , since we are able to give a complete set of generators , which can be regarded as elementary operations . a simple example of a @xmath187-matrix in @xmath328 is for some reflection invariant polynomial @xmath394 given by @xmath395 which we will call `` shear transformation '' . in particular , @xmath396 holds for all @xmath397 is a group homomorphism from the additive group @xmath398 into the group of centered symplectic cellular automata @xmath399 . ] . the symmetric polynomials @xmath400 , @xmath319 , and @xmath401 form a basis of the subring @xmath398 . thus every shear transformation can be decomposed into a finite product of elementary shear transformations @xmath402 with @xmath403 and @xmath170 . the local rule of the corresponding qca @xmath404 with @xmath405 is for @xmath406 given by @xmath407 for @xmath408 we have the single cell operation ( `` local shear transformation '' ) @xmath409 which correspond for @xmath28 to applying the phase gate @xmath410 to all single cells . another single cell operation is the `` local fourier transformation '' , which is in phase space given by the matrix @xmath411 with some constant @xmath412 ( for @xmath315 we will write @xmath413 ) . for @xmath28 we have @xmath315 and the corresponding qca switches the operators @xmath58 and @xmath30 in each single cell and is therefore given by applying the hadamard matrix . note that all symplectic single cell transformations can be obtained by a product of local shear and local fourier transformations , which is a generalization to higher cell dimensions of the fact that local clifford operations are generated by hadamard and phase gate . the symplectic transformations @xmath414 are elementary symplectic cellular automata in the sense of the following theorem . the proof , which is technically slightly more involved , is given in the appendix [ app-3 ] . [ structure - thm-2 ] the group of centered symplectic cellular automata @xmath399 is generated by the set @xmath415 . a more concrete formulation of the statement of theorem [ structure - thm-2 ] is that every one - dimensional centered symplectic cellular automaton @xmath107 is a finite product of shear transformations and local fourier transforms of the following form : @xmath416 with reflection invariant polynomials @xmath417 and constants @xmath418 . let us consider in the qubit case ( @xmath28 ) the symplectic cellular automaton @xmath419 and , since @xmath420 for @xmath28 , we have @xmath421 . the corresponding cqca is given by @xmath422 the basic idea for deriving a decomposition like ( [ decomp ] ) is to reduce the support of the first column of @xmath107 by applying a shear transformation from the right . we get @xmath423 this matrix is obviously equal to @xmath424 and we have @xmath425 which is indeed a decomposition in accordance with ( [ decomp ] ) . for @xmath28 all the generators @xmath426 and @xmath427 are their own inverses , so the time evolution of these operations alternates between the identity and a single timestep of the automaton . especially these qcas show no propagation , because the neighborhood of the iterated automaton does not increase with the number of timesteps . a nontrivial time evolution only occurs , if the symplectic cellular automaton is composed of at least two different generators . in this chapter we are looking for translationally invariant clifford operations with periodic boundary conditions on an @xmath3-dimensional lattice . these boundary conditions are given by an @xmath3-dimensional torus @xmath428 , which is determined by @xmath3 independent lattice vectors @xmath429 ( see figure [ torus ] ) , and all lattice points which differ by these vectors are identified . the number of ( not identified ) lattice points is given by @xmath430 . we denote here by @xmath431 the ring of polynomials @xmath432 such that the @xmath433-variables fulfill the periodic boundary conditions @xmath434 . this guaranties that the product of two polynomials from @xmath431 is again an element from @xmath431 . but algebraically there are large differences between @xmath165 and @xmath431 : @xmath431 is not a division ring , because there are zero divisors and there are in general other invertible elements than @xmath435 . the reflection @xmath436 is again given by replacing @xmath285 by @xmath437 or in other words we substitute @xmath438 by @xmath439 . the symplectic form @xmath189 is then of the same form as in the infinite lattice case . a @xmath138-dimensional torus defined by @xmath440 and @xmath441 . ] now we have to say , how a clifford qca ( or a symplectic cellular automaton ) is defined on these systems . in the general theory of qcas @xcite , the neighborhood of a qca with periodic boundary conditions is not allowed to be too large in comparison with the torus . this guaranties that the qca can be extended to the whole lattice . since this case is covered by restricting the existing clifford qcas to periodic boundary conditions , we drop all locality conditions , and we take the same structure as in corollary [ cor - sca ] as definition of a symplectic cellular automaton : _ a @xmath187 matrix @xmath442 with entries in @xmath431 is a symplectic cellular automaton if the column vectors fulfill @xmath275 and @xmath276 . _ with this definition it is possible to state an analogous version of theorem [ structure - thm-1 ] also for periodic boundary conditions . but the proof is quite different from the infinite lattice case . for a phase space vector @xmath443 the following is equivalent : 1 . [ equiv1 - 1 ] there exists a uniquely determined state @xmath349 with @xmath350 for all @xmath444 . [ equiv1 - 2 ] @xmath445 is a maximally isotropic @xmath431-subspace . [ equiv1 - 3 ] there is a symplectic cellular automaton @xmath107 with @xmath446 . [ equiv1 - 1 ] . @xmath447 [ equiv1 - 2 ] . a stabilizer state on @xmath448 qudits is uniquely determined , if and only if , the minimal number of generators of the stabilizer group equals @xmath448 @xcite . here we have the @xmath449 generators @xmath347 . these are independent , if and only if , they generate a maximally abelian algebra , or equivalently , if @xmath445 is a maximally isotropic subspace . [ equiv1 - 2 ] . @xmath352 [ equiv1 - 3 ] . since we have a finite dimensional space , there exists a symplectic basis , and any basis of a maximally isotropic subspace can be extended to a symplectic basis @xcite . for this construction we turn to the original phase space ( by inverse `` fourier transform '' ) and define by @xmath450 basis vectors of a subspace . since we know by [ equiv1 - 2 ] . that this space is isotropic , these vectors fulfill @xmath451 and therefore @xmath452 . then there exists a dual vector @xmath453 with @xmath454 and we define @xmath455 . we get that @xmath456 holds and so we have @xmath344 . we have to verify that we can choose @xmath343 , such that @xmath354 holds . but if @xmath343 is a solution to @xmath344 the same is true for @xmath457 for any @xmath458 and we can find similar to the case of theorem [ structure - thm-1 ] an appropriate @xmath216 with @xmath360 . [ equiv1 - 3 ] . @xmath352 [ equiv1 - 2 ] . suppose that @xmath107 is a symplectic cellular automaton with @xmath446 . then @xmath107 induces a homomorphism between the maximally isotropic subspace @xmath459 and @xmath445 with @xmath460 . since @xmath107 is invertible and preserves the symplectic form @xmath189 , it follows that any maximally isotropic subspace is mapped onto a maximally isotropic subspace , which implies that @xmath445 is maximally isotropic . as an example of a translationally invariant stabilizer state , we consider translationally invariant graph states @xcite . the graph is encoded by its adjacency matrix @xmath461 , and the isotropic subspace that determines the graph state is given by the phase space vectors @xmath462 with @xmath463 . translation invariance of the graph state implies that the matrix elements @xmath464 depend only on the difference @xmath465 , so there is a function @xmath466 such that @xmath467 holds . thus after fourier transform the phase space vector , which generates the maximally isotropic subspace , is given by @xmath468 and we can choose a suitable symplectic cellular automaton @xmath107 with @xmath446 by @xmath469 a translationally invariant graph state on the 1d torus @xmath470 . ] ' '' '' figure [ fig-1 ] represents a translationally invariant graph state on the 1d torus @xmath470 . the adjacency matrix @xmath471 is given by @xmath472 and obviously only depends on the difference @xmath465 of the variables @xmath473 . applying the fourier transform , yields the polynomial @xmath474 . the symplectic cellular automaton , which creates the graph state as explained above is given by the matrix @xmath475 note that @xmath107 is reflection invariant , since @xmath476 is a reflection fix - point . we are going to present another example to show that the phase space vectors do not have to be reflection invariant , because not all invertible elements are monomials . but the invertibility of a fixed polynomial depends on the size of the torus and so a fixed phase space vector @xmath76 may define a translationally invariant stabilizer state for some @xmath477 , but it is possible that there exists @xmath478 , such that @xmath479 is not maximally isotropic , and therefore @xmath76 does not characterize a unique stabilizer state for @xmath478 . we consider for @xmath28 the phase space vector @xmath480 on a one dimensional torus of variable size . note that @xmath481 holds for all @xmath477 , so @xmath76 generates an isotropic subspace . the corresponding tensor product of pauli operators is given by @xmath482 and is obviously not reflection invariant . let us first have a look at the case @xmath483 . it is easy to show that @xmath484 is not invertible . we define @xmath485 and have that @xmath486 , but @xmath487 . so @xmath488 is not maximally isotropic , and there is no unique stabilizer state . for @xmath489 the inverse of @xmath484 is given by @xmath490 , so @xmath76 and @xmath491 generate the same subspace , which is actually maximally isotropic . so @xmath76 is indeed a valid column of a symplectic automaton , but starting from the `` all spins up '' state both cqcas corresponding to @xmath492 prepare the same stabilizer state . we have analyzed the structure of clifford quantum cellular automata that act on a @xmath3-dimensional lattice of @xmath4-level systems . the results which can be achieved depend on the dimension of the lattice and whether we put periodic boundary conditions or working with the infinite lattice . we have characterized the group of cqcas in terms of symplectic cellular automata on a suitable phase space . with the help of fourier transform , this phase space can be identified with two - dimensional vectors of laurent - polynomials , and symplectic cellular automata can be described by two - by - two matrices with laurent - polynomial entries . we have shown that these entries must be reflection invariant and that up to some global shift the determinant of the matrix must be one , so the group of cqcas is isomorphic to the special linear group of two - by - two matrices with reflection invariant polynomials as matrix elements . we have proven that there is a correspondence between 1d cqcas and 1d translationally invariant stabilizer states . for a fixed translationally invariant pure stabilizer state @xmath493 , which is in particular a product state , every other translationally invariant pure stabilizer state @xmath149 can be created by applying an appropriate cqca @xmath494 to the chosen product state : @xmath495 . pure stabilizer states can be also characterized by maximally isotropic subspaces . we have characterized the phase space vectors , which generate maximally isotropic subspaces , namely their components must be coprime and reflection invariant . in the one - dimensional case we have also more clarified the group structure of cqcas . as we have shown , each one - dimensional cqca can be decomposed into a product of elementary shear automata and local fourier transforms , so the group of cqcas is generated by this set of operations . for periodic boundary conditions the techniques from infinitely extended lattices can be applied to a certain extend . according to the discussion of translationally invariant stabilizer states on the 1d lattice , we have proven that there is an analogous correspondence between cqcas and translationally invariant stabilizer states with periodic boundary conditions even in any lattice dimension . hv is supported by the `` dfg forschergruppe 635 '' . [ iso - inbedding ] let @xmath496 and @xmath220 be an isotropic , but not maximally isotropic @xmath165-subspace . then there exists a phase space vector @xmath497 such that @xmath498 is maximally isotropic . 1 . @xmath500 ( analogously @xmath501 ) : then @xmath502 and @xmath209 is not invertible since this would force @xmath503 to be maximally isotropic . so we can set @xmath504 . 2 . @xmath505 ( analogously @xmath506 ) with @xmath216 reflection invariant : then we have @xmath507 . we set @xmath508 and get that @xmath509 is a maximally isotropic subspace since @xmath510 implies @xmath511 . @xmath512 and @xmath513 : then @xmath514 with @xmath216 invertible , so @xmath76 is reflection invariant for some @xmath515 . because @xmath220 is not maximally isotropic we can find @xmath224 with @xmath516 . since @xmath208 and @xmath209 are nonvanishing this implies @xmath517 for some @xmath518 . we can choose @xmath223 and @xmath509 to be maximally isotropic . for a polynomial @xmath309 the coefficient of the monomial @xmath438 is @xmath519 . recall that `` degree''of a polynomial in @xmath309 is defined by @xmath520 and that the support is defined by @xmath521 . [ lem - reduce ] let @xmath186 be a symplectic cellular automaton which is invariant under the reflection at the origin : @xmath522 and @xmath523 . if the degrees of column vectors fulfill @xmath524 then there exists a shear transformation @xmath426 , with reflection invariant @xmath309 , such that the symplectic transformation @xmath525 satisfies @xmath526 and @xmath527 . since @xmath76 and @xmath343 are reflection invariant , the degree is an even integer and we introduce @xmath528 , @xmath529 , as well as @xmath530 . we conclude from the identity @xmath531 that @xmath532 is valid . this implies that @xmath533 for some @xmath534 . now @xmath535 which implies that @xmath536 now we observe @xmath537 from which we conclude that @xmath538 . if @xmath539 we can find a shear transformation @xmath540 such that @xmath541 fulfills @xmath542 . we can proceed this reduction until the @xmath543th step with @xmath544 . the resulting symplectic cellular automaton @xmath545 then satisfies @xmath546 . if @xmath547 then , there is an appropriate constant @xmath548 such that @xmath549 holds with @xmath527 . here @xmath216 is the reflection invariant polynomial @xmath550 if @xmath551 , then the shear transformation @xmath552 is not applied and we get @xmath553 with the polynomial @xmath554 . let @xmath555 be a symplectic cellular automaton which is invariant under the reflection at the origin . then , by lemma [ lem - reduce ] , there exists a symplectic cellular automaton @xmath556 and a shear transformation @xmath557 such that @xmath558 and @xmath539 . thus we can iterate this reduction process until @xmath559 is a constant symplectic transformation ( corresponding to a qca with single cell neighborhood ) , which can be decomposed into a product of local shear transformations @xmath560 and local fourier transforms @xmath561 with @xmath562 . this yields the following decomposition of @xmath555 : @xmath563 krger , o. and werner , r. f. : _ gaussian quantum cellular automata _ , in : n. cerf , g. leuchs , e.s . polzik ( ed . ) , `` quantum information with continuous variables of atoms and light '' , imperial college press ( 2007 ) .

we present numerical simulations of active fluid droplets immersed in an external fluid in 2-dimensions using an immersed boundary method to simulate the fluid droplet interface as a lagrangian mesh . we present results from two example systems , firstly an active isotropic fluid boundary consisting of particles that can bind and unbind from the interface and generate surface tension gradients through active contractility . secondly , a droplet filled with an active polar fluid with homeotropic anchoring at the droplet interface . these two systems demonstrate spontaneous symmetry breaking and steady state dynamics resembling cell motility and division and show complex feedback mechanisms with minimal degrees of freedom . the simulations outlined here will be useful for quantifying the wide range of dynamics observable in these active systems and modelling the effects of confinement in a consistent and adaptable way .

1605.01621

active fluids are ubiquitous in biology and include the cell cytoskeleton @xcite , bacteria films and even schools of fish @xcite . the common trait of these systems is that they are driven out of equilibrium at the level of their constituents , which interact hydrodynamically with their neighbours . in this paper we use a continuum model of active fluids developed in @xcite where the constituent fluid particles generate local force dipoles . these systems demonstrate aspects of cytoskeletal dynamics in cells or collections of microswimmers such as bacteria , which generate dipolar stresses in their surrounding medium . recent experimental work has shown that it is possible to confine these systems to droplets or vesicles _ in vitro _ to isolate their dynamics and observe emergent behaviour @xcite . with these systems as motivation we consider two limits of the active fluid model , firstly the case of an isotropically ordered active fluid confined to a droplet interface and secondly the case of a droplet of active fluid with strong local polar ordering . we are able to capture the full dynamics of these two systems in a single computational model that predicts interesting phase behaviour in both cases . we have developed numerical simulations to model active fluids confined within and on the interface of deformable droplets in 2-dimensions . the simulations are based on the immersed boundary method introduced by peskin @xcite to model interactions between fluids and elastic interfaces in biological systems . this method models the droplet interface as a 1d lagrangian mesh of points which interacts with a fixed fluid mesh via a numerical analogue of the dirac delta function ( as outlined in the methods section ) . the advantage of this method over phase field methods ( such as @xcite ) is that we can track and numerically conserve quantities defined on the interface naturally in the simulation and potentially model a wide ranges of interface properties . moreover , we can easily incorporate external boundaries in a self - consistent manner . however , a limitation of this method is that it is significantly more difficult to consider droplet separation or combination , phenomena which are captured automatically by phase field methods . applying the immersed boundary method to active fluids is a natural step as the properties and geometries of biological interfaces can be complex . we simulate the coupled dynamics of active particles on the interface as well as in the droplet bulk in a self consistent way . firstly , this allows us to exactly conserve the mass of active particles within the drop ( as shown in methods ) . this leads to interesting droplet dynamics including transitions from stationary to motile and back to stationary as a function of activity ( see results : active polar fluid ) . secondly , we show that this method can also consistently model an active polar fluid droplet and compare the results of these to previous studies in the literature as well as characterising previously unpredicted states . furthermore , we have made improvements to previous fluid immersed boundary models by ensuring conservation of droplet area ( to be consistent with incompressibility ) and to reduce numerical instabilities in the boundary due to the finite sized mesh ( as shown in methods ) . these are necessary improvements in order to model these active systems where flows are generated locally . the structure of the paper is as follows . we first introduce the analytical equations describing the system in the following section , and then the numerical implementation of these is presented in the methods section . following this we present tests of the simulations and compare with other simulation data . we then present some interesting results from these simulations for the two cases of active droplets discussed . these include spontaneous symmetry breaking leading to steady swimming states , droplet deformation , and oscillating states . in the discussion section we describe potential uses for these simulations and extensions to these that could be performed in the future to investigate even more biologically relevant systems . in particular we discuss various ways that confinement can be added to this model . furthermore , we outline how to extend these simulations to 3 dimensions and how we expect the results presented here to generalise to the 3d case . finally , we also discuss potential experimental realisation of these simple systems _ in vitro _ and how these simulations can be used to simulate multicomponent active droplets or droplets with active and passive components . we construct numerical simulations of active fluid droplets using an immersed boundary ( ib ) method for modelling fluid - fluid interactions . the ib method was originally formulated for modelling fluid - structure interactions in biological problems , where the structure can be represented by elastic fibres @xcite . in our simulations , we follow the work of @xcite which adapts this to fluid - fluid interactions in a 2d domain . we begin by considering an infinitely thin , closed interface @xmath0 represented by the position vector @xmath1 where @xmath2 is a lagrangian coordinate along the curve @xmath0 and @xmath3 is time . the force density on this interface due to the surface tension is given in 2d by : @xmath4}\end{aligned}\ ] ] where @xmath5 is the surface tension , @xmath6 is the tangent vector to the surface and @xmath7 . this acts as an external force on the fluids , which can be translated to a 2d force density via the 2d dirac delta function as follows @xmath8 where @xmath9 is the position vector in the medium . the 2d dirac delta is defined by the identity @xmath10 where @xmath11 is the total fluid domain , consisting of fluids labelled 0 and 1 . in the case of incompressible stokes flow , which we assume here , force balance in the fluid dictates : @xmath12 where @xmath13 is the fluid viscosity , @xmath14 is the flow velocity and @xmath15 the pressure . at the length and velocity scales of the cell cytoskeleton , inertia is negligible . an estimate of the reynolds number using the parameter tables in the appendix and the density of water @xmath16 kg m@xmath17 gives @xmath18 ( assuming the maximum velocity scale is @xmath19 ) . finally , if we assume that the fluids are completely immiscible , then the interface velocity will be the same as the velocity in the medium at that point , such that : @xmath20 these simple equations model the evolution of a newtonian fluid droplet embedded in another newtonian fluid , with a varying surface tension @xmath21 at the interface . in @xcite surface tension decreases with surfactant concentration @xcite , however here we consider the case where it depends on a concentration of active matter on the interface generating contractile stresses . thus the model is similar to models of anti - surfactants on an interface that locally increase surface tension @xcite . the difference in an active system is that energy is consumed locally by the particles and converted to work , and it is assumed that the chemical fuel concentration can be treated as a constant everywhere . we consider an active liquid crystal dispersed on the interface such that the fluid particles are ordered along the tangent line of the interface . this is the 2d limit of the case of a 3d fluid droplet with an isotropically ordered liquid crystal confined to the plane of the interface ( the 3d case is discussed analytically in @xcite ) . as in the model of the actin cortex in @xcite we consider that this active fluid is contractile enough to generate a negative pressure in the ( 1d ) boundary layer for small concentrations , and that a passive pressure will become stronger at larger concentration . this is in qualitative agreement with experiments on reconstituted crosslinked actomyosin _ in vitro _ which , within a range of myosin and crosslinker density , initially contracts and increases in density until a steady state size is reached @xcite . in this case , we consider the lowest order contribution to be linear in @xmath22 as we are using a one - component description . these contractile forces , when confined along the tangent line of the droplet interface , give rise to a change in surface tension equivalent to the stress in the active fluid : @xmath23 where @xmath24 is the bare surface tension of the fluid drop . the activity @xmath25 is a phenomenological parameter quantifying the strength of the active contractile @xmath26 or extensile @xmath27 forces with passive repulsion force proportional to @xmath28 . it will be useful to consider results with respect the effective activity @xmath29 where @xmath30 is the stationary steady state concentration . note also that there is potentially a passive pressure term linear in @xmath31 absorbed into @xmath25 , but as all of these terms are phenomenological this does not change the model . in general the order of the active term in powers of @xmath22 requires details of the microscopic interaction responsible , but for the scope of this paper we consider this linear case only . as in @xcite conservation of mass on the moving interface @xmath32 gives : @xmath33 where @xmath34 is the stretching factor , relating the length of a boundary element @xmath35 to its initial length @xmath36 such that @xmath37 . eq implies that : @xmath38 the second term in eq is the lagrangian advection term . diffusion of the mass on the closed interface is incorporated with the addition of the term @xmath39 where @xmath40 is the diffusion coefficient on the interface . including this term , the evolution of the concentration @xmath22 becomes : @xmath41 the final term @xmath42 captures any binding to or unbinding from the boundary , which we define in eq . we explicitly consider the coupling of the interface concentration to a bulk concentration @xmath43 in the internal fluid . we only consider this concentration as active when bound to the interface , because an isotropic stress generates no flow in an incompressible medium . in the case of the cytoskeleton active fluid , our system could model a two component system of motor proteins and actin filaments , where the isotropic filament network is concentrated on the interface and remains fixed while a population of motor proteins in the droplet can bind and unbind from this network . then the equations describing an internal bulk concentration can be incorporated into this model by defining a level - set function @xmath44 such that : @xmath45 } { \rm d } l = \begin{cases } 1 \qquad { \rm if } \quad { \boldsymbol{x } } \in \omega_0\\ 0 \qquad { \rm if } \quad { \boldsymbol{x } } \notin \omega_0 \end{cases } \ , .\end{aligned}\ ] ] note that length element @xmath46 changes as the boundary deforms . then , following @xcite we define the concentration field of the unbound active particles as @xmath47 and hence the advection - diffusion equation describing this quantity is : @xmath48}}{\partialt } } + { \nabla \cdot}{{\left ( { \boldsymbol{v}}({\boldsymbol{x}},t)h({\boldsymbol{x}},t)\rho({\boldsymbol{x}},t ) \right ) } } = d_f{\nabla \cdot}{{\left ( h({\boldsymbol{x}},t)\nabla \rho({\boldsymbol{x}},t ) \right ) } } - q({\boldsymbol{x}},t)\end{aligned}\ ] ] where : @xmath49}\delta({\boldsymbol{x}}-{\boldsymbol{x}}){\left| { \frac{\partial{\boldsymbol{x}}}{\partials } } \right|}{\rm d}s \ , . \end{aligned}\ ] ] the unbound concentration of active material at the boundary @xmath50 therefore is defined as : @xmath51 we assume for this model that the binding rates obey a linear scaling with local concentration . however it is likely that this is not a realistic assumption given the large gradients in concentration field that the simulations predict . investigating and incorporating more realistic binding behaviour goes beyond the scope of this work and would demand a model informed by the microscopic details of the system in question . + eq ensures that the total mass of active material is conserved such that : @xmath52}}{\partialt}}{\rm d}^2{\boldsymbol{x } } + \int_c { \frac{\partialc}{\partialt } } { \left| { \frac{\partial{\boldsymbol{x}}}{\partials } } \right|}{\rm d}s = 0 \,.\end{aligned}\ ] ] thus these equations describe advection - diffusion inside the droplet with containing boundary conditions . we independently incorporate an active fluid in the bulk of the drop . as mentioned in the previous section , an isotropic active fluid can not generate flow if it is incompressible , so we consider the case of an active liquid crystal with strong polar ordering . thus , we define a unit polarisation vector @xmath53 which describes the direction of local polar ordering . as shown in @xcite we can model the passive dynamics of this system by the frank free energy functional with appropriate boundary terms . in this context we can write this as : @xmath54}{\rm d}^2{\boldsymbol{x } } \,.\end{aligned}\ ] ] the first term in eq is the one constant approximation of the frank free energy density . the second term is used to replace the lagrange multiplier which ensures @xmath55 as used in @xcite . this term is minimised when @xmath53 is a unit vector and so for large @xmath56 this is realised everywhere inside the droplet except near to any topological defects . this can be seen by writing the polarisation as @xmath57}$ ] so that in the large anchoring limit the free energy is minimised for @xmath58 everywhere , and @xmath59 must satisfy the ginzburg - landau equation @xcite : @xmath60 this ordinary differential equation ( ode ) is non - linear due to the @xmath61 term and hence we are only able to solve it numerically . assuming that @xmath62 at the droplet boundary ( strong anchoring ) and @xmath63 at the origin we find the solutions plotted in fig [ fig : rdep ] . in the limit @xmath64 this term fulfils the role of the lagrange multiplier @xmath65 and strictly enforces @xmath62 on @xmath66 and @xmath63 on @xmath67 . in the simulations , we have to use finite @xmath56 , hence there is some finite gradient of polarisation across the interface and around the defect , which sets the effective core size of the defect in the system . . as @xmath56 increases the distribution tends towards @xmath62 everywhere.,scaledwidth=70.0% ] the final term in eq is minimised for filament polarisation @xmath53 aligned with the outward normal direction near the interface . we consider this polar boundary condition as actin networks are found to order in such a way near the cell membrane in experiments @xcite . this term is the source of polar ordering in the simulations as all the other terms are symmetric for @xmath68 . + this model now looks similar to many phase field models of ( active and passive ) liquid crystal emulsions in the literature @xcite . the key difference in the immersed boundary formulation is that the interfacial width does not depend on the surface tension and is a numerical approximation to a sharp interface , and the level - set function @xmath44 is enslaved to the position of this interface . this means we can usefully change the properties of the boundary without requiring a change to any other part of the calculation . thus , if we assume that activity is confined to the drop we can write the active stress as @xcite @xmath69 the passive distortion stress in the liquid crystal is defined in terms of the molecular field @xmath70 @xcite @xmath71 the molecular field @xmath72 acts to minimise the free energy functional @xmath73 with respect to @xmath53 , and @xmath74 is the coupling constant between the flow gradient and polarisation . the final term is the ericksen stress , a generalisation of the hydrostatic pressure for complex fluids , which in this case is given by @xcite : @xmath75 in addition , there is a stress term due to the polarisation across the interface given by ( as used in @xcite ) : @xmath76 the stress then is incorporated into the force balance so that eq becomes : @xmath77 finally the evolution of the polarisation field is governed everywhere by @xcite : @xmath78 where the vorticity tensor @xmath79 and strain rate @xmath80 act to reorient the polar particles . in this section we look at some preliminary tests performed to check the accuracy of the simulations and results that are pertinent to the research questions posed in the introduction . namely , we quantify symmetry breaking transitions and steady state behaviour including droplet motility and deformation which resemble aspects of cytoskeletal dynamics driven by active stresses . we also look at the role of certain parameters on the phase transitions we observe and relate this to similar systems discussed in the literature . we normalise the length and time scales of the simulations with respect to the initial droplet radius @xmath81 and the viscosity @xmath13 and following some checks of accuracy choose step sizes of @xmath82 , @xmath83 and @xmath84 . we choose a grid length of @xmath85 , which gives @xmath86 , which has prime factors of @xmath87 , @xmath88 , and @xmath89 , resulting in efficiency savings in the fft algorithm . a full list of the parameter value ranges used in these simulations can be found in the methods section . we relate these values to si units by setting the length , time and energy scales to cytoskeleton relevant values such that @xmath90 and @xmath91 @xcite , and @xmath92 @xcite . we first test the fluid simulation code alone without any active terms to check that it agrees with the expected behaviour of a simple fluid drop in 2d . we consider a droplet with an initially deformed shape given by : @xmath93}{\left [ \cos(\theta^0_{k+1/2}),\sin(\theta^0_{k+1/2 } ) \right]}\end{aligned}\ ] ] where @xmath94 ( see fig [ fig : testshape ] ) . we can then measure the evolution of this shape at time @xmath95 using the fourier coefficients : given by eq . blue arrows show the initial flow field , scaled for visibility . * ( b ) * evolution of the shape factor @xmath96 ( see eq ) with respect to the simulation time @xmath3 ( log scale ) for values of the surface tension @xmath5 ( in simulation units ) spanning two orders of magnitude.,scaledwidth=80.0% ] @xmath97 } } { \left ( \theta_{k+1}^n - \theta_{k}^n \right)}\\ \label{fcoeff2 } b^n_l = \sum_{k=0}^{n-1 } \frac{1}{2}{\left ( { \left| \tilde{{\boldsymbol{x}}}^n_k \right| } + { \left| \tilde{{\boldsymbol{x}}}^n_{k+1 } \right| } \right ) } { \sin{\left [ \frac{l}{2}(\theta_k^n + \theta_{k+1}^n ) \right ] } } { \left ( \theta_{k+1}^n - \theta_{k}^n \right)}\end{aligned}\ ] ] where @xmath98 is the point position in the coordinate system with the droplet centre of mass at the origin and @xmath99 . generally , we compute the first few orders of fourier coefficients @xmath100 , but in this particular test we just present @xmath96 as the shape index as this is initially set to @xmath101 by eq . the evolution of this shape index is presented in fig [ fig : testshape](b ) for several values of surface tension @xmath5 . as one would expect the droplet returns to a spherical shape , and the time it takes to do so is linearly proportional to the surface tension . reducing the grid spacing from @xmath82 to @xmath102 makes no discernible difference to the droplet evolution and nor does increasing the grid size @xmath103 . thus our choice of @xmath104 appears sufficient to render interactions with the droplet s periodic image negligible . we also note that once the droplet reaches steady state , there is still some flow in the system even though analytically the flow should be zero . this arises due to inaccuracies in the mapping to the cartesian mesh . we find that these flows are very small and do not affect the droplet shape . however , these flow are on average only an order of magnitude smaller than those produced by the random perturbation that we apply , so they could affect the symmetry breaking direction . in this section we present the steady states observed in the active isotropic interface model . first , we consider the case where the active fluid is bound to the interface , setting @xmath105 and quantify the steady states observed . following this we increase this binding rate and observe the differences this coupling can make to the observed steady state . in all of these cases , in order to enable the droplet to break symmetry we apply a perturbation to the system at time @xmath106 . we do this by altering the concentration at the interface at the beginning of the time - step by a small amount . in order to conserve mass , we perturb the concentration using a fourier series : @xmath107}\end{aligned}\ ] ] where @xmath108 is as defined in eqs and and the coefficients @xmath109 and @xmath110 are assigned randomly generated values between @xmath111 and @xmath112 . we find that for all values of contractile activity tested above a threshold value , a swimming steady state is observed with the droplet remaining circular ( see fig [ fig : abbspeed ] ) as we predict analytically in @xcite . we find it necessary to use a non - zero value of @xmath113 on the interface in order to ensure numerical stability and physically reasonable results . in fig [ fig : abbspeed ] we show how the dependence of steady state droplet swimming speed against effective activity @xmath114 only depends weakly on the parameter @xmath113 . the effect on droplet speed of the parameter @xmath113 is weak because while increasing @xmath113 reduces the magnitude of the gradients in @xmath22 ( and hence @xmath5 ) along the interface , these gradients are spread over more of the droplet interface and still generate flow towards the concentration peak . for different values of @xmath113 . inset : droplet swimming steady state , coloured interface shows concentration ( black to yellow : low to high ) and vector field shows flow field scaled for visibility . simulation parameters : @xmath115.,scaledwidth=60.0% ] we also see from fig [ fig : abbspeed ] that the activity dependence of the speed is typical of active steady state phenomena @xcite initially showing a linear dependence on activity near the threshold and a weaker dependence at higher activity values . this suggests an optimum swimming speed efficiency , above which extra swimming speed begins to cost more active energy . the phase diagram in fig [ fig : abphase](a ) shows how the activity threshold depends on the diffusion of @xmath22 on the interface . this dependence is approximately linear and is well characterised by the linear stability analysis presented in @xcite . interestingly , fig [ fig : abphase](b ) shows that even though diffusion increases the active energy required to break symmetry , faster diffusion rates result in faster swimming droplets at the same activity values . further , we see that at large enough activity , we observe the formation of two peaks in the concentration profile , which quickly coalesce to form one . . simulation parameters : @xmath116 . ] setting the binding rates @xmath117 to be non - zero can completely alter the dynamics of the system . in the case of slow bulk diffusion or binding rate compared to the active flow timescale , we see that this coupling simply dampens the instability discussed in the previous section and reduces the final droplet speed . this is because the active concentration is recycled by the internal fluid , unbinding more at peak concentrations and diffusing slowly through the drop ( see fig [ fig : bulkvel ] ) . we fix the bulk diffusion @xmath118 and increase the binding rate @xmath119 and observe that the feedback from the internal concentration changes the observed behaviour . the flow field in the swimming drop reference is rearward near the interface and forward through the middle . this means that the active particles unbinding at the droplet rear are advected through the middle of the drop and this feedback can lead to the nucleation of a second peak in the concentration . in the simulations this results in a continuous transition with respect to @xmath114 from a steady motile state to ` wandering ' motile states ( characterised by the periodic nucleation and coalescence of a second concentration peak , snapshot shown in fig [ fig : bulkvel](b ) ) and stationary two - peak solutions ( fig [ fig : bulkvel](c ) ) . movies of these simulations can be found in the supplementary information . over a range of contractile activity values for different binding rates @xmath120 . insets show steady state configurations with concentration on interface shown by colour density ( black to yellow ) and blue arrows show the velocity . higher binding rates reduce the droplet speed and enables the formation of two concentration peaks ( see insets ( b ) and ( c ) ) . simulation parameters : @xmath121 , @xmath122 , @xmath123 , @xmath124 . ] we see that the wandering state is really a transition phase from steady motility to a stable two - peak configuration . this result is very interesting as it suggests a simple feedback mechanism such as advection - diffusion of material through the droplet is sufficient to stabilise the two - peak solution . physical intuition suggests that two peaks of this nature should always be unstable due to the hydrodynamic interaction between them . thus these stable solutions highlight a simple physical mechanism that is generic to such droplet systems . however , as outlined in @xcite , this transition is not simply a function of the effective peclt number in the drop , since faster diffusion in the drop decreases the activity required to spontaneously form multiple peaks in concentration on the interface . however , it is clear that the binding rate and advection terms need to be large to stabilise the two - peak solution . we can test that the simulations of a droplet of active polar fluid are working by comparing to the hybrid lattice boltzmann ( lb ) simulations published in @xcite . by adjusting our simulation parameters to be the same as in the hybrid lb simulations of an active polar fluid we can compare the steady states directly . as in reference @xcite we set the following parameters as : @xmath125 , @xmath126 , @xmath127 , @xmath128 , @xmath129 , @xmath130 , @xmath131 , @xmath132 and @xmath133 . we initialise both sets of simulations with a small polarisation in the positive @xmath134 direction inside the droplet , and perturb the systems by applying a small normal anchoring @xmath135 for the first 5000 time - steps . we set the surface tension to @xmath136 which is approximately the effective surface tension in the phase field model in @xcite . we also note that , the active stress in the hybrid lb simulations is proportional to @xmath137 where @xmath138 inside the droplet , and @xmath139 outside . therefore , when we compare activity values in fig [ fig : lbcomp](a ) , the activity values used in the hybrid lb simulations are multiplied by @xmath87 for consistency . we can see from fig [ fig : lbcomp ] that we have good agreement between the two sets of simulations , with similar steady state velocity trends and droplet shape over a range of contractile activity values . we see that the symmetry breaking threshold agrees well between the simulations , however the non - linear dependence of droplet speed on activity is slightly different . this may in part be due to the estimates of surface tension which have no analytical analogue in the lb simulations , as well as the only approximate incompressibility achieved by lb ( which is exact in our model ) . ( purple outside , yellow inside ) and the black arrows show the polarisation field.,scaledwidth=80.0% ] in this section we consider specifically the case with a large polar anchoring term @xmath140 in order to investigate the effect of this coupling between internal polarisation and shape . we observe that with this boundary condition we only see an active motile state for extensile activity @xmath141 . this is in contrast to simulations by giomi _ @xcite which show that , with normal anchoring an active _ nematic _ droplet shows swimming states for contractile activity . this suggests that it is the difference between nematic and polar order that is responsible for the difference . the stationary state of the system with polar order consists of an aster with defect @xmath142 at the droplet centre , whereas in the nematic case it consists of an elongated droplet with @xmath143 defects along the long - axis of the droplet . this difference in stationary state therefore changes the geometry such that the stability threshold for motion changes sign . this change appears to be robust for the values of @xmath74 we have tested ( @xmath144 corresponding to ` rod - like ' and one to ` disc - like ' flow aligning particles @xcite ) . as predicted in @xcite the activity threshold for the onset of motion does not depend on the surface tension , however the droplet steady state speed as we can see in fig [ fig : shapevel ] the motile steady state is elongated and the shape deformation depends on the surface tension . in fig [ fig : shapevel ] we plot the @xmath145 mode shape parameter @xmath146 from eqs and ( essentially measuring droplet elongation ) and final droplet velocity against the surface tension . clearly , due to the strong anchoring at the interface , the polarisation profile is coupled to the boundary shape , so when the surface tension is small the polarisation profile is able to deform the droplet more and reach a faster steady state . as surface tension is increased the deformation reduces and goes towards a circular droplet , and in this limit the droplet velocity begins to plateau at its minimum value . this shows a positive feedback between the shape and the motion in these droplets , but in this case it does not affect the symmetry breaking threshold . this defect driven motion is similar in appearance to those predicted for crawling active droplets in @xcite , despite the difference in geometry and flow boundary conditions . mode shape parameter ( calculated as @xmath146 from eqs and at time @xmath147 , left axis ) plotted against droplet surface tension . the simulations use @xmath148 and @xmath149 . inset shows the swimming steady state for @xmath150 where the shading is a schlieren texture representation of the polarisation field for crossed @xmath134-@xmath151 polarisers such that dark regions represent areas where the polarisation is parallel or antiparallel to the @xmath134or @xmath151 direction ( or isotropic as is the case outside the droplet ) . the blue arrows show the flow field in the fluid ( averaged over neighbouring points for clarity ) , while the blue line traces the droplet interface . other parameters were set as listed in tables [ fluid params ] and [ apfparams ] with @xmath152.,scaledwidth=70.0% ] for contractile activity we see that the lowest order symmetry breaking mode corresponds to a non - motile state where the droplet appears to initiate division . the rate of deformation of the droplet interface is exponential and is modelled well by the linear stability analysis at small times . in some cases we see that this deformation is halted at some steady state ` dumbbell ' shape , where the defect is elongated into a line defect along the long axis of the droplet and can separate into 3 distinct defects ( of charge + 1,-1 and + 1 respectively ) shown in fig [ fig : deformed ] . at larger activity values or lower @xmath5 we see that the opposite sides of the immersed boundary are almost brought into contact . however , these simulations do not yet have the capability of modelling division of the droplet as this is not a part of the ib model , which is a shortcoming in comparison to phase field models . thus in these cases the simulation method breaks down . in most cases we test the deformation does stop and we observe further symmetry breaking of the internal polarisation field such that the droplet front enlarges and the droplet swims with the 3-defect configuration . however we do not characterise these dynamics here as they are complex and depend strongly on the value and sign of the liquid crystal coupling parameter @xmath74 . and surface tension @xmath153 at 3 different times in the simulation ( corresponding to @xmath154 , @xmath155 and @xmath156 respectively in simulation units ) using the same format as the inset in fig [ fig : shapevel ] . all other parameters are the same used to obtain fig [ fig : shapevel ] . ] in principle one could impose conditions that would allow division of the boundary , such that there were two separate immersed boundaries in the simulation . to do this , one would need to calculate the separation between non - neighbouring boundary points @xmath157 and @xmath158 , and once they were within a certain distance ( say @xmath36 ) separate the boundary into 2 boundaries one from @xmath159 the other from @xmath160 . it is well known that active systems can display turbulent like behaviour at zero reynolds number when the the equilibrium relaxation time is long compared to the timescale of the active dynamics . in terms of activity , the threshold for this behaviour is proportional to @xmath161 where @xmath162 is the one frank elastic constant and @xmath163 is the system size @xcite . these turbulent - like states have been observed experimentally in bacterial suspensions and reconstituted cytoskeletal networks @xcite where the active flows can spontaneously create and annihilate polarisation defects in pairs and generate vortices in the flow . we observe the onset of such dynamics in an active polar fluid droplet by reducing the elastic constant in our simulations to @xmath164 . we find that due to the strong anchoring condition at the interface , and relatively high surface tension the polarisation field remains approximately fixed at the boundary . however , reducing the constant @xmath162 reduces the equilibrium interaction between these filaments and those near the droplet centre which show dynamics that are only weakly coupled to the interface . this transition to turbulent motion is easily observed if we consider the case of contractile activity which for low activity values stabilises the defect at the droplet centre . snapshots of the dynamics are displayed in fig [ fig : turbulence ] . we see that due to the low @xmath162 value the stationary configuration becomes unstable but the flows are not large enough to significantly deform the drop . the result is unsteady dynamics of the polarisation field in the centre of the drop that are mostly decoupled from the interface dynamics . this results in stretching and distortion of the original @xmath165 defect into a line and then further pairwise creation of @xmath166 defects ( shown in fig [ fig : turbulence ] by the points where four black stripes intersect ) . as we scale the droplet size up , or decrease @xmath162 , one would expect that these turbulent dynamics would become faster as the effective energy cost of deformation becomes smaller relative to the activity . note that due to the polar nature of the system modelled here , only defects of integer topological charge can be generated , so these have fundamentally different dynamics to the half integer defects observed in active nematics . , surface tension @xmath167 and @xmath164 at 4 different times in the simulation ( corresponding to @xmath16825 , 75 , 125 and 250 respectively in simulation units ) . the shading represents the schlieren texture as used in fig [ fig : deformed ] . the blue arrows show the flow field in the fluid ( averaged over neighbouring points for clarity ) , while the blue line shows the droplet interface . we also use a finer mesh than that used to obtain fig [ fig : deformed ] , with @xmath169 and @xmath170 in order to resolve the defect dynamics more clearly . ] this method can be extended to a full 3d simulation , but the steps required to do so go beyond the scope of this paper . the equations related to the fixed fluid grid , such as the fluid flow equations and polarisation dynamics equations remain unchanged in 3d and the same combination of fast fourier transform and runge - kutta methods ( see methods ) can be utilised to solve for the dynamics of these quantities . however , the fluid - fluid interface is a 2d surface in the generic 3d case , and so this requires a different representation from the 1d lagrangian mesh used here . a direct extension would be to represent the fluid interface as a 2d triangular mesh . then the boundary points will evolve in exactly the same way as before . however , rather than removing single points from the interface ( as outlined in methods ) one would have to remove individual triangles below a threshold area and replace the mass contained within that triangular area . similarly , triangles with excessive eccentricity or area should be split along their short axis into two triangles to ensure stability . in general one could use any standard re - meshing algorithm ensuring that the concentration on the interface is numerically conserved and the spatial distribution is unchanged . such a mesh has been used to model surfactant coated drops in @xcite . an alternative method is to only model the immersed boundary points implicitly via the cartesian fluid mesh ( known as the immersed interface method @xcite ) . then one has to explicitly consider the stress jump in the normal direction at the interface in order to impose the boundary force @xcite . this normal force is then imparted on the cartesian mesh within the interfacial region ( defined by the numerical delta function range ) . similarly , the boundary region and associated level - set function are advected due to flow normal to the interfacial region . thus the approach is closely related except that the boundary mesh is implicitly defined rather than an explicit lagrangian mesh . we expect that the dynamics reported in this paper will be representative of what would be seen in 3d . in @xcite a linear stability analysis of these two systems predicts the same qualitative symmetry breaking behaviour despite the flow being quantitatively changed by the geometry . the main qualitative difference one would expect to observe is a deformation of the droplets with an active interface in 3-dimensions , and an investigation of how this deformation couples to concentration dynamics would be an interesting investigation for future work . in this paper we have detailed the numerical method that we have used to simulate the evolution of active droplets in 2d . we have used an immersed boundary ( ib ) method to model the fluid boundary as a lagrangian mesh of points . this allows us to explicitly simulate a concentration of isotropic active contractile particles on the interface that alter the surface tension . further , we use the immersed boundary to define a level - set function which defines the inside and outside of the drop . using this we can define an active polar liquid crystal phase in the droplet interior with anchoring at the droplet interface . our results reveal steady state migrating states in both types of active droplet . in all cases we investigate the steady state droplet velocity , which scales non - linearly with activity , levelling off at larger activity values . this means there is a generic reduction in swimming efficiency as activity is increased in these systems as observed for the flow rate in other active droplet and bulk systems @xcite . the first system we consider is a contractile isotropic active fluid on the droplet interface and observe that without any binding and unbinding the only observed active dynamic steady state is a swimming droplet with one peak in concentration . we see that for small binding rates the steady state remains unchanged but reduces the droplet velocity . at higher binding rates feedback from the fluid flow stabilises a second concentration peak and the droplet can reach a stationary two - peak configuration for large enough activity . this shows how a simple feedback mechanism can stabilise more complex behaviour , since at low binding rates if two peaks are formed in the concentration they always attract and coalesce to form a single peak at the droplet rear . these higher order states show that this active interface shares much of its internal dynamics with models of thin layers of active fluids @xcite , and so provides a useful insight into how such a system evolves immersed in an external fluid . thus , it is a useful step in modelling the dynamics of the actin cortex in cells which many active thin film models are based on . secondly , we simulate an active polar fluid droplet with strong anchoring at the interface . we see that if when we take the frank elastic constant to be significant ( or equivalently a small droplet ) we see transitions to highly ordered dynamic states , including motion , rotation and symmetric deformation . due to the geometry of the droplets and the polar anchoring condition , we find strikingly different behaviour for the contractile and extensile regimes , with extensile activity promoting translational symmetry breaking and contractile promoting symmetric modes of deformation . using a smaller frank constant ( equivalent to a larger droplet ) we observe the onset of turbulent dynamics , discussed in depth for active nematics in @xcite . due to the confinement inside the droplet the dynamics are still constrained and we see that the number of defects in the polarisation field rarely exceeds 3 for the range of activities simulated here . however , the dynamics are unsteady due to the slow relaxation of the molecular field compared to the active flow timescale . the results discussed here are consistent with previous studies of other active fluid droplet systems @xcite . additionally , given the increase in experimental progress in producing stable active gels and fluids in vitro @xcite , and confining such systems to droplets and vesicles @xcite it is possible that these simplified models can be tested experimentally . in particular the active fluid interface model may be achievable by coating a fluid droplet or the inside of a vesicle with a stable isotropic crosslinked actin layer . active contraction of the layer could be introduced by trapping myosin motor complexes inside the droplet / vesicle using existing microfluidic techniques . such motor complexes may have different binding kinetics than the linear case assumed here , however one would expect to observe qualitatively similar behaviour . a useful extension of these simulations that is beyond the scope of the work we present here would be to simulate these systems in 3 dimensions . from our analysis in @xcite we would expect much of the observed dynamics to remain qualitatively the same in 3d , with the exception that we would expect deformation of the droplets under surface tension gradients . in particular it would be interesting to simulate the droplet deformation when we predict steady two peak solutions in the concentration of an active isotropic fluid on the droplet interface ( see fig [ fig : bulkvel ] ) which could equally manifest as a ring or two poles of higher concentration in this geometry . the simulations presented here can also be used to model interactions with external boundaries or elastic solids in a self - consistent way through the ib method . similarly , one could investigate the dynamics of multiple active droplets simply by adding more lagrangian boundary meshes . these boundaries interact via their coupling to the fluid flow , and hence one would be able to simulate confinement of these droplets without necessarily sacrificing the periodic solution method outlined in the methods section . a preliminary investigation into such external boundaries is discussed in @xcite . finally , the modular nature of the code means that these simulations are capable of simulating a coupled active interface and an internal active polar fluid . the dynamics of such a system will be even more complex and so the results presented here will be relevant to understanding those systems in future . similarly , this will be relevant also to systems of passive liquid crystal drops propelled by active surfactants as studied in @xcite . in conclusion , the results we present here demonstrate complex dynamics in confined active fluids with minimal degrees of freedom . these simple one - component models demonstrate steady state dynamics akin to those observed in cells , cell fragments and reconstituted cell components that are driven by the active cell cytoskeleton . the numerical method we outline in this paper , based on the immersed boundary method enables continuum level simulation of these systems and has a wide range of potential applications in this field . in this section we outline explicitly the numerical method used in the simulations and the parameter value ranges studied . the bulk fluid domain @xmath171 is defined on a staggered periodic cartesian grid of points with spacing @xmath172 in each direction . the 3 overlaid grids correspond to positions @xmath173 , @xmath174 , and @xmath175 where @xmath176 and the same for @xmath151 . the points @xmath177 and @xmath178 run from @xmath179 and @xmath180 to @xmath181 and @xmath182 ( where @xmath103 are the lengths of the simulation box in the @xmath134 and @xmath151 directions and @xmath183 is the total number of grid points in each direction ) . the interface @xmath0 is described by a staggered grid of lagrangian points @xmath184 where @xmath185 with @xmath186 . these points have corresponding cartesian coordinates @xmath187 and @xmath188 . the step size is defined as @xmath189 where @xmath190 is the initial boundary length and @xmath191 the initial number of boundary points . points can be added or removed from this mesh during the simulation , which will be outlined later in this section . fig [ fig : grid ] gives an outline of which quantities are defined on which grid . we integrate over time with step @xmath192 such that @xmath193 for @xmath194 . we will denote the time - step at which a quantity is defined by a superscript where necessary ( _ e.g. _ @xmath195 ) , if none is given then all quantities in that equation will be at the same arbitrary time - step @xmath95 . . generally , @xmath134-components of vectors are on the @xmath196 grid , @xmath151-components of vectors are on the @xmath197 grid and scalars are on the @xmath198 grid.,scaledwidth=60.0% ] staggered grids are commonly used for fluid simulations of this nature in order to achieve consistency between first and second order gradients without the need to use a 5-point stencil in 1d ( or 13-point in 2d ) . we define the gradient operator on the bulk fluid as @xmath199 so that if @xmath200 then the @xmath134-component of the function @xmath201 is defined on the grid @xmath202 and the @xmath151-component on @xmath203 . values of @xmath201 can then be interpolated to other lattices where required . the laplacian operator can then be defined as @xmath204 such that it is on the same grid as the function @xmath205 . similarly , the boundary derivative @xmath206 is defined on the @xmath207 lattice , such that @xmath208 . we use the 2d numerical delta function @xmath209 described in @xcite to couple boundary quantities to the bulk fluid and vice versa . this delta function is given by the product of two 1-dimensional functions @xmath210 : @xmath211}/h \right)}\phi{\left ( { \left [ y_j - y_k \right]}/h \right)}}{h^2}\end{aligned}\ ] ] where the function @xmath210 acts as a numerical analogue of the 1d analytical dirac delta function . it is desirable to use a function that is only non - zero for a few points , as this aids computational efficiency , hence a restriction regularly chosen is @xmath212 if @xmath213 or @xmath214 . in @xcite several constraints are imposed that one would expect from a delta function , namely that it be a continuous function in @xmath215 and : @xmath216}^2 & = \lambda\end{aligned}\ ] ] where @xmath217 is a constant . note that the condition in eq gives the identity @xmath218 but also ensures a symmetric distribution of @xmath210 . the derivation of @xmath210 is outlined in reference @xcite , here we just give the resulting function : @xmath219 from this it is straightforward to check that @xmath220 . it should be noted that this distribution maintains its properties for all real values of @xmath215 and is not restricted to integer values . + using this delta function one can distribute the force density on the boundary @xmath221 to the fluid . first , we discretise eq to give : @xmath222 where the tangent vector ( defined on the @xmath223 grid ) is given by @xmath224 the corresponding force density components in the fluid then can be defined as : @xmath225 where the @xmath226 and @xmath227 superscripts are used to denote components of the vector . note that the @xmath134 component is defined on the @xmath228 grid and the @xmath151 component on the @xmath229 , this is the same for the components of the velocity vector . we compute the velocity components @xmath230 and @xmath231 , and pressure @xmath232 by solving the incompressible stokes equation ( i.e. eq such that @xmath233 ) . to do this we employ a projection method where we first define the intermediate solutions @xmath234 and @xmath235 such that @xmath236 where @xmath237 is the sum of all the force densities acting on the fluid . first , we solve for @xmath234 ; utilising the periodicity of the lattice we can write any function @xmath238 in terms of a 2d discrete fourier transform ( dft ) : @xmath239 } \\ \label{dft2 } \hat{g}_{\alpha,\beta } & = \frac{1}{\sqrt{m_x m_y}}\sum_{i=0}^{m_x-1 } \sum_{j=0}^{m_y-1 } g_{i , j } \exp{\left [ -2\pi i{\left ( \frac{i\alpha}{m_x } + \frac{j\beta}{m_y } \right ) } \right ] } \,.\end{aligned}\ ] ] here we have used the notation @xmath240 to avoid confusion with the index notation . using eqs and we can re - write eqs and so that for each value of @xmath241 and @xmath242 : @xmath243 where @xmath244 using standard nag library functions @xcite to compute the fast fourier transforms ( ffts ) , we find the dfts of the components of @xmath245 . looping over all values of @xmath241 and @xmath242 we can use eqs and to assign values to the dft of @xmath246 . we then compute the inverse dft of @xmath247 to acquire values for @xmath246 . following this we calculate @xmath248 and compute its dft . then we can calculate @xmath249 similarly . finally , we relate these quantities back to the velocity and pressure as follows : @xmath250 this is a computationally efficient method for solving the stokes equation as each set of assignments only require @xmath251 operations . the time taken for the fft computations scales approximately as @xmath252 so this is the limiting complexity in this process . + once the bulk velocity is computed we can calculate the velocity of the interface using eq : @xmath253 where @xmath254 is the value of @xmath255 rounded down to the nearest integer and @xmath256 is the same but for @xmath257 . generally in ib methods this interface is updated using a simple forward first order euler scheme , _ i.e. _ @xmath258 . instead we use an explicit fourth order runge - kutta method as outlined later in this section . we find that this method improves the numerical stability of the algorithm compared to a forward euler update , particularly in the case where the boundary position is coupled to an internal polarisation field . we define a concentration of active particles @xmath259 on the boundary points @xmath223 which alters the surface tension @xmath260 according to eq . the concentration dynamics are then governed by eq which we discretise using the crank - nicholson scheme ( as in @xcite ) : @xmath261 the quantity @xmath262 contains the binding and unbinding rates for the concentration and is defined further below in eqs . and . we update the concentration @xmath22 after updating the boundary position @xmath263 so that @xmath264 is defined . then the set of eqs in can be represented by a periodic tridiagonal matrix : @xmath265 rearranging eq we can read off the expressions for @xmath266,@xmath267,@xmath268 , and @xmath269 : @xmath270 we solve this matrix directly using standard nag linear algebra functions @xcite . we test that the mass is conserved numerically by outputting the change in total mass in the simulation between time - steps and find that these values are randomly distributed around zero with maximum deviations on the order of machine precision . in order to numerically model the equations in the model section we need a numerical analogue for the level - set function @xmath44 . taking the gradient of both sides of eq we see that by definition : @xmath271 thus , as in @xcite , to find the numerical equivalent of the function @xmath44 we solve the following poisson equation : @xmath272}{\left| \partial_s { \boldsymbol{x}}_k \right|}\delta s \ , .\end{aligned}\ ] ] we again solve this equation using the same fft method used in to solve eqs , and . however , as we set the magnitude of the zero - wavenumber mode of the fourier transform to zero , this only defines @xmath44 up to a constant ( more exactly it defines @xmath44 so that its average value is zero ) . this can be corrected simply in the simulation by recording the value of @xmath44 at any point where @xmath273 and @xmath274 ( a point strictly outside the drop ) and then subtracting the value of @xmath44 at this point from the value of @xmath44 at all points in the simulation . this then results in a profile where all points outside the droplet are approximately equal to zero . however , as discussed in @xcite this can lead to numerical issues when dividing by @xmath44 , and so we redefine the level set function at all points as @xmath275 setting @xmath276 as a sufficiently small discrepancy to avoid noticeable ` leaking ' of the internal concentration . we can then define a concentration field that is contained within the droplet by the quantity @xmath47 . thus , in order to conserve mass and ensure that the concentration is contained within the droplet we can use the same equations as used in @xcite to describe a soluble surfactant . we update @xmath43 from eq using the following crank - nicholson scheme ( as used in @xcite ) : @xmath277 } & = - \frac{1}{2}{\boldsymbol{d}}_h \cdot { \left [ ( { \boldsymbol{v}}h\rho)^{n+1}_{i , j } + ( { \boldsymbol{v}}h\rho)^{n}_{i , j } \right ] } \notag\\ & + \frac{d_b}{2 } { \left [ { \boldsymbol{d}}_h\cdot{\left ( h{\boldsymbol{d}}_h\rho \right)}^{n+1}_{i , j } + { \boldsymbol{d}}_h\cdot{\left ( h{\boldsymbol{d}}_h\rho \right)}^{n}_{i , j } \right ] } \notag \\ \label{drhodts } & - \frac{1}{2}{\left ( \sum_{k=0}^{n-1}q_k^{n+1}\delta_h({\boldsymbol{x}}_{i , j } - { \boldsymbol{x}}_k^{n+1 } ) + \sum_{k=0}^{n-1}q_k^{n}\delta_h({\boldsymbol{x}}_{i , j } - { \boldsymbol{x}}_k^{n } ) \right)}\end{aligned}\ ] ] where @xmath278 is a central difference operator on the staggered grid ( as defined in @xcite ) . the exchange terms @xmath279 and @xmath280 are given by : @xmath281 as we update the boundary positions and concentration before this step , @xmath282 , @xmath283 and @xmath284 are already defined . analytically we know that the system should reach a stable steady state where @xmath285 inside the drop and @xmath286 on the interface . in these simulations however , the bulk concentration evaluated at the interface is @xmath287 , and so the equivalent steady state for this numerical system is given by : @xmath288 where the droplet shape and position is assumed fixed in this passive case and the angled brackets indicate the average value over the boundary such that : @xmath289 thus , we initialise the simulation with a concentration @xmath290 on the interface and @xmath291 everywhere except at @xmath292 where @xmath293 everywhere so that the concentration in the system is directly proportional to @xmath44 and is in the stable steady state configuration ( in the absence of activity or external flows ) . we define the polarisation field @xmath294 , @xmath295 and stress components @xmath296 from eqs - on the centred lattice @xmath297 and interpolate any gradients that are evaluated on the other lattices therein . first , we calculate the molecular field and the free energy density using from the identity @xmath298 : @xmath299}\nabla_{hc } h_{i , j}\\ \notag g^{\rm tot}_{i , j } & = \frac{k}{2}{\left [ { \left ( \nabla_{hc } p^{(x)}_{i , j } \right)}^2 + { \left ( \nabla_{hc } p^{(y)}_{i , j } \right)}^2 \right ] } + \frac{k}{4}c_b { \boldsymbol{p}}^2_{i , j}{\left ( { \boldsymbol{p}}^2_{i , j } - 2h_{i , j } \right)}\\ \label{freetot } & + \frac{w_1}{2}{\left [ { \left| \nabla_{hc}h_{i , j } \right| } + { \boldsymbol{p}}_{i , j}\cdot{\left ( \nabla_{hc}h_{i , j } \right ) } \right]}^2\end{aligned}\ ] ] where the central gradient operator @xmath300 is defined by @xmath301 . from these we calculate the components of the stress @xmath302 , and then finally the force density contribution from these stresses : @xmath303 as with the velocity field , the @xmath134-component of the force density is defined on the @xmath228 lattice and the @xmath151-component on the @xmath229 lattice . this force density is then added to the boundary force density in eq before calculation of the velocity . once the velocity is computed , the polarisation update equation can be computed from eq as : @xmath304 where @xmath305 } \notag \\ u^{(xx)}_{i , j } = & \partial_x v^{(x)}_{i-1/2,j } \notag \\ u_{i , j}^{(xy ) } = u_{i , j}^{(yx ) } = & \frac{1}{4}{\left [ { \left ( \partial_x v^{(y)}_{i , j-1/2 } + \partial_x v^{(y)}_{i , j+1/2 } + \partial_x v^{(y)}_{i-1,j-1/2 } + \partial_x v^{(y)}_{i-1,j+1/2 } \right ) } \right . \notag \\ & \left.+ { \left ( \partial_y v^{(x)}_{i-1/2,j } + \partial_y v^{(x)}_{i+1/2,j } + \partial_y v^{(x)}_{i-1/2,j-1 } + \partial_y v^{(x)}_{i+1/2,j-1 } \right ) } \right ] } \notag \\ u_{i , j}^{(yy ) } = & \partial_y v_{i , j-1/2}^{(y ) } \notag \,.\end{aligned}\ ] ] the polarisation is updated using the coupled fourth order runge - kutta method at the same time as the boundary is updated using eqs and , detailed in the `` algorithm overview '' section . as the immersed boundary in these simulations is the interface between two fluids , there is no resistance to neighbouring points getting too close or too far apart , as there would be in an elastic boundary . therefore , it is necessary to add and remove points from the simulation to keep the boundary well defined and maintain accuracy . we remove the point @xmath306 whenever @xmath307 and add a point between @xmath306 and @xmath308 when @xmath309 . see fig [ fig : addrem ] for a visual guide to this process . and * ( b ) * adding a point between @xmath306 and @xmath308 . circular points denote the boundary position nodes and crosses denote the concentration defined on the half - integer lattice.,scaledwidth=70.0% ] to remove a point @xmath306 we reassign @xmath310 for all @xmath311 , then reduce the total number of boundary points to @xmath312 . conservation of mass then dictates : @xmath313 and again @xmath314 for all @xmath315 . the dashed quantities all denote the new values after removal of the boundary point . + to add a point between @xmath306 and @xmath308 we reassign @xmath316 for all @xmath317 then add the point @xmath318 with concentration @xmath319 to conserve mass ( @xmath320 remains unchanged ) . we notice that when large flow fields are generated in the simulations , inaccuracies in the boundary positions can accumulate that need to be addressed . firstly , we observe ` buckling ' of neighbouring boundary points in areas where the boundary flow converges , which results in points not being removed from the simulation even though they are overlapping . thus , we also add a condition for the removal of a boundary point if @xmath321 where , after some testing of different boundary shapes , we arrived at @xmath322 , where we have defined @xmath323 to be the distance of the boundary point from the droplet centre of mass . therefore any excessive second order gradients in the boundary position are effectively smoothed out . however , we choose the limit so that smooth boundary deformations are still permitted , such as those observed in our results for an active polar fluid drop . + we observe that the droplets can lose significant area over the whole simulation time , despite the fluid incompressibility . this effect is observed in the fluid immersed boundary simulations presented in @xcite and @xcite but is very slight and gradual ( on the order of @xmath324 area loss over a whole simulation ) as the size of the flows is imposed externally . it is possible to observe much more rapid area loss in our simulations due to the continual addition and removal of interface points , and so to prevent this , and be consistent with the incompressibility condition , we calculate the area at each time step as @xmath325 we subtract this from the initial droplet area at time @xmath326 to get @xmath327 . in order to keep the shape fixed , we then alter the values of @xmath328 with a constant in the normal direction : @xmath329 . we then iterate this process , recalculating the area after this alteration and updating the value of @xmath330 accordingly until the area is conserved within a certain tolerance . we choose this tolerance to be @xmath331 and find that this step rarely takes more than 2 iterations , and in an individual step @xmath330 is at most the order of @xmath332 , which is several orders of magnitude below the average velocity of the points in these simulations . thus this area conservation step makes little difference to the evolution of the droplet . the final simulation code is modular in nature , so that we can turn features on or off . the bare code contains the simulation of a passive fluid drop in a periodic domain , then one can choose whether to add an active boundary and an active polar fluid separately . on initialisation we define the fluid boundary points @xmath328 , the concentration on the boundary @xmath333 ( if included ) and the filament polarisation @xmath294 ( if included ) . the order of execution of the algorithm is given below : 1 . boundary forces calculated according to eq . bulk forces calculated as outlined in eqs - ( if active polar fluid is included ) . 3 . solve the stokes equation following eqs to . calculate boundary velocity using eqs and . simultaneously update boundary and polarisation field using an explicit fourth - order runge - kutta method ( outlined below ) . correct the boundary position for area conservation as outlined in the previous section . update concentration at the interface by solving the system in eq update the bulk concentration @xmath43 using the crank - nicholson scheme of eq add or remove fluid boundary points as outlined by fig [ fig : addrem ] . in step 5 we apply a coupled fourth order runge - kutta method where we update the boundary and the polarisation field . at the start of the update we store @xmath334 and @xmath335 , then update them for a half time - step such that @xmath336 and @xmath337 where @xmath338 is the vector given by the right hand sides of eqs and . we then repeat steps 1 - 4 for this new polarisation and boundary position . repeating this process we can define : @xmath339 where @xmath340 and @xmath341 are again the right hand sides of eqs and evaluated at the @xmath342 and @xmath343 time - steps respectively . we also define @xmath344 similarly as this vector evaluated for the @xmath345 values of the polarisation and boundary condition . in between each fractional update we repeat steps 1 - 4 . then finally the updated polarisation and boundary are given by : @xmath346 we run the simulation , iterating over steps 1 - 9 usually @xmath347 times , unless we deem that longer is required . a full list of the parameter value ranges is given below : .fluid simulation parameters [ cols="^,^,^",options="header " , ] we acknowledge the epsrc for funding this work , grant reference ep- k503149 - 1 . we also thank suzanne fielding for useful discussion and insight , and davide marenduzzo and elsen tjhung for sharing the simulation source code used in @xcite . [ [ s1_video ] ] s1 video . + + + + + + + + + * ( a ) . * steady state swimming droplet from figure [ fig : bulkvel](a ) . * ( b ) . * ` wandering ' droplet from figure [ fig : bulkvel](b ) . * ( c ) . * steady symmetric droplet from figure [ fig : bulkvel](c ) . colour gradients show concentration on the interface and in the bulk , from black ( low ) to yellow ( high ) . vladescu i d , marsden ej , schwarz - linek j , martinez va , arlt j , morozov an , et al . . phys rev lett . http://dx.doi.org/10.1103/physrevlett.113.268101[doi : 10.1103/physrevlett.113.268101 . ] wottawah f , schinkinger s , lincoln b , ananthakrishnan r , romeyke m , guck j , et al . . phys rev lett . http://dx.doi.org/10.1103/physrevlett.94.098103[doi : 10.1103/physrevlett.94.098103 .

we presentco observations of 45 giant molecular clouds in m33 made with the bima array . the observations have a linear resolution of 20 pc , sufficient to measure the sizes of most gmcs in the sample . we place upper limits on the specific angular momentum of the gmcs and find the observed values to be nearly an order of magnitude below the values predicted from simple formation mechanisms . the velocity gradients across neighboring , high - mass gmcs appear preferentially aligned on scales less than 500 pc . if the clouds are rotating , 40% are counter - rotating with respect to the galaxy . gmcs require a braking mechanism if they form from the large scale radial accumulation of gas . these observations suggest that molecular clouds form locally out of atomic gas with significant braking by magnetic fields to dissipate the angular momentum imparted by galactic shear . the observed gmcs share basic properties with those found in the galaxy such as similar masses , sizes , and linewidths as well as a constant surface density of 120 pc . the size linewidth relationship follows , consistent with that found in the galaxy . the cloud virial masses imply that the co - to - h conversion factor has a value of and does not change significantly over the disk of m33 despite a change of 0.8 dex in the metallicity .

astro-ph0307322

many theories address molecular cloud formation , but few observations can distinguish between them . as a result , there is little consensus on which process dominates ( * ? ? ? * and references therein ) . one avenue for testing these theories is to study molecular clouds in a wide range of environments and to determine which aspects of the environment set the cloud properties . we have adopted this approach in a study of the local group galaxy m33 . * paper i ) completed a survey of the entire optical disk of m33 in @xmath7 with sufficient sensitivity to detect all molecular clouds more massive than @xmath8 . this survey represents the first _ flux - limited _ survey of giant molecular clouds ( gmcs ) in any spiral galaxy . while there are complete surveys of molecular gas in the milky way , blending of emission along the line of sight make it impossible to obtain a complete catalog . the only other flux - limited survey of a molecular cloud population in another galaxy was completed by the nanten group studying the large magellenic cloud @xcite . @xcite used the ovro interferometer to study m33 gmcs at high resolution ( 20 pc ) , but their observations were limited to clouds in the central 1.5 kpc of the galaxy that are associated with optical extinction and bright single - dish co detections . this paper is a high resolution follow - up to the survey of paper i , targeting fields known to contain bright gmcs over a large range of galactic radii . the @xmath9 observations presented in paper i have sufficient linear resolution ( 50 pc ) to resolve the emission into molecular clouds , but not to measure the cloud sizes . the higher resolution observations presented here ( 20 pc ) resolve most of the gmcs , and provide additional information about their masses , angular momenta and morphologies . these data show that the gmcs in m33 are similar to those found in the milky way . in addition , the data constrain cloud formation theories , and they suggest that the co - to - h@xmath5 conversion factor is constant across the galaxy . we observed 17 fields in @xmath0co @xmath10 using the bima array @xcite in fall 2000 and spring 2001 . we chose fields that contain the highest mass gmcs cataloged in paper i ( figure [ fields ] ) . the observations were made in the c - array configuration yielding a synthesized beam fwhm of @xmath11 . we observed the co line in the upper sideband with a velocity resolution of 1.016 km s@xmath12 . system temperatures typically ranged between 420 k and 730 k. table [ table1 ] lists the field center , synthesized beam size , integration time , and rms noise for each field . we tried to obtain @xmath13 jy beam@xmath12 rms noise in each field while maintaining good @xmath14 coverage . three additional fields ( fields 18 through 20 in table [ table1 ] ) were observed in conjunction with the survey presented in paper i and are discussed in more detail in that paper . because of the longer integration times and smaller synthesized beam sizes , the high resolution observations can detect lower mass clouds than the survey in paper i. cccccc field & @xmath15 & @xmath16 & beam size & source time & @xmath17 + name & ( 2000 ) & ( 2000 ) & ( ) & ( hours ) & ( jy beam@xmath12 ) + 1 & @xmath18 127 & 30 42 280 & 5.3 @xmath19 5.1 & 8.22 & 0.127 + 2 & @xmath18 876 & 30 39 878 & 5.8 @xmath19 5.4 & 0.872 & 0.199 + 3 & @xmath18 657 & 30 47 508 & 5.7 @xmath19 5.4 & 0.872 & 0.200 + 4 & @xmath20 235 & 30 25 398 & 5.8 @xmath19 4.9 & 0.255 & 0.205 + 5 & @xmath20 410 & 30 46 743 & 5.8 @xmath19 5.2 & 1.34 & 0.159 + 6 & @xmath18 382 & 30 40 381 & 5.8 @xmath19 5.2 & 1.34 & 0.160 + 7 & @xmath20 501 & 30 33 583 & 5.8 @xmath19 5.2 & 1.34 & 0.161 + 8 & @xmath18 107 & 30 42 280 & 6.4 @xmath19 5.4 & 1.88 & 0.178 + 9 & @xmath20 425 & 30 33 661 & 5.7 @xmath19 5.0 & 4.44 & 0.0912 + 10 & @xmath18 220 & 30 38 311 & 6.9 @xmath19 4.9 & 2.34 & 0.212 + 11 & @xmath18 135 & 30 33 442 & 6.5 @xmath19 6.0 & 1.33 & 0.262 + 12 & @xmath20 519 & 30 39 228 & 6.5 @xmath19 6.0 & 1.33 & 0.261 + 13 & @xmath18 342 & 30 46 167 & 6.6 @xmath19 6.3 & 1.45 & 0.221 + 14 & @xmath20 596 & 30 49 108 & 6.5 @xmath19 6.0 & 2.20 & 0.213 + 15 & @xmath18 164 & 30 39 181 & 6.7 @xmath19 6.2 & 1.92 & 0.196 + 16 & @xmath18 105 & 30 36 100 & 6.5 @xmath19 5.8 & 3.09 & 0.234 + 17 & @xmath18 950 & 30 49 120 & 6.7 @xmath19 6.1 & 16.0 & 0.0832 + 18 & @xmath20 583 & 30 41 699 & 6.6 @xmath19 6.6 & 4.00 & 0.155 + 19 & @xmath20 425 & 30 39 130 & 6.6 @xmath19 6.5 & 4.00 & 0.153 + 20 & @xmath20 522 & 30 37 190 & 6.6 @xmath19 6.5 & 4.00 & 0.155 + our data reduction followed the algorithm developed for the bima survey of nearby galaxies @xcite . only the c - array data were used in producing the final data cubes because of their significantly lower noise levels . observations of uranus ( fall 2000 ) and mars ( spring 2001 ) were used to establish the flux of the phase calibrator 0136 + 478 ( 3.6 jy in fall 2000 , 2.7 jy in spring 2001 ) . bootstrapping the quasar flux to the m33 data established the flux calibration . binning velocity channels set the final velocity resolution at 2.00 km s@xmath12 . we used uniform weighting of the @xmath14 data in the inversion , choosing a grid spacing that resulted in @xmath21 pixels for the final maps , and we cleaned the cube using a hgbom algorithm after correcting for the primary beam attenuation . since interferometers filter out zero - spacing flux from the observed object , complete flux recovery requires additional observations using a single dish telescope . to measure the total flux from the gmc fields , we observed 18 fields over the course of three nights using the university of arizona steward observatory 12-m telescope at kitt peak . the selected fields for the 12-m observations are displayed in figure [ fields ] . to check for faint emission , each field was observed to attain a signal - to - noise ratio at the peak of the co line of at least 10 . every two hours , jupiter or saturn was observed to optimize the pointing and focus of the telescope . during sunset , the pointing and focus checks were done before observing each new field to compensate for rapid variations in the thermal stresses on the telescope . both polarizations were observed with the 500-khz and 1-mhz filter banks in parallel mode . the median system temperature was 350 k for the observations and the median column of water was 0.7 mm . the observations used absolute off positions which targeted low - significance detections in the d - array survey ( see paper i ) . these observations help to establish the completeness limit in the survey by determining the fraction of real detections of low significance . the positions of the off fields were selected to be well separated in velocity from the strong molecular cloud emission in the on position . emission in the off fields appears as absorption features in the resulting spectra . only two of the off fields contained emission at the expected velocities . we reduced the data by flagging bad channels in every spectrum , combining the polarizations to reduce the noise , and fitting a quadratic baseline . the vane calibration of the antenna temperature was found to be accurate to better than 10% using continuum observations of jupiter and line observations of dr21(oh ) . all spectra were observed to an rms noise level of at most 10 mk in a 2.6 km s@xmath12 channel , equivalent to a molecular gas surface density of of 0.1 @xmath22 pc@xmath3 . finally , we observed a 5 @xmath19 5 pointing mosaic centered on the most massive cloud complex in the galaxy ( field 17 in table [ table1 ] ) to map all flux associated with this cloud . the single dish map was combined with both the c and d array bima data using miriad s imcomb routine . the difference between the fully - recovered data cube and the cube generated solely from the interferometer data shows the location and spatial structure of the emission that is detected only by the 12-m telescope . by comparing the interferometer and single dish data , we estimate the amount of flux that has been resolved out by the interferometer or not selected by our cloud identification techniques ( [ masking ] ) . we compare each of the 18 single dish spectra to a representative spectrum generated from the interferometer data by averaging together the spectra in the data cube and correcting for the beam profile of the 12-m . to improve the signal to noise in the interferometer spectrum , only pixels with significant ( @xmath23 ) co emssion are included in the average . both the c - array observations and the d - array observations discussed in paper i recover less than 100% of the flux as expected . the median ratios of the interferometer and single dish integrated intensities ( @xmath24 ) are : @xmath25 because the distribution of this ratio is asymmetric about its median , we quote the positive and negative dispersions separately . the 12-m observations indicate that a significant fraction of emission is resolved out or masked out ( see [ masking ] ) . however , comparing the peak antenna temperatures in the spectra shows that : @xmath26 while the widths of the interferometer lines are narrower : @xmath27 three effects can lead to a discrepancy between the interferometer and single dish data : ( 1 ) calibration errors , ( 2 ) extended co emission resolved out by the interferometer , and ( 3 ) limited sensitivity or masking effects in the interferometer data . calibration discrepancies would appear as a relative scaling of the observed spectra by a constant factor . however , the ratios of the peak temperatures and linewidths indicate better flux recovery in the middle of the line . we conclude that scaling alone can not not be the source of flux loss . observations of the off positions using the 12-m telescope indicate that there is no extended co emission on kiloparsec scales with sufficient surface brightness to account for the discrepancy between the interferometer and the single dish observations . each of the 18 single dish spectra have a @xmath28 noise level of 0.07 k km s@xmath12 ( 0.3 @xmath22 pc@xmath3 ) ; however , accounting for the observed difference requires a surface brightness of 1.0 k km s@xmath12 ( 4.3 @xmath22 pc@xmath3 ) . to search for even fainter emission , we averaged together the 16 empty off positions , shifting the velocities to align any faint emission at the velocity . no line is present in the resulting spectrum , which has a @xmath28 surface brightness sensitivity of 0.019 k km s@xmath12 ( @xmath29 pc@xmath3 ) . since the off positions are separated by projected distances of roughly a kiloparsec , we conclude that a co emission component on this scale can account for no more than 2% of the discrepancy between the interferometer and the single dish fluxes . the extra emission detected in the single - dish spectra is likely from low mass clouds surrounding each gmc lying below our clipping threshold . figure [ diffmap ] compares the map of field 17 generated from interferometer data alone with the map generated by merging interferometer data with the single dish map . the greyscale image is the fully recovered map and the contours represent the _ additional _ emission seen only in the fully recovered map . the relative levels of the contours to the greyscale are indicated at the bottom of the scale bar , showing the additional flux recovered is a very low surface brightness component . taking the fourier transform of the added emission shows that @xmath30 of the added flux would be detected by the interferometer with more integration time . the spatial filtering of the interferometer occurs only at @xmath14 distances of @xmath31 and does not account for the discrepancy between the two telescopes . most of the added emission is excluded from the interferometer map simply by virtue of being below the significance threshold for our masking methods . also plotted in figure [ diffmap ] are the locations of low mass clouds seen only in the c - array data , where the synthesized beam is well matched to the cloud sizes . two of these clouds appear in the distribution of diffuse emission suggesting clumpy structure within this envelope . the diffuse emission seems principally associated with the gmcs , and a discussion of what this observation implies for cloud formation appears in [ finis ] . we conclude that the interferometer observations detect the central clouds of complexes in the galaxy containing high mass gmcs ( @xmath32 , paper i ) . we defined clouds in the position - velocity interferometer data cubes as connected regions in three dimensions containing more than 50 pixels , each with a value greater than @xmath33 . our criteria guarantee that _ at least _ five statistically independent regions are included in each cloud because interferometer data cubes are oversampled by a factor of 3 in each spatial direction , though the velocity channels are nearly independent . for five joint @xmath34 detections , the likelihood of false detection is @xmath35 in these data cubes which have @xmath36 independent elements . analysis of the noise in the data indicates that the distribution is sufficiently gaussian to justify such a treatment . we simulated @xmath37 data sets mimicking the observation conditions , correlator configuration , and baselines of the array . the cloud extraction program found no false detections for these selection criteria . this algorithm extracts 45 molecular clouds from the data cubes , including all clouds in these fields cataloged in paper i. to account for low amplitude signal at the edges of the clouds , the regions were expanded by 1 pixel in both the position and the velocity directions . figure [ maps ] shows maps of the individual clouds . five of the 45 clouds are not completely recovered since they lie near the half power point of the primary beam . because their peaks are included in the maps , we estimate that the truncation will not increase their masses and areas by more than a factor of 2 . the observed parameters of the 45 clouds are listed in table [ cloudprop ] . the analysis uses a distance to m33 of 850 kpc @xcite . ccccccccccccccccc & @xmath38 & alternative & v@xmath39 & @xmath40 & @xmath41 & @xmath42 & @xmath43 & @xmath44 & @xmath45 & @xmath46 & @xmath47 + number & @xmath48 & names & ( km s@xmath12 ) & ( km s@xmath12 ) & ( pc ) & ( @xmath49 m@xmath50 ) & ( @xmath49 m@xmath50 ) & ( 0.01 km s@xmath12 pc@xmath12 ) & ( @xmath51 ) & pc km s@xmath12 & pc km s@xmath12 + 1 & @xmath52 & eprb19,ws1,2,4,5 & @xmath53 & 7.7 & 27 . & 31 . & 26 . & 8.38 & @xmath54 & 19.2 & 25.3 + 2 & @xmath55 & eprb132,ws3 & @xmath56 & 5.5 & & & 6.7 & 2.29 & @xmath57 & & 2.14 + 3 & @xmath58 & ws33,34 & @xmath59 & 6.7 & & & 3.6 & 9.58 & @xmath60 & & 6.18 + 4 & @xmath61 & eprb31,ws29 & @xmath62 & 6.1 & 22 . & 16 . & 4.2 & 10.1 & @xmath63 & 15.3 & 47.4 + 5 & @xmath64 & eprb31,ws29 & @xmath65 & 11 . & 48 . & 100 & 14 . & 13.6 & @xmath66 & 93.9 & 103 . + 6 & @xmath67 & eprb30,ws30 & @xmath68 & 4.9 & & & 2.9 & 11.9 & @xmath69 & & 10.3 + 7 & @xmath70 & ws10 & @xmath71 & 8.4 & 13 . & 17 . & 9.9 & 8.36 & @xmath72 & 4.00 & 13.4 + 8 & @xmath73 & eprb75,ws6 & @xmath74 & 4.9 & & & 2.9 & 3.49 & @xmath75 & & 2.44 + 9 & @xmath76 & eprb75,ws8,9 & @xmath77 & 4.4 & 13 . & 4.6 & 5.3 & 2.20 & @xmath78 & 1.06 & 4.22 + 10 & @xmath79 & eprb18,ws27,28 & @xmath80 & 6.5 & 33 . & 27 . & 12 . & 1.90 & @xmath81 & 6.43 & 8.87 + 11 & @xmath82 & eprb3,ws21,22,23 ( ? ) & @xmath83 & 9.6 & 32 . & 56 . & 41 . & 9.53 & @xmath84 & 30.0 & 46.0 + 12 & @xmath85 & eprb16 & @xmath53 & 7.6 & & & 6.7 & 10.6 & @xmath86 & & 12.8 + 13 & @xmath87 & & @xmath88 & 7.2 & 22 . & 22 . & 11 . & 3.29 & @xmath89 & 5.06 & 4.56 + 14 & @xmath90 & eprb75,ws7 & @xmath91 & 4.2 & 6.7 & 2.3 & 3.7 & 7.70 & @xmath92 & 1.04 & 12.6 + 15 & @xmath93 & eprb4,ws32 & @xmath94 & 10 . & 31 . & 61 . & 9.57 & @xmath95 & 27.5 & 45.6 + 16 & @xmath96 & & @xmath97 & 5.5 & 13 . & 7.3 & 7.0 & 1.25 & @xmath98 & 0.617 & 1.64 + 17 & @xmath99 & eprb46,ws16,17,18 & @xmath100 & 8.2 & 3.8 & 4.8 & 12 . & 14.0 & @xmath101 & 0.624 & 16.2 + 18 & @xmath102 & & @xmath103 & 6.0 & & & 3.4 & 4.70 & @xmath104 & & 3.08 + 19 & @xmath105 & eprb20 & @xmath106 & 4.8 & 49 . & 22 . & 23 . & 0.685 & @xmath107 & 4.99 & 0.0282 + 20 & @xmath108 & eprb7 & @xmath109 & 6.0 & 14 . & 9.4 & 15 . & 8.74 & @xmath110 & 4.96 & 10.5 + 21 & @xmath111 & eprb22 & @xmath112 & 5.8 & 42 . & 27 . & 12 . & 4.40 & @xmath113 & 24.2 & 20.4 + 22 & @xmath114 & eprb58 & @xmath83 & 4.6 & 10 . & 4.0 & 5.5 & 6.75 & @xmath115 & 2.04 & 10.4 + 23 & @xmath116 & eprb72 & @xmath117 & 5.1 & & & 6.2 & 7.77 & @xmath118 & & 11.2 + 24 & @xmath119 & eprb10 & @xmath120 & 4.2 & 14 . & 4.7 & 6.7 & 4.05 & @xmath121 & 2.47 & 4.75 + 25 & @xmath122 & eprb10 & @xmath123 & 6.7 & 18 . & 15 . & 15 . & 7.24 & @xmath124 & 7.26 & 13.9 + 26 & @xmath125 & eprb15 & @xmath126 & 7.2 & 7.8 & 7.7 & 8.5 & 11.9 & @xmath127 & 2.24 & 12.0 + 27 & @xmath128 & eprb15 & @xmath129 & 6.0 & 16 . & 11 . & 5.5 & 1.54 & @xmath130 & 1.16 & 4.96 + 28 & @xmath131 & eprb6 & @xmath132 & 6.5 & 32 . & 25 . & 27 . & 3.49 & @xmath133 & 10.7 & 6.92 + 29 & @xmath134 & eprb21 & @xmath135 & 7.4 & 38 . & 40 . & 21 . & 8.91 & @xmath136 & 40.0 & 47.0 + 30 & @xmath137 & eprb15 & @xmath138 & 7.3 & 26 . & 26 . & 16 . & 9.44 & @xmath139 & 19.3 & 33.9 + 31 & @xmath140 & eprb2 & @xmath141 & 6.5 & 29 . & 23 . & 32 . & 3.58 & @xmath142 & 9.06 & 17.6 + 32 & @xmath143 & & @xmath144 & 5.2 & & & 5.8 & 4.76 & @xmath145 & & 4.31 + 33 & @xmath146 & eprb74 & @xmath147 & 8.5 & 28 . & 38 . & 8.0 & 5.35 & @xmath148 & 12.4 & 18.7 + 34 & @xmath149 & & @xmath150 & 7.4 & 8.8 & 9.2 & 7.5 & 13.8 & @xmath151 & 3.29 & 16.7 + 35 & @xmath152 & & @xmath147 & 5.4 & 7.9 & 4.3 & 6.2 & 9.35 & @xmath153 & 1.77 & 19.1 + 36 & @xmath154 & & @xmath150 & 6.7 & 47 . & 41 . & 24 . & 3.96 & @xmath155 & 26.9 & 45.8 + 37 & @xmath156 & eprb1 & @xmath157 & 9.6 & 38 . & 67 . & 78 . & 7.37 & @xmath84 & 33.3 & 27.6 + 38 & @xmath158 & eprb108 & @xmath159 & 6.8 & 16 . & 14 . & 7.1 & 9.67 & @xmath160 & 8.03 & 14.6 + 39 & @xmath161 & eprb5 & @xmath162 & 9.0 & 21 . & 32 . & 32 . & 7.47 & @xmath163 & 9.61 & 29.3 + 40 & @xmath164 & & @xmath157 & 6.2 & 13 . & 9.3 & 9.6 & 7.93 & @xmath165 & 3.85 & 12.8 + 41 & @xmath166 & eprb8 & @xmath167 & 8.7 & 19 . & 27 . & 28 . & 6.97 & @xmath168 & 7.66 & 11.4 + 42 & @xmath169 & eprb9 & @xmath170 & 9.3 & 23 . & 38 . & 30 . & 8.22 & @xmath171 & 13.7 & 28.3 + 43 & @xmath172 & & @xmath173 & 6.3 & & & 4.6 & 9.95 & @xmath174 & & 7.18 + 44 & @xmath175 & eprb11 & @xmath176 & 6.4 & 18 . & 14 . & 6.7 & 10.8 & @xmath177 & 10.7 & 23.4 + 45 & @xmath178 & eprb11 & @xmath179 & 9.2 & 23 . & 38 . & 25 . & 8.56 & @xmath86 & 14.4 & 19.2 + we measure the gmc masses using two methods : ( 1 ) assuming a constant co - to - h@xmath5 conversion factor and ( 2 ) applying the virial theorem . the co - to - h@xmath5 conversion factor ( @xmath180 , * ? ? ? * ) converts between integrated co line intensity ( @xmath24 ) and molecular hydrogen column density ( @xmath181 ) . we use a conversion factor of @xmath182 @xcite . in converting the column density to a cloud mass , we include the mass of helium with a number fraction of 9% relative to hydrogen nuclei . this gives the luminous mass of the cloud : @xmath183 where @xmath184 is the projected surface area of the molecular cloud . our observations typically have @xmath185 ( 4.3 @xmath186 ) . errors in measuring masses using the @xmath180 factor arise not only from the uncertainty in the @xmath180 factor but also from the flux calibration and the noise in the data . determining the true masses of entire gmcs is further complicated by the presence of low surface brightness emission around the central clouds . for the 36 resolved clouds , we can also compute the mass using the virial theorem . the virial mass requires a measurement of the velocity width of the gmc and the cloud size . we used the intensity - weighted second moments of the position and the velocity to determine the size and linewidth respectively , following ( * ? ? ? * hereafter s87 ) . the radius of the cloud @xmath41 is defined as @xmath187 where @xmath188 and @xmath189 are the dispersions in right ascension and declination converted to projected distance and @xmath190 is the beam width . the coefficient scales the result to match the definitions of cloud radius given in s87 to facilitate comparison with milky way gmcs ( [ comparison ] ) . the scaling is chosen so that the radius of a circular cloud would be the same using either measurement method . our size measure is independent of the orientation of the coordinate system in which the moments are calculated . with the noise level of our data , this radius measurement works best for cloud radii larger than 10 pc , the projected beam radius . with this definition of @xmath41 , the virial mass of the cloud is then : @xmath191 this virial mass estimate implicitly assumes a density profile of @xmath192 . these methods reproduce the virial masses found by @xcite to within 40% for the four clouds that are well - resolved in both samples . the virial mass estimate assumes a spherically symmetric cloud so that the velocity dispersion in the radial direction corresponds to the size scale in the plane of the sky . therefore , the most reliable virial mass estimates come from clouds that appear round on the sky . we define the spatial filling fraction @xmath193 as @xmath194 which compares the area actually occupied by the cloud to the area that the cloud would occupy if it were circular with diameter equal to @xmath195 , the cloud s largest linear dimension . we restrict analyses involving virial masses to the 29 `` round '' clouds , defined as having @xmath196 , to minimize errors in the virial mass estimates . clouds with small filling fractions appear as elongated structures and suggesting that they may be decomposed into multiple , dynamically distinct gmcs ( _ e.g. _ clouds 5 , 10 , 19 , or 21 in figure [ maps ] ) . figure [ masscomp ] compares the virial and luminous ( @xmath180 factor ) masses for the 36 resolved clouds in our study . a linear fit between the virial and co masses for the 29 round clouds finds @xmath197 with a reduced @xmath198 value of 0.9 . the slope of unity implies that our adopted value for the @xmath180 factor is consistent with the virial masses . in [ fluxrec ] , we found that our observations underestimated the integrated intensity and linewidth of the co emission by 50% and 25% respectively . recovering more of the flux with higher sensitivity observations would increase the luminous mass as well as the size and linewidth used in the virial mass . these increases tend to offset each other so the ratio of the luminous mass to the virial mass remains roughly constant . to investigate how omitting 50% of the flux in a field affects our mass estimates , we modeled the clouds as azimuthally symmetric brightness distributions with a power law surface brightness ( @xmath199 ) . the linewidth scales with the size following the relation found in [ comparison ] . using the above methods , we calculated @xmath200 for the central 50% of the emission and for the entire cloud . for all values of @xmath201 between @xmath202 and 0 , the ratio for the central 50% is never more than 1.5 times higher than the ratio for the entire cloud . this estimate is an upper limit since some of the emission is likely from dynamically distinct , low mass clouds surrounding each gmc , implying the interferometer recovers significantly more than half the emission from the central cloud . we conclude that there is a good correspondence between virial and luminous masses derived from the interferometer data . the values of masses , radii and linewidths , are comparable to those found in the milky way ( s87 ) . the empirical relations between these quantities also match those found for the milky way molecular clouds and this similarity is discussed in detail in [ comparison ] . the high resolution co observations allow measurement of the velocity gradient across the molecular clouds . many mapping studies of milky way molecular clouds note such gradients and are summarized in @xcite . the clouds in m33 also show significant velocity gradients , with magnitudes comparable to milky way clouds . to measure the gradient , we derive the velocity centroid at each position in the cloud using the intensity weighted first moment of the masked spectrum . then , we fit a plane to the resulting velocity centroid surface . the coefficients of the fit define @xmath203 , the gradient in this surface and @xmath204 , the radial velocity of the cloud , after @xcite . the vector quantity @xmath205 can also be expressed as the gradient magnitude , @xmath206 , and the position angle of the gradient @xmath207 . the latter is measured east from north to facilitate comparison with the orientation of the galaxy . errors in quantities derived from the fit are propagated from the original data using the covariance matrix . the results of the gradient fits are given in table [ cloudprop ] . planes are good fits to the velocity centroid surfaces with the median value of reduced @xmath208 for the measured gradients . fitting other functions ( like paraboloids or radial collapse profiles ) to the centroid surface yield significantly higher values of @xmath198 . the magnitudes of the gradients are displayed in the histogram in figure [ gradhist ] . the magnitudes are comparable to the typical 0.1 km s@xmath12 pc@xmath12 for the clouds found in the milky way @xcite . in addition , there is no significant relation between position angle and gradient magnitude . figure [ gradfig ] shows position - velocity cuts along the derived gradients for four clouds in the data set . adjacent to each plot is a position - velocity cut perpendicular to the derived gradient . this figure demonstrates that planes are good fits to the data since the cut along the gradient shows a linear trend and the perpendicular cuts appear constant . most authors assume that the velocity gradients are the signature of cloud rotation @xcite . if the gradients are due to rotation , then the magnitude of the velocity gradient measures the angular velocity vector @xmath209 projected into the plane of the sky @xmath210 . the position angle of the gradient is @xmath211 . the gradient therefore provides much information about the angular velocity and rotation axis of a cloud . we find ( 1 ) the _ linear _ gradients observed are the signature of _ solid - body _ rotation . ( 2 ) the rotation energy for these clouds is @xmath212 of the cloud binding energy . ( 3 ) the position angles of the rotation axes show marginal alignment with the axis of the galaxy , as is seen in figure [ pahist ] . ( 4 ) there are many clouds with position angles separated by more than @xmath213 from that of the galaxy . ( 5 ) if the clouds are rotating , roughly 40% of the clouds are retrograde rotators ( [ pacorrs ] ) . ( 6 ) for the median gradient magnitude of @xmath214 km s@xmath12 pc@xmath12 , the rotation period of a molecular cloud would be 125 myr , significantly longer than the assumed molecular cloud lifetimes of 10 to 30 myr ( paper i , * ? ? ? we measure the angular momentum of the molecular clouds from the gradient fits and the cloud radii . for a solid body rotator with a power law surface mass density , the specific angular momentum , is intended will be specifically noted . ] @xmath215 where @xmath216 is a constant set by the mass profile and @xmath41 is the radius of the cloud . for a constant surface density cloud , @xmath217 , ranging from 0.33 to 0.5 for different density profiles @xcite this analysis adopts @xmath218 . the derived angular momentum is listed as @xmath46 in table [ cloudprop ] . the alignment of the rotation axis with respect to the observer affects the magnitude of the observed gradient . the value @xmath219 where @xmath220 is the true angular momentum of the cloud and @xmath221 is the inclination to the line of sight . without knowing the inclination , @xmath46 will underestimate the true angular momentum . to avoid assuming solid - body rotation , we calculated a second measurement of the angular momentum @xmath222^{1/2}\cdot \nonumber\\ & & ( v_k - v_0 ) \sin \vartheta_{i , j } \cdot \left[\sum_{i , j , k } t_a(x_i , y_j , v_k)\right]^{-1}\end{aligned}\ ] ] where @xmath223 and @xmath224 , are the centers of the cloud averaged over position and velocity space respectively and @xmath225 is the angle between the point @xmath226 in the cloud and the gradient judged from the center of the cloud . this value can be calculated even when the cloud is too small to be deconvolved accurately . even with this measurement , a cloud can be oriented with its angular velocity vector along the line of sight thereby obscuring any signature of rotation . the effects of inclination must ultimately be dealt with statistically under the assumption that the clouds are not preferentially oriented towards the observer . * bb00 ) show that turbulent velocity fields can also produce linear gradients in the observational domain . the required turbulent power spectrum , @xmath227 , has a value of @xmath228 between @xmath229 and @xmath101 , so that the largest size - scales have the most power , appearing as gradients . using their scaling for the magnitude of the velocity gradient as a function of size gives gradients on order 0.08 km s@xmath12 pc@xmath12 for the median cloud radius of 20 pc , comparable to the observed values . if the gradients are due to turbulence and not rotation , our measurements change only in that the angular momentum derived from the data will _ overestimate _ the true value of the angular momentum by a factor of 2 to 3 ( bb00 ) . the orientation of the net angular momentum vector is still represented by the position angle of the gradient . the measured angular momenta are significantly less than expected from simple theories of cloud formation . in the absence of external forces , the angular momentum of a molecular cloud should be equal to the angular momentum of the gas from which it formed . in a shearing galactic disk , the initial angular momentum for a forming cloud is ( see appendix [ angder ] ) : @xmath230 in this equation , @xmath231 is the distance from the center of the galaxy , @xmath232 is galactocentric radius where the cloud is found , and @xmath233 is the galactic rotation curve ( from * ? ? ? * ) . the radial accumulation length @xmath234 is half the radial extent of the parent atomic gas . the radial mass distribution of the collapsing gas determines the parameter @xmath235 . in the simplest case , the radial accumulation length is set by requiring the mass of atomic hydrogen in a cylinder stretching to infinity in the @xmath236 direction and of radius @xmath234 to equal the mass of the molecular cloud : @xmath237 . we adopt @xmath238 in equation [ jgal ] calculated for the collapse of a cylinder . using the measured masses of molecular clouds and the local value of @xmath239 , we find that _ in all but 3 of the 36 resolved molecular clouds , the predicted angular momentum is higher than the observed value , on average by a factor of 5_. we calculated the values of @xmath239 using the maps of @xcite after recalibrating this map to match the surface density of @xcite . if the velocity gradients are due to turbulence instead of rotation , bb00 argue that the observed angular momentum should be scaled down by a factor of 2 to 3 , which only increases the discrepancy between the predicted angular momentum and the observed value . statistically , projection effects can account for only 20% of the discrepancy . we refine this simple model as follows : ( 1 ) by assuming that the gas far from the galactic plane does not contribute to the forming molecular cloud and ( 2 ) by including some estimate of projection effects . instead of the progenitor material extending infinitely far from the galactic plane , we assume the gas is accumulated over a cylinder with its height equal to its diameter , centered on the location of the gmc . the three - dimensional distribution of atomic hydrogen is unknown , but we assume a scale height of @xmath240100 pc throughout the galaxy with @xmath241 . changing this value by a factor of 2 does not alter the results significantly . again , the dimensions of the cylinder are determined by requiring its volume to contain a mass of atomic gas equal to that of the molecular cloud . using the radius of the cylinder as the radial accumulation length @xmath234 , the predicted angular momentum is determined from equation [ jgal ] . we also assume the clouds are solid body rotators with random orientations and scale the predicted angular momentum down by a factor of @xmath242 . if , instead , we assume the clouds are strictly aligned with the galaxy , the correction only differs from @xmath243 by 3% . we compare this model to the measurements in figure [ spangcomp ] . the measured angular momenta are plotted as @xmath244 error bars as a function of galactic radius . figure [ spangcomp ] shows that the observations are always less than the predictions of this solid body model ( filled circles ) . the discrepancy is more than an order of magnitude for 20 of the clouds . recent work on star formation at galactic scales suggests that star formation preferentially occurs where the gas disk is toomre unstable @xcite . we calculate the radial accumulation length if the clouds actually formed from the instability . the @xcite instability is a gravitational instability in a shearing , thin fluid disk that bunches the gas into concentric rings , provided the disk is unstable according to the criterion @xmath245 where @xmath246 is the local epicyclic frequency , @xmath247 is the ( three dimensional ) velocity dispersion of the fluid , and @xmath248 is the local surface mass density of the fluid under consideration . the radial accumulation length is set by the original radial extent of the contracting rings : @xmath249 where @xmath250 is the most unstable scale for toomre collapse . the rotation curve determines the epicyclic frequency for the galaxy @xmath251 we assume that there is sufficient mass in the annulus to form molecular clouds which is true over most of the galaxy . figure [ spangcomp ] also shows the predicted values of the cloud angular momentum for gmcs formed using the toomre instability . these values only agree with the observed angular momenta for a few clouds in the inner 500 pc of the galaxy , _ where the disk is most stable _ according to the criterion in equation [ toom_eq ] . since all but two of the clouds disagree with the predictions of the toomre instability by a large margin , it seems unlikely that the clouds form by the action of this instability alone . moreover , the toomre instability predicts that cloud angular momentum should increase with galactocentric radius , which is not seen in figure [ spangcomp ] . the parker instability is another potential avenue of molecular cloud formation @xcite . the parker instability is a magneto - hydrodynamic ( mhd ) effect in gas coupled to a buoyant magnetic field . as the instability evolves , a section of the field bulges out of the plane of the galaxy , causing the atomic gas to flow down the gravitational potential well along the edge of the bulge and collect into molecular clouds . the characteristic length scale of the instability is @xmath252 , where @xmath253 is the scale height @xcite . the values for the predicted angular momentum are shown in figure [ spangcomp ] , for a 100 pc scale height . these values have been corrected for the pitch angle of the magnetic field ( @xmath254 , * ? ? ? * ) since the collapse occurs along the magnetic field lines , so that @xmath255 . the parker instability also gives a large discrepancy between the observed and predicted values . matching the observed values would require a gas scale height of @xmath256 pc , which would imply an unrealistically large stellar mass density . we conclude _ none of the simple theories produce the observed values of angular momentum in the molecular clouds_. figure [ spangcomp ] compares the theoretical and observed angular momentum for the resolved molecular clouds . even using the moment based measurement of angular momentum ( @xmath47 ) , we find an average discrepancy of more than a factor of 5 between observation and theory . this discrepancy shows that nearly all clouds suffer an angular momentum problem independent of assuming solid - body rotation . the failure of the solid body model is particularly troubling since this model gives the minimum accumulation length required to gather the mass of the gmc from the atomic hydrogen . this constraint will limit even the most sophisticated models of cloud formation . there are at least two possible solutions to the angular momentum problem . one is to gather material for molecular clouds from similar galactocentric radii , thereby minimizing the shear in the atomic gas , as may happen in the swing amplification instability @xcite . if clouds have small radial accumulation lengths , they must have long azimuthal accumulation lengths . agreement with the observed angular momenta requires @xmath257 pc . to form a cloud of @xmath258 out of atomic hydrogen with a typical surface density of @xmath259 pc@xmath3 requires azimuthal accumulation lengths of @xmath260500 pc . with the maximum non - circular velocities in the atomic gas @xmath261 10 km s@xmath12 @xcite , this implies a limit on the timescale of cloud formation of @xmath262 myr , larger than the 1020 myr cloud lifetimes implied in paper i. this timescale is a significant fraction of the time between the spiral arm crossings for most of the galaxy away from the co - rotation radius @xcite . consequently , such a long time scale for cloud formation has difficulties explaining preferentially finding clouds in spiral arms of galaxies . a second solution is to brake the material with external forces such as magnetic fields . magnetohydrodynamic effects seem to be critical for shedding angular momentum in the star formation process and may also act in the early stages of cloud formation . if the magnetic field is so dynamically significant as to produce an instability , it may also provide enough tension to brake the clouds during their formation . @xcite measures the mean field strength in m33 as @xmath263 , which gives an alfvn speed of @xmath2606 km s@xmath12 ( assuming a sech@xmath264 density profile to calculate the midplane volume mass density ) . this value is comparable to the sound speed in the warm neutral medium of most galaxies , suggesting that mhd braking and cloud formation via instabilities occur on comparable timescales . the braking time is set by the time needed for an alfvn wave to sweep through a volume of diffuse gas with a moment of inertia equal to that of the forming cloud @xcite . if the atomic gas has a constant density , then the alfvn wave must pass through a distance comparable to the accumulation length for the molecular cloud . instabilities cause cloud formation on a time scale comparable to the sound crossing time for the accumulation length . since the crossing times for sound and alfvn waves are comparable , we conclude that mhd braking is a viable option to slow the rotation of forming clouds . because only 10% of clouds show angular momentum approaching the expected values from any of the formation mechanisms ( e.g. figure [ spangcomp ] ) , the mhd braking must occur in the atomic gas or during a small fraction of the molecular cloud lifetime . we return to the implications of these observations for cloud formation in [ finis ] . the molecular clouds in this sample suggest a correlation between specific angular momentum and mass . fitting a power - law relationship between @xmath46 and @xmath265 finds @xmath266 , though there is significant scatter in the data ( @xmath267 , figure [ mags ] ) . a similar correlation appears in the milky way data where @xcite finds @xmath268 . if the molecular clouds form by accumulation of atomic gas , then the angular momentum should scale as the accumulation length squared ( equation [ jgal ] ) : @xmath269 . the mass of the cloud is @xmath270 , so a correlation is expected with @xmath271 . the variations in the mean hydrogen surface density are small compared to the order of magnitude spanned in mass ( [ comparison ] ) , so the angular momentum is expected to scale with mass for any accumulation model . that @xmath272 instead of @xmath273 may be due to the action of the braking mechanism , though @xmath273 is not strongly ruled out by the poor quality of the fit . such correlation between @xmath220 and @xmath274 is also expected for a turbulent velocity field . bb00 find that the velocity gradient scales with clouds size @xmath41 as @xmath275 so @xmath276 . the clouds in our sample have a constant surface density regardless of cloud size ( [ comparison ] ) . therefore , the mass of a cloud is proportional to its area @xmath277 and @xmath278 , close to the observed trend . there is no significant variation of specific angular momentum as a function of galactic radius , contrary to the predictions of all the accumulations mechanisms considered here ( figure [ spangcomp ] ) . this implies that the angular momentum of the clouds does not depend on any of the properties that change across the galactic disk , including galactic shear , surface mass density of stars and atomic gas , interstellar pressure , interstellar radiation field , or metallicity . the position angle of the angular momentum vector can be measured from the velocity gradients as discussed in [ velgrad ] . the data show correlations between the cloud position angles and that of the galaxy as well as among the individual gmcs . if the angular momentum is set entirely by the galactic shear , then the rotation axes of the gmcs should be aligned with the galaxy . the distribution of gradient position angles is shown in figure [ pahist ] , including rough divisions between prograde and retrograde rotators ( assuming the gradients represent rotation ) . because of the unknown inclination of the clouds relative to the line of sight , we analyze the distribution of position angles statistically . we assume that the clouds are distributed with orientations on the sphere set by the probability distribution function ( pdf ) : @xmath279^{1/\alpha-1 } \frac{\sin \theta}{2}\label{pdf}\ ] ] where @xmath280 is the angle between the angular momentum vector of the cloud and that of the galaxy , @xmath15 is a parameter that determines the degree of alignment between these vectors , and @xmath281 is the azimuthal angle . for @xmath282 , the orientation angles are randomly distributed on the sphere so there are an equal number of prograde and retrograde rotators . for @xmath283 , the clouds are closely aligned with the galaxy ( @xmath284 implies that 91% of the clouds are prograde rotators ) . this pdf is motivated solely for ease of simulation so that a single parameter @xmath15 determines how well aligned the clouds are with the galaxy . by determining the value of @xmath15 that most closely represents the observed distribution of position angles , we measure the degree of alignment between the clouds and the galaxy . we generated position angle distributions following this pdf using monte carlo simulations and then used a one - sided k - s test to measure the difference between the observed and simulated distributions . after testing a range of @xmath15 , we found the most probable value is @xmath285 ( with k - s likelihood @xmath286 ) , implying 61% of the clouds are prograde rotators . the likelihood of a random distribution on the sphere ( @xmath282 ) is significantly lower ( @xmath287 ) implying some alignment between the clouds and the galaxy . nevertheless , the wide range of position angles indicates any angular momentum imparted by galactic shear is randomized in some fashion . in addition to the cloud - galaxy correlation , nearby clouds show a higher degree of correlation between their gradient position angles . figure [ pacorr ] shows the difference in position angle between pairs of clouds as a function of cloud separation . this figure uses clouds with @xmath288 since these clouds have the most significant gradients . if the clouds are randomly oriented with respect to each other , the value of this statistic should be @xmath213 , but clouds with separations smaller than 500 pc show a significantly smaller value implying their angular momentum vectors are aligned . this alignment would be expected if these neighboring clouds formed from the same parent cloud of . if clouds are the product of large scale motions in the diffuse ism , this 500 pc length may represent the scale of the flows that give rise to the clouds . the clouds in the m33 observations have similar radii ( @xmath289 pc ) , masses ( @xmath290 ) and velocity dispersion fwhms ( 5 @xmath291 10 km s@xmath12 ) as those found in the milky way by s87 . we tested these similarities in more detail and found the gmcs in m33 were indistinguishable from those in the milky way . we emphasize comparison of our results with the work done by s87 because both studies use equivalent methods for determining the properties of the molecular clouds and the sensitivities are comparable : @xmath2600.7 k km s@xmath12 for s87 vs. 0.6 @xmath291 1.3 k km s@xmath12 for this work . we only use the 23 clouds that are well - resolved ( @xmath292 pc ) and round ( @xmath293 ) in these comparisons since the derived properties of these clouds are the most reliable . the results of paper i show that the mass spectrum for gmcs in m33 differs from that found in the milky way : @xmath294 for m33 vs. @xmath295 for the milky way ( s87 ) . despite differences in the mass distribution , clouds in m33 show the same power law relationships between size , mass and linewidth as are seen in the milky way ( larson s laws , after @xcite , see also s87 ) . we plot the size linewidth and the mass linewidth relationship for the clouds in m33 in figure [ larson ] along with the data from the milky way @xcite . where feasible , the cloud properties of the milky way data have been recomputed using the methods in [ cloudprops ] to ensure a common basis for comparison . fitting power - laws to the two relationships shows @xmath296 and @xmath297 . the fits are consistent with the relations seen in the milky way ( s87 ) in both normalization and scaling . however , cloud sizes in m33 are systematically smaller by @xmath298% in m33 compared to s87 , which is most likely due to differences in the size measurement technique . figure [ surfdens ] shows the mean surface mass density within molecular clouds as a function of cloud mass . this figure shows no significant scaling of surface density with mass . similar plots of surface density as a function of cloud radius and galactic radius also show no scaling . the slopes of linear fits between the mass density and cloud mass or cloud radius are consistent with zero . in addition , two statistical tests of correlation , the spearman rank order and the kendall @xmath299 tests show no significant correlation between mass , radius and surface density ( see * ? ? ? * ) . we conclude that gmcs have a surface mass density of @xmath300 ( @xmath301 ) with a dispersion of 60% , which does not vary with cloud mass or radius . this constant mass density is also seen for the gmcs in the milky way @xcite . the consistency of larson s laws between the milky way and m33 shows that the macroscopic properties of a gmc ( @xmath41 , @xmath302 , @xmath303 ) are the same in both galaxies for a given cloud mass . since these macroscopic properties follow similar trends , it is reasonable to expect that the internal properties of gmcs depend only on the mass of the cloud . the gmcs throughout the local group also follow larson s laws ; the lmc @xcite , andromeda @xcite , and dwarf elliptical galaxies @xcite all show some evidence that cloud properties are set by cloud mass . some unifying mechanism must establish the macroscopic properties of the cloud solely in terms of the cloud mass . since turbulence could provide a relationship between size and linewidth , the balance between turbulent support and self - gravity would naturally set the macroscopic properies of a cloud solely in terms of its mass @xcite . in m33 , we are presented with a unique opportunity to study the effects of galactic environment on the co - to - h@xmath5 conversion factor ( @xmath180 ) since there are no systematic differences due to inhomogeneous observation techniques or multiple galaxies . many studies report a variation in the @xmath180 factor as a function of metallicity _ e.g. _ @xcite and references therein . there is no evidence for such a trend in m33 . in figure [ xfac2 ] , we plot the ratio of the luminous to virial mass @xmath200 as a function of metallicity for the 23 well - resolved ( @xmath292 pc ) , round ( @xmath196 ) gmcs in our study . the metallicity at each point is obtained using the position of the cloud and the metallicity gradient ( @xmath304}/dr$ ] ) of @xcite . fitting a line to the data gives : @xmath305-[\mathrm{o / h}]_{\odot}\right)+(0.88 \pm 0.09)\ ] ] accounting for errors in both the metallicity ( 0.1 dex ) and the mass ratio @xcite . the value of @xmath198 for the fit is 0.9 . the value of @xmath198 using the relationship of ( * ? ? ? * dotted line ) as a fit gives 3.31 , showing poor agreement ( @xmath306 ) with their quoted trend . the slope of our fit is consistent with _ no variation in the conversion factor over a range of 0.8 dex in metallicity _ ( a factor of 6 ) . @xcite note that the ratio between virial and co masses does not change systematically with radius in the milky way . in [ mass_sec ] , we argue that interferometer observations overestimate the ratio @xmath200 by _ at most _ a factor of 1.5 . this can not account for the discrepancy with the @xcite data , which we attribute to the difficulties in synthesizing a homogeneous data set from multiple observational studies using a variety of analysis techniques . the studied gmcs range in galactocentric radius from 170 pc to 4000 pc , and there are significant changes in galactic properties over this range . the interstellar radiation field changes value by nearly an order of magnitude over this range in galactic radius @xcite . other variations include changes in the cosmic ray flux and the midplane pressure of the gas . such robustness of the @xmath180 factor as a gmc mass measure is observed in the milky way ( s87 ) , but variations among galaxies may still produce different values in the conversion factor . for example , the @xmath180 factor may change for gmcs in even lower metallicity systems such as the smc that have an average metallicity below the minimum 8.1 observed in m33 . this study of m33 suggests several attributes of the molecular cloud formation process . ( 1 ) molecular clouds form directly from atomic hydrogen rather than the agglomeration of small molecular clouds . ( 2 ) the formation process makes both low mass and high mass molecular clouds in regions of cloud formation . ( 3 ) the formation events are local , _ i.e. _ length scales less than 500 pc . ( 4 ) the progenitor gas has its initial angular momentum dissipated by mhd effects . ( 5 ) the similarity of molecular clouds across many environments implies that the molecular mass distribution sets the properties of molecular clouds and star formation . our results complement those of paper i where it is argued that molecular clouds form directly from atomic hydrogen . in that paper , the striking correspondence between molecular clouds and dense atomic filaments strongly suggests cloud formation from atomic gas . this work strengthens that suggestion since we find no evidence for diffuse molecular gas that could serve as a precursor to gmcs ( [ fluxrec ] ) . additionally , the atomic gas is significantly more massive than the molecular component on a global as well as a local ( @xmath307 pc ) scale . if diffuse _ molecular _ gas were the precursor material , the angular momentum problem would be greatly exacerbated , scaling predicted angular momenta up by a factor of @xmath308 . we also find that the flux not recovered in paper i survey can be accounted for by halos of molecular gas around the detected gmcs , comprised in part by low mass molecular clouds ( [ fluxrec ] ) . these low mass clouds may form in the same event as the high mass cloud at the center of the interferometer field . such low mass molecular clouds would have easily been detectable in the off fields of the 12-m observation and may be restricted to regions where high mass molecular clouds also form . the position angle correlations in [ pacorrs ] suggest that the formation of molecular clouds occurs on length scales less than 500 pc . a common formation event could produce aligned molecular clouds , analogous to the formation of the solar system producing aligned rotation in the planets . correlations in the orientation of angular momentum can be predicted in simulations of turbulent cloud formation that include realistic models of galactic shear . such predictions should show the observed degree of alignment with the galaxy and correlations among the gmcs . the angular momentum problem can be solved by appealing to mhd braking . forming the molecular clouds from a narrow range of radii requires a long formation time . formation mechanisms that accumulate gas supersonically can not have any angular momentum dissipated by alfvn waves since the accumulation time is smaller than crossing times for magnetosonic waves . in contrast , the magnetic field is sufficient to brake forming molecular clouds and braking occurs over timescales comparable to those of instabilities that could potentially form molecular clouds ( [ mhdsec ] ) . molecular clouds appear to have their macroscopic properties ( @xmath231 , @xmath302 ) set by their masses ( [ comparison ] ) . in addition , there appears to be a constant star formation rate per unit molecular mass within galaxies @xcite . if this is true , then the properties of star formation on a galactic scale depend on the amount of atomic gas converted into molecular clouds and little else . although there are similarities in the properties of molecular clouds for a fixed mass , there is good evidence that the mass distribution of molecular gas into individual clouds varies among galaxies . different mass spectra may be the hallmark of multiple formation mechanisms converting atomic into molecular gas . the steep mass spectrum for the molecular clouds in m33 ( paper i ) represents the _ only _ difference in the molecular cloud population between m33 and the milky way . the mass distribution of molecular gas is thus the dominant factor controlling star formation on a local and galactic scale . we have presented high resolution follow - up observations to a survey of gmcs in m33 made by @xcite , using the bima array and the uaso 12-m millimeter wave telescope . a total of 45 individual gmcs were detected . the uaso 12-m observations measured the total flux in 18 fields , providing information on the flux lost by the interferometer and the presence of a diffuse component of molecular gas . from our observations , we made the following conclusions . there is no evidence for a diffuse molecular component traced by co spanning the disk of m33 . we place a @xmath28 surface mass density limit of @xmath309 pc@xmath3 on the presence of such a component . this is significantly less than the typical surface density of atomic hydrogen , 7 @xmath22 pc@xmath3 . most of the co flux in the galaxy is associated with gmcs with smaller clouds clustered around the large cloud . low mass clouds may appear in other parts of the galaxy but are not ubiquitous . the velocity gradients of the molecular clouds are approximately linear . the magnitudes of the gradients are comparable to those found in the milky way . if the gradients are due to rotation , the rotation period is significantly longer than a cloud lifetime ( for @xmath310 myr ) . gmcs show significantly smaller angular momenta than are predicted by simple formation theories . the discrepancy is , on average , a factor of 5 and ranges up to two orders of magnitude . if the velocity gradients are due to turbulent motions , the discrepancy widens by at least a factor of 2 . both the toomre and parker instabilities predict angular momentum values that are discrepant from observations . the specific angular momentum is related to mass as @xmath311 . this value is consistent with the derived value for the galaxy of @xmath312 @xcite . there is no significant variation of the specific angular momentum with galactocentric radius , though such variations are predicted by all the large scale accumulation theories . the clouds appear somewhat aligned with the rotation of the galaxy , though a random distribution of position angles is not strongly excluded . if the velocity gradients are from rotation , only @xmath313 of the clouds are prograde rotators . the projected velocity gradients of neighboring , high - mass clouds are preferentially aligned . this correlation vanishes for separations larger than 500 pc . this alignment may be a signature of the large scale mechanisms that dictate the formation of molecular clouds and should be seen in turbulent cloud formation models . the observed molecular clouds are similar to those found in the milky way . we find a size linewidth relationship of @xmath314 and a mass line width relationship of @xmath315 . both of these relationships are indistinguishable from those found in the milky way . there is also no detectable variation in the column density of the molecular clouds with the mass or radius of the clouds . equivalent virial masses and luminous masses imply that the @xmath180 factor for m33 is equal to @xmath316 . there is no significant variation of the @xmath180 factor with metallicity over a range of 0.8 dex . these observations support a model for cloud formation in m33 using atomic gas as the progenitor material for molecular clouds . the formation process accumulates the atomic gas over a small distance ( @xmath317 pc ) with significant braking by magnetic fields . we thank an anonymous referee whose comments improved the paper . we gratefully acknowledge the help of edvige corbelli for the surface density and rotation curve data used in her paper . we thank adam leroy for extremely useful discussions of data reduction and analysis . our discussions with ellen zweibel regarding mhd braking during cloud formation have proven especially fruitful . observations at the uaso 12-m were aided by great help from the operator team consisting of sean keel , jon carlsen , john downey and headed by paul hart . we are also grateful to tom dame who provided us with swift checks on our data using the cfa 1.2 m telescope . er s work is supported in part by a nsf graduate fellowship and was completed with great encouragement from elizabeth tan . this research has been made possible by extensive use of nasa s astrophysics data system ( ads ) and the nasa / ipac extragalactic database ( ned ) . consider the progenitor mass for a molecular cloud distributed in galactocentric radius with a linear mass density @xmath318 . the specific angular momentum of this material is @xmath319 if the cloud s final distance from the center of the galaxy is @xmath232 , then we define @xmath320 and perform a taylor expansion of the velocity curve around @xmath232 to get @xmath321 \left(r+r_c\right ) dr\ ] ] where @xmath322 multiplying out the integral gives , @xmath323 \int \lambda(r ) r\ dr + d(r_c ) \int \lambda(r ) r^2\ dr\right]\ ] ] where the first term represents the angular momentum of a particle orbiting the center of the galaxy and the other two terms are the spin angular momentum of the cloud , the measured quantity in these observations . in the absence of external forces , the cloud will collapse to the center of mass , which implies @xmath324 . in this case , the imparted angular momentum is @xmath325 where the matter is accumulated from a region of width @xmath326 . we define a constant @xmath235 set by the moment of the mass distribution in terms of the dimensionless parameter @xmath327 : @xmath328 for collapse in a circular region of radius @xmath234 around @xmath232 with uniform surface mass density , @xmath329 . , r. 2000 , in the interstellar medium in m31 and m33 . proceedings 232 . we - heraeus seminar , 22 - 25 may 2000 , bad honnef , germany . edited by elly m. berkhuijsen , rainer beck , and rene a. m. walterbos . shaker , aachen , 2000 , p. 171 - 178 , 171178 , k. , vogel , s. n. , wilson , c. d. , & dame , t. m. 2000 , in the interstellar medium in m31 and m33 . proceedings 232 . we - heraeus seminar , 22 - 25 may 2000 , bad honnef , germany . edited by elly m. berkhuijsen , rainer beck , and rene a. m. walterbos . shaker , aachen , 2000 , p. 37 - 40 , 3740

in confined helimagnetic nanostructures , skyrmionic states in the form of incomplete and isolated skyrmion states can emerge as the ground state in absence of both external magnetic field and magnetocrystalline anisotropy . in this work , we study the dynamic properties ( resonance frequencies and corresponding eigenmodes ) of skyrmionic states in thin film fege disk samples . we employ two different methods in finite - element based micromagnetic simulation : eigenvalue and ringdown method . the eigenvalue method allows us to identify all resonance frequencies and corresponding eigenmodes that can exist in the simulated system . however , using a particular experimentally feasible excitation can excite only a limited set of eigenmodes . because of that , we perform ringdown simulations that resemble the experimental setup using both in - plane and out - of - plane excitations . in addition , we report the nonlinear dependence of resonance frequencies on the external magnetic bias field and disk sample diameter and discuss the possible reversal mode of skyrmionic states . we compare the power spectral densities of incomplete skyrmion and isolated skyrmion states and observe several key differences that can contribute to the experimental identification of the state present in the sample . we measure the fege gilbert damping , and using its value we determine what eigenmodes can be expected to be observed in experiments . finally , we show that neglecting the demagnetisation energy contribution or ignoring the magnetisation variation in the out - of - film direction although not changing the eigenmode s magnetisation dynamics significantly changes their resonance frequencies substantially . apart from contributing to the understanding of skyrmionic states physics , this systematic work can be used as a guide for the experimental identification of skyrmionic states in confined helimagnetic nanostructures .

1604.08347

dzyaloshinskii - moriya interactions @xcite ( dmi ) may occur in magnetic systems that lack some type of inversion symmetry . the inversion asymmetry can be present in the magnetic system either because of a non - centrosymmetric crystal lattice @xcite ( helimagnetic material ) or due to the interfaces between different materials which inherently lack inversion symmetry . @xcite consequently , the dmi can be classified either as bulk or interfacial . the dmi favours magnetic moments at neighbouring lattice sites to be perpendicular to each other ( in plane that is perpendicular to the dzyaloshinskii vector ) , which is in contrast to the symmetric ferromagnetic exchange interaction which tends to align them parallel . when acting together , these two interactions mutually compete and find a compromise in the twist between two neighbouring magnetic moments , which allows a rich variety of different magnetisation textures . one of them is a skyrmion configuration with very promising properties @xcite for the development of future high - density , power - efficient storage @xcite and logic @xcite devices . after it was predicted @xcite that magnetic skyrmions can emerge in the presence of dmi , skyrmions were observed in magnetic systems with both bulk @xcite and interfacial @xcite types of dmi . however , all studies of helimagnetic ( bulk dmi ) materials required an external magnetic field to be applied in order to stabilise skyrmions . recently , a systematic micromagnetic study @xcite reported all equilibrium states that can emerge in confined helimagnetic nanostructures and identified the lowest energy ( ground ) states . in particular , this study reported that in confined helimagnetic nanostructures two different skyrmionic states can emerge as the ground state in absence of both external magnetic field and magnetocrystalline anisotropy . one state does not contain a complete spin rotation , whereas the other state contains one full spin rotation along the disk sample diameter , plus an additional magnetisation tilting at the boundary due to the specific boundary conditions . @xcite we refer to these configurations as incomplete skyrmion ( isk ) and isolated skyrmion ( sk ) states , respectively . @xcite in addition , the same study showed that the higher - order target ( t ) state with two complete spin rotations along the disk sample diameter can emerge as a metastable state at zero external magnetic field . understanding the dynamic response of skyrmionic states in confined helimagnetic nanostructures is of importance both from the aspect of fundamental physics as well as for their manipulation . in this work , we explore the dynamics of all three equilibrium skyrmionic states using a full three - dimensional model which includes the demagnetisation energy contribution and does not assume the translational invariance of magnetisation in the out - of - film direction . a similar dynamics simulation study was performed for the isolated skyrmion breathing eigenmodes @xcite in confined two - dimensional samples with interfacial dmi ; and high - frequency skyrmion spin excitations were analytically studied in thin cylindrical dots . @xcite the low - frequency ( two lateral and one breathing ) eigenmodes were reported in two - dimensional simulations of a hexagonal skyrmion lattice , @xcite where the demagnetisation energy was neglected . later , microwave absorption measurements explored the low frequency eigenmodes in cu@xmath0oseo@xmath1 , @xcite fe@xmath2co@xmath3si , @xcite and mnsi @xcite helimagnetic bulk samples . in the case of a magnetic bubble @xcite ( magnetisation state stabilised due to the strong uniaxial anisotropy in the absence of dmi ) analytic , @xcite simulation , @xcite and experimental @xcite studies reported the existence of two low frequency gyrotropic eigenmodes , suggesting that the skyrmion possesses mass . in contrast , a recent analytic work @xcite suggests that only one gyrotropic eigenmode exists in the confined dmi - induced skyrmion state , whereas another low - frequency lateral eigenmode is interpreted as an azimuthal spin - wave mode @xcite . using our full three - dimensional model , employing the eigenvalue @xcite method , we compute all existing ( both lateral and breathing ) eigenmodes below @xmath4 in three different skyrmionic states . in addition , using the ringdown @xcite method , we determine which eigenmodes can be excited using two different experimentally feasible excitations ( in - plane and out - of - plane ) . in contrast to the magnetic bubble , in the confined dmi stabilised skyrmionic states we find the existence of only one low frequency gyrotropic eigenmode . we also demonstrate the nonlinear dependence of eigenmode frequencies on the external magnetic bias field and the disk sample diameter , and show that the gyrotropic eigenmode might be the reversal mode of skyrmionic states . after we identify all eigenmodes of incomplete skyrmion ( isk ) and isolated skyrmion ( sk ) ground states , we compare their power spectral densities ( psds ) in the same sample at different external magnetic field values . we discuss the comparisons and observe several key differences that can contribute to the experimental identification of the state present in the studied sample . although we base this study on the specific helimagnetic material fege , in order to make this study relevant to any helimagnetic material , we need to determine as many resonant frequencies as possible that can be detected using a specific excitation . because of that , we need to reduce the linewidth and allow sufficient separation between peaks in the power spectral density ( computed using the ringdown method ) . consequently , in the first part of this work , we use the gilbert damping @xcite @xmath5 . after we identify all resonance frequencies and corresponding eigenmodes using @xmath6 , we experimentally measure the real value of fege gilbert damping and use it to determine which ( out of all previously identified eigenmodes ) can be experimentally detected in the fege sample . finally , we investigate how the demagnetisation energy contribution and magnetisation variation in the out - of - film direction affect the dynamics of skyrmionic states . we report that although the eigenmode magnetisation dynamics is not significantly affected , the resonance frequencies change substantially , which indicates that ignoring the demagnetisation energy or modelling the thin film helimagnetic samples using two - dimensional meshes is not always justified . we simulate a thin film helimagnetic cubic b20 fege disk with @xmath7 thickness and diameter @xmath8 , as shown in fig . [ fig : methods](a ) . the thin film sample is in the @xmath9 plane and perpendicular to the @xmath10 axis . an external magnetic bias field @xmath11 is applied uniformly and perpendicular to the sample ( in the positive @xmath10 direction ) . ( a ) a thin film fege disk sample with @xmath7 thickness and diameter @xmath8 . an external magnetic bias field @xmath11 is applied uniformly and perpendicular to the sample ( in the positive @xmath10 direction ) . ( b ) a cardinal sine wave excitation magnetic field @xmath12 , used in the ringdown method , is applied for @xmath13 in either in - plane ( @xmath14 ) or out - of - plane ( @xmath15 ) direction . ( c ) the fourier transform of excitation field @xmath16 shows that all eigenmodes ( allowed by the used excitation direction ) with frequencies lower than @xmath17 are excited approximately equally . ] the total energy of the system we simulate contains several energy contributions and can be written as @xmath18 \,\text{d}^{3}r.\ ] ] the first term @xmath19 $ ] is the symmetric exchange energy density with material parameter @xmath20 . the unit vector field @xmath21 , with cartesian components @xmath22 , @xmath23 , and @xmath24 , represents the magnetisation field @xmath25 , where @xmath26 is the saturation magnetisation . the second term @xmath27 is the dzyaloshinskii - moriya energy density with material parameter @xmath28 , which is obtained by including lifshitz invariants suitable for materials of the crystallographic class t , such as the cubic b20 fege ( p@xmath29 space group ) used in this study . the coupling of magnetisation to an external magnetic field @xmath11 is defined by the zeeman energy density term @xmath30 , with @xmath31 being the magnetic constant . the @xmath32 term is the demagnetisation ( magnetostatic ) energy density . because @xmath32 is crucial for the stability of skyrmionic states in confined helimagnetic nanostructures , @xcite we include its contribution in all subsequent simulations . the last term is the magnetocrystalline anisotropy energy density @xmath33 , and because it does not play an important role in the stability of skyrmionic states in the studied system , @xcite we assume the simulated material is isotropic and neglect the magnetocrystalline anisotropy energy contribution . the fege material parameters we use are the following : @xcite saturation magnetisation @xmath34 , exchange energy constant @xmath35 , and dzyaloshinskii - moriya energy constant @xmath36 . in our model , we do not assume any translational invariance of magnetisation in the out - of - film direction , which significantly changes the energy landscape both in infinitely large thin films @xcite and in confined thin film nanostructures . @xcite the relevant length scales in the simulated system are the exchange length @xmath37 and helical length @xmath38 . we choose the finite element mesh discretisation so that the maximum spacing between two neighbouring mesh nodes is below @xmath39 , which is significantly smaller than the exchange length @xmath40 . the magnetisation dynamics is governed by the landau - lifshitz - gilbert ( llg ) equation @xcite @xmath41 where @xmath42 , with @xmath43 and @xmath44 is the gilbert damping . we compute the effective magnetic field @xmath45 using @xmath46}{\delta \mathbf{m}},\ ] ] where @xmath47 $ ] is the total magnetic energy functional , given by eq . ( [ eq : total_energy ] ) . we validated the boundary conditions by running a series of simulations and reproducing the results reported by rohart and thiaville @xcite . we implemented the presented model in the finite element method framework and developed a micromagnetic simulation tool finmag ( successor of nmag @xcite ) . for the low - level finite element operations , we use fenics project @xcite and for the adaptive step time integration we use sundials / cvode solver . @xcite for visualisation , we use matplotlib @xcite and paraview . @xcite we study the dynamic properties of skyrmionic states using two different methods : eigenvalue method @xcite and ringdown method . @xcite in both eigenvalue and ringdown methods , we firstly compute an equilibrium magnetisation configuration @xmath48 by integrating a set of dissipative time - dependent equations , starting from a specific initial magnetisation configuration , until the condition of vanishing torque ( @xmath49 ) is satisfied . the details on selecting the initial magnetisation configurations can be found in ref . . we perform all relaxations in this work down to the maximum precision limited by the unavoidable numerical noise . because the magnetisation dynamics is not of interest in the relaxation process , we set the gilbert damping in this stage to @xmath50 . we perform the eigenvalue method computations in a finite element framework , motivated by the analytic procedure by daquino _ et al . _ @xcite the perturbation of the system s magnetisation from its equilibrium state @xmath48 can be written as @xmath51 , where @xmath52 and @xmath53 because of the imposed micromagnetic condition @xmath54 . if this perturbation expression is inserted into the undamped llg equation , we obtain @xmath55 by using a taylor expansion @xmath56 , where @xmath57 , and knowing that @xmath58 and @xmath59 , we get @xmath60,\ ] ] where all @xmath61 terms and higher are neglected . when the system is in its equilibrium , because @xmath62 and @xmath63 , the equilibrium effective field can be written as @xmath64 , where @xmath65 . now , if all vector fields are discretised on the finite elements mesh , eq . ( [ eq : eigenvalue_method2 ] ) becomes @xmath66.\ ] ] using the matrix @xmath67 with property @xmath68 , eq . ( [ eq : eigenvalue_method3 ] ) can be written as @xmath69 where @xmath70 $ ] . this linear differential equation has a full set of solutions that can be expressed as @xmath71 , where @xmath72 is a constant vector field . using this ansatz , eq . ( [ eq : eigenvalue_method4 ] ) becomes the eigenvalue problem @xmath73 we solve this eigenvalue problem using python wrappers for the arpack @xcite solvers which are implemented in the scipy @xcite package , which results in a set of resonant frequencies @xmath74 and eigenvectors @xmath72 from which we express the magnetisation dynamics as @xmath75 . in the ringdown method , similar to the eigenvalue method , we firstly relax the system to its equilibrium magnetisation state @xmath48 . after that , we perturb the system from its equilibrium by applying a time - dependent @xmath76 external magnetic field excitation @xcite over @xmath77 , where @xmath78 is the maximum excitation field value , @xmath17 is the cut - off frequency , @xmath79 is the direction in which the excitation is applied , and @xmath80 is the unnormalised cardinal sine function @xmath81 the time - dependence of the used excitation @xmath16 is shown in fig . [ fig : methods](b ) . computing the fourier transform of @xmath16 shows that using this excitation enables us to excite all eigenmodes ( which are allowed by the direction of excitation @xmath79 ) in the @xmath82 $ ] range approximately equally , as demonstrated in fig . [ fig : methods](c ) . we compute the @xmath78 value so that @xmath83 is the excitation amplitude at any frequency . @xcite more precisely , the maximum value of the cardinal sine wave excitation is @xmath84 . we apply the excitation in two experimentally feasible directions : ( i ) in - plane @xmath85 and ( ii ) out - of - plane @xmath86 . after the system is perturbed from its equilibrium state , we simulate the magnetisation dynamics for @xmath87 and sample the magnetisation field @xmath88 at all mesh nodes @xmath89 at uniform time steps @xmath90 ( @xmath91 ) . although the excitation is sufficiently small so that the perturbation from the equilibrium state can be approximated linearly , in order to make sure we do not introduce any nonlinearities to the system s dynamics with the excitation , we delay sampling by @xmath92 after the excitation field is removed . finally , we analyse the recorded magnetisation dynamics @xmath88 using : ( i ) spatially averaged and ( ii ) spatially resolved methods . @xcite we subtract the time - independent equilibrium magnetisation configuration @xmath93 from the recorded magnetisation dynamics and perform the fourier analysis only on the time - dependent part @xmath94 . in the spatially averaged analysis , we compute all three spatially averaged magnetisation components @xmath95 , @xmath96 , at all time steps @xmath97 . after that , we apply a discrete fourier transform and sum the squared fourier coefficient moduli ( which are proportional to the power ) to obtain the power spectral density ( psd ) : @xmath98 where @xmath99 is the number of time steps at which the magnetisation dynamics was sampled . on the other hand , in the spatially resolved analysis , we firstly compute the discrete fourier transform at all mesh nodes ( separately for all three magnetisation components ) and then compute the psd as the spatial average of the squared fourier coefficient moduli : @xcite @xmath100 where @xmath101 is the number of finite element mesh nodes . because the power values in psd are in arbitrary units ( a.u . ) , we normalise all psds in this work so that @xmath102 , where @xmath103 , and show them in the logarithmic scale . we choose to analyse dynamics of skyrmionic states in the @xmath104 $ ] range in order to avoid the presence of artefact peaks in psds due to aliasing @xcite as a consequence of discrete time sampling limitations . although , the frequency resolution in the eigenvalue method is determined by the machine precision , the frequency resolution for the ringdown method is @xmath105 , where @xmath106 is the number of sampling points during the sampling simulation stage . we study the dynamics of all three equilibrium skyrmionic states that can be observed at zero external magnetic field in confined thin film helimagnetic disk samples with diameters @xmath107 . more precisely , we explore the resonance frequencies and corresponding eigenmode magnetisation dynamics of the ground state incomplete skyrmion ( isk ) and isolated skyrmion ( sk ) states , as well as the metastable target ( t ) configuration . the difference between these states is in how many times the magnetisation configuration covers the sphere . a quantity that is usually used to determine whether a magnetisation configuration covers the sphere is the skyrmion number @xcite @xmath108 . however , in confined helimagnetic nanostructures , an additional tilting of magnetisation at the sample edges @xcite in the winding direction opposite to the skyrmion configuration reduces the absolute skyrmion number value @xcite . this does not allow us to determine what skyrmionic state is present is the sample because @xmath109 for all of them . to address this , a scalar value @xcite @xmath110 in a three - dimensional sample is defined as @xmath111 this scalar value has no mathematical or physical interpretation , and is defined merely to support the classification of skyrmionic states in confined nanostructures . using the eigenvalue method , we find all existing eigenmodes by computing their resonance frequencies and magnetisation dynamics . however , this method does not allow us to determine what eigenmodes can be excited using a particular excitation . therefore , we employ the ringdown method for an in - plane and an out - of - plane excitation and overlay the resulting spatially averaged and spatially resolved power spectral densities ( psds ) with the resonance frequencies obtained from the eigenvalue method . if the eigenvalue method resonance frequency coincides with a psd peak , this implies that the corresponding eigenmode can be `` activated '' using a specific excitation and we mark it using a triangle ( @xmath112 ) symbol . all other eigenmodes , that can not be activated using a particular excitation , we mark with a circle ( @xmath113 ) symbol . throughout this work , we study the magnetisation dynamics below @xmath4 . we analyse the target state dynamics in supplementary section s1 . the first magnetisation configuration that we study is the incomplete skyrmion ( isk ) state . the magnetisation component @xmath24 of the isk state , along the sample diameter , does not cover the whole @xmath114 $ ] range , which is required for the skyrmion configuration to be present in the sample , and because of that , the scalar value @xmath115 is in the @xmath116 $ ] range . @xcite in other works , this state is called either the quasi - ferromagnetic @xcite or the edged vortex @xcite state . the incomplete skyrmion state is in an equilibrium for all studied disk sample diameters @xmath117 and at all external magnetic bias field values . @xcite we explore the resonance frequencies and corresponding eigenmode magnetisation dynamics in a @xmath118 diameter disk sample at zero external magnetic field , where the isk state is not only in an equilibrium , but is also the ground state @xcite ( global energy minimum ) . firstly , we compute all existing eigenmodes using the eigenvalue method and show the magnetisation dynamics of all identified eigenmodes in supplementary video 1 , and their schematic representations in supplementary section s3 . then , we excite the system using an in - plane excitation and show the spatially averaged ( sa ) and spatially resolved ( sr ) power spectral densities ( psds ) overlaid with the eigenvalue method resonant frequencies in figs . [ fig : isk_psd](a ) and [ fig : isk_psd](b ) , respectively . in these two psds , we identify five peaks ( a , d , e , f , and h ) and schematically represent their corresponding eigenmode magnetisation dynamics in fig . [ fig : isk_psd](e ) . the lowest frequency and the most dominant eigenmode a at @xmath119 consists of a dislocated incomplete skyrmion state core ( where @xmath120 ) revolving ( gyrating ) around its equilibrium position in the clockwise ( cw ) direction . schematically , we represent the skyrmionic state core with a circle symbol , together with a directed loop if it gyrates around its equilibrium position . consequently , we classify the eigenmode a as the gyrotropic ( translational ) mode . the eigenmode f at @xmath121 is the second most dominant eigenmode . its magnetisation dynamics consists of a ring contour , defined by the constant magnetisation @xmath10 component distribution , revolving around the sample centre in the counterclockwise ( ccw ) direction . this eigenmode is not gyrotropic because the isk state core remains at its equilibrium position . the eigenmode h at @xmath122 , present in both sa and sr psds , is composed of the isk state core together with two @xmath24 contour rings revolving in the cw direction . however , the inner contour revolves out - of - phase with respect to both the outer contour and the isk state core . because of that , we depict the inner contour ring using a dashed line and both the isk state core loop and the outer contour ring using a solid line as a way of visualising the mutually out - of - phase dynamics . the eigenmode d is present only in the sr psd at @xmath123 and consists of the isk state core and a contour ring revolving in the cw direction , but mutually out - of - phase . so far , all identified eigenmodes are lateral , but in the sa psd at @xmath124 , we also identify a very weak eigenmode e with radially symmetric magnetisation dynamics . although we expect that all eigenmodes present in the sa psd are also present in the sr psd , this is not the case for eigenmode e. we believe this is the case because this breathing eigenmode can be excited with an out - of - plane excitation , but emerges in simulations with an in - plane excitation due to the numerical noise , which is consistent with its small amplitude . this eigenmode , together with other breathing eigenmodes , will be discussed subsequently when we excite the isk state using an out - of - plane excitation . now , we perturb the incomplete skyrmion state from its equilibrium using an out - of - plane excitation and show the spatially averaged ( sa ) and spatially resolved ( sr ) power spectral densities ( psds ) in figs . [ fig : isk_psd](c ) and [ fig : isk_psd](d ) , respectively . using this excitation , we identify five eigenmodes ( a , b , c , e , and g ) and schematically represent their magnetisation dynamics in fig . [ fig : isk_psd](e ) . the most dominant eigenmode e is present in both sa and sr psds at @xmath123 . its magnetisation dynamics consists of a @xmath24 contour ring that shrinks and expands periodically , while the overall magnetisation configuration remains radially symmetric . because of that , we classify this eigenmode as a breathing mode . the second most dominant eigenmode is the gyrotropic mode a , which was also observed when the system was excited using an in - plane excitation , suggesting that it can be experimentally detected independent of the used excitation direction . the last eigenmode g present in both sa and sr psds at @xmath125 consists of two contour rings breathing mutually out - of - phase . more precisely , when one contour shrinks , another one expands , and vice versa . we schematically illustrate this out - of - phase breathing using dashed and solid lines depicting the contours . finally , the eigenmodes b and c , visible only in the sr psd at @xmath126 and @xmath127 , respectively , can be understood as a particular magnetisation configuration rotating in the sample in the cw direction without dislocating their core , as shown in fig . [ fig : isk_psd](e ) . after analysing the incomplete skyrmion power spectral densities for @xmath128 and @xmath129 , we now explore how the resonance frequencies depend on the external magnetic bias field @xmath130 and the disk sample diameter @xmath8 for both an in - plane and an out - of - plane excitation . firstly , we fix the disk sample diameter at @xmath118 and reduce the external magnetic field from @xmath131 to @xmath132 in steps of @xmath133 . more precisely , we start at @xmath134 initialising the system using incomplete skyrmion configuration with positive core orientation isk@xmath135 , relax the system to its equilibrium , and then run both eigenvalue and ringdown simulations . after that , we reduce the external magnetic field by @xmath133 , relax the system to its equilibrium using the relaxed ( equilibrium ) state from the previous simulation as initial configuration , and run dynamics simulations . we iterate until we reach @xmath132 . we show the resulting @xmath130-dependent power spectral density ( psd ) maps for an in - plane and an out - of - plane excitation in figs . [ fig : isk_sweep](a ) and [ fig : isk_sweep](b ) , respectively . in these @xmath130-dependent psd maps , a discontinuity in resonance frequencies at @xmath136 is evident . this is the case because for @xmath128 and @xmath137 , the isk state with positive ( @xmath120 ) core orientation ( isk@xmath135 ) is in an equilibrium . however , for @xmath138 , the isk@xmath135 is not in an equilibrium anymore and the isk state reverses its orientation to the negative ( @xmath139 ) direction ( isk@xmath140 ) in order to reduce its zeeman energy . this is consistent with the incomplete skyrmion hysteretic behaviour studies . @xcite secondly , we change the disk sample diameter @xmath8 between @xmath141 and @xmath142 in steps of @xmath143 at zero external magnetic bias field and show in figs . [ fig : isk_sweep](c ) and [ fig : isk_sweep](d ) the @xmath8-dependent psd maps for an in - plane and an out - of - plane excitation , respectively . in psd maps , we show the spatially resolved psds , computed using eq . ( [ eq : spatially_resolved ] ) , because in comparison to the spatially averaged psds , they exhibit more resonance peaks . @xcite we show two plots for every psd map : one for the complete studied frequency range ( @xmath144 ) and another plot in order to better resolve the low - frequency ( @xmath145 ) part of the psd map . in the case of an in - plane excitation , three lateral eigenmodes ( a , d , and f ) are visible in the @xmath130-dependent psd map , shown in fig . [ fig : isk_sweep](a ) , and in the isk@xmath135 range their frequencies nonlinearly and monotonically increase with @xmath130 . eigenmodes d and f are not as dominant as eigenmode a in the psd map below approximately @xmath146 , which results in the lack of sufficient contrast for them to be visible . now , if we change the direction of excitation , five eigenmodes ( a , b , c , e , and g ) are visible in the @xmath130-dependent psd map , as shown in fig . [ fig : isk_sweep](b ) . in the isk@xmath135 range , eigenmodes a , b , and c are visible only between @xmath147 and approximately @xmath148 . in fig . [ fig : isk_psd](d ) at zero external magnetic field , eigenmodes a , b , and c have very similar frequencies which makes it difficult for experimentalists to determine which eigenmode the resonance frequency they measure belongs to . from the out - of - plane @xmath130-dependent psd map in fig . [ fig : isk_sweep](b ) , we observe that this difficulty can be resolved by reducing the external magnetic field towards the switching field . more precisely , the frequencies of eigenmodes a and b both decrease , but only the frequency of eigenmode a approaches zero . on the contrary , the frequency of eigenmode c increases by reducing @xmath130 . also , the separation between a , b , and c peaks in the psd increases , by either increasing or decreasing external magnetic field value . we show the dependence of their frequencies at high external magnetic fields in supplementary section s2 using the eigenvalue method and demonstrate that no eigenmode crossing occurs . in addition , as we previously discussed , by changing the excitation to an in - plane direction , only eigenmode a would be present . the eigenmodes e and g are visible in the whole examined range of @xmath130 , and their frequencies increase nonlinearly and monotonically with external magnetic bias field . interestingly , the frequency of eigenmode a approaches zero near the switching field @xmath149 , suggesting that this gyrotropic eigenmode might be the reversal ( zero ) mode of the incomplete skyrmion state in the studied sample . by varying the disk sample diameter @xmath8 , for an in - plane excitation , we observe the gyrotropic eigenmode a frequency increasing between @xmath141 and @xmath150 ( where it reaches its maximum ) , and then decreasing with @xmath8 , as shown in fig . [ fig : isk_sweep](c ) . another visible eigenmode in the psd map above approximately @xmath151 , for an in - plane excitation , is the eigenmode f whose frequency monotonically decreases with @xmath8 . in the case of an out - of - plane excitation , we identify seven ( a , b , c , e , g , x@xmath152 , and x@xmath0 ) eigenmodes in the psd map shown in fig . [ fig : isk_sweep](d ) . the magnetisation dynamics of all these eigenmodes was discussed before , except x@xmath152 and x@xmath0 , because they were not present in the psds below the maximum studied frequency @xmath4 for @xmath128 . the eigenmode a frequency dependence is the same as for an in - plane excitation and another six eigenmodes ( b , c , e , g , x@xmath152 , and x@xmath0 ) frequencies monotonically decrease with the disk sample diameter . in this section , we explore the dynamics of an isolated skyrmion ( sk ) state , for which the magnetisation @xmath10 component covers the @xmath114 $ ] range once ( plus the additional magnetisation tilting at the boundaries due to the specific boundary conditions @xcite ) along the disk sample diameter , and consequently , the scalar value @xmath110 is in the @xmath153 $ ] range . @xcite the sk state is in an equilibrium @xcite for @xmath154 and @xmath155 . we study the sk state dynamics for a @xmath156 diameter disk sample at zero external magnetic bias field , where the sk state is not only in an equilibrium , but is also the ground state . @xcite after we perform the eigenvalue method computations , we excite the system using an in - plane excitation and show the spatially averaged ( sa ) and spatially resolved ( sr ) power spectral densities ( psds ) in figs . [ fig : sk_psd](a ) and [ fig : sk_psd](b ) , respectively . the magnetisation dynamics of all existing eigenmodes computed using the eigenvalue method are shown in supplementary videos 2 and 3 , and their schematic representations in supplementary section s3 . in both sa and sr psds , we identify nine peaks ( eigenmodes a , b , d , e , g , i , j , l , and m ) , and show their schematic representations in fig . [ fig : sk_psd](e ) . the lowest frequency eigenmode at @xmath157 is the gyrotropic eigenmode a. its magnetisation dynamics consists of a dislocated sk state core ( where @xmath158 ) gyrating around its equilibrium position in the ccw direction . in both psds , the eigenmode b at @xmath159 is the most dominant one , and consists of a contour ring ( defined as a constant magnetisation @xmath10 component distribution ) revolving in the cw direction . the eigenmode d at @xmath160 is composed of both the sk state core and a magnetisation contour ring revolving in the ccw direction , but mutually out - of - phase . at @xmath161 , we identify an eigenmode e with two magnetisation contour rings revolving mutually out - of - phase in the cw direction . similarly , the eigenmode g at @xmath162 also consists of two contour rings revolving mutually out - of - phase , but now in the ccw direction . the four remaining eigenmodes ( i , j , l , and m ) are significantly weaker in both psds when compared to the power of previously discussed eigenmodes . their magnetisation dynamics , shown in fig . [ fig : sk_psd](e ) , are all lateral and contain different combinations of revolving contour rings and revolving sk state core . now , we change the excitation to be in the out - of - plane direction . the computed spatially averaged and spatially resolved power spectral densities , overlaid with the resonance frequencies obtained from the eigenvalue method , are shown in figs . [ fig : sk_psd](c ) and [ fig : sk_psd](d ) , respectively . in this case , we observe five peaks ( eigenmodes a , c , f , h , and k ) in both psds , and a significantly weaker lateral eigenmode g ( previously discussed ) in sa psd . we show the schematic representation of their magnetisation dynamics in fig . [ fig : sk_psd](e ) . similar to the incomplete skyrmion state , the gyrotropic eigenmode a can also be activated with an out - of - plane excitation . the lowest frequency breathing eigenmode c at @xmath163 consists of a single contour ring that shrinks and expands periodically . an eigenmode f at @xmath164 is composed of two contour rings breathing mutually out - of - phase . similar to the eigenmode c , the eigenmode h at @xmath165 consists of a single breathing contour , but now with a smaller contour diameter ( smaller @xmath24 ) . at @xmath166 , we identify the highest frequency breathing eigenmode k in the studied frequency range , which contains three breathing contours , where the inner and the outer contours breathe out - of - phase with respect to the middle one . so far , we analysed the isolated skyrmion state dynamics for @xmath167 and @xmath129 . now , our objective is to determine how the resonance frequencies depend on disk sample diameter and external magnetic field . we compute the @xmath130-dependent psd map in two parts in order to obtain how the resonance frequencies change in the entire range of external magnetic field values where the isolated skyrmion state with negative core orientation ( sk@xmath140 ) is in equilibrium . more precisely , the system initialised with sk@xmath140 at high external magnetic fields relaxes to the incomplete skyrmion state with positive core orientation @xcite isk@xmath135 . consequently , if we keep reducing @xmath130 and use the equilibrium state from previous simulation iteration as initial state , we could not reach the sk@xmath140 state . therefore , we firstly fix the disk sample diameter to @xmath156 , set @xmath129 , initialise the system using sk@xmath140 configuration , relax the system , and run dynamics simulations . then we increase the external magnetic field by @xmath133 using the equilibrium state from previous simulation as the initial state at new value of external magnetic field . we iterate this until we reach @xmath131 . similarly , starting from zero external magnetic field , we reduce @xmath168 in steps of @xmath133 , until @xmath169 is reached . we show the @xmath130-dependent psd maps for an in - plane and an out - of - plane excitation in fig . [ fig : sk_sweep](a ) and [ fig : sk_sweep](b ) , respectively . in these psd maps , two discontinuities in resonant frequencies at @xmath170 and @xmath171 are present . the first discontinuity occurs because decreasing @xmath130 causes the sk state core with negative ( @xmath158 ) orientation ( sk@xmath140 ) to switch to the positive ( @xmath120 ) direction ( sk@xmath135 ) at the switching field @xmath172 . on the other hand , the discontinuity at @xmath171 occurs because , above this value , the sk@xmath140 is not in an equilibrium anymore and the system relaxes to the incomplete skyrmion state with positive core orientation ( isk@xmath135 ) . secondly , at @xmath129 , we vary @xmath8 between @xmath141 and @xmath142 in steps of @xmath143 and show the @xmath8-dependent psd maps in figs . [ fig : sk_sweep](c ) and [ fig : sk_sweep](d ) for an in - plane and an out - of - plane excitation , respectively . now , a single discontinuity in resonance frequencies is present at @xmath173 , below which the disk sample diameter becomes too small to accommodate the full magnetisation rotation and the isk state emerges . the external magnetic bias field and disk sample diameter values at which the discontinuities occur are consistent with equilibrium @xcite and hysteretic behaviour @xcite studies . for an in - plane excitation , in the @xmath130-dependent psd map , shown in fig . [ fig : sk_sweep](a ) , five previously discussed eigenmodes ( a , b , d , e , and g ) are visible in the @xmath130 range where the sk@xmath140 state is in an equilibrium . the frequency of gyrotropic eigenmode a firstly increases , reaches its maximum at approximately @xmath174 , and then decreases down to its minimum at approximately @xmath175 , after which it keeps increasing with @xmath130 . in comparison to the other eigenmodes , its frequency varies over a much smaller range ( less than @xmath176 ) over the entire @xmath130 range where the sk@xmath140 state is in an equilibrium . similar to the incomplete skyrmion state , the frequency of gyrotropic eigenmode a approaches zero near the switching field @xmath172 , suggesting that this eigenmode might govern the isolated skyrmion reversal process . the eigenmode b frequency increases approximately linearly up to @xmath177 , after which it continues increasing nonlinearly . the frequency of eigenmode d , firstly decreases , reaches its minimum at approximately @xmath178 , and then continues increasing nonlinearly with @xmath130 . the frequencies of eigenmodes e and g exhibit more complicated behaviour where two extremes ( maximum and minimum ) are present in their @xmath130-dependences . when an out - of - plane excitation is used , we observe five previously discussed eigenmodes ( a , c , f , h , and j ) in the @xmath130-dependent psd map , shown in fig . [ fig : sk_sweep](b ) . the eigenmode a now becomes invisible in the psd map below @xmath179 . the breathing eigenmode c frequency increases monotonically over the entire sk@xmath140 field range . the frequency dependences of eigenmodes f , h , and j , exhibit more complicated behaviour having both local maximum and minimum in their @xmath130-dependences . in the @xmath8-dependent psd map , shown in fig . [ fig : sk_sweep](c ) , obtained when an in - plane excitation is used , five previously discussed eigenmodes ( a , b , d , e , and g ) are present . in contrast to the frequencies of eigenmodes d , e , and g that monotonically decrease with @xmath8 over a wide range of frequencies , the eigenmodes a and b frequencies vary in a much smaller ( less than @xmath176 ) range over entire studied @xmath8 range . eigenmodes d , e , and g become invisible in the psd map below approximately @xmath180 . in fig . [ fig : sk_sweep](d ) , we show the @xmath8-dependent psd map for an out - of - plane excitation , where five eigenmodes ( a , c , f , h , and j ) are visible . similar to the eigenmodes a and b , the lowest frequency breathing eigenmode frequency changes over a much smaller range than the frequencies of eigenmodes f , h , and j , when the disk sample diameter is changed . one of the challenges in the study of skyrmionic states in confined helimagnetic nanostructures is the detection of what state emerged in the studied sample . in this subsection we discuss how measuring resonance frequencies can contribute to the identification of the emerged state . previously , in sections [ sec : isk ] and [ sec : sk ] , we studied the dynamics of both incomplete skyrmion ( isk ) and isolated skyrmion ( sk ) states in disk samples with diameters for which these states were the ground states . now , we compare the power spectral densities ( psds ) of isk and sk states in a @xmath181 diameter disk sample with @xmath7 thickness at different external magnetic field values @xmath168 ( between @xmath182 and @xmath183 in steps of @xmath184 ) . in this sample size and at all simulated external magnetic field values , both isk and sk states are in equilibrium . more specifically , the sk state is metastable and the isk state is the ground state . @xcite we show the comparison of spatially resolved isk and sk psds at different external magnetic field values for an in - plane and an out - of - plane excitation in fig . [ fig : comparison ] . because in a @xmath181 diameter disk sample there are no dominant isk and sk eigenmodes that can be easily detected in experiments above @xmath185 , we now limit our discussion of psds below @xmath185 in order to better resolve the key differences , that can contribute to the identification of the present state . firstly , in the case of an in - plane excitation ( left column in fig . [ fig : comparison ] ) , the frequency of isk gyrotropic eigenmode a ( the lowest frequency isk eigenmode ) , increases with @xmath130 . on the contrary , the sk gyrotropic eigenmode a ( again the lowest frequency sk eigenmode ) frequency remains approximately the same . furthermore , by increasing the external magnetic field the sk eigenmode b frequency increases , and consequently , the frequency difference between two lowest frequency sk eigenmodes @xmath186 increases in a wide range of frequencies . in contrast , the frequencies of two lowest frequency isk eigenmodes a and d both increase with @xmath130 , so that the frequency difference @xmath187 between them changes over a small range of frequencies ( remains approximately the same ) . however , at low values of external magnetic field , it could be difficult to measure the isk eigenmode d due to its relatively small amplitude . in that case , between @xmath184 and @xmath188 , the frequency of dominant isk eigenmode f does not change , so the @xmath189 difference reduces with @xmath130 for about @xmath190 . when we excite the system using an out - of - plane excitation ( right column in fig . [ fig : comparison ] ) , at @xmath129 , several resonance frequencies below @xmath190 are present , which does not allow a clear identification of the emerged state by measuring resonance frequencies in that region . however , by increasing the external magnetic field , the low frequency part of psds simplifies . more specifically , the sk eigenmode a frequency again does not change , while the sk breathing eigenmode c , and therefore the difference @xmath191 , increase with @xmath130 . in addition , for a sk state above @xmath184 , the frequency of eigenmode f remains approximately the same , and therefore , the difference @xmath192 decreases with @xmath130 . on the contrary , isk eigenmodes a , b , and c disappear from the psd after @xmath193 , whereas the frequency difference @xmath194 between two most dominant isk eigenmodes e and g remains approximately the same , since their frequencies both increase . the dependences of resonant frequencies in this sample with @xmath195 are in a good agreement with the psd maps shown in fig . [ fig : isk_sweep](a ) and fig . [ fig : sk_sweep](a ) and eigenvalue computed results in supplementary section s2 . this suggests that these identification differences can probably be applied to different sample sizes . at @xmath196 , we approach the transition from sk to isk state and additional peaks in sk state psds , shown in fig . [ fig : comparison ] , occur . in the previous analysis of skyrmionic states dynamics , we intentionally used the small gilbert damping value @xmath197 as used in other eigenmode studies , @xcite in order to allow enough separation between peaks in the power spectral densities ( enabled by the reduced linewidth ) and identify all eigenmodes that can be excited using a particular experimentally feasible excitation . however , in experiments , which eigenmodes can be observed strongly depends on the real value of gilbert damping . therefore , in this section , we measure the fege gilbert damping value @xmath198 and repeat our simulations in order to determine what eigenmodes are expected to be experimentally observed in helimagnetic fege confined nanostructures . we perform the ferromagnetic resonance measurements in a fege thin film with @xmath199 thickness , grown on the si substrate in the ( 111 ) direction and capped with a @xmath200 thin ge layer . @xcite we show the linewidth @xmath201 ( half width at half maximum ) measurement points at different resonance frequencies @xmath74 , together with a first degree polynomial fit in fig . [ fig : fege_damping_measurements ] . the polynomial fit allows us to decompose the @xmath201 dependence into a frequency independent inhomogeneously - broadened component @xmath202 and an intrinsic damping - related part : @xcite @xmath203 where @xmath198 is the gilbert damping and @xmath204 is the gyromagnetic ratio . from the slope of the polynomial fit and using the frequency - dependent term that reflects the `` viscous '' damping of the precessive magnetisation motion associated with the fmr , we find @xmath205 . the linewidth @xmath201 ( half width at half maximum ) measurement points at different resonance frequencies @xmath74 for a fege thin film , together with a first degree polynomial fit from which the gilbert damping was extracted . ] the comparison of power spectral densities computed using three - dimensional and two - dimensional models in absence of demagnetisation energy contribution with the psd obtained using a full simulation model for an isolated skyrmion state in the case of ( a ) in - plane and ( b ) out - of - plane excitation . simulated sample is a @xmath156 diameter disk with @xmath7 thickness at zero external magnetic field . ] now , we use the measured @xmath206 and repeat the ringdown simulations for the two skyrmionic ground states that can exist in the studied system . we show the spatially resolved power spectral density of an incomplete skyrmion state in a @xmath118 diameter disk sample at zero external magnetic bias field for an in - plane and an out - of - plane excitation in figs . [ fig : fege_damping_psd](a ) and [ fig : fege_damping_psd](b ) , respectively . we observe that , when the system is excited using an in - plane excitation , only the gyrotropic eigenmode a is present in the psd . on the other hand , for an out - of - plane excitation , we identify two eigenmodes in the psd shown in fig . [ fig : fege_damping_psd](b ) . the first one is the gyrotropic eigenmode a , which is also present in the in - plane psd , and another one is the lowest frequency breathing eigenmode e. the psds of the isolated skyrmion state in a @xmath156 diameter disk sample at zero external magnetic bias field are shown in figs . [ fig : fege_damping_psd](c ) and [ fig : fege_damping_psd](d ) for an in - plane and an out - of - plane excitation , respectively . now , only the three lowest frequency isolated skyrmion eigenmodes are present . the gyrotropic eigenmode a and eigenmode b can be identified when the system is excited using an in - plane excitation . on the other hand , for the out - of - plane excitation , only the breathing eigenmode c is present . usually , in the simulations of skyrmionic states dynamics in helimagnetic samples , for simplicity , the demagnetisation energy contribution is neglected and/or a helimagnetic thin film sample is modelled using a two - dimensional mesh . it has been shown that the demagnetisation energy contribution @xcite and the magnetisation variation in the out - of - film direction @xcite radically change the energy landscape . consequently , using these assumptions when the static properties of skyrmionic states are explored is not justified . because of that , in this section , we investigate how these two assumptions affect the dynamics of the isolated skyrmion ( sk ) state in studied helimagnetic nanostructure . firstly , we repeat the isolated skyrmion state simulations in a @xmath156 diameter disk sample at zero external magnetic bias field , but this time we set the demagnetisation energy contribution @xmath207 in eq . ( [ eq : total_energy ] ) artificially to zero . secondly , again in the absence of demagnetisation energy contribution , we simulate the sk state dynamics under the same conditions , but this time using a two - dimensional mesh to model a thin film sample ( i.e. not allowing the magnetisation variation in the out - of - film direction ) . we show the comparison of power spectral densities computed using three - dimensional and two - dimensional models in absence of demagnetisation energy contribution with the one computed using a full model in figs . [ fig : sk_nd2d_psd](a ) and [ fig : sk_nd2d_psd](b ) , for an in - plane and an out - of - plane excitations , respectively . we observe that although the magnetisation dynamics of identified eigenmodes do not change significantly , the resonance frequencies of some eigenmodes change substantially . in the 3d simulations in absence of demagnetisation energy , while the frequency of eigenmode d remains approximately the same , the frequencies of eigenmodes a and b increase by 71% and 18% , respectively . on the other hand , the frequencies of eigenmodes e and g decrease by 14% and 21% , respectively . furthermore , power spectral densities in fig . [ fig : sk_nd2d_psd](b ) , computed for the out - of - plane excitation , show that the frequency of breathing eigenmode c increases by 17% , whereas the frequency of eigenmode f decreases by 34% . if the thin film sample is modelled using a 2d mesh , which does not allow the magnetisation to vary in the @xmath10 direction , the frequencies of lateral eigenmodes a , b , and g do not change significantly in comparison to the 3d model in absence of demagnetisation energy contribution . although the frequency of eigenmode d does not change in the 3d simplified ( @xmath208 ) model , neglecting the sample thickness , increases its frequency by 20% . the frequency of eigenmode e increases so that it is approximately the same as in the full 3d model . in comparison to the 3d simplified model , the frequency of breathing eigenmodes c and f further increase by 19% and 7% , respectively . in the low frequency region of fig . [ fig : sk_nd2d_psd](b ) we observe several eigenmodes that are not present in the three - dimensional model . although there are theoretically less eigenmodes in a two - dimesional sample ( the number of existing eigenmodes equals to the number of degrees of freedom in the system ) , we do not know at what frequencies they will occur . we believe the reason for this is that although the state is the same , its dynamics quilitatively changes at low frequencies due to the missing thickness dimension . we also observe eigenmodes computed using the eigenvalue method that agree with the peaks in psd shown in fig . [ fig : sk_nd2d_psd](b ) . using the eigenvalue method , we computed all eigenmodes with frequencies below @xmath4 for the incomplete skyrmion , isolated skyrmion , and target states in helimagnetic thin film disk samples at zero external magnetic field . because which eigenmodes are present in the power spectral density strongly depends on the excitation used to perturb the system from its equilibrium state , we performed the ringdown simulations using two different experimentally feasible excitations ( in - plane and out - of - plane ) . we demonstrated that in all three simulated states , two lateral and one breathing low - frequency eigenmodes exist , as previously demonstrated in two - dimensional skyrmion lattice simulations @xcite and microwave absorption measurements in bulk helimagnetic materials @xcite . however , only one lateral eigenmode is gyrotropic , with the skyrmionic state core gyrating around its equilibrium position . the other lateral eigenmode we observed is not gyrotropic because it consists of a single contour ring ( defined by the magnetisation @xmath10 component distribution ) revolving around the static skyrmionic state core . the existence of only one gyrotropic eigenmode is in accordance with the recent analytic ( rigid skyrmion two - dimensional model ) findings by guslienko and gareeva , @xcite but in contrast to magnetic bubbles where two gyrotropic eigenmodes were found . @xcite because the two gyrotropic eigenmodes with opposite gyration direction in a magnetic bubble imply it possesses mass , our findings suggest that the confined dmi - induced skyrmionic states in the studied system are massless . the low - frequency breathing eigenmode we observe , where a single magnetisation @xmath10 component contour ring shrinks and expands periodically , is in accordance with findings in ref . . for the incomplete skyrmion and the isolated skyrmion states we found that the resonance frequencies depend nonlinearly on both the disk sample diameter and the external magnetic bias field . we observed that the frequency of the gyrotropic eigenmode approaches zero near the switching field @xmath147 ( where the reversal of skyrmionic state core occurs ) for both incomplete skyrmion and isolated skyrmion states , suggesting that this eigenmode might be the reversal mode of studied skyrmionic states . we find that when the skyrmionic state core orientation reverses , the revolving direction of all lateral eigenmodes changes , which confirms that the revolving direction depends on the direction of the gyrovector as shown in ref . . after we identified all existing eigenmodes of isk and sk ground states , we compared their psds in the same sample at different external magnetic field values . we identified several characteristics that can contribute to the experimental identification of the state that emerged in the sample by measuring the resonance frequencies . in the identification and analysis of eigenmodes , we used a small gilbert damping value in order to provide enough separation between peaks in the psd . however , which eigenmodes are expected to be observed in experiments strongly depends on the real gilbert damping value @xmath198 . therefore we measured @xmath198 in the fege thin film , and carried out ringdown simulations with this @xmath198 . we showed that for the incomplete skyrmion , two eigenmodes ( gyrotropic and breathing ) are present in the psd computed using an out - of - plane excitation , whereas only the gyrotropic eigenmode is present in the psd computed using an in - plane excitation . in the isolated skyrmion case , two lateral eigenmodes are present in the psd obtained using an in - plane excitation , whereas a single breathing eigenmode is present in the psd computed after using the out - of - plane excitation . our simulations took into account the demagnetisation energy contribution , which is usually neglected for simplicity in both analytic and simulation works . to explore the importance of model assumptions , we carried out further systematic simulation studies in which we set the demagnetisation energy contribution artificially to zero . we also repeated the simulations under the same conditions on 3d and 2d meshes ( with and without permissible magnetisation variation in the out - of - film direction , respectively ) . we found that although the magnetisation dynamics of eigenmodes does not change significantly , their frequencies change substantially . this suggests that ignoring the demagnetisation energy contribution or approximating a thin film helimagnetic sample using a two - dimensional mesh is not always justified . this work provides a systematic dynamics study of skyrmionic states in confined helimagnetic nanostructures . we report all eigenmodes present in the sample as well as which eigenmodes can be observed using particular experimentally feasible excitations . apart from contributing to fundamental physics , this work could support experimentalists to determine what magnetisation configuration is present in the confined helimagnetic sample by measuring ferromagnetic resonance spectra . this work was financially supported by the epsrc doctoral training centre ( dtc ) grant ep / g03690x/1 , opendreamkit horizon 2020 european research infrastructure project ( 676541 ) , and the epsrc programme grant on skyrmionics ( ep / n032128/1 ) . d.c .- o . acknowledges the financial support from conicyt chilean scholarship programme becas chile ( 72140061 ) . c.c . acknowledges the support from a junior research fellowship at gonville and caius college . the scholarship of c.s.s was supported by the stfc rutherford appleton laboratory and the hitachi cambridge laboratory . c.h.m . acknowledges support by epsrc through the standard research grant ep / j007110/1 . we also acknowledge the use of the iridis high performance computing facility , and associated support services at the university of southampton , in the completion of this work . we thank robert l. stamps for useful discussions . 63ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1016/0022 - 3697(58)90076 - 3 [ * * , ( ) ] link:\doibase 10.1103/physrev.120.91 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.44.1538 [ * * , ( ) ] link:\doibase 10.1016/s0304 - 8853(97)01044 - 5 [ * * , ( ) ] link:\doibase 10.1038/nphys2045 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.134425 [ * * , ( ) ] link:\doibase 10.1126/science.1240573 [ * * , ( ) ] link:\doibase 10.1126/science.1195709 [ * * , ( ) ] link:\doibase 10.1038/ncomms1990 [ * * , ( ) ] link:\doibase 10.1088/0022 - 3727/44/39/392001 [ * * , ( ) ] link:\doibase 10.1038/nnano.2013.29 [ * * , ( ) ] link:\doibase 10.1038/srep09400 [ * * , ( ) ] http://www.jetp.ac.ru/cgi-bin/e/index/e/68/1/p101?a=list [ * * , ( ) ] link:\doibase 10.1016/s0304 - 8853(98)01038 - 5 [ * * , ( ) ] link:\doibase 10.1038/nature05056 [ * * , ( ) ] link:\doibase 10.1126/science.1166767 [ * * , ( ) ] link:\doibase 10.1038/nmat2916 [ * * , ( ) ] link:\doibase 10.1038/nature09124 [ * * , ( ) ] link:\doibase 10.1126/science.1214143 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.113.077202 [ * * , ( ) ] link:\doibase 10.1038/srep17137 [ * * , ( ) ] link:\doibase 10.1103/physrevb.88.184422 [ * * , ( ) ] link:\doibase 10.1103/physrevb.90.064410 [ * * , ( ) ] link:\doibase 10.1002/pssr.201510419 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.017601 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.109.037603 [ * * , ( ) ] link:\doibase 10.1038/ncomms3391 [ * * , ( ) ] link:\doibase 10.1038/nmat4223 [ * * , ( ) ] link:\doibase 10.1109/20.951053 [ * * , ( ) ] link:\doibase 10.1103/physrevb.74.214406 [ * * , ( ) ] link:\doibase 10.1103/physrevb.76.104426 [ * * , ( ) ] link:\doibase 10.1109/lmag.2015.2413758 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.109.217201 [ * * , ( ) ] link:\doibase 10.1103/physrevb.79.224429 [ * * , ( ) ] link:\doibase 10.1038/nphys3234 [ * * , ( ) ] link:\doibase 10.1109/lmag.2016.2616333 [ * * , ( ) ] link:\doibase 10.1016/j.jcp.2009.05.026 [ * * , ( ) ] link:\doibase 10.1063/1.1852191 [ * * , ( ) ] link:\doibase 10.1103/physrevb.87.094424 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1109/tmag.2004.836740 [ * * , ( ) ] link:\doibase 10.1109/tmag.2007.893843 [ * * , ( ) ] link:\doibase 10.1007/978 - 3 - 642 - 23099 - 8 [ _ _ ] , edited by , , and , , vol . ( , , ) p. link:\doibase 10.1145/1089014.1089020 [ * * , ( ) ] link:\doibase 10.1063/1.4822377 [ * * , ( ) ] link:\doibase 10.1109/mcse.2007.55 [ * * , ( ) ] @noop _ _ ( , ) http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.48.6660{&}rep=rep1{&}type=pdf[__ ] ( ) pp . link:\doibase 10.1109/mcse.2007.58 [ * * , ( ) ] link:\doibase 10.1109/mcse.2011.37 [ * * , ( ) ] link:\doibase 10.1109/tmag.2012.2206820 [ * * , ( ) ] link:\doibase 10.1016/j.jmmm.2016.08.009 [ * * , ( ) ] @noop _ _ ( , ) p. link:\doibase 10.1038/nnano.2013.210 [ * * , ( ) ] link:\doibase 10.1103/physrevb.87.014401 [ * * , ( ) ] link:\doibase 10.1209/0295 - 5075/101/37001 [ * * , ( ) ] link:\doibase 10.1063/1.4962726 [ * * , ( ) ] link:\doibase 10.1103/physrevb.92.144402 [ * * , ( ) ] link:\doibase 10.1109/tmag.2003.815450 [ * * , ( ) ] link:\doibase 10.1063/1.2197087 [ * * , ( ) ] link:\doibase 10.1051/epjconf/20147502002 [ * * , ( ) ] link:\doibase 10.1103/physrevb.92.020403 [ * * , ( ) ] link:\doibase 10.1063/1.4914496 [ * * , ( ) ]

we review the physics of polarization observables in high energy reactions in general and discuss the status of _ nlo _ one loop corrections to these observables in specific . many high order radiative corrections exists for rates but not many _ nlo _ radiative corrections exist for polarization observables . the radiative correction calculations for polarization observables are somewhat more complicated than those for rates . they tend to be smaller than those for the rates . in most of the examples we discuss we include mass effects which significantly complicate the radiative correction calculations . we elaborate a general scheme which allows one to enumerate the number of independent density matrix elements in a reaction and provide explicit examples of angular decay distributions in self analyzing decays that allow one to experimentally measure the density matrix elements . we provide examples of reactions where certain density matrix elements are only populated at _ nlo _ or by mass effects . in our discussion we concentrate on semileptonic bottom and top quark decays which are linked to leptonic and decays through a fierz transformation . mz - th/03 - 09 + hep - ph/0306082 + june 2003 + * one loop corrections to polarization observables * + j.g . krner and m.c . mauser + institut fr physik , johannes gutenberg universitt + staudinger weg 7 , d55099 mainz , germany _ invited lecture at the international school `` heavy quark physics '' , + dubna , russia , 27 may 5 june 2002 + to appear in the proceedings ( lecture notes in physics ) _

hep-ph0306082

most of the _ nlo _ radiative corrections to rates have been done quite some years ago . in fact , two loop _ nnlo _ calculations for rates are now becoming quite standard . in @xmath3annihilation one is even pushing hard to determine the _ nnnnlo _ corrections to the @xmath4 ratio @xcite . contrary to this , many _ nlo _ radiative correction calculations to polarization observables involving also massive quarks have only been done in the last few years . it is clear that higher order radiative corrections to unpolarized observables will always be at the center of attention because data on unpolarized observables are as a rule much more accurate than data for polarized observables . nevertheless , as experimental data on polarization observables has been accumulating over the years there is an evolving need for radiative corrections to polarization observables . one of the reasons that polarized radiative correction were lagging behind unpolarized radiative corrections is that the computational effort in the calculation of radiative corrections to polarization observables is larger than that for rates . for once , one can not sum over the spins of intermediate states whose polarization one wants to calculate . one therefore can not make use of the powerful unitarity method to calculate radiative corrections from the absorptive parts of higher order loop graphs . further , the definition of polarization observables brings in extra momentum factors in the integrands of the requisite phase space integrals which makes life more difficult . this is particularly true if the masses of particles in the process can not be neglected . for example , in the process @xmath5 the longitudinal component of the polarization vector of the top quark along the @xmath6direction is given by @xmath7 where the denominator factor @xmath8 comes in for normalization reasons ( @xmath9 is the momentum of the @xmath10 ) . it is quite clear that such square root factors lead to nontrivial complications in the phase space integrations . similar square root factors appear when projecting onto the polarization states of the @xmath10 . inclusion of mass effects as e.g. in the semileptonic decay @xmath11 ( @xmath12 ) or in the leptonic decay @xmath13 ( @xmath14 ) render the analytical calculations considerably more complicated . the villain is the kllen function @xmath15 which is brought in by extra three momentum factors . in the case that e.g. @xmath16 the kllen function simplifies to @xmath17 which leads to an enormous simplification in the phase space integrals . physically speaking mass effects become large in regions of phase space where the massive particles become nonrelativistic . for example , in the leptonic decay of the muon @xmath18 , the mass of the electron can not be neglected in the threshold region where the energy of the electron is small . this is illustrated in fig . [ plot1 ] where the longitudinal polarization of the electron is plotted against the scaled energy @xmath19 of the electron @xcite . we have chosen a logarithmic scale for @xmath20 in order to enhance the threshold region . in the threshold region the longitudinal polarization deviates considerably from the naive value @xmath21 . the radiative corrections in the threshold region can be seen to be quite large . all our results are given in closed analytic form . in the days of fast numerical computers one may rightfully ask what the advantage of having closed form expressions is . one of the advantages is that all requisite mass and momentum limits can be taken in analytic form . thus the analytic results can be subjected to tests against known limiting cases providing for the necessary checks of the full results . a second issue is user friendliness . the analytic expressions are simple enough to be incorporated in numerical programs by the prospective user . parameter values such as masses and coupling constants can be varied at will by the user in his own program without having to refer to numerical programs written by others . we have checked that our analytical formulas are numerically stable even in the small and large mass limits . _ nlo _ corrections to rates can be quite large . _ nlo _ corrections to polarization observables have a tendency to be somewhat smaller . the reason is that polarization observables are normalized quantities . they are normalized with regard to the total rate . the numerators and denominators in the relevant polarization expressions tend to go in the same direction . the reason is that the _ nlo _ numerators and denominators are dominated by the soft gluon or soft photon contributions which are universal in the sense that they multiply the relevant born term expressions and thus cancel out in the ratio . for example , for @xmath5 the @xmath22 correction to the rate is @xmath23 and @xmath24 for the polarization observables @xcite . at _ nlo _ some density matrix elements become populated which vanish at leading order . an example is again the decay @xmath5 where the rate into the right handed @xmath10 vanishes ( @xmath25 ) for @xmath26 at _ lo _ but where @xmath27 at _ nlo_. generally speaking , standard model radiative corrections can populate density matrix elements elements that vanish at leading order . consequently radiative corrections change angular decay distributions as do new physics effects . the lesson to be learned is clear . before ascribing a given polarization effect to new physics one has to make sure that it does not result from radiative corrections of old physics . in the above example a non zero @xmath27 could result from an admixture of a right handed charged current ( new physics ) or from radiative corrections . only the precise knowledge of the magnitude of the radiative corrections allows one to exclude or conclude for new physics effects . in this review we will mostly be concerned with radiative corrections to the current induced transitions @xmath28 , @xmath29 and @xmath30 . it should be clear that the additional gluons ( or photons ) in radiative correction calculations only couple to the @xmath31 ( or @xmath30 ) side of the relevant semileptonic ( or leptonic ) transitions . this implies that the structure of the _ nlo _ radiative corrections is the same in the two classes of processes . the current induced transition @xmath30 is not in the standard model form but in the charge retention form . however , the charge retention form can be linked to the standard model form by the remarkable property that the @xmath32 interaction is an eigenvector under fierz crossing ( see sect . 3 ) . one therefore has @xmath33 [ \bar{e } \gamma_{\alpha } ( { 1 \hspace{-0.95 mm } { \rm l}}- \gamma_5 ) \nu_{e } ] + \mbox{h.c . } \\[2 mm ] & = & \frac{g_f}{\sqrt{2 } } [ \bar{\nu}_{\mu } \gamma^{\alpha } ( { 1 \hspace{-0.95 mm } { \rm l}}- \gamma_5 ) \nu_{e } ] [ \bar{e } \gamma_{\alpha } ( { 1 \hspace{-0.95 mm } { \rm l}}- \gamma_5 ) \mu ] + \mbox{h.c . } \end{aligned}\ ] ] all of the above three transitions are therefore governed by the same matrix elements . in particular the _ nlo _ qed and qcd radiative corrections have the same structure for all three transitions . in table 1 we provide a list of the processes for which the radiative corrections to the rates and the polarization observables can all be calculated from essentially the same set of _ nlo _ matrix elements . in this review we shall only present a few sample results from the processes in table [ tab1 ] because the results are to numerous to fit into a review of the present size . instead we attempt to share our insights into the general features of spin physics and illustrate these with sample results taken from the processes in table [ tab1 ] . .list of processes based on the same current induced matrix elements [ cols= " < , < , < " , ] [ tab4 ] it should be clear that one needs to do polarization measurements in order to disentangle the full structure of particle interactions . polarization measurements are particularly simple when the particle whose polarization one wants to measure decays . the angular decay distribution of the decay products reveals information on the state of polarization of the decaying particle . the information contained in the angular decay distribution is maximal when the particle decay is weak . the fact that the angular decay distribution reveals information on the polarization of the decaying particle is sometimes referred to as that the particle decay is self analyzing . there are principally two ways to obtain angular decay distributions which we will refer to as the non covariant and the covariant methods . in the non covariant method one makes use of rotation matrices whereas in the covariant method one evaluates scalar products of four vectors involving momenta and spin four vectors in given reference frames . we shall discuss one example each of the two methods . as an example for the non covariant method we write down the angular decay distribution of polarized top decay @xmath34 followed by @xmath35 . the angular decay distribution can be obtained from the master formula @xmath36 where @xmath37 is the density matrix of the top quark which reads @xmath38 @xmath39 is the magnitude of the polarization of the top quark . the @xmath40 are helicity matrix elements of the hadronic structure function @xmath41 . the sum in ( [ masterformula ] ) extends over all values of @xmath42 and @xmath43 compatible with the constraint @xmath44 . the second lower index in the small wigner @xmath45function @xmath46 is fixed at @xmath47 for zero mass leptons because the total @xmath48quantum number of the lepton pair along the @xmath49 direction is @xmath47 . because there exist different conventions for wigner s @xmath50functions we explicate the requisite components that enter ( [ masterformula ] ) : @xmath51 , @xmath52 and @xmath53 . the fact that one has to specify the phase convention of the wigner s @xmath50functions points to one of the weaknesses of the non covariant method to obtain the correct angular decay distributions : one has to use one set of consistent phase conventions to obtain the correct signs for the angular factors . for someone not not so familiar with the angular momentum apparatus this is not always simple . for the polar angle dependencies the correctness of a sign can always be checked by using physics arguments . this is more difficult for the azimuthal signs . including the appropriate normalization factor the four fold decay distribution is given by @xcite @xmath54 where the helicity structure functions @xmath55 , @xmath56 etc . are linear combinations of the helicity matrix elements @xmath40 we have taken the freedom to normalize the differential rate such that one obtains the total @xmath57 rate upon integration _ and not _ the total rate multiplied by the branching ratio of the respective @xmath10 decay channel . the polar angles @xmath58 and @xmath59 , and the azimuthal angle @xmath60 that arise in the full cascade type description of the two stage decay process @xmath61 are defined in fig . [ plot4 ] . for better visibility we have oriented the lepton plane with a negative azimuthal angle relative to the hadron plane . for the hadronic decays of the @xmath6 into a pair of light quarks one has to replace @xmath62 by @xmath63 in fig . [ plot4 ] . we mention that we have checked the signs of the angular decay distribution ( [ diffrate ] ) using covariant techniques . at first sight it seems rather strange that the angular analysis is done in two different coordinate systems , namely the top quark rest system and the @xmath6 rest system . this runs counter to common wisdom that invariants should always be evaluated in one reference system . some insight into the problem may be gained by using the orthonormality and completeness relation of polarization vectors to rewrite the contraction of the hadron and lepton tensors in a form which exhibits the correctness of using two different reference systems for the cascade decay . the orthonormality and completeness relation read @xmath64 \mbox{completeness : } & \hspace{8 mm } & \sum_{m , m ' = s , \pm , 0 } \hspace{-6 mm } \epsilon^{\mu}(m ) \epsilon^{\ast \nu}(m ' ) g_{mm ' } = g^{\mu \nu } \quad \mu , \nu = 0 , 1 , 2 , 3 , \qquad \end{aligned}\ ] ] where @xmath65 , @xmath66 and @xmath67 , @xmath68 . the scalar or time component of the polarization four vector is denoted by @xmath69 . on using the completeness relation one then has @xmath70 & = & \big ( l_{\mu'\nu ' } \epsilon^{\mu'}(m ) \epsilon^{\ast \nu'}(n ) \big ) \big ( h_{\mu\nu } \epsilon^{*\mu}(m ' ) \epsilon^{\nu}(n ' ) \big ) \ , g_{mm ' } g_{nn'}. \end{aligned}\ ] ] the point is that the two lorentz contractions appearing on the second line of ( [ lorentztrick ] ) can be evaluated in two different lorentz frames . the leptonic invariant @xmath71 can be evaluated in the @xmath72 cm frame ( or in the @xmath10 rest frame ) while the hadronic invariant @xmath73 can be evaluated in the rest frame of the top quark . another advantage of this method is that one can easily incorporate lepton mass effects as is mandatory for @xmath29 or the rare @xmath74 transitions if the charged lepton or leptons in the final state are @xmath1 leptons . this technique was used to derive angular decay distributions including leptonic polarization effects for semileptonic and rare bottom meson decays in @xcite and @xcite . note also that the correct phase choice for the polarization vectors is no longer crucial since the polarization vectors always appear as squares in the orthonormality and completeness relations . as an example of the covariant method we discuss the angular decay distribution of polarized @xmath0 decay into a polarized @xmath75 . from helicity counting as described in sect . 4 one knows that there are altogether five spin dependent structure functions and one spin independent structure function describing the leptonic decay of a polarized muon into a polarized electron . we thus define a spin dependent differential rate in terms of six invariant structure functions @xmath76 . accordingly one has @xcite @xmath77 & + & a_5 ( s_{\mu } \!\cdot\ ! s_{e } ) + \frac{1}{m_{\mu}^2 } a_6 \ , \epsilon_{\alpha \beta \gamma \delta } \ , p_{\mu}^{\alpha } \ , p_{e}^{\beta } \ , s_{\mu}^{\gamma } \ , s_{e}^{\delta}. \end{aligned}\ ] ] eq . ( [ invrate ] ) will be evaluated in the rest system of the muon where @xmath78 and @xmath79 @xmath80 . the velocity of the electron is denoted by @xmath81 where @xmath82 and @xmath19 denotes the scaled energy of the electron . in the rest frame of the @xmath83 the polarization four vectors of the @xmath83 and @xmath84 are given by @xmath85 s_{e}^{\alpha } & = & ( \frac{\vec{n}_e \!\cdot\ ! \vec{p}_e}{m_e } ; \vec{n}_{e } + \frac{\vec{n}_e \!\cdot\ ! \vec{p}_e}{m_e ( e_e + m_e ) } \vec{p}_e ) , \end{aligned}\ ] ] where the polarization three vector @xmath86 of the @xmath83 and the quantization axis @xmath87 of the spin of the @xmath84 in their respective rest frames read ( see fig . [ plot5 ] ) @xmath88 and @xmath89 eq . ( [ restframepolmu ] ) holds for @xmath90 polarized muons . for partially polarized muons with magnitude of polarization @xmath39 the representation ( [ restframepolmu ] ) has to be multiplied by @xmath39 such that @xmath91 . the scalar products in ( [ invrate ] ) can then be evaluated with the result @xmath92 including the correct normalization one finally arrives at the angular decay distribution @xcite @xmath93 & + & g_5 p \sin \theta_p \sin \theta \cos \chi + g_6 p \sin \theta_p \sin \theta \sin \chi ) , \end{aligned}\ ] ] where the relation between the invariant structure functions @xmath76 and the frame dependent spectrum functions @xmath94 can be found in @xcite . @xmath95 is the unpolarized spectrum function , @xmath96 and @xmath97 are single spin polarized spectrum functions referring to the spins of the @xmath83 and @xmath84 , resp . , and @xmath98 , @xmath99 and @xmath100 describe spin spin correlations between the spin vectors of the muon and electron . @xmath100 represents a so called @xmath101odd observable . this is evident when rewriting the angular factor multiplying @xmath100 in ( [ diffrate ] ) in triple product form , i.e. @xmath102 @xmath103 . as stated before @xmath100 is identically zero in the standard model since , on the one hand , the weak coupling constant @xmath104 is real and , on the other hand , the loop contributions do not generate imaginary parts . in the terminology of sect . 4 @xmath105 are diagonal structure functions , while the azimuthally dependent structure functions @xmath99 and @xmath100 are non diagonal . as mentioned before , the structure function @xmath99 is proportional to the electron mass and therefore vanishes for @xmath106 . naturally , the angular decay distribution ( [ diffrate ] ) can also be derived in the helicity formalism . in the two examples discussed in this section we chose to orient the z axis along the three momentum of one of the particles . in this case one says that the angular analysis is done in the helicity system . of course , other choices of the orientation of the z axis are possible . a popular choice is the transversity system where the z axis is normal to a given plane spanned by the momenta of the decay products . a more detailed discussion of the choice of frames can be found in @xcite where the angular decay distribution in the semileptonic decays @xmath107 including lepton mass effects was investigated . we mention that the transversity system is the preferred choice when discussing the _ cp _ properties of the final state @xcite . we present our results in terms of the three vector current amplitudes @xmath108 @xmath109 and the three axial vector current amplitudes @xmath110 @xmath109 defined by ( @xmath111 , @xmath112 ) @xmath113 \langle b(p_b ) |j^a_{\mu } | t(p_t ) \rangle & = & \bar{u}_b(p_b ) \big\ { \gamma_{\mu } f_1^a + p_{t , \mu } f_2^a + p_{b,\mu } f_3^a \big\ } \gamma_5 u_t(p_t ) . \end{aligned}\ ] ] in this section we choose to label the current transition according to the transition @xmath28 as in @xcite . as emphasized before , up to the colour factor identical expressions are obtained for the @xmath114 transition if the leptonic four fermion interaction is written in the charge retention form . for the form factors one obtains @xcite @xmath115 & + & \frac{1}{2 } \ln \bigg ( \frac{\lambda^4}{m_b^2 m_t^2 } \bigg ) \ln ( w_1 w_{\mu } ) + \ln \bigg ( \frac{w_1 ^ 3}{w_{\mu } } \bigg ) \ln \bigg ( \frac{w_{\mu } ( 1 - w_1 ^ 2)}{w_{\mu } - w_1 } \bigg ) \bigg ] - \ln \bigg ( \frac{\lambda^4}{m_b^2 m_t^2 } \bigg ) \nonumber \\[2 mm ] & - & \frac{m_t^2 - m_b^2}{2 q^2 } \ln \bigg ( \frac{m_b^2}{m_t^2 } \bigg ) - 4 + \ln ( w_1 w_{\mu } ) \bigg ( \frac{m_t^2 \sqrt{\lambda}}{2 q^2 } - \frac{(m_t + m_b)^2 - q^2}{m_t^2 \sqrt{\lambda } } \bigg ) \bigg\ } , \nonumber \\[2 mm ] f_2^v & = & \frac{\alpha_s}{4 \pi } c_f \frac{m_t - m_b}{q^2 } \bigg\ { 2 - \bigg ( \frac{m_t + 2 m_b}{m_t - m_b } - \frac{m_t^2 - m_b^2}{q^2 } \bigg ) \ln \bigg ( \frac{m_b^2}{m_t^2 } \bigg ) \\[2 mm ] & - & \bigg ( \frac{m_t^2 \sqrt{\lambda}}{q^2 } - \frac{m_b}{m_t - m_b } \frac{q^2 + ( m_t - m_b)(3 m_t + m_b)}{m_t^2 \sqrt{\lambda } } \bigg ) \ln ( w_1 w_{\mu } ) \bigg\ } , \nonumber \\[2 mm ] f_3^v & = & f_3^v ( m_t , m_b ) = f_2^v ( m_b , m_t ) , \end{aligned}\ ] ] where we have denoted the scaled ( small ) gluon mass by @xmath116 . we define @xmath117 ( @xmath118 , @xmath119 ) and use the abbreviations @xmath120 the axial vector amplitudes @xmath121 can be obtained from the vector amplitudes by the replacement @xmath122 , i.e. one has @xmath123 @xmath124 . our one loop amplitudes are linearly related to the one loop amplitudes given in @xcite after correcting for a typo in @xcite ( see also @xcite ) . note that the infrared singularities proportional to @xmath125 and the would be mass singularities ( also called collinear singularities ) proportional to @xmath126 all reside in the born term form factors @xmath127 and @xmath128 . they are eventually cancelled by the corresponding singularities in the tree graph contribution . a look at the arguments of the log and dilog functions in [ oneloop ] shows that the one loop contribution is purely real as remarked on earlier . in the case of the electroweak radiative corrections to @xmath5 there are altogether @xmath129 different vertex correction diagrams in the feynmant hooft gauge @xcite as compared to the one vertex correction diagram in the qcd and the @xmath0decay cases discussed in this section . in addition to the massive one loop three point functions one has to calculate the many massive one loop two point functions needed in the renormalization program . we have recalculated all one loop contributions analytically and have checked them analytically and numerically with the help of a xloops / ginac package that automatically calculates one loop two point and three point functions @xcite . our one loop results agree with the results of @xcite . the results are too lengthy to be reproduced here in analytical form . the full analytical results will be given in a forthcoming publication @xcite . as emphasized earlier on , the _ nlo _ tree graph contributions for the qed and qcd cases are identical up to _ nlo _ except for a trivial colour factor in the case of qcd . differences set in only at _ nnlo _ where three gluon coupling contributions come in . we choose to present and discuss the _ nlo _ tree graph contributions in the qed radiative corrections to leptonic @xmath0decay as written down in @xcite . we begin with the born term contribution . this is an exercise that everyone has probably gone through before in the unpolarized case . the point that not everyone is familiar with is that it is very simple to include the spin of fermions at the born term level . all that is needed is the substitution @xmath130 for the fermion s momenta where the notation @xmath131 is explained in the following . the @xmath130 rule is best explained by looking at the trace expression of the charge side tensor which reads @xmath132 the dependence on the polarization four vectors of the @xmath83 and @xmath84 has been retained in ( [ borntrace ] ) . since only even numbered @xmath133matrix strings survive between the two @xmath134factors in ( [ borntrace ] ) one can compactly write the result of the trace evaluation as @xmath135 where @xmath136 \bar{p}_{e}^{\alpha } & = & p_{e}^{\alpha } - m_{e } s_{e}^{\alpha } , \end{aligned}\ ] ] and where @xmath137 and @xmath138 are the polarization four vectors of the @xmath83 and @xmath84 . in the qed case one has the simplifying feature that the contribution of the last term antisymmetric in @xmath139 can be dropped . the reason is that the dependence on the momentum directions of the @xmath140 and @xmath141neutrinos is completely integrated out in the differential rate . thus the neutrino side of the interaction can only depend on the spatial piece of the second rank tensor build from the momentum transfer to the neutrinos ( the neutrinos are treated as massless ) which is symmetric in @xmath142 . upon contraction with the charge side tensor the antisymmetric piece drops out . unfortunately the @xmath130 trick no longer works at _ nlo _ order . but still , by using the @xmath131notation , the result can be presented in a very compact form . the result for the _ nlo _ @xmath114 tree graph contribution reads @xmath143 & + & \bigg ( \frac{k \!\cdot\ ! \bar{p}_{\mu } + m_{\mu}^2 } { k \!\cdot\ ! p_{\mu } } - \frac{p_e \!\cdot\ ! \bar{p}_{\mu } } { k \!\cdot\ ! p_e } \bigg ) \frac{k^{\alpha } \bar{p}_e^{\beta } + k^{\beta } \bar{p}_e^{\alpha } - k \!\cdot\ ! \bar{p}_e g^{\alpha \beta } } { k \!\cdot\ ! p_{\mu } } \nonumber \\[1 mm ] & + & ( k \!\cdot\ ! \bar{p}_{\mu } ) \frac{p_e^{\alpha } \bar{p}_{e}^{\beta } + p_e^{\beta } \bar{p}_{e}^{\alpha } - m_e^2 g^{\alpha \beta } } { ( k \!\cdot\ ! p_e)(k \!\cdot\ ! p_{\mu } ) } - ( k \!\cdot\ ! \bar{p}_e ) \frac{p_{\mu}^{\alpha } \bar{p}_{\mu}^{\beta } + p_{\mu}^{\beta } \bar{p}_{\mu}^{\alpha } - m_{\mu}^2 g^{\alpha \beta } } { ( k \!\cdot\ ! p_e)(k \!\cdot\ ! p_{\mu } ) } \nonumber \\[1 mm ] & + & ( k \!\cdot\ ! \bar{p}_e ) \frac{p_e^{\alpha } \bar{p}_{\mu}^{\beta } + p_e^{\beta } \bar{p}_{\mu}^{\alpha } - p_e \!\cdot\ ! \bar{p}_{\mu } g^{\alpha \beta } } { ( k \!\cdot\ ! p_e)^2 } - ( k \!\cdot\ ! \bar{p}_{\mu } ) \frac{p_{\mu}^{\alpha } \bar{p}_e^{\beta } + p_{\mu}^{\beta } \bar{p}_e^{\alpha } - p_{\mu } \!\cdot\ ! \bar{p}_e g^{\alpha \beta } } { ( k \!\cdot\ ! p_{\mu})^2 } \bigg\ } \nonumber \\[1 mm ] & - & \frac{e^2}{2 } \bigg ( \frac{m_{\mu}^2}{(k \!\cdot\ ! p_{\mu})^2 } \!+\ ! \frac{m_e^2}{(k \!\cdot\ ! p_{e})^2 } \!-\ ! \frac{2 p_e \!\cdot\ ! p_{\mu } } { ( k \!\cdot\ ! p_e ) ( k \!\cdot\ ! p_{\mu } ) } \bigg ) ( \bar{p}_e^{\alpha } \bar{p}_{\mu}^{\beta } + \bar{p}_e^{\beta } \bar{p}_{\mu}^{\alpha } - \bar{p}_e \!\cdot\ ! \bar{p}_{\mu } g^{\alpha \beta } ) . \end{aligned}\ ] ] the momentum of the radiated photon is denoted by @xmath144 . for the aforementioned reason we have dropped the contribution of antisymmetric terms . when the antisymmetric @xmath145tensor pieces are kept , and when one replaces @xmath146 and @xmath147 to account for colour , one recovers the _ nlo _ qcd corrected hadronic tensor for the @xmath28 transition listed in @xcite . in the last line of ( [ matrix - element ] ) we have isolated the infrared singular piece of the charge side tensor which is given by the usual soft photon factor multiplying the born term contribution . technically this is done by writing @xmath148 the remaining part of the charge side tensor in ( [ matrix - element ] ) is referred to as the hard photon contribution . it is infrared finite and can thus be integrated without a regulator photon mass . analytic phase space integrations without a regulator photon mass are much simpler . the regulator photon mass would introduce a new mass scale into the problem which , as emphasized before , would complicate the phase space integrations . in the phase space integration over the photon momentum the infrared singular piece is regularized by introducing a ( small ) photon mass resulting in a logarithmic mass divergence in the photon mass . since the integrand of the soft photon piece is much simpler the phase space integration can be done analytically by carefully taking the appropriate mass zero limits in the integrations . the infrared divergence shows up as a logarithmic mass divergence in the photon mass . in addition , since the soft photon piece factors the born term tensor , the born term tensor can be pulled out of the phase space integrations . the resulting singular soft photon piece is therefore universal in the sense that it is the same for all spin structure functions , i.e. once the soft photon integration has been done for the rate the work is done . eventually the infrared singular piece is cancelled by the corresponding singular piece in the one loop contributions . the approximation where only the soft photon ( or gluon ) piece is retained in the _ nlo _ tree graph hadron tensor is called the soft photon ( or gluon ) approximation . from what has been said before it is clear that a _ nlo _ radiative correction calculation is much simplified in the soft photon ( or gluon ) approximation . in the next section we shall take a specific example , namely the process @xmath149 , to investigate the quality of the soft gluon approximation . it is interesting to note that the _ nlo _ hadronic tensor in the elektroweak corrections to the process @xmath5 is only marginally more complicated than the corresponding tree graph contribution ( [ matrix - element ] ) although the number of contributing diagrams has doubled to four @xcite ( see fig . [ plot6 ] ) . we do not list the finite piece of the hadronic tensor in this review but only write down the soft photon factor which multiplies the born term tensor just as in ( [ matrix - element ] ) . when discussing the decay @xmath5 one has to of course include the antisymmetric @xmath145tensor piece in the born term tensor . the soft photon contribution reads @xmath150 & - & \frac{2 q_t q_b \ , p_t \!\cdot\ ! p_b } { ( p_t \!\cdot\ ! k ) ( p_b \!\cdot\ ! k ) } - \frac{2 q_t q_w \ , p_t \!\cdot\ ! q } { ( p_t \!\cdot\ ! k ) ( q \!\cdot\ ! k ) } + \frac{2 q_b q_w \ , p_b \!\cdot\ ! q } { ( p_b \!\cdot\ ! k ) ( q \!\cdot\ ! k ) } \bigg ) , \end{aligned}\ ] ] where the soft photon amplitude is given by ( see e.g. @xcite ) @xmath151 @xmath152 , @xmath153 and @xmath154 are the electric charges of the top quark , the bottom quark and the @xmath6boson , resp . , in units of the elementary charge @xmath75 . gauge invariance of the soft photon amplitude is easily verified when replacing @xmath155 in ( [ softphotonamp ] ) . it is then just a statement about charge conservation @xmath156 . it is noteworthy that by setting @xmath157 and @xmath158 in ( [ softphotonfactor ] ) , one recovers the soft photon contribution in ( [ matrix - element ] ) . in fact , the whole _ nlo _ electroweak tree graph tensor in @xcite reduces to the corresponding _ qed tensor with the above charge replacements . in the same vein , the replacements @xmath146 and @xmath147 , to account for colour , and the replacements of the charge factors by @xmath157 and @xmath158 will bring one from the electroweak case to the qcd case . in the soft gluon ( or photon ) approximation one keeps only the soft gluon ( or photon ) piece in the tree graph contribution but includes the full one loop contribution . new structure is thus only generated by the one loop contribution since the soft gluon piece has born term structure . the structure contained in the hard part of the tree graph contribution is lost in the soft gluon approximation . a brief glance at the corresponding hard photon piece in ( [ matrix - element ] ) shows that it is considerably more difficult to do the analytic phase space integration for the hard part than for the soft photon piece . in particular , the integration of the hard photon piece has to be done separately for each density matrix element . contrary to this , the integration of the soft photon piece has to be done only once since the born term factor can be factored out of the integral . technically , the soft gluon ( or photon ) approximation is much simpler than the full calculation if done analytically . of course , if the calculation is done numerically , the integration of the hard gluon ( or photon ) part causes no additional problems . when the soft gluon ( or photon ) approximation is used this is done at the cost of loosing interesting structure contained in the hard gluon ( or photon ) part . in order to be able to judge the quality of the soft gluon approximation we have calculated the rate for @xmath160 and have compared the results of the full calculation with the results of the soft gluon approximation using different cut off values for the gluon energy @xcite . [ plot7 ] shows that the soft gluon result is below the full result for the whole range of cut off values @xmath161 ( @xmath162 ) . for maximal cut off values @xmath163 the soft gluon approximation is @xmath164 , @xmath165 and @xmath166 below the full result at @xmath167 , @xmath168 and @xmath169 , respectively . as emphasized above , much of the rate that is being missed by the soft gluon approximation has interesting structure . we have already discussed various aspects of the decay @xmath5 in previous sections . in this section we concentrate on unpolarized top decay . in particular , we want to discuss the mass dependence of the longitudinal piece of the @xmath10 boson the measurement of which could lead to an independent determination of the mass of the top quark . the angular decay distribution for unpolarized top decay can be obtained by setting @xmath170 in ( [ diffrate ] ) . the contribution of the three remaining structure functions @xmath56 , @xmath55 and @xmath171 can be disentangled by a measurement of the shape of the lepton energy spectrum from top decay . given enough data , one can hope to determine the longitudinal contribution with @xmath172 accuracy @xcite . at _ lo _ the mass dependence of the longitudinal contribution is given by @xmath173 which gives @xmath174 using @xmath175 and @xmath176 . _ nlo _ corrections to the longitudinal contribution were calculated in @xcite ( qcd ) and in @xcite ( electroweak and finite width ) . curiously enough the electroweak and finite width corrections tend to cancel each other in the structure functions . in fig . [ plot8 ] we show the top mass dependence of the ratio @xmath177 . the born term and the corrected curves are practically straight line curves . the horizontal displacement of the two curves is @xmath178 . one would thus make the corresponding mistake in the top mass determination from a measurement of @xmath177 if the born term curve were used instead of the corrected curve . if we take @xmath175 as central value , a @xmath172 relative error on the measurement of @xmath177 would allow one to determine the top quark mass with @xmath179 accuracy . such a top mass measurement would be a welcome alternative to the usual invariant mass determination of the top quark mass since the @xmath177 measurement is a completely independent measurement of the top quark mass . certain extensions of the minimal standard model with two higgs doublets ( as e.g. the minimal supersymmetric standard model ) contain charged higgs bosons such that the decay @xmath180 would be possible if kinematically allowed . the lagrangian for the decay reads @xmath181 for the differential decay rate one obtains @xmath182 where @xmath58 is the angle between the polarization vector of the top quark and the higgs . the @xmath183dependence in ( [ difrate ] ) is determined by the asymmetry parameter @xmath184 . at _ lo _ one obtains @xmath185 @xmath186 at this point we want to emphasize that one needs the polarization information contained in ( [ difrate ] ) to be able to determine the relative sign of the two coupling constants @xmath187 and @xmath188 . the rate is proportional to @xmath189 and is therefore insensitive to the relative sign of the coupling constants . we now specify to the so called model 2 where the coupling structure is expressed in terms of @xmath190 , and where @xmath191 and @xmath192 are the vacuum expectation values of the two neutral components of the two higgs doublets . in model 2 the coupling constants are given by @xmath193 b & = & \frac{g_w}{2 \sqrt{2 } m_w } v_{tb}(m_t \cot \beta - m_b \tan \beta ) . \end{aligned}\ ] ] the weak coupling factor @xmath194 is related to the usual fermi coupling constant @xmath104 by @xmath195{2 } \sqrt{g_f}$ ] . in fig . [ plot9 ] we show a plot of the asymmetry parameter @xmath196 as a function of the mass ratio @xmath197 for a fixed value of @xmath198 . the asymmetry parameter is large and negative over most of the range of mass ratios . the radiative corrections can be seen to be quite substantial . consider the decay @xmath199 where the current @xmath200 can be thought of as representing an off shell @xmath10 containing spin @xmath201 and spin @xmath202 pieces . taking @xmath203 the longitudinal polarization vector of the spin @xmath201 piece is given by @xmath204 while the scalar polarization vector reads @xmath205 . in the limit @xmath206 the longitudinal polarization vector becomes increasingly parallel to its momentum . in fact , one finds @xmath207 . in this limit the amplitude @xmath208 dominates over the transverse amplitudes @xmath209 since @xmath210 . at the same time one finds @xmath211 in this limit this provides a useful check on the high energy limits of the relevant unpolarized and polarized rate functions . next consider the scalar projection of the @xmath28 current transition at lowest order . using the dirac equation one obtains ( @xmath212 ) @xmath213 & = & \bar{u}_b ( m_t ( { 1 \hspace{-0.95 mm } { \rm l}}+ \gamma_5 ) + m_b u_t({1 \hspace{-0.95 mm } { \rm l}}- \gamma_5 ) ) u_t . \end{aligned}\ ] ] the second line of ( [ scalarproj ] ) has the coupling structure of a goldstone boson in the standard model . it is a very instructive exercise to follow this argument through to one loop order including renormalization @xcite . in fact , this generalizes to any order in perturbation theory : the scalar piece of the charged transition current has the coupling structure of a goldstone boson . together with the above statement that the longitudinal rate dominates at high energies and that it is equal to the scalar rate in this very limit one arrives at the goldstone boson equivalence theorem . in the high energy limit the coupling of a gauge vector boson to a fermion current is equivalent to that of a goldstone boson . this simplifying feature has frequently been used in the literature to dramatically reduce the effort needed in the computation of high energy processes involving vector gauge bosons . for example , the goldstone equivalence theorem was used in @xcite to calculate the bosonic two loop electroweak radiative corrections to the decay @xmath214 in the limit of a large higgs mass . the same approximation was used to calculate the dominant bosonic two loop electroweak radiative corrections to the decay of a heavy higgs into pairs of @xmath6 and @xmath215 gauge bosons @xcite . as concerns the coupling of charged higgs bosons in @xmath216 discussed in sect . 10 the coupling structure of the charged higgs is equal to that of a goldstone boson in the standard model if one specifies to the so called model @xmath201 with @xmath217 ( see e.g. @xcite ) . in this case one has @xmath218 the same relation holds true for the corresponding polarized top rates . all of the statements made in this section have been explicitly verified using our _ nlo _ results on @xmath5 @xcite and @xmath216 @xcite . we have already discussed various aspects of the leptonic decays of the @xmath0 and the @xmath1 lepton in previous sections . a complete solution to the problem has been given in @xcite where , for the first time , all mass and polarization effects have been included at _ nlo_. partial _ nlo _ results on polarization can be found in @xcite . in this section we concentrate on one aspect of the problem , namely on the so called anomalous contribution that , in the chiral limit , flips the helicity of the final state lepton at _ nlo_. collinear photon emission from a massless fermion line can flip the helicity of the massless fermion contrary to naive expectation . this has been discussed in a variety of physical contexts . this is a `` @xmath219 '' effect where the @xmath220 in the numerator is a spin flip factor and the @xmath220 in the denominator arises from the collinear configuration . in the limit @xmath106 the helicity flip contribution survives whereas it is not seen in massless qed . we shall discuss this phenomenon in the context of the left chiral @xmath114 transition . at the born term level an electron emerging from a weak @xmath221 vertex is purely left handed in the limit @xmath222 . naively , one would expect this to be true also at @xmath223 or at any order in @xmath224 because in massless qed photon emission from the electron is helicity conserving . let us make this statement more precise by looking at the string of @xmath133matrices between the initial state @xmath0 spinor and final state @xmath75 antispinor of the left chiral @xmath114 transition . using @xmath225 , and the fact that in massless qed every photon emission brings in two @xmath133factors ( one from the vertex and one from the fermion propagator ) one finds @xmath226 where @xmath227 stands for the ( even ) @xmath133 matrix string brought in by the emission of @xmath228 photons . we have commuted the left chiral factor @xmath229 to the left end where it then projects out the helicity state @xmath230 from the electron antispinor thus proving the above assertion . let us take a closer look at the anomalous helicity flip contribution in leptonic @xmath114 decays by considering the unnormalized density matrix element @xmath231 of the final state electron which is obtained by setting @xmath232 in ( [ diffrate ] ) ( remember that @xmath99 vanishes for @xmath106 and @xmath233 in the standard model ) . one has @xmath234 contrary to naive expectations one finds non vanishing right handed @xmath235 contributions which survive the @xmath106 limit when one takes the @xmath106 limit of the _ nlo _ contributions to ( [ plusplus1 ] ) @xcite . in fact , one finds @xmath236 - \big [ ( 1 - x)^2 ( 1 + 2 x ) \big ] p \cos \theta_p \big).\ ] ] the result is rather simple . in particular , it does not contain any logarithms or dilogarithms . the simplicity of the right handed contribution becomes manifest in the equivalent particle description of @xmath0decay where , in the peaking approximation , @xmath0decay is described by the two stage process @xmath237 followed by the branching process @xmath238 characterized by universal splitting functions @xmath239 @xcite . the symbols @xmath240 and @xmath241 stand for a helicity non flip and helicity flip of the helicity of the electron . in the splitting process @xmath242 is the fractional energy of the emitted photon . the off shell electron in the propagator is replaced by an equivalent on shell electron in the intermediate state . since the helicity flip contribution arises entirely from the collinear configuration it can be calculated in its entirety using the equivalent particle description . the helicity flip splitting function is given by @xmath243 , where @xmath244 , and where @xmath245 is the energy of the emitted photon . @xmath246 and @xmath247 denote the energies of the initial and final electron in the splitting process . the helicity flip splitting function has to be folded with the appropriate @xmath222 born term contribution . the lower limit of the folding integration is determined by the soft photon point where @xmath248 . the upper limit is determined by the maximal energy of the initial electron @xmath249 . one obtains @xmath250 & = & \frac{\alpha}{2 \pi } \gamma_0 \int_x^1 dx ' ( x ' - x ) \big ( ( 3 - 2 x ' ) + ( 1 - 2 x ' ) p \cos \theta_p \big ) \nonumber \\[2 mm ] & = & \frac{\alpha}{12 \pi } \gamma_0 \big ( ( 1 - x)^2 ( 5 - 2 x ) - ( 1 - x)^2 ( 1 + 2 x ) p \cos \theta_p \big ) , \end{aligned}\ ] ] which exactly reproduces the result ( [ plusplus2 ] ) . numerically , the flip spectrum function is rather small compared to the @xmath22 no flip spectrum function . however , when averaging over the spectrum the ratio of the @xmath22 flip and no flip contributions amounts to a non negligible @xmath251 , due to cancellation effects in the @xmath22 no flip contribution . we have discussed _ nlo _ corrections to a multitude of polarization observables in different processes including nonzero mass effects . the results are available in compact analytical form . they are ready for use in physics simulation programs which are reliable even at corners of phase space where mass effects become important . present and planned experiments ( tevatron run2 , lhc , belle , babar , @xmath1charm factories ) will be sensitive to these _ nlo _ standard model effects . standard model _ corrections to polarization observables are needed as background for possible new physics contributions . last but not least the _ nlo _ results are needed for _ nlo _ sum rule analysis involving polarization observables . all calculations are based on the same one loop and tree graph matrix elements . results on rates agree with previous calculations . results on polarization observables agree with previous calculations where available . we have checked various limits and found agreement with previous mass zero calculations , the goldstone boson equivalence theorem and the equivalent particle description of the anomalous helicity flip contribution . because of the various checks and the fact that we have essentially used one set of matrix elements as input to the calculations we feel quite confident that our results are correct . we would like to thank the organizers of this meeting , d. blaschke , m.a . ivanov and s. nedelko for providing a most graceful conference setting . we acknowledge informative discussions and e mail exchanges with a.b . arbuzov , f. berends , w. van neerven , f. scheck , k. schilcher , h. spiesberger and o.v . we would like to thank b. lampe , m. fischer , s. groote and h.s . do for their collaborative effort . m.c . mauser is supported by the dfg ( germany ) through the graduiertenkolleg `` eichtheorien '' at the university of mainz . baikov , k.g . chetyrkin and j.h . khn : phys . lett . * 88 * 012001 ( 2002 ) p.a . baikov , k.g . chetyrkin and j.h . khn : hep - ph/0212299 m. fischer , s. groote , j.g . krner and m.c . mauser : + hep - ph/0203048 ( 2003 ) , to be published in phys . d m. fischer , s. groote , j. g. krner , b. lampe and m. c. mauser : + phys . b * 451 * 406 ( 1999 ) m. fischer , s. groote , j.g . krner and m.c . mauser : + phys . d * 63 * 031501 ( 2001 ) m. fischer , s. groote , j.g . krner and m.c . mauser : + phys . d * 65 * 013205 ( 2002 ) h.s . do , s. groote , j.g . krner and m.c . mauser : phys . d * 67 * 091501 ( 2003 ) j.g . krner and m.c . mauser : hep - ph/0211098 m. fischer , s. groote , j.g . krner and m.c . mauser : phys . b * 480 * 265 ( 2000 ) m. fischer , s. groote , j.g . krner and m.c . mauser : + `` inclusive decays @xmath252 at @xmath22 including @xmath253 + and @xmath254 polarization effects '' , to be published s. groote , j.g . krner and j.a . leyva : phys . d * 56 * 6031 ( 1997 ) j.g . krner , a. pilaftsis and m.m . tung : z. phys . c * 63 * 575 ( 1994 ) s. groote , j.g . krner and m.m . tung : z. phys . c * 70 * 281 ( 1996 ) s. groote and j.g . krner : z. phys . c * 72 * 255 ( 1996 ) s. groote , j.g . korner and m.m . tung : z. phys . c * 74 * 615 ( 1997 ) a. brandenburg , m. flesch and p. uwer : phys . d * 59 * 014001 ( 1999 ) v. ravindran and w.l . van neerven : phys . b * 445 * 214 ( 1998 ) ; + nucl . phys . b * 589 * 507 ( 2000 ) a. czarnecki , m. jezabek and j.h . khn : nucl . b * 351 * 70 ( 1991 ) a. czarnecki , m. jezabek , j.g . krner and j.h . khn : + phys . lett . * 73 * 384 ( 1994 ) a. brandenburg : phys . b * 388 * 626 ( 1996 ) g. mahlon and s. parke : phys . b * 411 * 173 ( 1997 ) w. bernreuther , a. brandenburg , z.g . si and p. uwer : + phys . * 87 * 242002 ( 2001 ) w. bernreuther , a. brandenburg , z.g . si and p. uwer : + phys . b * 509 * 53 ( 2001 ) s. parke and y. shadmi : phys . b * 387 * 199 ( 1996 ) s. groote , j.g . krner and j.a . leyva : phys . b * 418 * 192 ( 1998 ) m.m . tung , j. bernabeu and j. penarrocha : phys . b * 418 * 181 ( 1998 ) j.g . krner , g. kramer , g. schierholz , k. fabricius and i. schmitt : + phys . b * 94 * 207 ( 1980 ) k. fabricius , i. schmitt , g. kramer and g. schierholz : + phys . * 45 * 867 ( 1980 ) a. brandenburg , l.j . dixon and y. shadmi : phys . d * 53 * 1264 ( 1996 ) k. hagiwara , k.i . hikasa and n. kai : phys . lett . * 47 * 983 ( 1981 ) k. hagiwara , k.i . hikasa and n. kai : phys . d * 27 * 84 ( 1983 ) k. hagiwara , k. i. hikasa and n. kai : phys . rev . lett . * 52 * 1076 ( 1984 ) j.g . krner , b. melic and z. merebashvili : phys . d * 62 * 096011 ( 2000 ) k. hagiwara _ et al . _ [ particle data group collaboration ] : + phys . rev . d * 66 * 010001 ( 2002 ) w.g . dharmaratna and g.r . goldstein : phys . d * 41 * 1731 ( 1990 ) ; w. bernreuther , a. brandenburg and p. uwer : phys . b * 368 * 153 ( 1996 ) w.g . dharmaratna and g.r . goldstein : phys . d * 53 * 1073 ( 1996 ) j.g . krner , g. schuler , g. kramer and b. lampe : + z. phys . c * 32 * 181 ( 1986 ) 181 j. c. pati and c. h. woo : phys . d * 3 * 2920 ( 1971 ) j.g . krner : nucl.phys.*32 * 282 ( 1970 ) j.h . khn , a. reiter and p.m. zerwas : nucl . b * 272 * 560 ( 1986 ) j.g . krner and g.a . schuler : z. phys . c * 46 * 93 ( 1990 ) a. faessler , t. gutsche , m.a . ivanov , j.g . krner and v.e . lyubovitskij : + eur . j. directc * 4 * 18 ( 2002 ) i. dunietz , h. r. quinn , a. snyder , w. toki and h. j. lipkin : + phys . d * 43 * 2193 ( 1991 ) g.j . gounaris and j.e . paschalis : nucl . b * 222 * 473 ( 1983 ) k. schilcher , m.d . tran and n.f . nasrallah , nucl . b * 181 * 91 ( 1981 ) , + erratum _ ibid . _ b * 187 * 594 ( 1981 ) c. bauer and h. s. do : comput . phys . * 144 * 154 ( 2002 ) a. denner and t. sack : nucl . b * 358 * 46 ( 1991 ) h.s . do , s. groote , j.g . krner and m.c . mauser : to be published future electroweak physics at the fermilab tevatron : report of the tev2000 group , + edited by d. amidei and r. brock , fermilab - pub-96/046 , + http://www-theory.fnal.gov/tev2000.html s. weinberg : `` the quantum theory of fields . vol . 1 : foundations , '' + ( cambridge university press , cambridge 1995 ) s. groote , j.g . krner : `` analytical results for @xmath22 radiative corrections + to @xmath255 up to a given gluon energy cut '' , to be published a. czarnecki and s. davidson : + proceedings of the lake louise winter institute , 330 ( 1993 ) j.g . krner , k. melnikov and o.i . yakovlev : phys . d * 53 * 3737 ( 1996 ) a. frink , b.a . kniehl , d. kreimer and k. riesselmann : + phys . d * 54 * 4548 ( 1996 ) b. falk and l.m . sehgal : phys . b * 325 * 509 ( 1994 ) w.e . fischer and f. scheck : nucl . b * 83 * 25 ( 1974 ) a.b . arbuzov : phys . b * 524 * 99 ( 2002 )

circuit cavity quantum electrodynamics ( qed ) is proving to be a powerful platform to implement quantum feedback control schemes due to the ability to control superconducting qubits and microwaves in a circuit . here , we present a simple and promising quantum feedback control scheme for deterministic generation and stabilization of a three - qubit state in the superconducting circuit qed system . the control scheme is based on continuous joint zeno measurements of multiple qubits in a dispersive regime , which enables us not only to infer the state of the qubits for further information processing but also to create and stabilize the target state through adaptive quantum feedback control . we simulate the dynamics of the proposed quantum feedback control scheme using the quantum trajectory approach with an effective stochastic maser equation obtained by a polaron - type transformation method and demonstrate that in the presence of moderate environmental decoherence , the average state fidelity higher than can be achieved and maintained for a considerable long time ( much longer than the single - qubit decoherence time ) . this control scheme is also shown to be robust against measurement inefficiency and individual qubit decay rate differences . finally , the comparison of the polaron - type transformation method to the commonly used adiabatic elimination method to eliminate the cavity mode is presented .

1311.6321

entanglement is regarded as one of the key resources for various applications in quantum information processing . while entanglement of bipartite systems is well understood @xcite , the characterization of multipartite entanglement is still an interesting research topic . it has been shown @xcite that there are two inequivalent non - biseparable classes of three - qubit entanglement states , the greenberger - horne - zeilinger ( ghz)class and the w class , which can not be transformed into each other by stochastic local operations and classical communications . the w state is central as a resource in quantum information processing and multi - party quantum communication as its entanglement is persistent and robust even under particle loss @xcite . the three - qubit entangled w state has been experimentally generated and demonstrated in systems of trapped ions @xcite , optical photons @xcite , superconducting phase qubits @xcite , and coupled nonlinear oscillator arrays @xcite . circuit qed system @xcite in which superconducting qubits based on josephson junctions are coupled to a high - q microwave transmission line resonator acting as a quantum bus has been demonstrated to be a promising solid - state quantum computing architecture . due to the great controllability of the superconducting qubits and microwaves in the circuit system , the circuit qed system , a solid - state analogy of quantum optics cavity qed , also has excellent potential as a platform for quantum control especially quantum feedback control experiments @xcite . in this paper , we present a simple and promising quantum feedback control scheme for deterministic generation and stabilization of a three - qubit @xmath0 state in the superconducting circuit qed system . generation and manipulation of entangled states are important tasks of quantum information processing . besides the scheme based on unitary dynamics to generate entangled states @xcite , there are proposals of entanglement generation by measurement @xcite . although measurements can generate entangled states that are otherwise difficult to obtain , the specific or target entangled states created are primarily probabilistic . furthermore , the measurement - alone approach can not stabilize and protect the generated entangled states from deterioration . one possible way to resolve this problem is to employ the technique of quantum feedback control @xcite . there have been proposals of using quantum feedback control to stabilize and generate two - qubit bell states in circuit qed @xcite . rist et al . @xcite recently presented an experimental demonstration of a superconducting two - qubit bell state produced by feedback based on parity measurements . the case for three - qubit entangled ghz in circuit qed has also been investigated @xcite , but a somewhat complicated method of an _ alternate - flip - interrupted _ zeno scheme and quantum feedback control technique with efficient measurement and rapid single - qubit rotations are required to produce and maintain the _ pre - ghz state _ with high fidelity . however , how to generate and stabilize the other inequivalent class of three - qubit states , namely the w state , in circuit qed has , to our knowledge , not been reported . here , we present a simple measurement and feedback control scheme that is feasible with current circuit qed technology to produce and stabilize the w state of @xmath2 . our scheme does not assume fast single - qubit rotations and is robust against measurement inefficiency and individual qubit decay rate differences . a successful experimental implementation and realization of using quantum feedback control to generate and stabilize a multi - qubit entangled w state as presented here will be an impressive demonstration in circuit qed experiments . moreover , previous investigations were performed in a parameter regime of a strongly damped resonator cavity so that one can adiabatically eliminate the cavity mode by enslaving the cavity to qubit dynamics @xcite . here , we go beyond this so - called bad - cavity limit by using a polaron - type transformation @xcite to trace out the cavity mode in our analysis . this allows us to work in a parameter regime in which the w state can be maintained with higher fidelity . the obtained effective ( stochastic ) master equation for the qubit degrees of freedom provides us with more intuitive understanding and physical insight into the qubit dynamics of the continuous quantum measurement and quantum feedback control process . note that our method can be extended straightforwardly to the generation and stabilization of a @xmath3-qubit w - type state that is a quantum superposition with equal expansion coefficients of all possible pure states in which exactly one of the qubits is in an excited state @xmath4 , while all other ones are in the ground state @xmath5 . the paper is organized as follows . we describe a three - qubit circuit qed setup and its corresponding model hamiltonian in sec . [ sec : model ] . the procedure of using polaron - type transformation to eliminate cavity field to obtain an effective master equation for the qubit degrees of freedom alone conditioned on continuous homodyne detection is also presented in this section . the quantum feedback control strategy to generate and stabilize the @xmath0 state of @xmath2 is described in sec . [ sec : qfc ] . the results of the average fidelity for the generation and stabilization of the @xmath6 state are presented in sec . [ sec : fidelity ] . the dependence of the average fidelity on the qubits decay rates @xmath7 , dispersive coupling strength @xmath8 , probe beam amplitude @xmath9 , feedback strength @xmath10 , and measurement efficiency @xmath11 are discussed . in sec . [ sec : comparison ] , we compare the polaron - type transformation method with the adiabatic elimination method to eliminate the cavity mode . a short conclusion is given in sec . [ sec : conclusion ] . ( color online ) schematic illustration of three qubits in circuit qed quantum feedback control setup ( mw : microwave drive ) . ] we consider a circuit qed setup in which three cooper pair boxes considered as qubits are coupled to a common field of a one - dimensional microwave transmission line resonator ( tlr ) treated as a cavity ( see fig . [ fig : setup ] ) . the system can be described well by the tavis - cummings model @xcite and the hamiltonian driven by a measurement signal is described by @xmath12.\label{eq : hamiltonian}\end{aligned}\ ] ] here , the operators @xmath13@xmath14 and @xmath15@xmath16 are respectively the lowering ( raising ) operators of the @xmath17th qubit and the microwave inside the cavity , @xmath18 is the transition frequency of the @xmath17th qubit , @xmath19 is the cavity frequency , @xmath20 is the strength of the @xmath17th qubit interacting with the cavity field , and @xmath9 and @xmath21 are the amplitude and frequency of the measurement drive . in the dispersive regime , where @xmath22 , we can eliminate the direct qubit - resonator coupling by using the unitary transformation @xcite @xmath23.\label{xformop}\ ] ] keeping terms in the hamiltonian up to second order in the small parameter @xmath24 and moving to a frame rotating with the measurement signal frequency @xmath21 for the cavity field and qubits , we obtain the hamiltonian @xcite @xmath25a^{\dagger}a+ \epsilon\left(a+a^{\dagger}\right)+\sum_{j}\epsilon\lambda_{j}\sigma_{j}^{x}\nonumber \\ & & + \sum_{j}\frac{\tilde{\omega}_{j}}{2}\sigma_{j}^{z}+\sum_{j > i}j_{ij}^{q}\left(\sigma_{i}^{-}\sigma_{j}^{+}+\sigma_{i}^{+}\sigma_{j}^{-}\right ) . \label{eq : effhamiltonian}\end{aligned}\ ] ] here , the detuning frequency between cavity and the measurement drive @xmath26 , the dispersive coupling strength @xmath27 , the dispersive - shifted qubit frequency @xmath28 , and the strength of qubit - qubit interaction mediated by the cavity field @xmath29/2 $ ] . one can see that in this dispersive limit , the qubit - resonator interaction induces a qubit - state - dependent shift on the resonator frequency . if we set @xmath30 , i.e. , the driving frequency to be in resonance with the cavity frequency , the measurement of the resonator frequency shift can be translated into the measurement of the phase shift between the incident and transmitted microwave drives . thus the information about the qubit state can be inferred from the homodyne signal coming from the transmitted microwave through the cavity or tlr . the optimal signal to noise ratio for single - qubit dispersive readout is achieved for @xmath31 @xcite while it is a little bit involved when multi - qubit joint measurement is considered [ see discussions related to eqs . ( [ eq : jointmeasop ] ) and ( [ eq : jmop2 ] ) and to the red dashed curve in fig . [ fig : fine](a ) ] . the evolution equation for the density matrix of the qubit and cavity system conditioned on the continuous homodyne detection in the joint rotating frame can be written as @xcite @xmath32+\sum_{j}\gamma_{j}\mathcal{d}\left[\sigma_{j}^{-}\right]\rho_{c}\nonumber \\ & & + \kappa\mathcal{d}\left[a \right]\rho_{c } + \sqrt{\kappa\eta}\mathcal{h}[ae^{-i\phi}]\rho_{c}\xi(t),\label{eq : me}\end{aligned}\ ] ] where the effect of the baths on the system describing the qubit and cavity decays is denoted by the decoherence superoperator terms given in the lindblad form @xmath33\rho = c\rho c^{\dagger}-\frac{1}{2}\left(c^{\dagger}c\rho+\rho c^{\dagger}c\right),\label{eq : lindblad}\ ] ] and @xmath34 and @xmath35 are respectively the cavity and individual qubit decay rates . the last term in the conditional master equation ( [ eq : me ] ) is the homodyne measurement unraveling term that describes the back action and stochastic nature of the quantum measurements , @xmath36 is the measurement efficiency ( @xmath37 corresponds to a perfect detector or efficient measurement , and @xmath38 represents the fraction of detections which are actually registered by the detectors ) , @xmath39 is the phase of the local oscillator that is mixed with the transmitted microwave in the homodyne measurement , and the measurement superoperator @xmath40\rho = c\rho+\rho c^{\dagger}-\left\langle c+c^{\dagger}\right\rangle \rho,\label{eq : measop}\ ] ] where @xmath41 means the quantum average of the operator @xmath42 . the stochastic nature of the random measurement outcomes is characterized by @xmath43 , a gaussian white noise with the ensemble average properties of @xmath44=0 $ ] , @xmath45=\delta(t - t')$ ] , where @xmath46 $ ] denotes an ensemble average over different realizations of the noise . the use of a gaussian white noise term here assumes that the local oscillator has no more noise than a coherent state , a good assumption at microwave frequencies . the measured homodyne current ( in units of frequency ) is proportional to @xmath47 although eq . ( [ eq : me ] ) can be used to study the conditional dynamics and measurement backaction , it provides little direct insight about how the evolution of the qubits depends on the continuous measurement outcomes . however , if the cavity field can be traced out and an effective ( stochastic ) master equation for the qubit degrees of freedom only can be obtained , more intuition and understanding to the qubit dynamics of the continuous quantum measurement process can be gained and thus help facilitate the successful development and design of further manipulation and control strategies for the qubit system , e.g. , the quantum feedback control strategy presented later in this article . to obtain the effective stochastic master equation for the qubits degrees of freedom only , a common method is the so - called adiabatic elimination procedure valid in the limit where the damping of the cavity is much larger than both the dispersive coupling strengths and the qubits decay rates , i.e. , @xmath48 and @xmath49 . here , we go beyond this limit and use a polaron - type transformation @xcite to trace out the cavity field . we will compare these two approaches in sec . [ sec : comparison ] . the polaron - type approach assumes only that the state of the qubits varies slowly within the measurement time during which the cavity field evolves to a steady coherent state depending on the qubit state . this assumption can be justified if the cavity field decay rate is much faster than the qubit decay rate @xmath50 . in this case , the unconditional master equation , i.e. , eq . ( [ eq : me ] ) averaged over the white noise process , indicates that a coherent state remains a coherent state with the amplitude @xmath51 of the cavity coherent state @xmath52 at @xmath30 satisfying @xmath53 when the qubits are in a basis state @xmath54 , where @xmath55 , @xmath56 and @xmath57 represent respectively the ground and excited states of a single qubit , and @xmath58 . then the elimination of the cavity ( tlr ) degrees of freedom is carried out by going to a frame defined by the transformation @xcite @xmath59\label{eq : polaron}\ ] ] with @xmath60 $ ] the displacement operator of the tlr , @xmath61=\exp\left[\alpha a^{\dagger}-\alpha^{*}a\right],\label{eq : polaron01}\ ] ] and @xmath62 are projection operators onto the respective basis ( logical ) states of the three - qubit hilbert space . in this transformed reference frame , the cavity field is displaced to start from a vacuum state with zero photon , i.e. , @xmath63|0\rangle_{tlr}=|\alpha_x\rangle_{tlr}$ ] . for simplicity , we take @xmath64 and @xmath65 . this implies that we assume three identical qubits with one wavelength separation apart in the tlr cavity ( see fig . [ fig : setup ] ) . this assumption also helps generate the @xmath66 state by continuous measurements as the @xmath66 state in this case is a simultaneous eigenstate of the system hamiltonian ( i.e. , when without consideration of the qubits decay ) and the homodyne measurement operator [ see eq . ( [ eq : jmop2 ] ) and further discussion below it ] . then following the calculations in refs . , we obtain an effective master equation for the qubits degrees of freedom alone conditioned on continuous homodyne detection as @xcite @xmath67\rho_{c}\left(t\right)\xi\left(t\right)-\imath\sqrt{\kappa\eta}\left[c_{\phi-\pi/2},\rho_{c}\left(t\right)\right]\xi\left(t\right),\label{eq : sme}\end{aligned}\ ] ] where @xmath68 is given by @xmath69 % + \sum_{j}\gamma_{j}\mathcal{d}\left[\sigma_{j}^{-}\right]\rho_c \nonumber \\ & & + \sum_{j}\gamma_{j}\mathcal{d}\left[\sigma_{j}^{-}\right]\rho_c + \kappa\mathcal{d}\left[\sum_{j}\lambda\sigma_{j}^{-}\right]\rho_c+\sum_{xy}\left(\gamma_{d}^{xy}-\imath a_{c}^{xy}\right)\pi_{x}\rho_c\pi_{y}. % \equiv\mathcal{l}\rho , \label{eq : ms}\end{aligned}\ ] ] in writing eq.([eq : ms ] ) , we have transformed to a frame rotating with the qubits transition frequency @xmath70 . this is also a suitable frame for applying an additional microwave drive with a frequency in resonance with the qubits transition frequency in order to coherently control the qubits as discussed in the context of quantum feedback control in the next section . as a result , the @xmath71 term in eq . ( [ eq : effhamiltonian ] ) now acquires time - dependent oscillating factors with frequency @xmath72 ( as we have set @xmath73 ) in the commutator of the first term in eq . ( [ eq : ms ] ) . the third term in eq . ( [ eq : ms ] ) represents the purcell effect at the damping rate @xmath74 which can be reduced by operating the qubits in the dispersive regime @xcite , while the fourth term contains both the measurement - induced dephasing @xmath75 and the ac stark shift @xmath76 given by @xmath77,\label{eq : dephasing}\\ a_{c}^{xy } & = & \left(\chi_{x}-\chi_{y}\right)\text{re}\left[\alpha_{x}\alpha_{y}^{*}\right].\label{eq : acstark}\end{aligned}\ ] ] the last two terms of eq . ( [ eq : sme ] ) come from the last term of the conditional master equation ( [ eq : me ] ) in the displaced polaron - type frame , in which the cavity field is transformed into @xmath78 the measured homodyne current from eq . ( [ eq : homodyne_i ] ) becomes @xmath79 here the joint measurement operator @xmath80 is given by @xcite @xmath81 where @xmath82 @xmath83 where the vectors are @xmath84 and @xmath85 , @xmath86 , and @xmath87 is the measurement rate for the polarization of @xmath88 . note that the conditional stochastic master equation ( [ eq : sme ] ) after being averaged over all possible measurement records reduces to the unconditional , deterministic master equation , i.e. , eq . ( [ eq : sme ] ) but without its last two unraveling noise terms . the outcomes of the homodyne current ( [ eq : current ] ) depend on the choice of the local oscillator phase @xmath39 . we would like to generate the entanglement state @xmath66 by quantum measurement . thus we choose the phase to be @xmath89 such that @xmath66 state is one of the eigenstates of the measurement operator @xcite @xmath90 where @xmath91 and @xmath92 in this case , there are four measurement outcomes of @xmath93 : @xmath94 , @xmath95 , @xmath96 , and @xmath97 which correspond respectively to the all - qubit excited state @xmath98 , the two - qubit excited states @xmath99 , the single - qubit excited states @xmath100 , and the ground state @xmath101 . one can see from eq . ( [ eq : ha ] ) and the last two terms of eq . ( [ eq : sme ] ) that in addition to the measurement operator @xmath102 providing the qubit state information , performing the homodyne measurement at @xmath89 also produces a stochastic phase represented by the term associated with @xmath103 where @xmath104 note that @xmath105 generates different relative phase kicks only between two groups of states , i.e. , between the group of @xmath106 and the group of @xmath107 . in other words , no relative random phase kick between the constituent basis states of the @xmath66 state ( thus no additional unwanted dephasing between them ) also helps generate and stabilize the target @xmath66 state . when the coherent state amplitudes are steady , the rates @xmath108 , @xmath109 and @xmath110 become @xmath111 , \label{eq : rate0}\\ \sqrt{\kappa\gamma_{1 } } & = & \sqrt{\gamma_{\rm m}}\left[\frac{24\left(\chi/\kappa\right)^{2}}{1 + 40\left(\chi/\kappa\right)^{2}+144\left(\chi/\kappa\right)^{4}}\right],\label{eq : rate1}\\ \sqrt{\kappa\gamma_{2 } } & = & \sqrt{\gamma_{\rm m}}\left[\frac{-4\left(\chi/\kappa\right)}{1 + 40\left(\chi/\kappa\right)^{2}+144\left(\chi/\kappa\right)^{4}}\right],\label{eq : rate2}\end{aligned}\ ] ] where @xmath112 is the effective measurement rate obtained from the adiabatic elimination method @xcite . in general , the measurement rate is related to the decoherence rate as the decrease in the off - diagonal element is associated with the gradual projection onto one of the corresponding measurement eigenstates . for example , in the steady state , the measurement rate @xmath113 for efficiency @xmath37 is equal to twice of the decoherence rate , i.e. , @xmath114 , for the case by the adiabatic elimination method [ see eq.([eq : asme ] ) ] @xcite . similarly , there is a relationship between the measurement rates defined through eq . ( [ eq : jmop2 ] ) and the corresponding decoherence rates in eq . ( [ eq : dephasing ] ) . we may define for efficiency @xmath37 the ratio between them as @xmath115},\ ] ] where the numerator , @xmath116 , is the measurement rate to distinguish between two eigenstates @xmath117 and @xmath118 of the measurement operator @xmath119 ( also proportional to the separation between two measurement outcomes corresponding to the states @xmath117 and @xmath118 ) , and the denominator is the measurement - induced dephasing in eq . ( [ eq : dephasing ] ) . when the coherent amplitudes have reached steady state , it can be shown that @xmath120 in other words , the measurement rates to distinguish between the states @xmath121 , @xmath122 , @xmath123 , and @xmath124 are twice of their corresponding measurement - induced dephasing rates . we wish to apply the measurement - guided quantum feedback control to generate and stabilize the @xmath66 state , thus distinct values of measurement current , which reveal quickly the information of corresponding qubit states are favorable . the separations between adjacent measurement outcomes are @xmath125 that depend on the value of @xmath126 . generally , larger separation between measurement outcomes implies quicker corresponding measurement eigenstate readout . let us focus on the smaller separation of @xmath127 between the measurement outcome of the ground state and that of the @xmath66 state ( or the single - qubit excited states ) . when the ratio @xmath126 is decreased from @xmath128 to @xmath129 , the separation between the outcomes of the ground state and the single - qubit excited state [ see the red - dashed curve in fig . [ fig : fine](a ) ] increases initially , and then reaches a local maximum around @xmath130 . it then vanishes around @xmath131 and later reaches another local maximum around @xmath132 . however , a larger @xmath8 value leads to a larger damping rate , @xmath133 , of the purcell effect which deteriorates the average fidelity to generate and stabilize the @xmath134 state . as a result , the fidelity at @xmath135 is expected to be smaller than that at @xmath136 . we will show later that the average fidelity at @xmath135 is also smaller than that at , say , @xmath137 even though its measurement outcome separation is larger [ see fig . [ fig : fine ] and the discussion related to it in sec . [ sec : fidelity ] ] . thus for the simulations presented in this article , the dispersive coupling strength is chosen to be @xmath136 and/or @xmath138 . these are readily accessible parameter values @xcite . ( color online ) measured homodyne currents @xmath139 ( @xmath140 realizations that yield roughly the same steady outcome values are grouped and averaged ) for an initial cavity state in a vacuum state and an initial qubits state in the separable state @xmath141 of eq . ( [ eq : factorstate ] ) with the qubits decay rates @xmath142 . the four measurement outcomes in solid lines are for the case of polaron - type transformation and they correspond to the qubits collapsing respectively onto states @xmath143 , @xmath144 , @xmath6 , and @xmath101 from the top to the bottom , while the measurement outcomes in dashed lines corresponds to the case of adiabatic elimination . the parameters used are @xmath145 , @xmath136 , @xmath146 , @xmath37 , and @xmath142 . ] suppose we start to evolve the conditional master equation ( [ eq : sme ] ) with an initial state of the cavity in a vacuum state and the qubits in a separable state @xmath147 ideally , it is expected that the qubits under continuous measurements will collapse gradually onto one of the eigenstates of the joint - qubit measurement operator @xmath102 stochastically in each individual realization . indeed , by ignoring the decay rates of the qubits , i.e. , by setting @xmath148 , the initial qubit state @xmath141 of eq . ( [ eq : factorstate ] ) will collapse onto the states @xmath98 and @xmath149 with probability @xmath150 each , and onto the states @xmath151 and @xmath2 with probability @xmath152 each . this is shown in fig . [ fig:1 ] where the averaged measured currents @xmath139 obtained by categorizing and averaging @xmath140 realizations that yield roughly the same steady outcome values [ as implied by eq . ( [ eq : factorstate ] ) ] are presented . one can see that after the time of about @xmath153 , the cavity field has evolved from the initial vacuum state to correspondingly distinguishable coherent states and four distinct measurement outcomes ( solid lines in fig . [ fig:1 ] ) are observed . the measurement outcomes maintaining at certain values for a considerably long time indicate that the qubits have collapsed onto and stayed in the corresponding states of @xmath143 , @xmath144 , @xmath6 and @xmath101 . however , this scheme of producing entangled @xmath66 or @xmath144 states by measurement only is probabilistic , and in the presence of qubit relaxation , the probabilistically generated entangled state will jump into other states . to generate the @xmath66 state deterministically and stabilize it against the influence of the environments , we employ the adaptive quantum feedback control technique based on quantum state estimation . the conditional stochastic master equation with this kind of the feedback control scheme becomes @xcite @xmath154\rho_{c}\left(t\right)\xi\left(t\right)-\imath\sqrt{\kappa}\left[c_{-\pi/2},\rho_{c}\left(t\right)\right]\xi\left(t\right ) \nonumber \\ & & -i[h_{\rm fb}(t),\rho_c(t)].\label{eq : feedbacksme}\end{aligned}\ ] ] here @xmath155 is the feedback control hamiltonian with control parameters designed from an estimation of @xmath156 . the advantage of the quantum state estimation scheme is that the feedback control can be designed from an optimal control method to ensure the qubit system passing through the more efficient trajectory by optimizing the targeted objective or minimizing the cost function . we choose the objective function to be the fidelity @xmath157 $ ] , where @xmath158 , and choose the feedback control hamiltonian to be @xmath159 , i.e. , only single qubit rotations . the strategy @xcite to determine the feedback strengths @xmath160 at each point in time is chosen optimally by maximizing the fidelity @xmath157 $ ] . by considering the dynamics of the feedback hamiltonian only , @xmath161 $ ] , the time evolution of the fidelity is @xmath162 = \left\langle-\imath\left[\rho_{w^{- } } , h_{\rm fb}\right]\right\rangle_{c}\nonumber \\ & = & f_{1}\left\langle-\imath\left[\rho_{w^{- } } , \sigma_{1}^{x}\right]\right\rangle_{c}+f_{2}\left\langle-\imath\left[\rho_{w^{- } } , \sigma_{2}^{x}\right]\right\rangle_{c}+f_{3}\left\langle-\imath\left[\rho_{w^{- } } , \sigma_{3}^{x}\right]\right\rangle_{c}.\label{eq : ft}\end{aligned}\ ] ] to maximize the fidelity , i.e. , to make @xmath163 $ ] positive and maximal , the optimal feedback coefficients to kick the qubit system back to the desired state @xmath158 are determined by @xmath164\rangle_{c}\right)$ ] , where sgn(@xmath165 ) denotes the sign function that extracts the sign of a real number @xmath165 , and @xmath10 is the maximum feedback strength that can be applied . this is a bang - bang feedback control scheme , meaning that the feedback strengths are always at the maximum or minimum values @xcite . ( color online ) time evolutions of the average fidelity @xmath166 of @xmath6 state ( over 1000 realizations or trajectories ) as a function of time for different initial states of @xmath101 , @xmath6 , @xmath144 , @xmath101 and @xmath167 . the parameters used are @xmath168 , @xmath145 , @xmath136 , @xmath146 , @xmath37 , and @xmath169 . ] considering qubits decay rates @xmath170 as in ref . @xcite , we demonstrate in fig . [ fig : initial ] that the entangled @xmath6 state can be generated and stabilized with high average fidelity @xmath171 for various initial qubits states by our feedback control strategy with a moderate feedback strength of @xmath168 . the average fidelity @xmath166 is obtained by averaging @xmath172 over @xmath140 realizations or trajectories . the various initial states are @xmath173 , @xmath66 , @xmath174 , @xmath175 , eigenstates of the joint measurement operator @xmath102 , and the separable state @xmath176 of eq . ( [ eq : jmop2 ] ) . they all reach the average fidelity of about @xmath177 in a time scale of about a few @xmath178 , in about the same time scale for the cavity field to evolve into distinguishable coherent states ( without feedback control ) in fig . [ fig:1 ] . other parameters used in fig . [ fig : initial ] are the coupling strength @xmath146 , and the dispersive coupling strength @xmath136 . figure [ fig : frelax](a ) shows the time evolutions of the average fidelity of the @xmath6 state for different qubit decay rates . the brown - dashed line is the ensemble average ( unconditional ) result of the qubit system evolving from the @xmath6 state without feedback control for the qubit decay rates of @xmath169 . the fidelity without feedback control deteriorates about linearly with time . this can be understood from a typical measured current record in a single realization of the experiment as shown in fig . [ fig : frelax](b ) . the qubits initially in the @xmath6 state corresponding to the result of @xmath179 for the parameters chosen here have probability @xmath180 to make a sudden jump into the ground state @xmath101 corresponding to @xmath181 in the time interval @xmath182 $ ] . if there is no feedback control , the qubit after the sudden jump will then stay in the ground state as indicated by the red - dashed curve in fig . [ fig : frelax](b ) . in contrast , even when the initial qubit state is the ground state @xmath101 , the qubit with feedback control will be driven to the @xmath6 state and be stabilized for a sufficient long time . this is also shown in fig . [ fig : frelax](a ) in which the ensemble averaged fidelities with application of feedback control ( even when the decay rate of the qubits is @xmath183 ) are stabilized with values above that of the brown - dashed line ( with @xmath184 ) when time @xmath185 . we discuss in the following the dependence of the average fidelity on the dispersive coupling strength @xmath8 , the probe beam amplitude @xmath9 , the feedback strength @xmath10 , and the measurement efficiency @xmath11 . the black - circle solid line in fig . [ fig : fine ] is the average fidelity @xmath166 versus the dispersive coupling strength @xmath8 for the probe field @xmath145 , the feedback strength @xmath168 , and the decay rate of the qubits @xmath186 . the dependence of the fidelity on @xmath8 is similar to the red - dashed curve which represents the separation between the measurement output signal @xmath187 that corresponds to the qubits state being in @xmath6 and the output signal that corresponds to the qubits state of @xmath101 . this is because larger separation means better state distinguishibility and thus helps the conditional qubits state estimation in the quantum feedback control scheme . one can observe that @xmath6 and @xmath101 become indistinguishable from the measurement current around the point @xmath188 , and thus the fidelity drops sharply around there as well . one can also notice from fig . [ fig : fine ] that the separation reaches maximum values at @xmath136 and @xmath135 , but the fidelity is higher at @xmath136 . this is because the collective damping rate @xmath133 of the purcell effect of the third term in eq . ( [ eq : ms ] ) increases with the values of @xmath8 . for example , for @xmath146 , the purcell collective damping rate of @xmath189 at @xmath135 is larger than the individual qubit decay rate , set to be @xmath186 here , while the purcell collective damping rate of @xmath190 at @xmath136 is much smaller than @xmath191 and thus does not play an important role . as a result , the average fidelity is lower for the case of a higher @xmath8 value when the corresponding separation of the measurement outcomes is the same . another observation from fig . [ fig : fine ] is that the average fidelity does not change with the dispersive coupling strength @xmath8 as sharply as the separation of the measurement outcomes does . when the separation of the measurement outcomes above a certain value ( about @xmath192 for the parameters chosen here ) , the average fidelity does not vary much [ see the behaviors of the separation of the measurement outcomes and the average fidelity around @xmath136 , where the purcell effect is not significant as compared to the individual qubits decay ] . this may also explain why the average fidelity at @xmath137 is larger than that at @xmath135 . the purcell collective damping rate of @xmath193 at @xmath137 , which is smaller than the individual qubit decay rate @xmath186 , is smaller than that of @xmath189 at @xmath135 . although the separation of the measurement outcomes at @xmath137 is also smaller there , its value is larger than @xmath192 . as a result , the average fidelity at @xmath137 is larger than that at @xmath135 . we perform most of our simulations choosing @xmath136 and/or @xmath137 . the dependence of average fidelity on the measurement drive amplitude @xmath9 is shown in fig . [ fig : fine ] . since the information gain rates [ c.f . ( [ eq : rate0 ] ) and ( [ eq : rate1 ] ) ] is proportional to @xmath194 , the bigger the value @xmath9 is , the larger the separation between measurement outcomes is and the quicker the conditional state collapse to one of the joint measurement operator eigenstates is . it is thus expected the average fidelity will also become better as @xmath9 increases as shown in fig . [ fig : fine ] . one may be temped to think that the arbitrarily quick readout or arbitrarily high fidelity can be achieved by simply increasing @xmath9 . but it was pointed out @xcite that the lowest - order dispersive approximation of hamiltonian eq . ( [ eq : effhamiltonian ] ) become accurate when the average photon number in the cavity is much smaller than the critical photon number of @xmath195 . the number of photon is proportional to @xmath196 . this puts a limit on how large the external drive @xmath9 could be for eq . ( [ eq : effhamiltonian ] ) to hold valid . in addition , note that the time - dependent second term @xmath197 in the first commutator of eq . ( [ eq : me ] ) also increases with @xmath9 . this term in the hamiltonian , in addition to qubit decay channel , will cause the qubits to flip or change their state during the process when the continuous measurement tries to localize the qubits to one of the joint measurement operator eigenstates . however , the value of @xmath198 we choose is small and for typical value of @xmath9 , the coefficient @xmath199 of this term is much smaller than the frequency @xmath200 ( as we have set @xmath73 ) of the oscillating factors . thus the effect of this term to mix different measurement eigenstates is small . we choose @xmath145 for most of the simulations presented in this paper although increasing @xmath9 further will improve the fidelity a little bit . figure [ fig : fine ] shows that increasing the feedback control strength improves the average fidelity in general . suppose a measurement outcome indicating deviation from the desired @xmath6 state happens , the applied feedback control has to overcome the effect of the localization due to the continuous measurement in order to move the qubits back to the target @xmath6 state . when the feedback control strength is smaller , the procedure to produce and stabilize the @xmath6 state takes a longer time with a lower fidelity . the qubits decay rates in fig . [ fig : fine ] are chosen to be @xmath169 . when the feedback control strength is @xmath168 , the fidelity reaches the value of @xmath177 . further increase of the feedback control strength , i.e. , @xmath201 , does not improve appreciably the fidelity . this indicates that if deviation occurs , the correction of the feedback control at @xmath168 is fast enough to kick the qubits back to the @xmath6 state . we thus choose @xmath168 for most of our numerical simulations . in practice , there exists inefficiency in the measurements which arises when the detectors sometimes miss detection or the measurement microwave photons does not go to the detectors due to lost . however , high measurement efficiency is not very essential for our feedback control scheme . although the fidelity decreases as the value of the measurement efficiency @xmath11 decreases as shown in fig . [ fig : fine ] , the fidelity is still above @xmath1 for @xmath11 as low as @xmath202 for the case of @xmath136 and qubits decay rates @xmath169 . the value of @xmath38 implies appearance of an additional non - unraveling dephasing term in the quantum trajectory ( stochastic master ) equation @xcite . however , this term causes dephasing only among @xmath143 , @xmath144 , @xmath6 and @xmath101 for the initial qubits states chosen in our simulations . as a result , it affects only the detailed dynamics of the qubits but does not destroy or prevent the controlled evolution toward the target entangled state @xmath6 . therefore , our feedback control scheme to generate and stabilize the @xmath6 state does not require high measurement efficiency and thus can be implemented experimentally with high fidelity with measurement efficiency available in current circuit qed experiments that use parametric amplification before the homodyne detection with an iq mixer @xcite . for example , refs . have achieved an effective quantum efficiency of @xmath203 . ( color online ) time evolutions of the average fidelity @xmath166 of the @xmath6 state with ( @xmath168 ) and without ( @xmath204 ) quantum feedback control for three sets of different individual qubits decay rates . the initial state for the case with feedback control ( @xmath168 ) is the ground state @xmath173 while it is the @xmath6 state for the case without feedback control ( @xmath204 ) . other parameters used are @xmath145 , @xmath136 , @xmath146 , @xmath37 , and @xmath205 . ] our feedback control scheme is robust even when the decay rates of the qubits are different . this is shown in fig . [ fig:9 ] where the time evolutions of the average fidelity of the @xmath6 state with and without feedback control for three sets of qubits decay rates are shown . the average fidelity is determined roughly by the average decay rate in each set as the behavior of the fidelity in each set is similar to that when the decay rates of the three qubits were equal to the average decay rate . the average fidelities of the @xmath6 state generated initially from the ground state @xmath173 with feedback control strength @xmath168 outperform those evolving from an initial @xmath6 state without feedback control ( @xmath204 ) after time @xmath206 . another commonly used procedure to eliminate the cavity field is the adiabatic elimination method . both the adiabatic method and the polaron - type transformation method assume that @xmath207 , but the adiabatic method employs an additional condition , i.e. , to assume that the damping of the cavity is much larger than the dispersive coupling strength , i.e. , @xmath208 . in the limit of @xmath209 , the term @xmath210 in eq . ( [ eq : coherent ] ) is ignored and the tlr cavity field reaches its steady coherent state rapidly with an amplitude equal to @xmath211 . as a consequence , the coherent state amplitude is assumed to be the same for all the qubits basis states . this is in contrast to the case in the polaron - type transformation method where the time - dependent coherent state amplitudes @xmath51 shown in eq . ( [ eq : coherent ] ) depend on @xmath212 and thus on the qubits basis states @xmath117 . the steady - state information gain rates in the limit of @xmath209 from eqs . ( [ eq : rate0 ] ) and ( [ eq : rate1 ] ) become @xmath213 and @xmath214 ( note also that @xmath215 ) . as a result , the measurement operator from eq . ( [ eq : jmop2 ] ) becomes @xmath216 . the effective conditional ( stochastic ) master equation ( [ eq : sme ] ) in the case of adiabatic elimination also reduces to @xmath217\sigma_{j}^{z } + \sum_{j}\epsilon\lambda(\sigma_{j}^{+}e^{i\delta t}+\sigma_{j}^{-}e^{-i\delta t } ) + \sum_{j > i}\chi\left(\sigma_{i}^{-}\sigma_{j}^{+}+\sigma_{i}^{+}\sigma_{j}^{-}\right ) , % + \chi\left|\alpha\right|^{2}\sum_{j}\sigma_{j}^{z } , \rho_{c}^{e}\left(t\right)\right]\nonumber\\ & & + \sum_{j}\gamma_{j}\mathcal{d}\left[\sigma_{j}^{-}\right]\rho_{c}^{e}\left(t\right)+\kappa\mathcal{d}\left[\sum_{j}\lambda\sigma_{j}^{-}\right]\rho^{e}_{c}\left(t\right)\nonumber\\ & & + \frac{\gamma_{\rm e}}{2}\mathcal{d}\left[\sum_{j}\sigma_{j}^{z}\right]\rho^{e}_{c}\left(t\right ) + \frac{\sqrt{\eta \gamma_{\rm m}}}{2}\mathcal{h}\left[\sum_{j}\sigma_{j}^{z}\right]\rho^{e}_{c}\left(t\right)\xi\left(t\right).\label{eq : asme}\end{aligned}\ ] ] here , the second term in the first commutator term and the fourth term of eq . ( [ eq : asme ] ) are reduced respectively from eqs . ( [ eq : acstark ] ) and ( [ eq : dephasing ] ) of eq . ( [ eq : ms ] ) . note again that the measurement rate here is twice of the decoherencee rate , @xmath218 . one can clearly see that the adiabatic elimination procedure is a special case of polaron - type transformation in the limit of @xmath219 . the measurement outcomes of the average homodyne currents obtained by categorizing and averaging @xmath140 realizations that yield roughly the same steady outcome values for the case of adiabatic elimination are plotted in fig . [ fig:1 ] to compare with the case of polaron - type transformation with the same parameters . the qubits are initially in the separable state @xmath141 of eq . ( [ eq : factorstate ] ) with the qubits decay rates set to zero , i.e. , @xmath142 , and the cavity state evolves from an initial vacuum state . one can see that the measurement outcomes in dashed lines for the adiabatic elimination case approach to their corresponding steady values more quickly . moreover , the four measurement outcomes in the adiabatic elimination limit become @xmath220 , @xmath221 , @xmath222 , and @xmath223 ; as a result , the steady value corresponding to @xmath143 ( @xmath101 ) is overestimated , i.e. , becomes larger ( smaller ) . thus neglecting the contribution of @xmath214 in the case of adiabatic elimination for the parameter of @xmath224 used in fig . [ fig:1 ] is not really valid . for the adoptive feedback control by state estimation method , it is important to use the correct conditional stochastic master equation to estimate the system state conditioned on the measured current . otherwise , wrong state estimation information will give rise to bad feedback control result . we have tested numerically that when @xmath225 both the adiabatic elimination method and the polaron - type transformation method give the same result for the typical parameters chosen in our simulation in the absent of feedback control . however , when @xmath226 , discrepancy in conditional qubits trajectories starts to emerge . however , fig . [ fig : fine ] indicates that in the presence of feedback control the average fidelity is below @xmath227 for @xmath170 with this value of @xmath228 . for the value of @xmath229 , the fidelity to stabilize the @xmath66 can be maintained at @xmath177 using the feedback control master equation ( [ eq : feedbacksme ] ) obtained by the polaron - type transformation method . in other words , in the parameter regime where the adiabatic elimination does not apply , the conditional stochastic master equation ( [ eq : asme ] ) can not be used ; otherwise the wrong information about the system state will lead to low - fidelity feedback control results . if one would use the conditional master equation ( [ eq : asme ] ) obtained by the adiabatic elimination method to perform the feedback control scheme calculation by adding a feedback hamiltonian commutator term for the case of @xmath230 , high average fidelity of @xmath231 could be achieved . but this is not correct as the conditional master equation ( [ eq : asme ] ) is not really valid when @xmath224 . in fact , if we nevertheless use the conditional master equation ( [ eq : asme ] ) obtained by the adiabatic elimination method for the state estimation to determine the sign of the feedback strength and then use the polaron - type feedback control master equation ( [ eq : feedbacksme ] ) ( mimicking the real experimental situation ) to evolve and calculate the fidelity , the average fidelity is found to be below @xmath232 . this is because the signs of the feedback control strength estimated by eq . ( [ eq : asme ] ) obtained by the adiabatic elimination method for the value of @xmath229 are often wrong . in conclusion , we have presented a simple and promising quantum feedback control scheme for deterministic generation and stabilization of a three - qubit @xmath134 state in a superconducting circuit qed setup , taking into account the realistic conditions of decoherence and decay . our scheme is based on continuous joint zeno measurements of multiple qubits in a dispersive regime and the application of multi - qubit adaptive feedback control . the dispersive measurement not only enables qubit state estimation for further information processing but also allows , together with the feedback control , for the generation and stabilization of the target entangled @xmath134 state starting from separable input states or from the ground states of the qubits . the feedback control hamiltonian can be realized by applying , besides the measurement drive , an additional control microwave drive with a frequency in resonance with the qubits transition frequency . we have employed the polaron - type transformation method to eliminate the cavity field to obtain an effective stochastic master equation for the qubits degrees of freedom alone , and simulated the dynamics of the proposed quantum feedback control scheme using the quantum trajectory approach . it is demonstrated that in the presence of moderate environmental decoherence , the average entangled state fidelity higher than @xmath1 can be achieved and maintained for a considerable long time ( much longer than the qubits decoherence time ) with our scheme . in our discussion , we have assumed to have identical qubits with transition frequencies @xmath64 and couplings @xmath65 . although the couplings @xmath233 of the qubits to the cavity field may , in a realistic experiment , not be the same , one is able to tune , due to their great tunability , the qubit transition frequencies to be pretty much the same by external voltages or magnetic fields . in other words , experimentally the detuning @xmath234 for all the qubits can be tuned to be equal while the dispersive coupling strengths @xmath235 are left slightly different . we have tested numerically that a mismatch smaller than @xmath236 ( we set @xmath146 ) in the coupling strengths @xmath233 changes insignificantly the fidelity to achieve the desired @xmath134 state . however , a mismatch of @xmath237 ( @xmath238 ) in @xmath233 results in a fidelity change , say for @xmath239 case , from 0.98 to 0.93 ( 0.84 ) . taking the values of @xmath34 to be about @xmath240 mhz yields a tolerant mismatch in @xmath233 , which will not affect the desired fidelity , to be about @xmath241 mhz . the required values for the physical parameters are achievable in current experiments . our method can also be extended straightforwardly to generate and stabilize an @xmath3-qubit ( with @xmath242 ) w - type state with one excitation shared across @xmath3 qubits in superposition . we have also compared the polaron - type transformation method with the adiabatic elimination method to eliminate the cavity field . it is shown that the adiabatic elimination procedure is a special case of polaron - type transformation in the limit of @xmath219 . our feedback control scheme is also shown to be robust against measurement inefficiency and individual qubit decay rate differences . recently , quantum feedback experiments stabilizing fock states of light in a cavity by using sensitive atoms , crossing the field one at a time as quantum nondemolition probes of its photon number have been reported @xcite . an experiment of stabilizing rabi oscillation of a superconducting qubit in a cavity using quantum feedback control via homodyne measurements has also been demonstrated @xcite . although the measurement efficiency was estimated to be about @xmath243 in the quantum feedback control experiment of a superconducting qubit , the rabi oscillations was shown to persist indefinitely . our feedback control scheme to generate and stabilize entangled state can still achieve high fidelity even when measurement efficiency is as low as @xmath243 . furthermore , processing data in real time using fast field - programmable gate array ( fpga ) electronics in circuit qed setup has been demonstrated @xcite , and this will facilitate the performance of quantum state estimation in real time in our scheme . thus our quantum feedback scheme has great potential to be realized experimentally in the near future . hsg acknowledges support from the national science council in taiwan under grant no . 100 - 2112-m-002 - 003-my3 , from the national taiwan university under grants no . 102r891400 , no . 102r891402 and no . 102r3253 , and from the focus group program of the national center for theoretical sciences , taiwan . 60ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1103/physrevlett.77.1413 [ * * , ( ) ] link:\doibase 10.1103/physreva.62.062314 [ * * , ( ) ] link:\doibase 10.1103/physreva.78.060301 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.86.910 [ * * , ( ) ] http://dx.doi.org/10.1038/nature04279 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.92.077901 [ * * , ( ) ] http://dx.doi.org/10.1038/nature09418 [ * * , ( ) ] a. a. gangat , i. p. mcculloch , and g. j. milburn , e - print arxiv:1304.4065 . link:\doibase 10.1103/physreva.69.062320 [ * * , ( ) ] http://dx.doi.org/10.1038/nature02851 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.95.060501 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.94.123602 [ * * , ( ) ] link:\doibase 10.1103/physreva.74.042318 [ * * , ( ) ] link:\doibase 10.1103/physreva.75.032329 [ * * , ( ) ] link:\doibase 10.1103/physreva.76.012325 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.080502 [ * * , ( ) ] v. bonzom , h. bouzidi , and p. degiovanni , eur . j. d , * 47 * , 133 ( 2008 ) . link:\doibase 10.1103/physreva.79.013819 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.102.200402 [ * * , ( ) ] http://dx.doi.org/10.1038/nature09416 [ * * , ( ) ] link:\doibase 10.1103/physreva.81.062325 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.104.100504 [ * * , ( ) ] link:\doibase 10.1103/physreva.84.061805 [ * * , ( ) ] l. dicarlo , j. m. chow , j. m. gambetta , lev s. bishop , b. r. johnson , d. i. schuster , j. majer , a. blais , l. frunzio , s. m. girvin and r. j. schoelkopf , nature * 460 * , 240 ( 2009 ) . m. d. reed , l. dicarlo , s. e. nigg , l. sun , l. frunzio , s. m. girvin , and r. j. schoelkopf , nature * 482 * , 382 ( 2012 ) . a. fedorov , l. steffen , m. baur , m. p. da silva , and a. wallraff , nature * 481 * , 170 ( 2012 ) . nadav katz et . al . , phys . * 101 * , 200401 ( 2008 ) . r. vijay , d. h. slichter , and i. siddiqi , phys . 106 * , 110502 ( 2011 ) . r. vijay et al . , nature * 490 * , 77 ( 2012 ) . k. w. murch et . al . , phys . * 109 * , 183602 ( 2012 ) . k. w. murch , s. j. weber , k. m. beck , eran ginossar , and i. siddiqi , e - print arxiv:1301.6276 . k. w. murch , s. j. weber , c. macklin , and i. siddiqi , e - print arxiv:1305.7270 . d. rist , m. dukalski , c. a. watson , g. de lange , m. j. tiggelman , ya . m. blanter , k. w. lehnert , r. n. schouten , and l. dicarlo , e - print arxiv:1306.4002 . http://stacks.iop.org/0957-4484/21/i=27/a=274015 [ * * , ( ) ] s. spilla , r. migliore , m. scala , and a. napoli , j. phys . . phys . * 45 * , 065501 ( 2012 ) . link:\doibase 10.1103/physreva.79.052328 [ * * , ( ) ] l. s. bishop , l. tornberg , d. price , e. ginossar , a. nunnenkamp , a. a. houck , j. m. gambetta , j. koch , g. johansson , s. m. girvin , and r. j. schoelkopf , new j. phys . * 11 * , 073040 ( 2009 ) . link:\doibase 10.1103/physreva.81.040301 [ * * , ( ) ] link:\doibase 10.1103/physreva.47.642 [ * * , ( ) ] h . m. wiseman and g. j. milburn , quantum measurement and control ( cambridge university press , cambridge , 2009 ) link:\doibase 10.1103/physreva.76.010301 [ * * , ( ) ] m. sarovar , h .- s . goan , t. p. spiller and g. j. milburn , phys . a * 72 * , 062327 ( 2005 ) . link:\doibase 10.1103/physreva.82.032335 [ * * , ( ) ] link:\doibase 10.1103/physreva.82.012329 [ * * , ( ) ] link:\doibase 10.1103/physreva.83.042313 [ * * , ( ) ] k. keane and a. n. korotkov , e - print arxiv:1205.1836 . link:\doibase 10.1103/physreva.50.5256 [ * * , ( ) ] link:\doibase 10.1103/physreva.68.032301 [ * * , ( ) ] ; s. raghunathan , t. a. brun and h .- s . goan , phys . a * 82 * , 052319 ( 2010 ) link:\doibase 10.1103/physreva.71.042309 [ * * , ( ) ] link:\doibase 10.1103/physreva.77.012112 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.103.083601 [ * * , ( ) ] q .- q . wu , l. xu , q .- s . tan , and l .- l . . phys . * 51 * , 1482 ( 2012 ) . d. i. schuster , a. a. houck , j. a. schreier , a. wallraff , j. m. gambetta1 , a. blais , l. frunzio , j. majer , b. johnson , m. h. devoret , s. m. girvin and r. j. schoelkopf , nature * 445 * , 515 ( 2007 ) . j. m. fink , m. gppl , m. baur , r. bianchetti , p. j. leek , a. blais , and a. wallraff , nature * 454 * , 315 ( 2008 ) . lev s. bishop , j. m. chow , jens koch , a. a. houck , m. h. devoret , e. thuneberg , s. m. girvin , and r. j. schoelkopf , nat . phys . * 5 * , 105 ( 2009 ) . link:\doibase 10.1103/physreva.65.042301 [ * * , ( ) ] m . a. castellanos - beltran , k. d. irwin , g. c. hilton , l. r. vale , and k. w. lehnert , nat . phys . * 4 * , 929 ( 2008 ) . f. mallet , m. a. castellanos - beltran , h. s. ku , s. glancy , e. knill , k. d. irwin , g. c. hilton , l. r. vale , and k. w. lehnert , phys . 106 * , 220502 ( 2011 ) . c. sayrin , i. dotsenko , x. zhou , b. peaudecerf , t. rybarczyk , s. gleyzes , p. rouchon , m. mirrahimi , h. amini , m. brune , j. m. raimond , and s. haroche , nature * 477 * , 73 ( 2011 ) ; x. zhou , i. dotsenko , b. peaudecerf , t. rybarczyk , c. sayrin , s. gleyzes , j. m. raimond , m. brune , and s. haroche , phys . * 108 * , 243602 ( 2012 ) ; b. peaudecerf , c. sayrin , x. zhou , t. rybarczyk , s. gleyzes , i. dotsenko , j. m. raimond , m. brune , and s. haroche , phys . a * 87 * , 042320 ( 2013 ) . d. bozyigit , c. lang , l. steffen , j. m. fink , c. eichler , m. baur , r. bianchetti , p. j. leek , s. filipp , m. p. da silva , a. blais , and a. wallraff , nat . phys . * 7 * , 154 ( 2011 ) .

we present results from detailed imaging of the centrally dominant radio elliptical galaxy in the cooling flow cluster abell 2597 , using data obtained with the wide field and planetary camera 2 ( wfpc2 ) on the _ hubble space telescope ( hst)_. this object is one of the archetypal `` blue - lobed '' cooling flow radio elliptical galaxies , also displaying a luminous emission - line nebula , a compact radio source , and a significant dust lane and evidence of molecular gas in its center . we show that the radio source is surrounded by a complex network of emission - line filaments , some of which display a close spatial association with the outer boundary of the radio lobes . we present a detailed analysis of the physical properties of ionized and neutral gas associated with the radio lobes , and show that their properties are strongly suggestive of direct interactions between the radio plasma and ambient gas . we resolve the blue continuum emission into a series of knots and clumps , and present evidence that these are most likely due to regions of recent star formation . we investigate several possible triggering mechanisms for the star formation , including direct interactions with the radio source , filaments condensing from the cooling flow , or the result of an interaction with a gas - rich galaxy , which may also have been responsible for fueling the active nucleus . we propose that the properties of the source are plausibly explained in terms of accretion of gas by the cd during an interaction with a gas - rich galaxy , which combined with the fact that this object is located at the center of a dense , high - pressure icm can account for the high rates of star formation and the strong confinement of the radio source . submitted to _ the astrophysical journal _

astro-ph9906335

rich clusters of galaxies generally contain a diffuse intracluster medium ( icm ) having temperatures @xmath0 k and densities @xmath1 @xmath2 , as inferred from observations at x - ray and other wavebands ( e.g. , ; ) . an important question related to the overall physics of the icm concerns the central regions of clusters ( @xmath3 kpc ) , where the inferred icm densities and pressures in some cases are sufficiently high that cooling to @xmath4 k can occur on timescales shorter than the cluster lifetime ( e.g. , ; ) . these `` cooling flow '' clusters often exhibit intense optical emission - line nebulae associated with the centrally dominant ( cd ) galaxies at their centers , together with blue continuum excess emission , and the strength of these effects appears to correlate with the cooling rate or central pressure of the x - ray emitting gas (; ; ; ; ; ; , , ; , ; ; ) . the majority of cooling flow cluster cd galaxies are also radio - loud , hosting moderately powerful but relatively compact radio sources @xcite . several intriguing questions have been posed concerning cds in cooling flows . for example , one question concerns the energy source for the line emission : the optical line luminosities are generally one to three orders of magnitude higher than would be expected from the x - ray - derived cooling rates , if each proton in the cooling gas experienced only a single recombination ( e.g. , ; ) . this has led to the consideration of numerous other potential sources of heating , including photoionization by young stars or the agn , auto - ionizing shocks , photoionization by x - rays from the icm , or cosmic - ray heating ( e.g. , and references therein ) . likewise the blue continuum emission is of considerable interest , and suggested possibilities for its origin have included scattered nuclear light , recent star formation , or inverse compton emission . finally , a crucial point to investigate concerns the inter - relationships between these various components : the cd elliptical , the emission - line gas , the blue continuum , the radio source , and the surrounding dense high - pressure icm : how are these related to one another , and do they have underlying common physical relationships ? we present results from a detailed study of a comparatively nearby strong cooling flow cluster , abell 2597 , using images obtained with the wide field and planetary camera 2 ( wfpc2 ) on board the _ hubble space telescope ( hst)_. previous radio and x - ray images have revealed that the cluster contains a moderately luminous radio source ( @xmath5 w hz@xmath6 , ; @xmath7 , ; @xmath8 , ) embedded in a luminous x - ray halo with @xmath9 ergs s@xmath6 (; ; ; ) . the host galaxy of the radio source is a cd elliptical containing an extended , luminous emission - line nebula (; ; ; ) . the galaxy is remarkable for its `` blue lobes '' of continuum excess emission that are approximately co - incident with the radio lobes ( , ) . spatially extended 21-cm absorption has also been detected against the radio lobes . although the radio source is comparatively compact ( with a size @xmath10 kpc ) , its power is comparable to that of much larger fr ii ( `` classical double '' ) sources , suggesting that the dense intracluster medium may play a role in its confinement . the source may be the nearest example of a compact steep - spectrum source . here we present data that fully resolve the blue emission in the lobes and reveal a complex network of emission - line and continuum filaments surrounding the central source . all distance - dependent quantities in this paper have been calculated assuming @xmath11 km s@xmath6 mpc@xmath6 and @xmath12 . we observed a2597 in july 1996 with the _ hst _ wfpc2 camera # 3 ( wf3 ) through the f410 m filter , with the total exposure time being `` cr - split '' into three equal , consecutive exposures to facilitate the removal of cosmic rays . this is effectively a narrow - band image , mostly containing flux from the redshifted [ ] @xmath133727 emission line . in this paper we also discuss new work that we have carried out on two archival broad - band images of a2597 , taken with the wfpc2 planetary camera ( pc1 ) through the f450w and f702w filters ( program i d # 6228 in the _ hst _ data archive ; see also ) . these images contain substantial continuum flux in addition to the redshifted [ ] @xmath133727 and h@xmath14 + [ ] @xmath156548,6583 emission lines , and therefore allow a detailed investigation of the emission - line and stellar continuum properties of this object , when combined with the f410 m image . in table [ tab : hst_obsns ] we present further details of the observations and filter characteristics . lcccccl 1996 july 27 & 2100 & f410 m & 4090.1 & 146.7 & wf3 & 0@xmath160996 1995 july 5 & 2500 & f450w & 4519.1 & 956.6 & pc1 & 0@xmath160455 1995 july 5 & 2100 & f702w & 6867.8 & 1382.5 & pc1 & 0@xmath160455 recalibration of the data was performed using the iraf / stsdas pipeline processing software for wfpc2 ( see also , ) , using updated calibration reference files where applicable . the removal of hot pixels with elevated dark currents was investigated using the iraf / stsdas task warmpix together with a table of hot pixels nearest in time to the observations , although it was found that most of these were adequately removed through the use of the appropriate dark frame . we verified that no significant alignment shifts were present between the multiple exposures in each of the three filters , before combining them using the iraf / stsdas task crrej which removes most of the cosmic rays as well as some remaining hot and cold pixels . the cosmic ray and bad pixel removal was checked by examining the residual difference between the cleaned image and each of the two original frames , and interpolations were performed across the remaining defects . sky subtraction in the broad - band f450w pc1 image was carried out by obtaining a mean value of the sky background in several empty regions of the frame , and subtracting this as a constant value from the entire image . the sky background for the f702w pc1 image was determined by scaling the f450w sky value using the sky count rates presented for these filters in the _ hst _ data handbook . this was necessary since the extended envelope of the elliptical is more prominent in the f702w image than in the f450w image , thereby precluding an accurate direct determination of the sky background . the sky count rate values ( 0.0071 and 0.0028 e@xmath17 s@xmath6 pixel@xmath6 for the f450w and f702w images respectively ) were sufficiently low that residual large - scale fluctuations in the images were at levels @xmath18 counts , well within the poisson noise limits . the two broad - band images were transformed to f450w and f702w magnitudes , @xmath19 and @xmath20 respectively , in the _ hst _ wfpc2 synthetic `` vegamag '' photometric system , defined in , making use of the iraf / stsdas synphot package to obtain accurate photometric parameters for the filters used . conversions to johnson _ b _ and _ r _ magnitudes were not carried out at this point , since substantial differences between the wfpc2 filter response functions and those of the johnson standard system would introduce photometric errors if significant color variations are present across the galaxy . in figure [ fig : f702w ] we present the broad - band f702w pc1 image of a2597 . it is clear that the central region of the elliptical is dominated by a complex network of filaments . some of the filaments appear clumpy and marginally resolved , particularly the arcs extending toward the north - east and the south . there is also some evidence of a dust lane crossing the nucleus , at approximately the same position angle as the major axis of the elliptical . a small object , possibly a companion galaxy , is present to the north of the elliptical , and numerous star clusters are also distributed throughout the elliptical , some of which may be resolved ( cf . ) . the filamentary structure dominates the narrow - band emission - line image of the galaxy , as shown in our f410 m wf3 image presented in figure [ fig : f450w_3 cm ] , which contains the redshifted [ ] @xmath133727 emission . we superpose contours from a 3.5 cm vla image of the radio continuum emission . the radio and optical images were aligned on the basis of the radio core position , which is assumed to coincide with the broad - band optical nucleus of the galaxy . having aligned the images on this basis alone , it is remarkable that the bright northern filamentary arc follows precisely along the outer contours of the radio lobe . the rest of the northern radio continuum is also apparently associated with lower - level filamentary emission . although there is less optical filamentation associated with the southern radio lobe , an emission - line filament is present near the eastern edge of this lobe . closer to the center , there is a bright , compact region of emission immediately south - west of the nucleus , which is co - incident with a radio hotspot as well as an abrupt change in the orientation of the radio lobe . we also note that some of the filamentary structures within the galaxy are apparently unassociated with the radio lobes , particularly the filaments extending away from the nucleus toward the east , and a more diffuse region surrounding the nucleus on the west . one of the interesting questions related to the properties of the filaments concerns the relative fractions of continuum and emission - line contributions to their total flux . we have investigated this by comparing the narrow - band f410 m image to the broad - band f450w image , both of which contain the redshifted [ ] @xmath133727 emission line but contain different amounts of stellar continuum flux . we used the iraf / stsdas package drizzle @xcite to transform the higher - resolution pc1 f450w image to the coarser pixel scale of the wf3 f410 m image , also taking into account the differences in orientation angles and geometrical distortions across the chips . we used the synphot package to flux - calibrate each of these two images in cgs units ( ergs s@xmath6 @xmath21 arcsec@xmath22 ) . a fundamental uncertainty inherent in this calibration is the lack of a detailed knowledge of the spectral energy distribution across the filter bandpass at each pixel in the image . two limiting cases are : ( 1 ) monochromatic emission , with all the flux being emitted at one wavelength ( appropriate for pixels dominated by a single emission line ) ; ( 2 ) a flat spectral energy distribution , specifically with constant flux density per unit wavelength @xmath23 . for each of the two filters , we generated a flux calibration factor for each limiting case using the synphot task bandpar . the difference between the two cases is negligible for the f410 m filter ( @xmath24 ) but is of order @xmath25 for the broader f450w filter . therefore we flux - calibrated the f410 m filter using the monochromatic limit , while the f450w filter was calibrated with the flat - spectrum approximation , which is justified by the fact that the flux in this broader filter is dominated by stellar continuum ( based on examining the spatially integrated ground - based spectra of ) . we note however that this approximation may break down for individual pixels with extremely high [ ] @xmath133727 equivalent widths , introducing a maximum error of @xmath25 ( for pure line emission ) , or more likely @xmath26 ( for comparable proportions of line and continuum emission ) . we then subtracted the flux - calibrated f410 m image from the geometrically transformed f450w image to investigate whether the filaments display any residual continuum emission . as described above , the uncertainties in flux calibration due to differences in the filter bandpasses are likely to be @xmath27 , therefore this subtraction provides a reliable indication of the relative contribution of emission - line and continuum emission to the filaments . we found that the extended filaments and arcs around the radio lobes disappeared almost entirely in the subtracted image , and therefore can be considered to consist predominantly of [ ] @xmath133727 line emission . the filaments nearer the galaxy nucleus , however , display a mixture of continuum and line emission and are apparently dominated in some regions by excess blue continuum , which we discuss in more detail in the following section . we have used the broad - band pc1 f450w and f702w images to investigate the detailed spatial distribution and morphology of excess blue emission in a2597 , which was first discovered in ground - based images by as marginally resolved _ u @xmath28 i _ continuum excess aligned along the radio axis . we first matched both pc1 images to a common point - spread function ( psf ) by convolving each image with the psf of the other , having computed the appropriate psf for each filter band using the _ hst _ tiny tim software @xcite . the images were then calibrated in the _ hst _ vegamag system , after which we calculated the @xmath29 color distribution of the galaxy , which we present in figure [ fig : m450w - m702w ] . we have also verified that smoothing the _ hst _ data to the resolution of the ground - based images reveals a good correspondence between the _ u_-band and f450w blue continuum . the blue emission in figure [ fig : m450w - m702w ] is resolved in detail into a number of separate components : ( 1 ) a band of extended , intense blue emission around the `` base '' of the northern radio lobe , immediately north - east of the nucleus ; ( 2 ) a group of blue point - like features south - west from the southern radio lobe ; ( 3 ) a blue arc that appears to trace the edge of the southern radio lobe ; ( 4 ) diffuse blue emission around the radio lobes , and specifically to the north - west of the nucleus . the possibility that the blue continuum is predominantly due to scattered nuclear light is ruled out on the basis of its low observed polarization ; optical synchrotron or inverse compton emission are also not feasible since none of the blue continuum sources correspond to strong radio hotspots . the most plausible remaining explanation is that the blue continuum is due to young stars , and in this section we examine various star formation scenarios that may be responsible for producing the extended blue continuum and the isolated blue clusters . in [ sec : implications ] we present a more detailed discussion of the relative plausibility of several possible triggering mechanisms for the star formation , together with various scenarios for the origin of the gas from which the stars are formed . our _ hst _ images resolve an intense band of blue continuum to the east and north of the nucleus , extending around toward the west , thereby covering a spatial scale of several kiloparsecs ( figure [ fig : m450w - m702w ] ) . some less intense blue emission is also associated with the southern radio lobe . resolving the blue emission allows a more accurate estimate of its intrinsic color and luminosity than was previously obtainable from the ground - based measurements obtained by , therefore we use our wfpc2 data to investigate the star formation processes involved . the intrinsic color of the blue excess is derived from the residual continuum in each band after subtracting the underlying elliptical fit , as described in more detail in [ sec : ell_fits ] . in that section we also discuss the effects of dust obscuration , which is present in the f702w band as well as f450w . since the strongest obscuration is confined to a few small localized patches , we are still able to discuss the overall properties of the remainder of the blue continuum excess . both the f450w and f702w bands are also contaminated by redshifted emission lines , [ ] @xmath133727 and h@xmath14 + [ ] respectively . however , the total integrated flux from the excess blue emission in the f450w image is substantially higher than the flux from the [ ] @xmath133727 emission in that region observed through the f410 m filter ; this also applies to the comparison between the f702w excess continuum and the h@xmath14 + [ ] flux scaled from the [ ] @xmath133727 flux using the mean h@xmath14 + [ ] / [ ] @xmath133727 ratio for this object . furthermore , the bulk of the line emission displays little spatial correspondence with the blue continuum . therefore in this analysis we neglect to first order the contribution from line emission to the continuum color ; spatially resolved spectrophotometry would be required to further quantify its contribution . we measure the intrinsic color of the blue excess to be in the range @xmath30 with a surface brightness ranging from @xmath31 mag arcsec@xmath22 ; the integrated absolute magnitude of the entire extended region of blue continuum excess is @xmath32 ( spread over a projected area of @xmath33 kpc@xmath34 ) . we compare the measured properties of the blue continuum to a set of stellar population synthesis models ( , private communication ) , which have been k - corrected to the redshift of a2597 and calibrated in the _ hst _ vegamag system using our _ hst _ wfpc2 filter bandpass transmission curves . we find it preferable to retain our measured magnitudes in the vegamag system and compare them with models that have also been calculated in this system , rather than attempting to transform the magnitudes to other systems such as the johnson / cousins system , because uncertainties in the color terms can translate into errors of several tenths of a magnitude . the models are shown in figure [ fig : charlot_models ] and represent two scenarios for star formation : ( 1 ) a constant star formation rate ( sfr ) , assuming an invariant imf ; ( 2 ) an instantaneous burst of star formation ( or `` simple stellar population '' , ssp ) . in each case we investigated two standard imfs for which models were available , namely and , each with upper and lower mass cutoffs of 0.1 and 100 @xmath35 respectively . we also examined models of different metallicities , specifically @xmath36 , 0.2@xmath36 and 0.02@xmath36 . further details of the model parameterizations are described in charlot & bruzual ( 1991 , 1999 ) and . having calculated _ hst _ colors and magnitudes for these various models , we note that the difference between the color evolution of @xmath36 and 0.2@xmath36models is minimal for the constant sfr models , thus in figure [ fig : charlot_models ] we only plot the 0.2@xmath36 models for the purposes of clarity . for the ssp models , the color evolution is more sensitive to metallicity but is relatively insensitive to the choice of imf , therefore the ssp models that we plot are for the salpeter imf only . in table [ tab : blue_extended ] we present the inferred star formation parameters that are required to match the observed range of colors and integrated absolute magnitude of the extended blue continuum distribution , for the various star formation scenarios that we have described . for each model , we tabulate the upper and lower limits of the range of epochs in its evolution that reproduce our observed @xmath29 color . for constant sfr models , the sfr for each of the two epochs is obtained by matching the model to the @xmath37 absolute magnitude . for the instantaneous burst models we tabulate the total mass of stars ( an sfr is not physically meaningful in this case since the burst is represented by a delta - function ) . we point out that the inferred ages are all highly dependent upon the measured colors , and should therefore be interpreted as _ upper limits _ due to the likely presence of reddening across most of the field . for example , a decrease in the intrinsic @xmath29 color by only a further 0.2 magnitudes will yield ages as low as @xmath38 yr for the continuous sfr models , and @xmath39 yr for the instantaneous burst models . however , due to our lack of accurate reddening estimates ( precluded by the presence of emission lines in both bands ) we are unable to quantify this further . rcc const . sfr : salpeter : @xmath36 & @xmath40 & @xmath41 0.2@xmath36 & @xmath42 & @xmath43 0.02@xmath36 & @xmath44 & @xmath45 scalo : @xmath36 & @xmath46 & @xmath47 0.2@xmath36 & @xmath48 & @xmath49 0.02@xmath36 & @xmath50 & @xmath51 & & @xmath52@xmath35 ) inst . burst : salpeter : @xmath36 & @xmath53 & @xmath54 0.2@xmath36 & @xmath55 & @xmath56 0.02@xmath36 & @xmath57 & @xmath58 scalo : @xmath36 & @xmath59 & @xmath60 0.2@xmath36 & @xmath61 & @xmath46 0.02@xmath36 & @xmath62 & @xmath63 the southern radio lobe displays substantially less blue emission than the northern lobe . its @xmath29 color is similar to that of the northern lobe but its flux is much less , with an integrated apparent magnitude of only @xmath64 . since it is fainter , errors in its measured properties are much more dominated by uncertainties in the underlying continuum , reddening correction , and contributions from line emission , therefore we do not feel justified in carrying out a comparison with star formation models to the same level of detail as the more prominent northern blue excess . we do note , however , one interesting feature of the southern blue emission , namely a blue `` arc '' that follows the outer contours of the southern lobe , from its western edge around to the south . this is unlikely to correspond to optical synchrotron emission , as the radio contours show no enhanced brightening at the location of the arc , thus the arc may also indicate emission from young stars although our data are of insufficient s / n to quantify this further . lccccc 1 & @xmath65 & @xmath66 & @xmath67 & @xmath68 & @xmath69 2 & @xmath70 & @xmath71 & @xmath72 & @xmath73 & @xmath74 3 & @xmath75 & @xmath76 & @xmath77 & @xmath78 & @xmath79 4 & @xmath80 & @xmath81 & @xmath82 & @xmath83 & @xmath84 we now discuss a group of four blue clumps , or knots , that are readily apparent south - west of the southern radio lobe . because of their relative isolation , they can serve as a valuable reference for comparison when discussing the star formation that has occurred north of the nucleus . in the original ( unsmoothed ) pc1 images they are marginally resolved , with fwhm values in the range @xmath85 . they apparently correspond to the southern _ u @xmath28 i _ blue continuum excess reported by . these features do not appear to be directly associated with the radio lobe , and in all likelihood they are young star clusters . in table [ tab : blue_compact ] we present the magnitudes and colors of these objects , calibrated in the _ hst_vegamag photometric system . our photometric analysis of these objects indicates that their colors are about @xmath86 magnitudes bluer than the surrounding elliptical continuum . as a consistency check , we also carried out photometry of a number of the brighter , redder objects around a2597 in the f702w image , and compared these with the _ r_-band photometry of the same objects presented by . we found that the magnitudes typically agreed to within @xmath87 mag , which is quite satisfactory given the uncertainties in the color terms that had been used by to transform their _ hst _ magnitudes to the johnson / cousins system . we have compared the colors and magnitudes of the blue clumps to the stellar synthesis models of , k - corrected to the redshift of a2597 and calibrated in the _ hst _ bandpasses as described in the previous section . since we are investigating individual clusters in this case , we carry out a comparison only with the instantaneous burst models , with the same range of metallicities and imfs as for the northern blue continuum . in the lower panel of figure [ fig : charlot_models ] we plot the evolution in color - magnitude space of three population synthesis models with a salpeter imf , metallicities of @xmath36 , 0.2@xmath36 and 0.02@xmath36 , and total mass @xmath88 @xmath35 , together with the observed colors and magnitudes of the blue clusters . the comparison suggests that the colors of the three bluest clusters are consistent with ages @xmath89 myr and metallicities above @xmath90 , with ages of the two redder of these three clusters ranging up to a maximum of @xmath91 myr for masses up to @xmath92 @xmath35 . the reddest of the four clusters is either much older ( @xmath93 yr ) with a mass @xmath94 @xmath35 , or otherwise is the same age as the other three , but has higher extinction and/or a metallicity significantly above solar ( however , this cluster is fainter than the other three and its properties are not as well constrained ) . in [ sec : implications ] we discuss in further detail the implications of the measured properties of the blue continuum regions for various physical mechanisms that can be responsible for triggering the star formation . in order to present a more detailed discussion of the various components that make up the central galaxy , including the filaments , the blue continuum , and possible dust obscuration , it is necessary to model and remove the underlying elliptical stellar continuum distribution this was done by means of the iraf / stsdas task ellipse ( based on an algorithm described by ) , which makes use of iterative sigma - clipping algorithms to reject the possible contribution of non - elliptical components to the fit . our data are complicated both by the presence of filaments in emission , and also by apparent dust absorption in the central regions of the galaxy . we initially attempted techniques similar to those described by and , involving a number of repeated fits to both the f702w and f450w images , which allows the possibility of correcting for patches of dust obscuration . however , this technique does not easily allow additional corrections for emission features , particularly the line emission from the filaments which is present in both the f450w and the f702w broad - band images of a2597 . therefore , we obtained the underlying elliptical continuum by fitting isophotes to a near - infrared h - band image of a2597 , recently obtained with the f160w filter on nicmos by . this band has the advantage of containing no line emission from the filaments , and also with the effect of dust obscuration being much reduced ( although still present ) . we fitted elliptical isophotes to this image by allowing the flux , ellipticity and position angle of each isophote to vary freely , and using sigma - clipping to reject features deviating by more than 3 @xmath95 from the mean level in each isophote ( for example corresponding to stars or clusters , or the dust obscuration region ) . inspecting the resulting residual image showed that this model provided a good fit to the overall galaxy profile , with the only significant residuals corresponding to localized features such as clusters or dust absorption patches . some low - level residual structure on large scales was also present , which is attributed to uncertainties in the flat - fielding applied in the reduction process . the elliptical model obtained from the nicmos data was then applied to the f702w and f450w images , by keeping the isophote shape parameters constant ( i.e. , ellipticity and position angle ) , but allowing the flux of each isophote to vary freely , and using sigma - clipping as before to remove contributions due to the filaments or dust patches . for isophotes with a semi - major axis @xmath96 we had to allow the isophote shape parameters to vary , since the smaller size of the nicmos image , together with its low - level flat - field effects , produced inaccuracies in the model at these large radii . however , for isophotes inside this radius the nicmos isophote shapes provided a very good fit to the f702w and f450w data . subtracting the resulting model from the input image yielded for each filter a first - iteration residual image which contained most ( but not yet all ) of the `` excess '' emission . because of the substantial contributions in emission from the filaments within the central @xmath97 , we carried out a second iteration , according to the following procedure . pixels deviating from 0 by more than 3 times the rms value in the residual image were subtracted from the original input image to yield a second input image . new isophotes were then fitted to this image , using the isophote shape parameters from the first iteration and allowing only the flux to vary , and again using sigma - clipping to reject non - elliptical emission features . this yielded a second residual image ; repeating this procedure further showed that the process had converged by the second iteration to provide an accurate representation of the underlying elliptical continuum flux distribution , together with the residual emission and absorption features . in figure [ fig : isophotes ] we present the final set of isophotal parameters fitted to the galaxy . the elliptical continuum color index across most of the galaxy is in the range @xmath98 which corresponds to _ b @xmath28 v _ @xmath99 , a typical value for ge galaxies ( e.g. , ) . the color is somewhat redder in the central region , possibly related to the presence of dust . one interesting feature of the galaxy is a change of @xmath100 in the orientation of the major axis , relative to the outer isophotes , for semi - major axis lengths @xmath101 . this also corresponds to an increase in ellipticity toward the center . this change may be due to the presence of dust located on either side of the nucleus , although in the f160w image the change is quite smooth and therefore we can not rule out the possibility that this may reflect a genuine shift in the underlying light distribution , for example due to a nuclear bar ( e.g. , ) . in figure [ fig : f702w_res ] we present the `` residual '' filamentary structure in the f702w band , produced by subtracting the elliptical galaxy model from the original image . we also overlay contours representing the vla 3.5 cm radio continuum image . the positive emission features in this image are likely to be dominated by h@xmath14line emission rather than stellar continuum , based on the fact that the filamentary structures in the f410 m image ( figure [ fig : f450w_3 cm ] ) are dominated by [ ] @xmath133727 emission . the filaments in figure [ fig : f702w_res ] are sampled with both higher s / n and better spatial resolution than those in figure [ fig : f450w_3 cm ] : the bright arcs are resolved into clumps in several locations , and diffuse emission can be clearly distinguished across the face of the northern radio lobe . in addition , the presence of strong , patchy dust obscuration is evident in a number of locations , particularly to the south - east and to the north of the nucleus . some weaker obscuration is also present west of the nucleus , just beyond the bright radio hotspot and associated emission - line filaments . lcccc n. arc - upper & @xmath102 & @xmath103 & @xmath104 n. arc - lower & @xmath105 & @xmath106 & @xmath107 n. lobe face & @xmath108 & @xmath109 & @xmath110 n. lobe ( total ) & @xmath111 & @xmath112 & @xmath113 sw knot & @xmath114 & @xmath106 & @xmath115 se arc & @xmath116 & @xmath117 & @xmath118 the distribution of optical line emission around the northern lobe of the radio source , as displayed in figures [ fig : f450w_3 cm ] and [ fig : f702w_res ] , can be divided into three regions : ( 1 ) the bright `` upper '' arc along the north - west of the lobe ; ( 2 ) a less distinct `` lower '' arc around the southern boundary of the lobe ; ( 3 ) diffuse , lower surface - brightness emission across the face of the radio lobe . the upper arc runs parallel to a bright ridge of enhanced radio emission along the edge of the lobe , both of which display a linear extent of @xmath119 ( 2.8 kpc ) . the arcs are dominated by emission - line radiation ; the lack of any detected blue continuum emission excludes non - thermal processes such as optical synchrotron emission or inverse compton scattering . we discuss these features in more detail in [ sec : n_lobe ] . the southern lobe displays a compact knot of high surface - brightness line emission co - incident with the strong radio hotspot immediately south - west from the nucleus . a prominent semi - circular filament of emission is also present se from the nucleus , along the eastern edge of the southern radio lobe . we find that these features are also dominated by line emission and not continuum radiation . we discuss the southern emission - line features further in [ sec : s_knot ] . in table [ tab : n_lobe_fluxes ] we present the integrated [ ] @xmath133727 fluxes of each of the three northern emission - line regions , together with the total luminosity of emission - line gas surrounding the northern radio lobe , as well as the bright emission - line knot immediately sw from the nucleus and the semi - circular se arc . we also derive the h@xmath120 luminosities for each of these regions from the measured [ ] @xmath133727 line fluxes by adopting a canonical value of h@xmath120 / [ ] @xmath133727 = 0.23 ( uncorrected for reddening ) , determined from the ground - based spectra of . implicit in this conversion is the assumption that the excitation and extinction properties of the nebula as a whole are also representative of the material associated with each radio lobe ; this approximation is necessitated by the current absence of spectra with higher spatial resolution . a prominent feature of this galaxy is the complex network of emission - line filaments associated with the radio lobes . in this section we examine scenarios in which the northern filaments trace physical interactions between the northern radio lobe and ambient gas . we derive physical parameters of the ionized and neutral gas around the radio lobe , including the neutral fraction and volume filling factor of the gas and its total mass , by combining constraints obtained from our new emission - line data with those from previous observations . we consider a geometry in which the radio lobe is surrounded by an approximately uniform shell of line - emitting gas ( except along the edge furthest away from the nucleus , where no significant emission is detected ) . this scenario is suggested by the similarities in flux and spatial extent of the upper and lower edge - brightened arcs . furthermore , the integrated flux of the diffuse emission across the face of the radio lobe ( which would include emission both from behind and in front of the lobe ) is approximately twice that of the flux in each arc , supporting an edge - brightened geometry for the shell . limits on the shell thickness or depth @xmath121 , obtained from the width of the arcs , are thus applicable to the line - emitting gas covering the face of the radio lobe . the width of the bright upper arc is only marginally resolved , covering less than @xmath122 pc1 pixels ( 130 pc ) , therefore we retain @xmath121 as a free parameter normalized to a fiducial value of 100 pc . the surface area of diffuse line emission across the face of the radio lobe is @xmath123 arcsec@xmath34 , thus the volume of the emission - line region in front of the radio lobe is @xmath124 kpc@xmath125 . given standard assumptions such as case b ionization conditions , uniform density emission - line gas , and an edge - brightened shell geometry , our observed h@xmath120 luminosity then provides a combined constraint on the volume filling factor @xmath126 and electron density @xmath127 of the warm ionized gas ( with electron temperature @xmath128 ) : @xmath129 we aim to derive the physical properties of the gas by combining this constraint with those obtained from observations . observed extended , broad absorption ( @xmath130 km s@xmath6 ) against the northern radio lobe . the kinematic properties are similar to those of the optical line - emitting gas , thus the northern radio lobe would be surrounded by a collection of clouds containing both neutral and ionized gas . in this section we derive physical properties of the gas for the following two plausible scenarios : . the is co - spatial with the ionized gas , with each cloud possessing an ionized `` skin '' responsible for the line emission ( e.g. , ) . in this case the ionized and neutral gas are swept up by the expanding radio lobe into a shell , thereby compressing the gas and determining its kinematics , but not necessarily ionizing it ( alternative ionization mechanisms are discussed by ) . for simplicity , in this scenario we consider the line - emitting gas around the northern lobe to be entirely photoionized by radiation originating from an external source ( e.g. , the active nucleus , hot stars , or radiation escaping from shocks in another region ) . the ionized and neutral gas are located in different regions , for example if the northern radio lobe expands into a previously existing ensemble of neutral clouds ( tidal debris or material condensed out of the cooling flow ) , fully ionizing each cloud by means of fast shocks driven into the gas . in this case , the radio lobe would be surrounded by an inner region of mostly ionized gas and an outer region of neutral clouds ( see for example the schematic presented in ) . the ionized gas would contain a mixture of cooling `` post - shock '' gas together with photoionized `` precursor '' gas that absorbs ionizing radiation emitted by the shocked gas . in this scenario , the neutral and warm ionized components occupy the same region but are represented by different volume filling factors . parameterizing the depth of the absorbing material by @xmath121 thus converts the observed column density ( @xmath131 @xmath21 , ) to a combined constraint on the density and volume filling factor : @xmath132 pressure balance is considered to hold between the neutral gas ( spin temperature @xmath133 k ) , warm ionized gas ( @xmath134 k ) and the hot ambient medium ( @xmath135 k ) . we define @xmath136 to represent the relative masses of neutral and warm ionized gas , and combine this with equations ( [ eqn : nf_1 ] ) and ( [ eqn : nf_2 ] ) to obtain the gas densities : @xmath137 taking @xmath138 @xmath2 @xcite to represent the overall line - emitting gas , we obtain the volume filling factor and thus the total mass of neutral and ionized gas in the shell , assuming only that its thickness @xmath121 remains approximately uniform : @xmath139 if the gas is distributed in the form of approximately spherical neutral clouds with ionized skins , then the above constraints yield the relative extent of the @xmath140 and @xmath141 cloud radius : @xmath142^{\frac{1}{3 } } \ , .\ ] ] we use this , together with the projected surface area of the northern lobe and the volume filling factor , to constrain the cloud covering fraction @xmath143 and cloud radius : @xmath144 combining this with the limits on the covering fraction of the extended absorption in front of the northern radio lobe ( @xmath145 ) and using our determination of the surface area of line - emitting gas in front of the radio lobe , yields an upper limit on the cloud radius : @xmath146 and corresponding numbers of clouds : @xmath147 a covering fraction of unity is obtained with @xmath148 clouds , with radii @xmath149 pc . we conclude our description of the physical properties of the gas in this scenario with a calculation of the inferred pressure associated with the gas : @xmath150 dyn @xmath21 . this is in very good agreement with the central x - ray pressure @xmath151 dyn @xmath21 derived by but is lower than the estimated minimum pressure of the northern lobe , @xmath152 dyn @xmath21 ( re - calculated with the cosmological parameters used in the present paper ) , thereby suggesting that the radio lobe is somewhat overpressured with respect to the surrounding gas . an alternative scenario for the distribution of neutral and ionized gas involves expansion of the radio lobe into a more extended region of neutral clouds . each cloud is fully ionized by the passage of the bowshock , analagous to the scenario envisioned for gps / css radio sources . the shell of warm ionized gas around the radio lobe would be surrounded by neutral clouds whose origin may be unassociated with the radio lobe ( for example , tidal debris around the host galaxy , or gas condensed from the cluster cooling flow ) . line emission is produced by cooling `` post - shock '' gas downstream from the shocks as well as `` precursor '' material immediately upstream from the shocks , photoionized by soft x - ray / uv radiation from the post - shock gas . the ionized gas clouds possess a bi - modal density distribution : the post - shock clouds have densities @xmath122 orders of magnitude above those of the photoionized precursor gas , while the relative total line radiation output of the two regions are approximately equal @xcite . we consider a case in which the expansion velocity of the emission - line shell surrounding the radio lobe is equated with a shock velocity @xmath153 km s@xmath6 ( i.e. , half of the observed optical linewidth ) . a relatively uniform expansion speed is suggested by the well - ordered morphology of the emission - line gas . mappings ii models show that the post - shock material in this case cools to gas with a density @xmath154 times greater than the precursor material (; ; ) . the electron density measured from the ratio by represents an average over all the emission from both regions , thus we use their value together with the ratio of post - shock / precursor line emission to infer a precursor density @xmath155 @xmath2 . using our observed h@xmath120luminosity , together with the relationships for shock and precursor emission , yields a required shock surface area of @xmath156 kpc@xmath34 to account for the observed line emission . this compares well with the apparent surface area of the shell ( @xmath157 kpc@xmath34 ) i.e. , the covering fraction of shocks across the emission - line shell would be of order unity for precursor gas of this density . the corresponding spatial scale on which the precursor gas is ionized would be of order @xmath158 pc ( depending on its volume filling factor ) ; beyond this distance the gas would remain largely neutral . having calculated the expected physical properties of the gas in each of the two preceding scenarios , we now compare their relative likelihood . in particular , we examine whether the radio lobe can supply the required energy to the surrounding gas . we begin with the total minimum energy content of the northern radio lobe , @xmath159 ergs , the lower limit on the radio source age @xmath160 yr from the synchrotron loss timescale , and the radio luminosity of the northern lobe @xmath161 ergs s@xmath6(derived from the observed radio properties by , and rescaled to the cosmological parameters used in this paper ) . we assume a fiducial value of @xmath162 for the conversion efficiency of the total jet kinetic power @xmath163 into radio emission , noting that this value is not well constrained observationally (; ) , i.e. , @xmath164 ergs s@xmath6 . in the first scenario , the energy transfer from the expanding radio cocoon to the gas is predominantly kinetic . given the cospatial nature of the gas components in this scenario , we equate the expansion speed of the optical line - emitting gas ( 300 km s@xmath6 ) with that of the . the kinetic energy of the neutral and warm ionized gas components is then : @xmath165 which is substantially less than the total minimum energy content of the northern radio lobe . in other words , the rate of energy input required to supply this amount of kinetic energy to the combined neutral and warm ionized gas components need only be @xmath166 of the total energy input into the radio lobe , therefore it is feasible that the expansion of the radio lobe can account for the gas kinematics of the northern emission - line filament system . for the second scenario , we investigate whether the energy input from the lobe expansion can feasibly account for the observed line luminosity through ionization by radiative shocks . if the rate of work done on the ambient medium by the expansion of the lobe corresponds directly to the total luminosity of the fully radiative shocks that are produced , then the resulting h@xmath120 line luminosity can be written in terms of the jet energy flux and shock velocity : @xmath167 where we have equated the shock velocity @xmath168 with the 300 km s@xmath6expansion velocity of the emission - line shell surrounding the radio lobe . despite the inherent uncertainties in deriving the radio lobe and jet energetic parameters , this value of @xmath169 is in good agreement with our measured total integrated line luminosity for the northern lobe , presented in table [ tab : n_lobe_fluxes ] . we note that this does not necessarily imply that shocks _ are _ the dominant ionization mechanism in the northern lobe ; instead we are stating that shocks , _ if present _ , correspond to a set of physically plausible values of parameter space . there is some question whether the overall properties of the entire eelr around a2597 can be explained in terms of shocks , given the constraints on the electron temperature and gas ionization state presented by . however , their results apply primarily to the integrated spectrum of the emission - line nebula , to which the northern lobe contributes only @xmath170 of the flux , thus local ionizing mechanisms in this region can readily be accommodated ; higher spatial resolution spectra will prove useful in further addressing such questions . thus , we have shown : ( 1 ) it is energetically feasible that the northern radio lobe can supply the kinetic energy of the surrounding gas , irrespective of the ionization mechanism ; ( 2 ) if the radio lobe expansion indeed produces auto - ionizing shocks in the gas , then the required shock luminosity and other properties of the gas are in good agreement with the observations . the bright radio hotspot 0@xmath165 south - west of the nucleus is a region of great interest since this location corresponds to a sharp change in the direction of the radio jet toward the east , which also appears to dramatically affect the resultant morphology of the entire southern radio lobe . postulated the presence of a dense deflecting clump of cool gas ( @xmath4 k ) , but the available ( ground - based ) optical data were of insufficient spatial resolution to verify its existence . our _ hst _ images clearly resolve a knot of high surface - brightness optical line emission associated with the radio hotspot . this object is therefore a highly suitable candidate for conducting a study of the effects of a direct jet / gas cloud interaction . here we use our images to infer the physical properties of the gas clouds , including their size , mass and density ( degree of inhomogeneity ) . two plausible scenarios exist for describing the encounter between the jet and the gas : either the radio jet propagates into a pre - existing region of cool dense gas , or an initially straight jet is deflected as a result of an interaction with one or more gas clouds moving directly into it ( for example , due to their orbital motion around the galaxy ) . the number of free parameters in each case prohibits a discussion of which scenario is more likely . rather , we examine fundamental physical properties of the gas that would be applicable to either scenario , using considerations based on energetics and momentum transfer , and following similar arguments to those previously presented by and . we consider a non - relativistic jet with fiducial speed @xmath171 km s@xmath6and kinetic power @xmath172 ergs s@xmath6 . the acceleration imparted by the jet to gas approximately at rest is related to the jet momentum flux , thus the minimum gas mass required to significantly deflect the jet over its lifetime ( equated to the radio source age @xmath173 yr ) is @xmath174 @xmath35 . if the gas cloud volume is approximately spherical ( with a diameter given by its transverse dimension , @xmath175 pc ) , the required mean density of hydrogen nuclei must exceed @xmath176 @xmath2 . we next examine the heating of the gas by the jet , using @xmath177 to represent the fraction of @xmath163 which is converted into heat energy . the cloud needs to be dense enough to cool at this rate or faster , otherwise it would be heated to high temperatures and disperse . writing @xmath178 as the rms average hydrogen number density of the cloud , and the cooling coefficient as @xmath179 ergs s@xmath6 cm@xmath125 , the balance between the cooling and heating rates requires @xmath181 comparing this with @xmath182 suggests that the cloud either exceeds the lower limit on its mass and/or is inhomogeneous . finally , the rms electron density @xmath183 in the cloud that is required to produce the observed h@xmath120 line luminosity is @xmath184 the similarity of the densities in equations ( [ eqn : n_rms ] ) and ( [ eqn : ne_rms ] ) might suggest that the deflecting cloud was composed purely of warm ionized gas at @xmath185 k. however , in photoionized gas at these temperatures the cooling coefficient is only @xmath186 ergs s@xmath6 cm@xmath125 and such gas could not radiate energy deposited by the jet unless the efficiency of deposition was low , @xmath187 . for typical conditions in photoionized emission line gas , the bolometric luminosity @xmath188 is about 25 times that in h@xmath120 , or @xmath189 ergs s@xmath6 , which is much smaller than @xmath163 . thus , either the efficiency of heating the deflecting cloud is low , or the cloud contains substantial amounts of atomic and/or molecular gas . the latter scenario is supported independently by the recent _ hst _ nicmos detection of significant molecular h@xmath190 emission co - incident with the knot . hence , we suggest that the observed properties of this emission - line knot are consistent with the presence of clumps of dense gas sufficiently massive to deflect the jet significantly over the lifetime of the radio source . we have so far demonstrated that the emission - line nebula in the cd elliptical of a2597 displays a close association with the radio source , and that several different properties of the gas are strongly suggestive of direct interactions between the radio lobes and the ambient medium . we have also shown that the blue continuum emission originates in a number of resolved locations within the central few kpc , and does not display a direct spatial correspondence with the emission - line filaments . we have identified a number of regions of recent or on - going star formation that can be modeled either as single - burst populations with ages of several times @xmath191 yr , or as continuous ( constant ) star formation with ages about an order of magnitude higher . we have also resolved an elongated region of strong dust obscuration immediately south - east of the nucleus , together with evidence of patches of dust in other locations nearby . in this section we explore the likely physical inter - relationships between these various components in cooling flow clusters , using a2597 as an example . studies to date suggest that the strength of the blue continuum excess and nebular line emission around cd galaxies in cooling flow clusters increases as a function of the inferred total cooling rate (; ; and references therein ) . cooling flow cluster cds possess abnormally high amounts of molecular gas and dust , and display more compact radio morphologies , relative to cd galaxies not associated with cooling flow clusters . thus , it is crucial to understand the underlying physical connection between the dense , high - pressure icm surrounding the elliptical , the blue continuum excess , the nebular line strength , the strong dust obscuration , and the radio source . in particular , is the presence of a cooling flow either a necessary and / or sufficient condition to explain these properties of the central galaxy , or do other processes also play a role ( e.g. , mergers or interactions ) ? we address this issue by breaking it down into the following questions : ( 1 ) what has triggered the recent star formation ; ( 2 ) what is responsible for triggering the agn ; ( 3 ) how are the triggering mechanisms related to the fuel supply in each case , and what is the relationship between the star formation mechanism , the triggering of the radio source , and the presence of a cooling flow ? generally , star formation in dense molecular clouds is triggered by compression of the clouds resulting from enhanced external pressure , or by deep molecular shocks associated with cloud - cloud collisions . here we examine the likelihood of various means of inducing these effects to result in the formation of stars . one triggering mechanism for star formation is direct interactions between the radio jets or lobes and dense gas clouds immediately surrounding them (; ; , ; ) . this scenario invokes the collapse of marginally unstable clouds in the ambient medium , resulting from overpressurization as they are overtaken by the expanding radio cocoon . it predicts that , while star formation is initiated along the radio lobe boundaries , the spatial correlation is short - lived due to differences between the orbits of the stars and the motion of the radio plasma . the relatively mild expansion velocity of the cocoon , compared to the working surface of the jet , prevents disruption of the clouds and facilitates their collapse . the large - scale asymmetry of the blue continuum in a2597 would correspond to an intrinsic asymmetry in the initial distribution of molecular gas clouds . the radio source age in this scenario would have to agree with the ages of the young stars , which we find to be in the range @xmath192 yr from the instantaneous burst stellar model synthesis ages . the lower limit on the radio source lifetime is @xmath193 yr from the synchrotron spectral ageing analysis , but given the uncertainties involved in this derivation the actual radio age can be substantially higher . thus , from timescale arguments it is plausible that the star formation is triggered by the radio source . this would imply that the radio source is intrinsically old but unable to penetrate the dense surrounding icm to scales of more than a few kpc . star formation in cooling flows is also proposed to occur via pressure - driven collapse of filaments or clouds condensing out of the cooling flow (; ; ) . the inferred cooling flow mass accretion rates for a2597 (; ) are about two orders of magnitude above our values of the sfr . such discrepancies are well - known from previous studies of a2597 and other cooling - flow clusters . possible solutions include erroneous estimates of the sfr , for example due to a time - variable sfr or changing imfs ; or mass deposition into material that does not necessarily form stars . these and other possibilities are discussed in detail by ; ; ; and . we do not rule this out as a potential source of fuel for star formation . it does not readily predict , however , the highly asymmetric structure of the blue continuum in a2597 . one means of obtaining asymmetry is via subcluster merging activity ( e.g. , , ; ) . however , a2597 displays little evidence of a recent cluster / subcluster merger , as the x - ray data suggest a relaxed system @xcite . the spatial scale of the blue continuum emission is also one to two orders of magnitude smaller , @xmath194 kpc , compared with the typical cooling radius of @xmath195 kpc at which the stars are expected to form . thus , cooling of the x - ray gas can provide a source of cooler material for star formation , but an additional mechanism appears to be required for triggering the concentrated massive star formation observed in a2597 . this leads us to discuss star formation triggered via accretion of a gas - rich companion galaxy by the elliptical . galaxy interactions are generally associated with triggering star formation ( e.g. , ; see also , and for reviews ) . detailed n - body / hydrodynamic models show that interactions between ellipticals and small gas - rich galaxies generally lead to gas accretion into the elliptical core @xcite , triggering a centrally condensed starburst as well as star formation in the tidal debris ( e.g. , , , ) . such features are observed in a wide range of galaxies , but specifically in large ellipticals ( e.g. , ; ; ; ) . the star - forming regions in a2597 show similar characteristics : intense star formation near the galaxy core , with smaller , more isolated blue clusters further out . the primary objection to transporting gas - rich galaxies into the cluster core is that ram - pressure stripping by the hot icm removes much of the relatively diffuse gas (; ) . however , dense clouds of molecular gas and dust have much lower filling factors and higher column densities than the component and are less affected by hydrodynamical processes (; ) . indeed , studies of the molecular content of galaxies in clusters show no significant dependence on location in the cluster , whereas the content decreases substantially toward the cluster core (; ; ) . therefore such an interaction can quite plausibly provide a fresh supply of gas , predominantly molecular . the stellar model ages of @xmath196 yr are also typical of the dynamical timescales involved in such interactions . thus , we propose that two plausible triggering mechanisms for the star formation are : ( 1 ) related directly to the passage of the radio source , or ( 2 ) related to accretion of a gas - rich companion galaxy . if the star formation is modeled in terms of a single burst then the resulting ages are in the range @xmath192 yr the low end of this range is comparable to the lower limit on the radio source age , thus we associate the single - burst models with triggering by the radio source . if , on the other hand , the star formation is modeled using a constant sfr , then the required ages of the stars are much higher , of order a few times @xmath197 yr . this is too high to be accounted for in terms of radio source triggering , but is consistent with the dynamical timescales typically required for interactions between galaxies . while our current data do not allow us to distinguish further between these two possible triggering mechanisms , we are able to conclude that triggering by the radio source is consistent with a single burst @xmath192 yr ago , while triggering by an interaction corresponds to continuous , on - going star formation over longer timescales up to @xmath93 yr . having discussed the star formation trigger mechanism , we now investigate the trigger for the agn itself . the majority of cd ellipticals in cooling flow clusters are known to host radio - loud sources , compared to a much lower fraction of cd s in non cooling flow clusters @xcite . a number of studies have discussed cooling flows as the primary explanation for the observed nuclear activity and star formation , through accretion of the cooling filaments into the central galaxy ( e.g. , ; , ) . however , cooling flows need not be the sole mechanism responsible for triggering and fueling the agn in particular , the long timescales of a steady cooling flow can not be reconciled with the comparatively short radio source lifetimes without invoking additional mechanisms to trigger the onset of the radio source . furthermore , many agn with similar properties are found in host galaxies not associated with cooling flow clusters . thus we are motivated to discuss the plausibility of an additional mechanism which is generally associated with agn triggering and fueling , namely interactions between galaxies ( e.g. , ; , and references therein ) . specifically , `` minor mergers '' between the host galaxy and a small gas - rich companion are an important means of dumping fuel into the central region (; ; ) . in such cases the host galaxy retains its original morphology more or less intact , with the primary change consisting of a fresh supply of @xmath196 @xmath35 of stars , gas and dust . the central elliptical in a2597 displays a number of features characteristic of a recent `` minor merger '' : ( 1 ) substantial dust which appears dynamically unsettled , located predominantly on one side of the nucleus ; ( 2 ) the extended complex of neutral , ionized and molecular gas , showing strong asymmetries on small scales ( the present work ) and large scales ; ( 3 ) a radio source with an age @xmath198 yr ; ( 4 ) small galaxies nearby , which if dynamically associated with the cd would have supplied the gas ; ( 5 ) the clusters of young stars . the high - pressure cluster environment associated with a cooling flow can enhance the star formation to the observed levels and confine the radio source . however , the actual _ triggering _ mechanism of the star formation and agn is plausibly accounted for by this type of interaction , likely with a galaxy containing substantial amounts of molecular gas as discussed in the previous section . when such a galaxy interacts with the central elliptical , some of its surviving molecular material would be disrupted and become visible as atomic / ionized gas , while the remainder is accreted into the core of the cd . in this scenario one expects comparatively little reddening in the diffuse line - emitting filaments , due to the relatively short sputtering times of dust in the high - pressure surrounding hot gas @xcite ; any remaining dust is concentrated in regions containing substantial amounts of dense molecular gas . indeed , we find that the strongest extinction in a2597 is concentrated in the compact , dusty regions immediately adjacent to the nucleus , located in the same vicinity as the molecular h@xmath190 emission detected by . it is therefore plausible that the observed distribution and physical properties of the dust and gas in a2597 can be accounted for by the capture of an infalling gas - rich galaxy which has retained a significant fraction of its original dense clouds of molecular gas and dust . this would also provide a supply of fuel for the current phase of activity of the agn . the age of the radio source is somewhat younger than the timescale associated with the interaction , but this likely represents the additional time required to funnel gas into the center of the galaxy , which in numerical simulations typically occurs well into the merger process ; alternatively the currently observed radio source may be only the latest in a series of multiple outbursts during the galaxy interaction process . finally , the location of this object in the center of a dense , high - pressure icm would account for the enhanced star formation , as well as the apparent strong confinement of the radio source . in this paper we have presented _ hst _ wfpc2 images that spatially resolve some of the most intriguing features of the cd elliptical in a2597 : a complex system of emission - line filaments occupies the inner few kpc of the galaxy , displaying a number of close spatial associations with the synchrotron - emitting radio lobes . the morphologies of several of the filaments are strongly suggestive of direct interactions between the radio source and ambient gas . in particular , we find that the detailed physical properties of the line - emitting gas around the northern radio lobe can be described in terms of a shell of neutral and ionized gas that surrounds the expanding radio lobe , and we show that the energetics of this shell can be readily accounted for by energy input from the expanding radio lobe . we also resolve a knot of high surface - brightness line emission approximately co - incident with a radio hotspot to the south - west of the nucleus , and we show that its properties are consistent with those of a dense , inhomogeneous clump of gas ( @xmath199 @xmath2 , @xmath174@xmath35 ) sufficiently massive to deflect the radio jet over the lifetime of the radio source and account for the severely disrupted morphology of the southern radio lobe . finally , the lack of detectable line emission in a number of locations around the radio lobes suggests an intrinsically asymmetric distribution of ambient gas about the galaxy . the excess blue continuum emission is clearly resolved by our _ hst _ images into a number of components , specifically an intense band of blue emission associated with the base of the northern radio lobe , as well as blue clusters south - west from the southern radio lobe , and a number of more diffuse regions distributed about the radio lobes . the morphological properties of the blue continuum ( together with its low observed polarization , ) rule out several explanations for its origin , including optical synchrotron or inverse compton emission , or scattered nuclear light , leaving recent star - forming regions as the most likely remaining possibility . comparing our photometry with stellar population synthesis models shows that these regions can be modeled either as single - burst populations with ages of several times @xmath191 yr , or as continuous ( constant ) star formation with ages about an order of magnitude higher , depending upon details of the assumed metallicity and imf parameters . the strong dust obscuration in the central region is resolved into a number of large clumps , each several hundred pc in size . most of these are distributed in a region extending away from the nucleus toward the south - east , while another strong clump of obscuration is evident immediately to the north of the nucleus , and some less heavily absorbed regions are present just west of the bright radio / emission - line hotspot in the south . a striking feature of the dust is its strong degree of clumping and highly asymmetric spatial distribution about the galaxy center . the cd elliptical in the cooling flow cluster a2597 is a complex system : it contains a powerful , compact radio source , surrounded by an extensive network of emission - line filaments with significant amounts of dust and prominent regions of recent or on - going star formation . we suggest that the high - pressure environment associated with the cooling flow is crucial to enhancing the star formation rate to the high levels observed . however , while a steady cooling flow can also provide a source of fuel for the star formation and the agn , its long timescales are difficult to reconcile with the short radio source lifetime , the substantial amounts of dust and molecular gas in the central few kpc , and the highly asymmetric distribution of the gas , dust and young stars without invoking an additional , more recent mechanism to account for these features . we propose that a recent `` minor merger '' or interaction in which the cd accreted gas and dust from a smaller gas - rich galaxy provides a plausible mechanism , specifically in terms of triggering the formation of the radio source and introducing large amounts of molecular gas and dust into the cd elliptical . the observed asymmetry of the gas and dust would indicate that the system is still in the process of virialization , i.e. , the event took place within the last @xmath196 yr . this dynamical timescale is also compatible with the timescales derived from the stellar models involving continuous star formation with a constant sfr , i.e. , a scenario in which the interaction is responsible for inducing on - going star formation . however , if the star formation is described in terms of a single burst then the models imply ages of the order @xmath192 yr , in which case the star formation would more likely have been triggered by the passage of the radio source . the age of the radio source is substantially less than the galaxy interaction timescale , and this either represents the additional time required to funnel gas into the center of the cd , or otherwise indicates that the present phase of radio activity is the latest in a series of multiple outbursts . the properties of a2597 are directly tied to the general question of the relationship between galaxy interactions , nuclear activity and star formation . if active galaxies in cooling flows such as a2597 are triggered in the same way as agn unassociated with cooling flows , then the primary reason for the difference in their properties , specifically the compact radio morphology and enhanced star formation , results from being located at the center of a dense , high - pressure icm . better understanding the effect of environment upon star formation would be especially desirable in the context of studying objects in a range of environments at low redshifts , as well as investigating the formation of stars and galaxies at early epochs where the environment is likely to play an important , perhaps dominant , role . we are grateful to stephane charlot and gustavo bruzual for kindly making available their latest stellar synthesis libraries with appropriate k - corrections and magnitudes calculated in the _ hst _ filters that were used to observe this object . a. m. k. would like to thank stefano casertano , brad whitmore , and paul goudfrooij for a number of valuable discussions , and andy fruchter for helpful explanations concerning use of the dither package . we thank michael wise for his involvement in the initial phase of the project . we also thank the anonymous referee for useful comments which helped to improve the paper . partial support for this work was provided by nasa through grant number from the space telescope science institute , which is operated by aura , inc . , under nasa contract . c. l. s. was supported in part by nasa asca grant and nasa astrophysical theory program grant .

we investigate the behavior of the zero - temperature quantum non - linear sigma model in dimensions in the presence of a damping term of the form , with . we find two fixed points : a spin - wave fixed point fp1 showing a dynamic scaling exponent and a dissipative fixed point fp2 with in the framework of the-expansion it is seen that there is a range of values where the point fp1 is stable with respect to fp2 , so that the system realizes a quantum critical behavior even in the presence of a dissipative term . however , reasonable arguments suggest that in this range is very narrow . in the broken symmetry phase we discuss a phenomenological scaling approach , treating damping as a perturbation of the ordered ground state . the relation of these results with the pseudogap effect observed in underdoped layered cuprates is discussed . , and renormalization group , non linear sigma model , quantum critical point , cuprates . 75.10.jm , 75.50.ee , 64.60.ak , 74.72.-h

cond-mat9811392

the presence of an antiferromagnetic ( af ) phase rapidly replaced by an anomalous metallic phase upon doping is one of the prominent features of the phase diagram of high temperature superconducting cuprates @xcite . indeed it was early suggested @xcite that the proximity to an insulating magnetic phase together with the nearly twodimensional structure of these materials could be responsible for the superconducting and anomalous normal - state ( i.e. non - fermi liquid ) properties of these systems via the creation of a quantum - disordered spin liquid ( resonating - valence - bond state ) . since then a great deal of attention has been devoted to the fascinating interplay between magnetism and charge degrees of freedom , leading to various theoretical proposals and to different scenarios for the metal - insulator transition at low doping . one scenario is based on the tendency of a magnetically ordered phase to segregate the additional holes due to doping @xcite . the tendency to phase separation is then frustrated by the coulombic repusion between the segregated carriers @xcite , thus leading to formation of hole - rich domain walls separating af domains in the form of stripe textures . within this scenario , pseudogaps naturally arise in the underdoped phase of the materials as a consequence of stripe fluctuations and local pair formation @xcite . obviously this will substantially affect the transition to the af insulating phase at low doping @xcite . the transition between a magnetically ordered phase and a metallic state can also be strongly affected by extrinsic ingredients like the disorder induced by dopant ions ( sr or zn in the case of @xmath8 ) . this may lead to the formation of local random magnetic moments giving rise to spin - glass ordering between the af insulating and the paramagnetic metallic ( pm ) phases @xcite . the stripe and the spin - glass scenarios clearly illustrate the complicated and rich nature that the af - pm transition may take . however , other proposals are more directly related to magnetism and to the contiguity between the af and the metallic phase . these proposals start from the seminal work of ref . @xcite showing that the materials with half - filled cuo@xmath9 planes are suitably described in terms of a twodimensional quantum heisenberg af model with very small interplanar coupling . following work @xcite suggested that , although the 2d model displays long - range order at @xmath10 , upon doping the added charges enhance the quantum spin fluctuations thus driving the system into a disordered state even at zero temperature ( quantum disordered phase ) @xcite . the occurrence of an af quantum critical point ( qcp ) provides a natural framework to intepret the scaling behavior at low temperatures and low frequencies of the of the @xmath11-integrated magnetic susceptibility @xmath12 . the susceptibility @xmath13 can be experimentally determined via magnetic resonance measurements of the nuclear copper spin - echo decay rate , @xmath14 and spin - lattice relaxation rate , @xmath15 @xcite . it is seen that in a range of temperatures @xmath16 , depending on doping , @xmath17 @xcite . assuming the scaling relations @xcite , @xmath18 and @xmath19 , with @xmath20 a dynamic critical exponent , @xmath21 an anomalous dimension , @xmath22 the antiferromagnetic ordering wave - vector , @xmath23 the correlation length , @xmath24 a scaling function , one gets @xmath25 @xcite . the constancy of this ratio is thus interpreted as the signature of a critical behavior corresponding to a @xmath3 value of the dynamic index @xcite . the strong critical spin fluctuations occurring at the af - qcp have also been claimed to provide a possible source of pairing and normal - state anomalies . despite the apparent simplicity of a scenario involving a direct transition between an insulating af phase and a pm phase , such a transition is far from being trivial and is still an open problem . the main difficulty is in the lack of a microscopic model being able to smoothly interpolate between an insulating af ordered phase with gapped charge excitation and a pm phase . on the one hand , the insulating af phase is suitably represented by the heisenberg model and , in particular , the long - wavelength and low frequency behavior of spin fluctuations in the cuo@xmath9 planes can be well described by the quantum non - linear sigma model @xcite , which at zero temperature and in two spatial dimensions is characterized by the action @xmath26 here @xmath27 is a three - component vector field subject to the real - space constraint @xmath28 which describes the local staggered magnetization , and @xmath29 is a coupling constant . it was first argued in @xcite that this description could be carried over to the case of the lightly doped compounds . in fact , one can think that in this case the @xmath27 field continues giving an effective description of the electron spins localized on the cu sites of the lattice , while the presence of itinerating holes provides a finite renormalization of the coupling constant . the theory described by the action ( [ pura ] ) , besides having the correct symmetries , realizes a zero - temperature transition for a given critical value @xmath30 of the coupling ( see @xcite for the quantum version and @xcite for the classical theory ) thus providing a good model for the physical picture of the af - qcp @xcite . from this point of view the @xmath3 value of the dynamic scaling exponent is a simple consequence of the lorentz invariance of ( [ pura ] ) . however , it is clear that the non - linear @xmath31 model does not properly describe the low - energy charge excitations , which are characteristic of a metallic state possibly providing a damping mechanism for spin - waves ( see , e.g. , case b of ref . @xcite ) . therefore the model in eq . [ pura ] provides a good description of a transition between an insulating ( charge - gapped ) af ( afi ) and an insulating paramagnet ( pi ) . , but it appears to be inappropriate when spin - waves can decay in low - energy particle - hole excitations . on the other hand , attempts have also been made to provide a microscopic basis for the transition to a pm phase . in this regard two distint cases a and b depending on the shape of the fermi surface in the metal were investigated in ref . @xcite within a spin - density - wave approach . in case a the af wavevector @xmath32 ( @xmath33 is the lattice spacing ) was not able to connect points of the fermi surface , so that momentum and energy conservation did not allow the decay of spin - waves in particle - hole pairs . in this case the transition was naturally characterized by a dynamic critical index @xmath34 . in case b , instead , there were ( `` hot '' ) points on the fermi surface connected by @xmath35 so that low - energy particle - hole pairs could be created by the decay of low - energy spin - waves . however , in both cases , a and b , the phase with af long - range order is metallic and therefore the model does not suitably describe a transition between an afi and a pm model would suitably describe the system also in this case despite the metallic character of both the ordered and disordered phases . ] . the case a above shares some common features with a mixed spin - fermion model without a three - body ( i.e. kondo - like ) direct coupling between the local moments and the fermion spins @xcite . this model has been shown @xcite to display a @xmath10 transition between a metallic ( fermi liquid ) phase with spin commensurate long - range order and a metallic ( also fermi liquid ) quantum disordered phase . owing to the absence of the three body coupling responsible for the direct spin - wave decay into particle - hole pairs , also the qcp of this model is characterized by @xmath34 . to the best of our knowledge no simple microscopic modelization is presently available to describe a direct afi - pm transition . of course the possibilities remain open that other extrinsic mechanisms ( stripes , disorder , ... ) provide intermediate steps between the afi and the pm phases . alternatively , one may also envisage that an afi to a pi phase transition occurs first at lower doping , followed at slightly larger doping by a pi - pm transition between two spin - disordered phases . together with the basic difficulty outlined above , the additional question arises concerning the observed @xmath34 critical behavior in the underdoped phase of the cuprates and the way this can be reconciled with the metallic character of this phase . in particular it is believed that the fermi surface of the cuprates in the metallic phase contains `` hot '' points connected by the af wavevector . therefore it should be possible for the spin - waves of arbitrarily low energy to decay into particle - hole pairs and get overdamped . this would lead to a relaxational behavior contrasting with the @xmath34 ( i.e. propagating ) critical behavior of the spin - waves . moreover , by accepting a @xmath36 critical behavior in the metallic phase , it remains to clarify how this may turn into the natural propagating behavior , which characterizes the spin excitations in the ordered side of the qcp . this latter difficulty was also encountered and stressed by the authors of ref . @xcite for their case b : within their mean - field description a @xmath36 behavior was obtained both in the disordered and the ordered af ( metallic ) phases . this suggested that the simple formulation b of the mean - field sdw model was only applicable on the disordered side of ( and not too close to ) the af - qcp . in the present paper we will not attempt to find a microscopic foundation to a direct afi - pm transition , but we will instead start from a semiphenomenological point of view by introducing a field - theoretical generalization of the non - linear @xmath31 model action ( [ pura ] ) to investigate a ) the condition to be fullfilled by a simple ( @xmath37-independent ) damping term in order to maintain a @xmath34 critical behavior ; b ) how in the @xmath38 case an additional crossover energy must behave in the ordered phase to provide a scale separating damped spin excitations at high energy and propagating spin excitations at low energy . the resulting theory will depend on the precise form of the effective damping term . we first discuss the simpler case of the metallic phase on the @xmath39 side of the qcp . within a microscopic model of spins coupled to free itinerant holes , integrating out the holes degrees of freedom in the random phase approximation gives a damping term of the form @xmath40 ( see _ e.g. _ @xcite ) . however , the choice of the adequate low frequency form of @xmath41 should also take in consideration the fact that in underdoped cuprates a significant loss of low - frequency spectral weight is observed at low temperatures in the distribution of quasi - particles , a phenomenon often referred to as `` pseudogap '' ( see _ e.g. _ @xcite ) . the origin of this effect is a debated issue , and has been alternatively interpreted : as the result of spin - fermion scattering @xcite , as due to the intrinsecally non - fermi liquid nature of the system ( luttinger liquid ) @xcite , as a signal of the formation of preformed pairs for @xmath42 @xcite , as due to stripe charge fluctuations @xcite . in all cases , a precise computation of these effects implies solving a difficult self - consistency problem going well beyond the rpa approach . we shall try here a rough , phenomenological approach , assuming that the damping coefficient @xmath43 can acquire at low frequencies an @xmath44-dependence @xmath45 , thus giving rise to an effective damping term of the form @xmath1 , with @xmath46 case corresponds to the case considered in ref . @xcite within a large - n framework . ] . _ _ _ _ once chosen a damping term @xmath47 we perform a renormalization group ( rg ) analysis of the resulting theory in the proximity of the qcp using the momentum - shell method and @xmath5-expansion . we observe the existence of two fixed points , which we shall denote fp1 and fp2 , showing respectively a @xmath3 , undamped spin - wave behavior and a @xmath48 , dissipative behavior . the point fp1 corresponds to the zero - temperature critical point of a pure quantum non - linear sigma model without any damping term , but it becomes unstable in a wide region of the parameters with respect to the dissipative fixed point fp2 . there exists a range of values @xmath49 for which one obtains a @xmath50 dynamic scaling exponent in the proximity of the stable quantum critical point , even in the presence of a damping term . in other words , there are `` soft '' damping terms which do not destroy the @xmath3 , spin - wave behavior . here @xmath51 is the dimension dependent lower bound for the exponents in the soft damping term , which we determine to be @xmath52 , where @xmath53 is the anomalous field dimension and @xmath54 is its value at the fp1 . it is worth noting that this result bears a resemblance to what is found in classical models with long - range interactions @xcite . in particular the scaling exponent of the damping stems from the non - analytic form of @xmath47 , which is not reconstructed by the rg corrections ( for a discussion of this point in classical models see ref . . therefore @xmath55 does not acquire singular corrections but for the anomalous field dimensions . within the framework of an expansion in @xmath56 , where @xmath57 is the spatial dimension of the system and @xmath58 is the bare value of the dynamic scaling exponent in the vicinity of the stable fixed point , one would get @xmath59 , where the 1 or 2 subscripts in @xmath60 refer to fp1 or fp2 respectively . however , the extrapolation of results valid for @xmath61 to the physical region @xmath7 should be done with great care . in particular , using the numerical results for the critical exponents of the @xmath62 model in 2 + 1 dimensions , it appears that @xmath51 is very near to two and the range of `` soft '' @xmath63 values should be a very narrow one . this means that within the here - considered model the appearance of a @xmath3 critical behavior in the cuprates can not be accounted for in the effective framework without assuming a value @xmath64 , _ i.e. _ , an almost linear decay of the quasi - particle spectral weight for low frequencies . this spectral weight suppression is a rather severe condition to be obtained from the magnetic scattering itself , as also suggested by a direct perturbative calculation within a spin - fermion model @xcite , where density of states and vertex corrections to the fermion bubble produced only a minor change of the power - law dependence of the damping term . it is worth noting that a substantial spectral weight reduction is however consistent with anderson s idea that a smooth connection between a charge - gapped insulator and a fermi - liquid metal is hardly conceivable . _ _ as far as the ordered phase is concerned , ( point ( b ) above ) it is necessary to take into account the presence of a non - zero staggered magnetization @xmath65 , which is accompanied by the presence of two quasi - particle bands separated by an energy @xmath66 . this suggests @xcite to introduce a damping term having two distinct asymptotics : @xmath67 for @xmath68 and @xmath69 for @xmath70 , with @xmath71 in the ordered ( as well as in the disordered ) phase , but in a first approximation we ignore this dependence mainly for simplicity reasons , but also because in the insulating phase , where a gap for the charge excitations providing spin - wave decay is present , it would be unnatural to expect a relevant q - dependence . ] the difficult point here is that the scale @xmath72 separating damped and undamped spin excitations vanishes at the transition . this prevents using a conventional rg approach , and we will resort to simple scaling arguments and phenomenological assumptions . our analysis does not attain the same degree of reliability of a traditional landau - ginzburg - wilson theory and has a heuristic character . we shall assume that @xmath73 has a power - law behavior , to be determined by matching the hydrodynamic behaviors on the two sides of the quantum transition . for @xmath74 , _ i.e._for @xmath38 , we find that ( i ) @xmath73 should close at the microscopic " scale as @xmath75 for @xmath76 , where @xmath77 is a correlation length , and that ( ii ) the spin - wave velocity @xmath78 vanishes as @xmath79 , to be consistent with the presence of damping . _ _ the paper is organized as follows : in sec . [ rengreq ] we derive the rg equations for the zero - temperature quantum non - linear sigma model with a damping term of generic form , following the scheme exposed in @xcite . this part is rather technical and it can be skipped in a first reading without loosing the general meaning of the paper . in sec . [ fixpanrgeq ] we perform the fixed point analysis of the rg equations for a damping term @xmath80 . here we show the existence of the two fixed points mentioned above , study their respective regions of stability and compute the relevant critical exponents , with particular regard to the value of the dynamic scaling exponent @xmath81 . in sec . [ ordphase ] we discuss damping in the ordered phase , introducing a phenomenological scale @xmath73 vanishing at the quantum critical point with a critical exponent that we determine by matching the hydrodynamic behaviors on the two sides of the transition . we also compute the critical behavior of the spin - wave velocity @xmath78 in the presence of damping . in sec . [ conclusions ] we present some conclusive remarks and discuss the connection between our model and other physically interesting systems , like macroscopic quantum tunnelling and spin chains with long range interaction . let us consider the euclidean action of the quantum non - linear sigma model @xcite at zero temperature in @xmath82 spatial dimensions , with a damping term @xmath41 of generic form and in the presence of a constant magnetic field @xmath83 : @xmath84 here @xmath27 is an @xmath85-component field satisfying the constraint @xmath86 in real space , @xmath87 , @xmath88 . the magnetic field is oriented along the @xmath85-th direction . the momentum and frequency integrations have been rescaled so that the adimensional spin velocity @xmath89 has bare value equal to one and the momentum integration is cut off at @xmath90 . with our rescaling , in ( [ azione1 ] ) the dimensionless coupling constant @xmath91 is related to the bare ( dimensional ) spin - wave velocity @xmath92 , the physical cut - off @xmath93 and the bare spin - stiffness constant @xmath94 by the relation @xmath95 @xcite . in the large @xmath96 ( here @xmath97 is the spin ) limit @xmath94 and @xmath92 are related to the coupling constant of the heisenberg model @xmath98 and the lattice spacing @xmath99 by @xmath100 and @xmath101 @xcite with @xmath102 . @xmath103 is also adimensional . action ( [ azione1 ] ) is formally equivalent to the hamiltonian of an anisotropic corresponding statistical mechanical model in @xmath104 dimensions . the critical point @xmath105 of the classical model is here reinterpreted as a zero - temperature quantum critical point , separating the ordered and the disordered ground state of the system , which are realized , respectively , for @xmath106 and @xmath107 . in order to study the long - distance , low - frequency behavior of the @xmath27 field in the proximity of the quantum critical point we shall use the renormalization group and the @xmath5-expansion , following refs . @xcite . in the broken symmetry phase we let @xmath108 , @xmath109 . keeping only terms up to @xmath110 , action ( [ azione1 ] ) takes the form @xmath111 where the inverse propagator , given by @xmath112 depends on the damping term @xmath47 , and the measure term @xmath113 is @xmath114 in ( [ azione2 ] ) the term proportional to the extended phase - space volume @xmath115 comes from the perturbative evaluation of the jacobian @xmath116 in ( [ rsd ] ) @xmath117 is an arbitrary frequency cut - off which does not enter the final result , and @xmath118 . let us denote the vertices appearing ( in the same order ) in ( [ azione2 ] ) as where the cut , double cut and triangle indicate respectively a @xmath119 , @xmath44 and @xmath41 insertion , and the dotted line stays for the first perturbative term of the jacobian . let then @xmath120 , with @xmath121 we shall use the momentum - shell method to derive recursion relations _ la _ wilson @xcite , integrating out the @xmath122 with @xmath123 , @xmath124 and all @xmath44 , and rescaling @xmath125 hereby introducing a dynamic scaling exponent @xmath81 and a wave function renormalization @xmath126 . we average with respect to the following propagator : @xmath127 the first perturbative term of the jacobian can be written as @xmath128 \int _ { k<1}{\,\mathrm{d}}\kappa \ , { \boldsymbol{\pi } } _ { { \boldsymbol{\kappa } } } \cdot { \boldsymbol{\pi } } _ { -{\boldsymbol{\kappa } } } \ ] ] and we will indicate it pictorially as _ _ after integration on the momentum shell the first term represents the modified first - loop jacobian contribution , while the second has to be absorbed in the mass renormalization and cancels the contribution from the second vertex in eq . ( [ azione2 ] ) ( see below ) , thus maintaining the theory massless at @xmath129 . the usual diagrams renormalizing the @xmath130 and @xmath131 terms both give a @xmath132 contribution , with @xmath133 in order to renormalize @xmath134 and @xmath83 we have to consider the following contributions : @xmath135 { \boldsymbol{\pi } } _ { { \boldsymbol{\kappa } } _ { 1}}\cdot { \boldsymbol{\pi } } _ { -{\boldsymbol{\kappa } } _ { 1}}\\ \nonumber & \simeq & \frac{1}{2 } \int _ { k_{1}<e^{-l}}{\,\mathrm{d}}\kappa _ { 1}\left [ \int _ { e^{-l}<k_{2}<1}\!\!\frac{f(\omega _ { 1}+\omega _ { 2 } ) -f(\omega _ { 2})}{\delta _ { ( f)}(k_{2},\omega _ { 2})}{\,\mathrm{d}}\kappa _ { 2}+ \frac{n-1}{2}ghli_{(f ) } \right ] { \boldsymbol{\pi } } _ { { \boldsymbol{\kappa } } _ { 1}}\cdot { \boldsymbol{\pi } } _ { -{\boldsymbol{\kappa } } _ { 1 } } \end{aligned}\ ] ] the symmetry of ( [ azione1 ] ) requires that @xmath83 scales as @xmath136 . since @xmath83 plays in ( [ azione2 ] ) the role of a mass , from one - loop mass renormalization we get that the field rescaling factor is @xmath137\ ] ] defining @xmath138 we finally get the rg equations : @xmath139 g\\ \label{rhatf } \frac{{\,\mathrm{d}}\hat f } { { \,\mathrm{d}}l}(\omega ) & = & \left ( 2-gi_{(f)}-z\omega \frac{{\,\mathrm{d}}}{{\,\mathrm{d}}\omega } \right ) \hat f ( \omega ) + gc_{(f)}(\omega ) \end{aligned}\ ] ] with @xmath140 equation ( [ rtildeg ] ) results as usual from field rescaling and renormalization of the @xmath130 term . the term @xmath141 in eq . ( [ rhatf ] ) describes the frequency rescaling and @xmath142 is related to the contributions in eq . ( [ diagrammi ] ) . let us now consider a damping term of the form @xmath143 , with @xmath144 a generic differentiable function behaving as @xmath145 for @xmath146 . one observes that for @xmath147 and @xmath148 integrable in a neighborhood of @xmath149 the term @xmath150 is twice continuously differentiable , so that for @xmath151 it is possible to expand @xmath152 with @xmath153\ ] ] in order to prove ( [ sviluppo],[kappaf ] ) it is enough to rewrite ( [ conv0 ] ) as @xmath154 and to differentiate the expression twice with respect to @xmath155 . @xmath156 is continuous but not differentiable in @xmath149 . the content of eq . ( [ sviluppo ] ) is that _ the non analyticity of the damping term in @xmath149 is not reproduced under renormalization__. this implies that the @xmath41 term renormalizes only according to its bare dimensions and to the contribution of the field rescaling factor . on the other hand , @xmath47 contributes to the renormalization of the @xmath131 term . _ _ _ according to eq . ( [ kappaf ] ) , @xmath157 is essentially a measure of the non linearity of @xmath134 and is not strongly dependent on its explicit form . for @xmath80 , with @xmath2 and @xmath43 a damping coefficient , equation ( [ rhatf ] ) , together with ( [ sviluppo],[kappaf ] ) , gives @xmath158\gamma \\ \label{grt3 } \frac{{\,\mathrm{d}}}{{\,\mathrm{d}}l}\frac{1}{c^{2 } } & = & 2(1-z)\frac{1}{c^{2}}+ 2gk_{\alpha } ( c,\gamma ) \end{aligned}\ ] ] with _ @xmath159 _ in appendix [ compcritind ] the limit for @xmath160 of these expression is discussed ( see ( [ loop2],[conv2 ] ) ) . _ _ we shall now study the rg equations obtained in the previous section for the zero - temperature critical point of the quantum non - linear sigma model with a damping term @xmath161 , described by the action ( [ azione1 ] ) . for the sake of clarity we rewrite here the rg equations ( [ rtildeg],[grt2],[grt3 ] ) : @xmath162g\\ \label{grt2 _ } \frac{{\,\mathrm{d}}\gamma } { { \,\mathrm{d}}l } & = & [ 2-\alpha z- gi_{\alpha } ( c,\gamma ) ] \gamma \\ \label{grt3 _ } \frac{{\,\mathrm{d}}}{{\,\mathrm{d}}l}\frac{1}{c^{2 } } & = & 2(1-z ) \frac{1}{c^{2}}+2gk_{\alpha } ( c,\gamma ) \end{aligned}\ ] ] where the integrals @xmath163 and @xmath164 were defined in ( [ loop],[conv ] ) . the dynamic scaling exponent @xmath81 was introduced when choosing to rescale frequencies as @xmath165 , @xmath166 being the parameter of the renormalization group . we look for fixed points of the rg transformation ( [ grt1_]-[grt3 _ ] ) . one immediately finds two fixed points , characterized by 00.00 @xmath167 , for @xmath168 ; @xmath169i_{\alpha } ( c _ { * } , \gamma _ { * } ) } } } , \ , \ , \ , c^{\alpha } _ { * } \gamma _ { * } = u_{*},\ , \ , \ , z_{2}={{\displaystyle \frac{2n-2-d}{(n-2)\alpha + 1 } } } $ ] , + for @xmath170 ; where @xmath171 is the solution of @xmath172 with @xmath173 , @xmath174 . strictly speaking , fp1 and fp2 are two _ lines _ of fixed points , since in both cases @xmath175 is arbitrary . in the case of fp2 either @xmath175 or @xmath176 can be arbitrarily chosen , and the remaining one is then determined through the constraint @xmath177 . _ _ concerning eq . ( [ vincolo ] ) , note that , letting @xmath178 : 1 . a unique solution @xmath171 exists for @xmath179 ; 2 . @xmath180 for @xmath181 ; 3 . no solution exists for @xmath182 . together with @xmath183 , this implies @xmath184 . these are all consequences of the following properties : @xmath185 is a monotonically decreasing function of @xmath186 , tending to a finite value for @xmath187 and vanishing for @xmath188 ; @xmath189 is an increasing function , vanishing for @xmath187 and unbounded for @xmath188 . the analogy with the statistical mechanical model shows that the perturbative expansion is in the parameter @xmath56 , where @xmath58 is the value of the dynamic scaling exponent obtained from purely dimensional considerations . for the @xmath190 case lorentz invariance imposes @xmath191 , while in the presence of a non - zero damping term @xmath192 with @xmath63 not too close to @xmath193 we expect the latter to be more relevant than @xmath131 , and @xmath194 . letting then @xmath195 be the one - loop perturbative correction to the bare dynamic exponent @xmath58 and @xmath21 the field anomalous dimension , one finds in the two cases : 00.00 @xmath196 , @xmath197 , @xmath198 , @xmath199 , @xmath200 @xmath201 , @xmath202 , @xmath203 , + @xmath204 , @xmath205 , + with @xmath206 the solution of @xmath207 . the point fp2 exists for @xmath208 . for @xmath209 one gets the unphysical situation @xmath210 . the additional condition @xmath211 amounts to @xmath212 , which is the condition of stability to order @xmath213 of fp2 with respect to fp1 . more precisely , at @xmath214 fp2 merges into fp1 ( see below ) . for @xmath215 one has @xmath216 ; keeping @xmath176 fixed this gives @xmath217 , as expected , since for fp2 one has @xmath218 at 0-th order in @xmath219 . for @xmath220 , ( [ loop ] ) shows that @xmath221 is finite , so that @xmath222 . the same is true in the @xmath223 case , but some care must be exerted . a direct computation using ( [ loop2],[conv2 ] ) leads to ambiguous results , since in this case the frequency wave - vector integration shell becomes singular : this is discussed more at length in appendix [ compcritind ] . we just observe here that for @xmath160 , @xmath215 implies @xmath224 and one has @xmath225 . one can send @xmath215 either with fixed @xmath226 , or with fixed @xmath227 , and in both cases the loop integral @xmath228 is finite at @xmath229 . in order to compute the critical exponent @xmath230 we linearize the rg equations ( [ grt1_]-[grt3 _ ] ) and find the eigenvalues of the linearized transformation ( see appendix [ compcritind ] for more detail ) : 00.00 @xmath231 , @xmath232 , @xmath233 for @xmath168 ; @xmath234 , @xmath232 , @xmath235 + for @xmath236 . this shows that the point fp2 always has one attractive direction in its region of existence , while @xmath237 depends on @xmath238 . ( 0,0 ) ( -26,55)(-61,30 ) ( -57,88)(0,0)[bl]@xmath239 ( -43,88)(0,0)[bl]@xmath240 ( -26,88)(0,0)[bl]@xmath241 ( -73,54)(0,0)[bl]@xmath242 ( -65,48)(0,0)[bl]@xmath243 ( -56,42)(0,0)[bl]@xmath244 ( -20,25)(0,0)[bl]@xmath245 ( -38,18)(0,0)[bl]@xmath246 ( -50,12)(0,0)[bl]@xmath247 ( -95,46)(0,0)@xmath248 ( -45,-4)(0,0)@xmath0 we can now assess the stability of the two fixed points . in the @xmath249 plane we distinguish the regions ( see fig . [ stability ] for the @xmath250 heisenberg case ) : 00.00 @xmath251 ; @xmath252 i.e. @xmath253 . in region fp1 there exists only the corresponding fixed point , which is stable . in region fp2 there appears a second , stable fixed point , while the point fp1 becomes unstable . the crossing between the regions fp1 and fp2 takes place for @xmath254 . notice that on the border line one gets @xmath255 and @xmath256 , so that when moving from the region fp2 to fp1 the critical exponents @xmath21 and @xmath81 vary _ continuously__. _ _ _ finally we report in fig . [ rgflow ] a numerical computation of the structure of the rg flow for typical values of the parameters . to this purpose it is convenient to introduce the rescaled variables @xmath257 , @xmath258 , so that ( [ grt1_]-[grt3 _ ] ) take the form @xmath259\}\tilde g \\ \label{gra2 } \frac{{\,\mathrm{d}}\tilde\gamma } { { \,\mathrm{d}}l } & = & [ 2-\alpha -\tilde g ( \tilde i _ { \alpha } ( \tilde\gamma ) + \alpha \tilde k _ { \alpha } ( \tilde\gamma ) ) ] \tilde\gamma \\ \label{gra3 } \frac{{\,\mathrm{d}}}{{\,\mathrm{d}}l}\frac{1}{c^{2 } } & = & 2 [ 1-z+ \tilde g \tilde k _ { \alpha } ( \tilde\gamma ) ] \frac{1}{c^{2 } } \end{aligned}\ ] ] where ( [ gra1 ] ) and ( [ gra2 ] ) are decoupled from ( [ gra3 ] ) and do not contain the dynamic exponent @xmath81 . this means that in the new variables the rg trajectories project on the @xmath260 plane and the two lines of fixed points project on two corresponding fixed points . this fact was not _ a priori _ obvious and greatly facilitates the analytical and qualitative study of the rg trajectories . _ _ ( 0,0)(-97,45)(0,0)@xmath261(-45,-4)(0,0)@xmath262(-70,9 ) ( -70,9)(-2,-1)5(-58,66 ) ( -58,66)(-2,-1)5 we notice the interesting fact that there exists a range of values @xmath263 for which one obtains a @xmath50 dynamic scaling exponent in the proximity of the quantum critical point , even in the presence of a damping term . in other words , there can be `` soft '' damping terms which do not spoil the spin - wave behavior of the fixed point . within the @xmath60-expansion for @xmath250 @xmath264 is given by @xmath265 ( cf . fig . [ stability ] ) . we shall argue about the extension of these results to the @xmath266 case in the next subsection . the above analysis of the stability of fp1 vs. fp2 bears a strong similarity with the study of the @xmath267 theory with long range interactions @xcite . for small @xmath268 and @xmath219 one has @xmath269 and @xmath270 and therefore @xmath271 , coherently with @xcite . the two fixed points studied in @xcite correspond to our fp1 and fp2 ; in our approach they are selected by the value of @xmath58__. _ the continuity of @xmath21 for @xmath272 appears as a consequence of their relative stability , as in @xcite . the computations here are considerably simpler , since first - loop calculations are sufficient to establish the general picture . another difference is in the presence of the dynamic exponent @xmath81 , which was absent in the classical works @xcite . _ _ _ it is worth stressing the fact , noticed after eq . ( [ sviluppo ] ) , that the non - analyticity of the term @xmath273 is not reproduced under renormalization , at least at first loop . as a consequence @xmath274 does not get singular corrections by itself , but only scales according to its bare dimension and to the anomalous field dimension ) below refer to the coupling @xmath275 of the dissipative term in eq . ( [ azione1 ] ) . in the critical region @xmath105 and the behavior of @xmath55 coincides with the behavior of @xmath275 . ] . the indication coming from @xcite is that this will be true also at higher orders . this would imply that , as an exact result , @xmath43 scales in the critical region according to the exponent @xmath276 irrespectively from the approximation . the value @xmath277 can then be obtained directly from the condition @xmath278 , giving @xmath279 which is in agreement with our one - loop analysis and is expected to remain valid at higher loop orders as well . this exact relation can be used to settle the problem of the crossover from fp1 to fp2 . this crossover takes place when @xmath256 , that is , using ( [ zeta ] ) , when @xmath280 , where @xmath281 is the anomalous field dimension for a given @xmath248 . continuity of @xmath282 through the crossover implies @xmath283 with @xmath284 being the critical exponent of the sigma model with zero damping in @xmath104 . for twodimensional systems , like the cuprates , @xmath285 . numerical estimates for the critical indices of the @xmath286 symmetric @xmath287 field theory obtained from summed perturbation series at fixed dimension @xmath288 ( see @xcite and table 25.4 of @xcite ) give @xmath289 and @xmath290 . in this respect the @xmath291-expansion greatly overestimates @xmath292 and underestimates @xmath293 in @xmath266 ( see fig . 1 ) . this suggests that , as a consequence of this small value of @xmath292 , since the relation ( [ zeta ] ) is probably an exact relation , for @xmath294 near one the physically relevant line of fixed point would be fp2 . it appears therefore that the @xmath3 behavior observed in the cuprates can be accounted for in this framework only in the hypothesis that some mechanism like the reduction of the spectral weight of low - frequency quasi - particle is so strong to effectively produce a @xmath273 damping term with @xmath63 very close to @xmath193 . in the ordered phase the effect of holes in the small - doping limit of the cuprates acts as a perturbation of the @xmath10 antiferromagnetically ordered ground state . the description of the interplay of spin and fermion degrees of freedom when approaching the qcp from the ordered side presents major difficulties . the simple introduction in the effective sigma model of a @xmath295 term is not satisfactory , since this term would be relevant also in the proximity of the ordered critical point @xmath296 , and would completely destroy the spin - wave picture of the ordered ground state . in the ordered phase the appearance of a non - zero staggered magnetization @xmath297 is accompanied by the presence of two quasi - particle bands , separated by an energy @xmath73 ( for a mean field analysis see _ e.g. _ ref . @xcite ) . as a consequence , one would expect the decay of spin - wave excitations in particle - hole pairs to be disfavored for @xmath298 , and an effective damping be felt only for @xmath299 . moreover , in an afi - pm transition @xmath73 should vanish at the qcp as well as @xmath297 does . were this not the case , since @xmath300 in the disordered phase one could hardly match the critical behavior on the two sides of the transition . we are not in a condition to directly link the scale @xmath73 to the staggered magnetization @xmath297 since the precise relation , if any , should emerge from the unknown microscopic dynamics of the quasiparticles . in this section we try to get some insight in this difficult problem using simple scaling arguments . we will conjecture for @xmath73 a scaling law with a critical exponent @xmath301 , computable in principle from the full theory including also the fermionic degrees of freedom , and we shall try to assess the value of @xmath301 in a self - consistent way , once the value of @xmath248 for the damping in the disordered phase is given . we emphasize that this procedure is not as reliable as the rg computation exposed in the preceding sections and has an heuristic character . _ _ a model damping term which could take into account the appearance of the scale @xmath73 is @xmath302 with @xmath71 , and @xmath303 . we shall later assume that @xmath73 vanishes at the qcp , so that the crossover between the two regimes is continuously shifted while approaching the transition . here we consider @xmath304 as a free parameter at the same level as @xmath305 and @xmath55 . assuming that changing the scale , @xmath47 approximately maintains the form ( [ model1 ] ) with varying values of @xmath305 and @xmath55 one finds from ( [ rhatf ] ) the characteristic exponents @xmath306 our phenomenological assumptions suggest that damping be relevant only at high frequencies . in this regime the spin - wave velocity @xmath78 goes as @xmath307 with the running coupling constant @xmath308 and the dynamic exponent @xmath20 . in the neighborhood of a critical point , @xmath20 is determined by the condition that @xmath308 does not renormalize . this corresponds to attributing all the renormalization of the @xmath309 term to the anomalous frequency scaling , @xmath310 . the running coupling constant @xmath308 differs from the physical spin - wave velocity @xmath78 , which is measured at any scale @xmath311 using a fixed system of physical units : one has @xmath312 . so , although @xmath308 does not renormalize by construction , when @xmath38 the physical velocity @xmath78 vanishes at the transition . in the case of the undamped sigma model the value @xmath34 is obtained as a consequence of lorentz invariance , so that there is no difference between @xmath308 and @xmath78 and the latter remains finite at the qcp . however , the damping term breaks lorentz invariance and can drive @xmath78 to zero at the qcp . ] . on the other hand , at any given scale the relevant @xmath155-modes are characterized by the condition @xmath313 . so , damping is relevant up to the scale @xmath314 where @xmath73 becomes of order @xmath315 . beyond this scale the gap for spin - wave decay is visible and @xmath47 is irrelevant . if @xmath73 at the microscopic " scale @xmath316 had a finite value , eventually @xmath47 would be always irrelevant . the simplest way to take into account the physical condition that @xmath73 should vanish approaching the transition is to assume that @xmath317 where @xmath318 and @xmath319 . this condition encodes in a simplified , phenomenological way the complicated interaction between damping and antiferromagnetic order , mediated in the full theory by the fermionic degrees of freedom . taking into account ( [ initcond ] ) and the scaling exponents ( [ gammarcritexp ] ) we get @xmath320 approximately at this scale damping becomes irrelevant and the critical behavior starts being characterized by @xmath34 . this gives @xmath78 vanishing as @xmath321 for @xmath294 not too close to two ( specifically for @xmath322 ) by approaching the quantum transition from the disordered phase one observes , as already discussed , a dissipative behavior dominated by the fixed point fp2 , with the scaling law @xmath323 . deep into the broken symmetry phase one instead has a propagating behavior , @xmath324 . this should match the critical behavior at the scale @xmath325 , with @xmath326 the corresponding critical index . continuity of the critical behavior in a neighborhood of the critical point imposes then @xmath327 , _ i.e. _ @xmath328 this is compatible with ( [ cswsimbars ] ) only for @xmath329 in terms of @xmath330 this condition on the critical behavior of the energy separation @xmath73 gives @xmath331 . it is not unreasonable to expect that the damped sigma model we considered contains some of the physics of the quantum critical point of the underdoped layered cuprates . for this reason we performed a rg analysis directly on this model , trying to maintain sufficient generality in the form of the effective damping term , in particular introducing an exponent @xmath332 that could account for the observed pseudogap behavior . in the broken symmetry phase we tried to get some insight on the interplay between damping and af ordering using simple scaling arguments . in both the ordered and the disordered phases , assuming continuity of the hydrodynamic behaviors at the quantum transition , we obtained that for @xmath74 ( where @xmath333 is expected to be very close to 2 ) it is not possible to account for the observed @xmath34 scaling law . it was observed in ref . @xcite that the @xmath34 scaling can be recovered at higher temperatures , when typical frequencies would become larger than the @xmath10 value of the damping term . however , experimentally it is the @xmath34 behavior which is observed at lower temperatures , while a relaxational @xmath36 behavior is observed at higher @xmath334 @xcite . it was argued @xcite that this could be due to the neglected @xmath334-dependence of the damping term @xmath55 . we propose in this paper that the same effect can be accounted for by assuming that the decay of the quasi - particle spectral weight for low frequencies produces a soft " damping term @xmath335 with @xmath336 . the @xmath34 scaling would then be a consequence of an almost linear decay of the quasi - particle spectral weight . we notice that this reduced weight is fully consistent with the observed occurrence of a pseudogap below a crossover temperature @xmath337 in the underdoped region of the phase diagram for the cuprates . the origin of this pseudogap might arise from extrinsic mechanisms like local cooper - pair formation and/or stripe fluctuations . alternatively a pseudogap could be the very consequence of the intrinsecally non fermi liquid nature of the system @xcite or of strong critical fluctuations occurring near the qcp . in this last regard singular af fluctuations have been shown within a spin - fermion model to strongly modify the quasiparticles properties @xcite in the proximity of the `` hot''points of the fermi surface separated by the af wavevector . since the particle - hole excitations around these points are responsible for the low - energy spin - wave decay , this naturally affects the damping of spin excitations , in a complicated interplay which deserves further investigation . we notice that the fluctuation - induced spectral weight reduction for the fermionic excitations is a definite signature of a non - fermi liquid metallic state . this is in agreement with general arguments put forward by anderson ( see _ e.g. _ , ref . @xcite ) on the difficulty in continuously connecting a charge - gapped insulating phase with a normal fermi - liquid phase . we remark , however , that anderson s arguments extend the presence of a non - fermi - liquid phase all over the metallic state as a consequence of low dimensionality . on the other hand , within the qcp scenario , the violation of the fermi - liquid properties is a consequence of critical fluctuations , which are only present around the critical point . as soon as one moves away from this point , the fluctuations loose their critical character thus providing simple perturbative corrections to the fermi liquid state . it is worth noting that the analysis carried out in this paper is not only related to the superconducting cuprates , but it has connections with other physically interesting problems , like the spin chains with long range interaction and macroscopic quantum tunnelling . by considering the model ( [ azione1 ] ) in @xmath338 , it is seen that the frequency axis is left as the only relevant `` direction '' ( cf . the technical considerations presented in appendix [ compcritind ] ) . as a first consequence , all the critical behavior should be reformulated in terms of frequency by suitably rescaling of the critical indices by @xmath339 . the frequency axis becomes then equivalent to a single space direction , via the dynamic critical index @xmath20 , thus establishing a connection with onedimensional classical models . fourier transforming back to ( imaginary ) times the damping term in ( [ azione1 ] ) gives rise to a long range interaction of the type @xmath340 . our results can be directly extended to the case @xmath341 . this leads to the comparison with the onedimensional ( classical ) @xmath342-component spin model with long - range interaction considered in ref . one can recognize that , once the spatial term is dropped in ( [ azione1 ] ) the connection is established through the coupling combinations @xmath275 and @xmath343 . this allows to rewrite the rg equation for @xmath344 in the same form as eq . ( 7 ) of ref . @xcite , with @xmath344 ( to be identified with @xmath334 ) having a fixed point ( fp2 in our model ) and @xmath345 . whitin the @xmath338 case , @xmath346 deserves special attention because of the connection with macroscopic quantum tunnelling @xcite . this connection is realized by observing that the @xmath347 non - linear @xmath31 model can be rewritten in terms of an angular degree of freedom and that the addition of the magnetic field produces a cos - like interaction ( sine - gordon model ) . in @xmath338 the model then describes a single degree of freedom with a kinetic ( @xmath348 ) term in a multivalley potential . in the presence of dissipation , a linear - in - frequency term can appear , leading again to long - range interactions on the ( imaginary ) time axis . models of this kind were considered by chakravarty @xcite in the case of a double well potential and successively by schmid @xcite to describe a dissipative quantum particle in a onedimensional periodic potential . in particular , the only difference between our eq . ( [ azione1 ] ) at @xmath346 and this latter model is the non - periodicity of the dissipative term of ref . @xcite . in the model of schmid , for vanishing @xmath103 ( @xmath296 in his notation ) , a line of fixed points for @xmath282 ( corresponding to our @xmath275 ) was found , with a critical point @xmath349 separating a regime with relevant @xmath103 from one with irrelevant @xmath103 . the long - range term introduced in @xcite is quadratic in the fields , so that the model for @xmath129 is free and @xmath282 ( i.e. @xmath275 ) does not renormalize non - linear @xmath31 model in @xmath7 for @xmath129 , where a singular point appears at a special value of the coupling ( see , for instance , @xcite ) ] . a ( dual ) fixed point was also found for very large @xmath103 , where the model of ref . @xcite was stated to be equivalent to the one of ref . we observe that , differently from the model of ref . @xcite our long - range term mantains the periodicity in the angular variable and provides an interaction . this renormalizes the coupling , thus allowing @xmath275 to flow under renormalization even at @xmath129 . one could conjecture that a new fixed point arises at finite @xmath103 . an adequate analysis would require a rg study at finite @xmath103 going beyond the @xmath60-expansion and taking into account instantonic corrections . this is an interesting open problem which deserves further investigation . * acknowledgments : * we gratefully acknowledge interesting discussions with c. di castro , a. chubukov and i. kolokolov . 10 for a review on magnetism in the cuprates see : a ) d. c. johnston , _ handbook of magnetic materials _ , - 237 , edited by k. h. j. buschow ( north holland , amsterdam 1997 ) , and b ) c. berthier , m. h. julien , m. horvati , and y. berthier , j. de phys . i * 6 * ( 1996 ) 2205 . for a recent review on magnetism in the cuprates with a particular attention to nmr - nqr experiments see : a. rigamonti , f. borsa and p. carretta , _ basic aspects and main results of nmr - nqr spectroscopies in high temperature superconductors _ , preprint 1998 . p. w. anderson , science * 235 * ( 1987 ) 1196 . v. j. emery , s. a. kivelson , and h. q. lin , phys . * 64 * ( 1990 ) 475 ; m. marder , n. papanicolau , and g. c. psaltakis , phys . b * 41 * ( 1990 ) 6920 . v. j. emery and s. a. kivelson , physica c * 209 * ( 1993 ) 597 . c. castellani , c. di castro , and m. grilli , z. phys . b * 103 * ( 1997 ) 137 . i. affleck and b. halperin , j. phys . a * 29 * ( 1996 ) 2627 . a. h. castro neto and d. hone , phys . * 76 * ( 1996 ) 2165 . j. h. cho , f. borsa , d. c. johnston , and d. r. torgeson , phys . b * 46 * ( 1992 ) 3179 . f. c. chou _ * 75 * ( 1995 ) 2204 . r. j. gooding , n. m. salem , and a. mailhot , phys . b * 49 * ( 1994 ) 6067 , and references therein . i. ya . korenblit , v. cherepanov , a. aharony , o. entin - wohlman , cond - mat/9709056 . s. chakravarty , b. i. halperin , and d. r. nelson , phys . b * 39 * ( 1989 ) 2344 . s. sachdev and j. ye , phys . * 69 * ( 1992 ) 2411 . a. v. chubukov and s. sachdev , phys . lett * 71 * ( 1993 ) 169 ; phys . lett * 71 * ( 1993 ) 2680(e ) ; a. v. chubukov , s. sachdev , and j. ye , phys . b * 49 * ( 1994 ) 11919 . a. sokol and d. pines , phys . * 71 * ( 1993 ) 2813 . a. sokol , in _ the los alamos symposium 1993 strongly correlated electronic materials _ , edited by k. s. bedell _ et al . _ ( addison - wesley , reading , ma , 1994 ) , pp . 490493 . t. imai , c. p. slichter , k. yoshimura , and k. kosuge , phys . * 70 * ( 1993 ) 1002 . m. takigawa , phys . b * 49 * ( 1994 ) 4158 . v. barzykin , d. pines , a. sokol , and d. thelen , phys . b * 49 * ( 1994 ) 1544 . v. barzykin and d. pines , phys . b * 52 * ( 1995 ) 13585 . a. v. chubukov , d. pines , and b. p. stojkovi , j. of phys . * 8 * ( 1996 ) 10017 . f. d. m. haldane , phys . lett . * 93a * ( 1983 ) 464 ; phys . * 50 * ( 1983 ) 1153 ; i. affleck , nucl . phys . * b257 * ( 1985 ) 397 . a. m. polyakov , phys . lett . * 59b * ( 1975 ) 79 . e. brzin and j. zinn - justin , phys . b * 14 * ( 1977 ) 3110 . j. zinn - justin , _ quantum field theory and critical phenomena _ ( oxford university press , new york , 1993 ) . s. sachdev , a. v. chubukov , and a. sokol , phys . b * 51 * ( 1995 ) 14874 . b. i. shraiman and e. d. siggia , phys . rev . lett . * 61 * , ( 1988 ) 467 ; phys . b * 42 * ( 1990 ) 2485 . s. sachdev , phys . b * 49 * ( 1994 ) 6770 . j. a. hertz , phys . b * 14 * ( 1976 ) 1165 . m. randeria and j .- c . campuzano , in _ proceedings of the international school of physics enrico fermi " _ ( varenna , 1997 ) . a. v. chubukov , d. k. morr , and k. a. shakhnovich , phyl . mag . b * 74 * ( 1996 ) 563 . p. w. anderson , phys . * 64 * , 1839 ( 1990 ) ; _ ibid . _ * 65 * , 2306 ( 1990 ) j. maly and k. levin , phys . b * 55 * , r15657 ( 1996 ) . m. e. fisher , s. ma , and b. g. nickel , phys . rev . lett . * 29 * ( 1972 ) 917 . j. sak , phys . b * 8 * ( 1973 ) 281 . a. aharony , in _ phase transitions and critical phenomena _ , edited by c. domb and m. s. green ( academic press , 1976 ) , vol . 6 , p. 358 . a. v. chubukov , phys . b * 52 * ( 1995 ) r3840 . d. r. nelson and r. a. pelcovits , phys . rev . b * 16 * ( 1977 ) 2191 . j. c. le guillou and j. zinn - justin , phys . * 39 * ( 1977 ) 95 ; phys . b * 21 * ( 1980 ) 3976 . p. monthoux and d. pines , phys . b * 49 * ( 1994 ) 4261 . r. hlubina and t. m. rice , phys . b * 51 * ( 1995 ) 9253 . g. baskaran and p. w. anderson , cond - mat/9706076 . j. m. kosterlitz , phys . * 37 * ( 1976 ) 1577 . a. o. caldeira and a. j. legget , phys . 46 * , ( 1981 ) 211 . the case @xmath350 and @xmath346 describes phase fluctuations in low dimensional superconductors . rg analyses on models , which are slightly different from the one considered in the present paper have been carried out by s. chakravarty , _ et al . b * 37 * , ( 1988 ) 3283 , v. j. emery and s. a. kivelson , phys lett . * 74 * , ( 1995 ) 3253 , and , more recently , by m. v. feigelman and a. i. larkin , cond - mat/9803006 . s. chakravarty , phys . * 49 * ( 1982 ) 681 . a. schmid , phys . . lett . * 51 * ( 1983 ) 1506 . for @xmath160 the expressions ( [ loop],[conv ] ) reduce to @xmath362 the same expressions can be obtained by a direct evaluation of ( [ loop0 ] , [ kappaf ] ) for @xmath223 . in particular the original expression ( [ loop0 ] ) shows that @xmath363 is in fact an analytic function of @xmath364 and @xmath43 . however , some problems arise when using expressions ( [ loop2 ] ) and ( [ conv2 ] ) to compute critical indices directly for @xmath365 instead of considering the @xmath366 limit of the general expressions given in section 3.1 . for @xmath365 in fact one expands around the critical dimension @xmath367 and the factor @xmath368 coming from angular integration vanishes . if the same computation is performed using field - theoretical methods , instead of wilson s integration on the momentum shell , no such problems arise : _ e.g. _ one has @xmath369 which correctly behaves as @xmath370 for @xmath371 , independently of the value of @xmath63 . notice that the @xmath372 factor coming from @xmath373 is cancelled by the factor @xmath374 and the @xmath375 divergence arises from @xmath376 . the problems with the wilson method are resolved if one makes a more symmetric choice of the integration shell , _ e.g. _ integrating over all the @xmath377 with @xmath378 ( which implies introducing a frequency cutoff ) . with this choice @xmath163 is finite for @xmath224 . we observed that @xmath379 computed with either choice of cutoffs have the same @xmath380 dependence , the only effect of the choice of the symmetric shell being that the integrals do not vanish for @xmath224 . in particular , the expressions for @xmath381 , @xmath382 given in the text are valid for @xmath383 : as a matter of fact , they are computed from the values ( [ iaka ] ) , which do not contain the angular integration factor @xmath373 .